CSH Associates

Risk reduction and ladies’ undies

You’ve probably seen the optimistic ads, typically of the latest blockbuster drug. “Cut your risk of [insert one of the following: dying/having a heart attack/developing diabetes] in half by taking the [insert one: new/improved/new and improved] Drug A.“ Sounds pretty good, doesn’t it? But be careful about first impressions: how data is presented may radically affect what you think – and even what you do.

A study by Jan Hux and David Naylor found that 88% of people said they would take a cholesterol-lowering drug that would result in a 34% reduction in heart attacks. But only half as many people – 42% — would take a drug that would result in 1.4% fewer people having heart attacks. In actual fact, in both cases the numbers were based on the same data: the difference lay in how the numbers were presented. The 34% simply referred to the relative reduction in risk ([Risk A-RiskB]/RiskA), whereas the 1.4% reflected the absolute risk reduction (Risk A-Risk B).

To illustrate the difference, think of bras.  Relative risk is sort of a push-up bra for health statistics: it can make a boring little stat look bigger and more impressive. Absolute risk, on the other hand, is bit like those soft, “natural” bras they sold in the 1970s – what you see is limited to what you have.

You’ll frequently hear about the difficulties in interpreting risk statistics in medicine – after all, this area involves making a lot of important decisions. For more information, you might want to check out the very accessible and understandable Know Your Chances, Understanding Health Statistics by Drs. Woloshin, Schwartz and Welch/

We depend on experts to cut through the statistical fog and clarify what risk really mean.  But even experts vary in their “statistical literacy;” like us, they can be entranced by those big, shiny numbers of relative risk reduction. And if you read Nassim Nicholas Taleb’s The Black Swan you’ll come to the conclusion that economists are so blinded by statistics that they can’t see what should be obvious (like sub-prime mortgages are risky business). Of course, this may reflect the fact that we are, after all, only human. Psychologists have shown that people are prone to overestimate the likelihood of getting good outcomes, and this effect is stronger when the need for the outcome is higher (referred to “the desirability effect”).  Avoiding a heart attack or cancer is a highly desirable outcome – particularly if you have a history of the illness in the family and you’re scared of the consequences.  Physicians are also vulnerable to the desirability effect, even if it’s the health of their patients, rather than their own, that’s on the line. So we have to be careful when reading through stats to avoid being blinded not only by science but by our own wishful thinking.

Take-away message/bottom line: If you want to exaggerate the size of a difference between two numbers, use relative risk.  If you want to minimize the size of the difference, use absolute risk.  If you want to really understand what is happening (e.g., if you’re the buyer of a service, as opposed to the seller), then ask to see both relative and absolute risks.

 

Is there such a thing as too many stats?

Some reports (particularly those produced by polling companies) contain the combination (e.g., cross-tabulations or cross-tabs) of literally hundreds of variables, with chi squares routinely for all columns.  This is a form of “throwing mud against the wall and seeing what sticks.”  It’s an oldie but goodie in the field of data torturing.

There’s a serious problem with testing everything that doesn’t run away (and even a few that do).  Let’s show this by using one of the most common statistical cut-offs: p<.05 (although in actual fact this is arbitrary, but that’s another discussion).  If you run one test, the odds that a finding is significant when it actually isn’t are 5% (this is referred to as a Type I error and just to confuse people, is often represented as alpha or ?).  But let’s say you run 150 comparisons and 100 are “significant” at .05.  The odds are that at least 5 of what you think are significant findings are not (i.e., they are Type I errors).  The probability of at least one incorrect conclusion is 99.4%.  Yikes.

There are ways of trying to get around this issue.  A low-tech approach is what is called a Bonferroni correction and involves upping the ante of statistical significance by dividing the cut-off of .05 by the number of comparisons you’re running.  So if you run 100 comparisons, to be significant you’d put the bar not at .05 but at .0005 (.05/100).  Of course, when you do that to reduce the risk of a Type I error, you increase the risk of a Type II error.  A Type II error is the probability of deciding a finding is not significant (e.g., there is no difference between the two groups) when in fact there really is a difference. Sophisticated methods have been developed to deal with the hazards of multiple comparisons, such as bootstrapping (creating a bunch of little samples out of your database), plotting results on a Poisson distribution or standardizing test statistics to z-scores.  Do not try this at home – speak to a statistician.  But probably the simplest answer is to avoid multiple comparisons by focusing and testing what the current evidence tells you should be important.  As the old saying goes, “An ounce of prevention is worth a pound of cure.”

Take-away message/bottom line:  The more comparisons are tested for statistical significance, the higher the risk of something appearing significant when in reality it’s just a chance finding.  Be wary where you’re faced with a sea of comparisons.

 

What is meant by ‘significant’?

If you read reports of research studies, you’ll typically see words to the effect that “the treatment was statistically significant” or “a significant number of respondents reported they did not believe politicians.”  What are the authors trying to tell you? To begin with, you need to tease out how the word significant is being used.  Is it being used in its lay or non-technical manner, i.e., as “meaning something,” “weighty,” or “noteworthy?”  Or is it being used in a technical sense?  And if it’s the latter, which type of technical sense?

Let’s look first at statistical significance.  Let’s say you find that, compared to Drug B, Drug A is associated with a 10% greater drop in blood pressure, with the difference being reported as having the statistical significance of p=.001.  It’s common to think that p=.001 means the chance is 1 in 1,000 that the difference we measured between Drug A and B could be due to chance.  But in actual fact, what it actually means is that there’s a 1% chance of observing a difference as large as this – even if there is no true difference between the two groups.  Can you get that slight difference in meaning?  I know, it’s as clear as a cell phone contract.  Let me put it another way: p values are a bit like a cheating husband – even if it appears he’s telling the truth, you should still take everything he says with a grain of salt.  (Or salt replacement, to keep my heart healthy friends happy.)

Another issue with statistical significance is that it is heavily influenced by sample size.  Small samples can’t detect small differences between groups.  Conversely, when samples are extremely large, you can pick up even the smallest of differences – even though they may not mean much in real life. This brings us to the issue of clinical significance (as its called in medical research) or practical or program significance (as it may be referred to in other fields).   Let’s go back to our experiment where Drug A was associated with a statistically significant drop in blood pressure (p<.001).  But is the drop big enough to make any difference in the health of the people who take it?  That’s what we mean by clinical significance and it’s something that typically has to be established by experts or by other criteria.  In some cases, what is considered “clinically significant” may be subjective and somewhat arbitrary.  And yes, sometimes you can have statistical significance without clinical significance or vice versa.  Oh my.

Take-away message/bottom line: When reading something, stop and consider what is really meant by “significant.”  Be aware of the difference between clinical and statistical significant.

 

Are the days of statistical significance numbered?

As I talked about in my last blog, there are different types of significance: statistical and clinical.  What is statistically significant may not always be clinically important.  In part this is a result of the fact that statistical significance reflects not only the size of the difference between groups but also the size of the group or groups you’re studying. In psychology, there’s a movement away from significance testing in favour of looking at effect sizes.

If you’re a techy type and want to learn more, check out Rex Kline’s book Beyond Significance Testing, Reforming Data Analysis Methods in Behavioral Research (APA, 2004).  He’s at Concordia, so it’s even a Canadian statistician.  If you’re not a techy, avoid Kline’s book unless you have a love of Greek symbols and a bad case of insomnia. Kline is part of the movement that recommends that instead of a black-and-white world of “this is significant” and “this is not,” you need to look at effect size.  Interpreting effect size is not a “one size fits all” process, as effect sizes may vary in different fields.  Above all, this book emphasizes that statistics should never be used as a substitute for thinking: don’t let the (significance test) tail wag the (research or evaluation) dog.

So what’s the take-away message for users of data?  First, when trying to interpret your data, consider the effect of sample (group) size on significance.  If you have an extremely small or large sample (e.g., <20 vs. several thousand), you’ll probably want to look more at effect sizes than p values. In fact, traditionally it’s considered a best practice not to report percentages if your n<20, so reporting p values may be even more problematic. Conversely, at the other end of the spectrum I’ve had data that, because of the huge size of the database (>45,000 records) produced significant p values for comparisons that actually had extremely small effect sizes.  Like teeny-tiny.  With such small effect sizes, p value be damned: I wouldn’t want to make communications or marketing decisions based on such relationships.

Take-away message/bottom line: In some cases, tests for statistical significance may not be the optimal way of analyzing data.  If you’re getting a report on a very small or a very large sample, you may want to ask for analysis of effect size.

 

Odds Ratios

The other day, I wanted to calculate the odds ratio (OR) of people in one database being different from those in the other database.  Since they were in different files, I couldn’t just run it through my statistics software and let it pop out an answer for me. There are a number of alternatives if you ever need an OR calculated.  First, you can do it by hand, which isn’t hard.  To calculate an odds ratio, you need to organize your data in a 2×2 table.  Let’s say ElissaPR and CarleneRS are trying to organize a three-star restaurant junket.  They have the choice of two cities.  You could look at percentages: 54% of the restaurants in City X are three-star (400/740) compared to 41% in City Y (114/278). So being smart gals, they’ll start packing for City X.  But what is they wanted to look at their odds of finding three-star restaurant if they go to X instead of Y?

Restaurants in:	Three-star	Nondescript	Total
City X	n=400 (A)	n=340 (B)	740
City Y	n=114 (C)	n=164 (D)	278
Totals:	504	514	1,018

The odds ratio compares the frequency of the occurrence of the characteristic you’re interested in between two groups.  It’s calculated by calculating (A x D) and dividing by (B x C).  In our case: AD =    (400 x 164)      =  65,600         = 1.69 BC       (340 x114)           38,760 Ok, so what does that mean?  It means the odds are 1.69 times higher of finding a three-star restaurant if they go to City X than if they went to City Y. If you don’t want to haul out your calculator and do ORs by hand, there are other options.  For example, it’s not hard to make Excel do the calculations for you.  Or even better, there are some handy-dandy and free Internet calculators. Here are a couple of links I’ve found (disclaimer: I can’t vouch for them but also have no ties to them). There are probably lots more. Just be sure to check to see how the tables are set up, as some might differ in the order in which you should enter your numbers. Calculator for confidence intervals of odds ratio in an unmatched case control study: <http://www.hutchon.net/ConfidOR.htm> 2×2 Contingency Table with OR, 95% CI, Phi coefficient of association, chi-square and fisher exact test: <http://faculty.vassar.edu/lowry/odds2x2.html> (What more can a girl ask for?)

Take-away message/bottom line: When comparing two groups, the Odds Ratio can show you the size of the effect of group membership.

 

How confident is your interval?  Understanding the Confidence Interval

Previously, we talked about Odds Ratios (OR), which is a nifty little statistic.  But if you’re reading something with ORs, check to see if each has an associated 95% Confidence Interval (CI).  Not CSI, but CI. Andy Field in Discovering Statistics Using SPSS (Sage Publications) defines a confidence interval as the range of values around a given statistic (such as an OR) that is believed to contain the true value of that statistic.  The “true value” means not just your sample statistic but the value you would get if, theoretically, you could calculate it for the entire world.  A 95% CI means that there is probability of 95% that the CI contains the true mean or OR. The CI is typically shown at 95% but it sometimes reported at 99% or even 80%.  What CI you choose will depend in part upon how important the sample estimate is for what you’re doing.  If you’re using polling numbers to decide whether or not to mortgage your house to support a run for political office, you’d never want a range with only an 80% chance of capturing the true measure of your popularity.  In this case, you’d probably want a 95% or even a 99% probability. In everyday life, we typically think of the 95% CI as the range that captures 95% of observations around your statistic.  But as you can see, the true meaning of the CI is slightly different.  (In Bayesian statistics, the counterpart to the CI is the credible interval.  I won’t go into that but you’ve got to love the name.) So how do you interpret a CI?  To give you an example, I’ll quote some calculations I did recently.  I found that compared to the general population of Canada, people who signed up for a certain health etool had an OR of having asthma of 1.12, with the 95% CI endpoints being 1.03 and 1.21.  This means there was a 95% probability that the effect of etool membership was as low as 1.03 (i.e., 3% higher) or as great as 1.21 (i.e., 21% higher).

Take-away message/bottom line: Confidence Intervals shows you the range that is believed to contain the true value.

 

It only looks bigger….

Recently, I was working with a group and someone suggested that we order the variables in a table by the p values of the regressions (e.g., .001, .01, and .05) so it would be “easy for the reader to see which variables are more important.”  The person who said this is very smart, but like most of us just got carried away with the prettiness that is presented by p values.  Let’s face it, when you’re plugging away analyzing some data, p values of .001 are like little gold nuggets sparkling in the midst of grit and grime (cooing “my precious” is not unknown).  Such findings must be more important than those with pedestrian p values of .05, mustn’t they? Unfortunately, the answer is no.  In his book Beyond Significance Testing, Reforming Data Analysis Methods in Behavioral Research (APA, 2004),Rex Kline lists the assumption that the p value is an indicator of the magnitude of an effect as one of the fallacies of significance testing.  In actual fact, p values reflect both the size of an effect and the sample size.  As I’ve said, if you have a large enough sample, even very small differences can end up with impressive p values.  With small samples, even a large difference between two groups may have only modest p values such as <.05. This has implications when you are reading research results.  A large study with thousands of participants may be able to produce findings with p values of <.001 without breaking a sweat, whereas studies involving small groups (like you traditionally find in psychology) may struggle to produced findings <.05.  I know, it’s not fair.  But the stats gods never promised anyone fairness. What are the practical implications when you’re reading something that include p values? Remember what p values do – and do not – tell you (as discussed in an earlier blog).

1. Consider the sample size, such as the number of participants or measurements.
2. Try to read at least some of the Methods section to find out how the study was conducted.  I know, the Methods section is pretty boring (if it’s any comfort to you, it’s as boring to write as it is to read), but it’s very important.  If the research used crappy methods, than the p values really don’t matter.  GIGO.
3. Make sure you consider the implications of a finding.  Earlier I talked about the difference between statistical and clinical (or programmatic) significance, and it’s something you need to keep in mind.  So a study found that men with eight-track tapes were 10% less likely to buy running shoes.  First of all, nobody is using eight tracks anymore, so who gives a shit?  And is this a real relationship between two variables or simply a chance finding that popped up because you threw so much statistical mud at the wall that something had to stick?

Take-away message/bottom line P values tell us the probability that a finding is due to chance but tell us nothing about the magnitude of an effect.  For that information, we need to use effect size statistics.  

 

Telephone polling – RIP?

What with call display, households pulling their landlines to avoid telemarketers (that’s me!) or people who don’t bother to have landlines (the young and the restless), the traditional telephone-based random-digit dialing poll appears to be in its death throes.  In the US, polling firms are experimenting with cell phone surveys, but honestly – do survey firms really expect that people will pay connection fees so the firms can make money? My personal opinion is that polling is a bit of a necessary evil.  On the one hand, I recognize that polling is often required in order to convince certain stakeholders, such as government ministries and departments.  And the media loves polls.  On the other hand, polling methodologies are so different from the sort of research I was trained in that I find it difficult to reconcile the two approaches.

Just so you know, polling firms do not use the word “random” in describing their samples.  The dialing is random but the selection of people is not: they work to fill “cells” so the final sample is an approximate representation of the general population by age, gender and part of the country.  And to be honest, they do a good job at that.  I’ve looked at results from a poll and compared the sample to the demographics of the general population (from past – and perhaps endangered — census data) and they are pretty similar.  Kudos on that aspect. The problem I have with telephone polling is that – despite the demographic similarities – you’re still talking about a sample of people who are willing to sit on the phone and answer a whole bunch of questions being asked by a complete stranger.  Who are these strange people?  Are they lonely?  Or bored?  Have they no Facebook?

There’s also the issue that polling firms tend to only give you data on people who complete the questionnaire.  Let’s face it, there must be lots who quit somewhere along the line.  As a rule, they also don’t tell you the number of people they had to contact to get their target sample number (in other words, the number of non-responders).  In traditional research, information on non-responders and incompletes is captured and used to a) calculate the response rate and b) determine if incompletes are significantly different from completers Several years ago, I described this situation in a presentation at a meeting of the American Evaluation Association (I was talking about some of the practical aspects of working with polling firms when you are trying to evaluate programs), and my audience of primarily university-based researchers looked at me in amazement.  To these people, this was an egregious error. Maybe in another blog I’ll go on to talk about some of the technical issues that you need to consider in interpreting polling data.  I think they’re important because, let’s face it, polling is a fact of life in most countries.  It’s not going to go away.

Take-away message/bottom line: Polls can be helpful but we need to interpret them keeping in mind both their strengths and weaknesses.  They are tools for decision-making – not decision-makers.  

 

It’s hip to be bi?

Jargon is the lifeblood of many professions and trades.  A doctor won’t say your arteries are blocked but that they are “occluded.”  A plumber refers to male threads and female fittings.  And we all know about the jargon of Twitter.  Statistics is not only rife with jargon, it’s practically lousy with it.  Sometimes, in the rush to sound “scientific” we make it harder for people to see how commonplace and simple statistics can be. Personally, I’m having trouble getting used to the trend of talking about “bivariate statistics.”  I understand what it’s supposed to mean – statistical procedures used to describe the relationship between two variables.  Some sources go on to state that the primary focus of bivariate statistics is the extent to which the two factors vary together (co-vary), which suggests correlations and regressions.  The statistical software giant, SPSS/PASW, lists bivariate statistics as including the t-test, ANOVA (ANalysis Of VAriance, rather than one episode of PBS’s great show, NOVA), correlations, and nonparametric tests such as rank sums.  It puts regression into another bucket.

Back in the olden days, as we scratched away on our parchment scrolls in the original Greek, the root “-variate” tended to be used almost exclusively in referring to linear regression.  So you could have univariate linear regression in which you look at the relationship between one dependent variable (the variable that you’re trying to explain) and one independent variable (the variable that you think may help to explain the dependent variable).  To give you an example, if you’re trying to see if income is determined by your number of years of education, the dependent variable would be income and the independent variable would be number of years of education.  The norm is to call this univariate regression because you are testing one relationship, whereas in fact two variables are involved – which echoes the definition of bivariate.  Isn’t terminology wonderful? The contrast to univariate regression is multivariate regression.  In multivariate regression, you have two or more dependent variables and any number of independent variables.   An example would be looking at the relationship between the independent variables of education and seeing if it affects both your income (dependent variable #1) and blood pressure (dependent variables #2). To complicate the matter, let’s not overlook multiple regression.  Multiple regression is simply a regression in which you have one dependent variable (as in univariate regression) but more than one independent variables.  If we were go back to our example, you might want to see if income is influenced by not only your own education but also your parents’ income (i.e., them who has, gets?).  So instead of two variables (bivariate), you now have three (again, should this be trivariate?) and you can add as many variables as need (quadvariate?, quintvariate?, septvariate? where will it end?).  Multiple regression is pretty common; multivariate regression probably less so. Occasionally, people may use the term “multivariate regression” when actually they mean multiple regression.

Take-away message/bottom line:  If you get a report and it talks about bivariate analyses, don’t panic.  It just means they looked at one independent and one dependent variable, and it probably involved common statistics such as t-tests, ANOVAs, correlations or univariate linear regression.

 

 Is it average or is it mean? 

For 99% of us, I suspect our first exposure to statistics was the class average.  You get your mark and then the average for the entire class is given so you can see where you stand (are you at, above or below average?). We all know how to calculate an average: add up the scores and then divide by the number of scores.  The problem with an average is that it doesn’t tell you anything about the variance in your group or study.  That’s why averages are one of the key topics in Darrel Huff’s classic, How to Lie With Statistics (written in 1954 but still in print and cited by Wikipedia as “the most widely read statistics books in history”).

An average is actually what statisticians refer to as a mean.  But unlike an average, a mean never goes anywhere without its standard deviation (sd).  It’s sort of like that friend you have who can’t image going to a restaurant or movie by herself.  If the standard deviation is not reported, you’ve got to wonder why. The standard deviation is a measure of the difference between each observation or score from the overall group mean.  If the standard deviation is small, then it means that values or observations tended to cluster fairly close to the mean.  If the standard deviation is large, then it means there’s a lot of spread: some values may be very small whereas others are very large Let’s say you calculate the mean income of a group of five people: a factory worker, a plumber, a salesman, a middle manager and Bill Gates.  Guess what – Bill’s income is going to swamp the results and you’ll end up with a mean that isn’t at all representative of the incomes of the other people.  And you’ll see at a glance that there’s a problem because the standard deviation will be large in comparison to the mean.   So when looking at means, don’t overlook the standard deviation. You can calculate the standard deviation by hand but it’s a bit of a pain: you need to subtract the mean from each number and square it, add them all up, divide by the number of scores – 1, and then find the square root (you see what I mean? – no pun intended).  A better option is to use a calculator with a sd option (if you’ve lost your instruction booklet, look online).  Or there are online free calculators.  I’ll give you links to a couple I’ve come across (disclaimer: I can’t vouch for them but also have no ties to them). Easy Calculation.com  http://easycalculation.com/statistics/standard-deviation.php (it seems they have a number of free statistics calculators, but there also have a zodiac sign calculator if you’re the only person in Canada not to know your sun sign) MathisFun.com http://www.mathsisfun.com/data/standard-deviation-calculator.html (officially I think it’s supposed to be for kids, but what the heck; just type over the existing example numbers and enjoy)

Take-away message/bottom line: An average is half a statistical: the arithmetic mean missing the standard deviation. You need the standard deviation to see how closely the numbers cluster around the mean.

 

Mean alternatives: median and mode

Previously, I talked a bit about means and the importance of the standard deviation in interpreting them. If your mean is craptorious, you have some options.  One is to get a bigger and better sample.  Another is to use another measure of central tendency such as the median. If you order a data set from smallest to largest, the median is the score in the middle.  If the data set has an even number of observations, the median is the average of the two scores that fall on either side of what would be the middle value.  In other words, the median is the number which divides the bottom half of the group from the upper half.  In lay terms, the mid-point. The median can be really helpful when you have a data set with a large amount of variance.  For example, time to arrive at hospital after a stroke can vary greatly, from a few minutes to over a day.  The mean arrival time would have a huge standard deviation associated with it because of the influence of the extremely slow responders.  In this case, you may want to look at the median to help you interpret the mean. Another statistic that can be used to describe a database is the mode.  The mode is the value or score that occurs most frequently.  So if out of 10 cases, six people arrived exactly at 33 minutes, the mode would be 33 minutes. People seem to love means, and although I’m partial to the little devils, they’re not always the best statistic to understand a database.  It’s important to understand that there are other options and you’ll probably look at several to figure out which is best.  For example, if your data doesn’t look at all like a bell curve (what statisticians call the normal distribution), then a mean or even a median may not be an accurate representation. And finally – means should never, ever, under penalty of death, be used when the numbers represent categories.  For example, if there was a category coding animal species and 1=lion, 2=tiger and 3=bear, you can’t calculate a mean animal species.

Take-away message/bottom line: There are several statistics other than the mean that look at central tendency.  Two are the median (the value that divides the sample into two equal halves) and the mode (the most frequent value).

 

More on means

It’s common for companies doing market or public opinion polling to report means.  Sometimes, they’ll also report medians.  But you shouldn’t just automatically churning out means or medians.  If you really want to understand your data, it’s helpful to start by looking at it by means of a graph, such as a scatterplot or a histogram.  What you find may surprise you. You’ve probably heard about the bell curve.  Most (inferential) statistics are based upon the assumption that the data being analyzed have a normal distribution, which is represented by that bell curve.  But what if your data doesn’t fit that nice pattern? This histogram (a sort of bar chart) shows a made-up distribution of weights.  I created it to represent a sample of Hollywood actresses, so the weights are pretty low.  Superimposed on the histogram is the normal curve or bell curve.  If I ask SPSS to spit out descriptive statistics, it tells me that the mean is about 117 lbs with a standard deviation of 10 lbs, and a median of 116.0.

But do the mean and median accurately tell you what’s going on with this sample?  Not really.  If you look at the bars, you’ll see that there’s actually not one but two peaks.  One is around 110 lbs (leading ladies) and another around 128 lbs (actors who play best friends).  Moreover, the first peak is skewed to the left, meaning that instead of being a nice bell curve, leading ladies tend towards the anorexic. In this case, just looking at the mean or even the median could be misleading.  If you really want to understand what is going on, it’s helpful to visually look at your sample.  Instead of one nice bell curve, you could have two or even three peaks.  The bell may be skewed to the left or the right, be tall and skinny, or short and fat.  Each means something different – and you may miss it if you just look at means and medians.

Take-away message/bottom line: If your polling firm or market research house gives you means or medians, ask if they looked at the data to see its actual distribution.  It might not conform to the standard bell curve – which is important to know.

 

Margin of error in polling data

As we’re heading into an election, there’s going to be a lot of reporting of public opinion polls in the media.  So this might be an opportune time to talk about margin of error. Previously, I talked about the confidence interval and gave you its definition: the range of values around a given statistic (such as an OR) that is believed to contain its true value.  That’s a little different than the folk or lay definition that you’ll frequently hear: the range that contains 95% of the observations.  Mea culpa, mea culpa, I’ve used that (erroneous) definition myself when trying to explaining the margin of error in polling data.  It’s just easier for people to grasp that concept and at least they’re getting the point that the statistic you’re quoting is not absolute but actually involves a range. So you read a news report that says something like, “when surveyed, 43% of Canadians thought politicians should be shot, the survey having a margin of error of +3%, 19 times out of 20” (a pretty common way of reporting polling results).  What it’s actually telling you is that there is a 95% probability (19/20) that the proportion of Canadians who think politicians should be shot is somewhere between 40% and 46%.  This may perhaps show you just how tenuous statistics can be: they are merely estimates of what is happening out there in the big, bad world, created by using a small sample of that world. The other thing you need to realize is that what is typically reported is an estimate of the margin of error for the entire poll.  It reflects the maximum sampling variation of any percentage based on all respondents from that poll.  Typically, that maximum variance is based on a 50% proportion or prevalence. The margin of error for any one question within the poll may vary somewhat, particularly when:

• the percentage or proportion reported moves further and further away from 50% (the margin of error actually declines as the proportions becomes more extreme)

• the number of respondents (the sample n) varies (e.g., you start trying to look at subgroups within the sample, such as women vs. men, or a lot of people refused to answer a particular question)

For most purposes, the overall margin of error should be sufficient.  But if a particular estimate is really, really critical to your operation, you might want to ask your vendor to calculate its specific margin of error.  And in interpreting it, don’t forget its real definition.

Take-away message/bottom line :The margin of error for polls is typically an estimate for the entire poll.  For the stat in question, it tells you the top and the bottom of the range for which there is a 95% probability.

 

About Statistics

A September 29, 2010 blog in UK newspaper The Guardian may be worth your while.  Here’s the link: http://www.guardian.co.uk/science/blog/2010/sep/29/statistics-lies-abuse The blog by Matt Parker, entitled The Simple Truth About Statistics, makes the point that “in the age of the internet, there is no reason why anyone should be fooled by statistics.”   His thesis is that the internet gives people the ability to check on statistics and to look at actual source journal articles or reports.  He says, “By their very nature, statistics can only be misused when the audience doesn’t bother checking them.”  But I think he misses an important point.  The target of most media reports containing statistics – such as the general public, politicians, physicians, and other decision-makers – typically don’t have the time to do the sort of fact-checking that he recommends.  Yes, we should do it, but with information being thrown at us at hyper-speed, who has the time?

I think the real responsibility for the accuracy of statistics lies 50% with those who create them and 50% with those who report them.  I’ve chunked numbers and generated statistics for a number of media releases and for me there are two key priorities: make sure the statistics are as accurate as possible (given that to err is human and sometimes you have only limited data with which to work) and explain them as clearly and simply as possible to reporters.  It can be a difficult balancing act – in large part because words like “significance,” “probability,” “odds, and “likelihood” have both technical and lay meanings.  It’s important that you don’t assume that reporters will have any training in statistics or will be able to interpret your numbers to their readers effectively or accurately. At the same time, reporters also have a responsibility to ask enough questions to ensure they understand what they are reporting. This can be difficult when they are facing deadlines and under the gun to find new, startling or interesting numbers to please their editors.  After all, the news business is a for-profit enterprise.  Whether through TV, newspapers or the internet, each business is trying to generate a profit by attracting readers.  Boring news does not attract readers.  For example, a story stating that a new compound reduced the rate of tumours in ten lab rats from 3% to 1.5% is not very attractive.  But a story saying that an experimental compound cuts cancer rates in half is likely to catch your eye.  It’s show business, folks. So am I advising people to ignore all stats in media reports?  No, that would be throwing the baby out with the bathwater.  But it’s important to take a somewhat skeptical stance if the stat in question 1) may influence important decisions, 2) contradicts everything we’ve previously known, or 3) sounds too good to be true.  If one of those applies and you have the time, it may be worth your while to do a bit of your own fact-checking.

Take-away message/bottom line: Don’t assume that all statistics reported in news reports or stories are accurate or have been interpreted correctly.

 

A picture is worth…..  sites for charting/graphing options

Charts can be terrific in showing and explaining data.  Of course, that’s if they are done correctly – a badly designed chart can be misleading and inaccurate. I recently (via a Guardian blog) happened upon a nifty little chart that shows some of the common chart options for data.  If you want to see it, go here: <http://sites.google.com/site/abhishektwr/choosing_a_good_chart.pdf?attredirects=0> There’s also a good, if somewhat complex, “Periodic Table of Visualization Methods” at: <http://www.visual-literacy.org/periodic_table/periodic_table.html>.  It’s an interactive table, so be sure to move your mouse over the cells to see what it suggests.  This table isn’t limited to statistical data, but includes suggestions on how to visualize information, concepts, strategies, and metaphors.  So you also see things like stakeholder maps, cycle diagrams, Gantt charts, concept fans, etc.  Can keep you busy. If you want to use charts or graphs for more than just presenting data, you might want to check out VizThink (<http://vizthink.com>).  VizThink describes itself as “a global community for visual thinkers and communicators who like to get beyond words and believe that visuals can be an effective tool whether you’re just trying to work through your ideas or working to get your message across as simply as possible.”  Whew.  Perhaps being a visual thinker leads to run-on sentences (as in James Joyce)? Of course logic models are a common form of concept modeling, where you trace the hypothetical (or wishful) relationship between your program inputs (e.g., health brochures), outputs (health consultations) and outcomes (short- and long-term health impact).  You can create logic models using specialized software of something like Visio.  The University of Wisconsin has templates in Word and Excel if you want something simpler (<http://www.uwex.edu/ces/pdande/evaluation/evallogicmodelworksheets.html> ).

Take-away message/bottom line: Sometimes, charts can be more effective in communicating ideas that text or tables.

 

Tips for making clear and effective graphs

A picture really is worth a thousand words.  Studies have shown that when people read reports, their eyes naturally gravitate to charts and pictures.  As most people scan instead of diligently reading your wonderful prose, it’s well worth your while to spend time producing effective graphs.  If you’re like me and tend to stick with plain old PowerPoint when charting data, here are some suggestions.

• Many sources suggest you avoid pie charts, stacked bar graphs and stacked line graphs, as they can be difficult to interpret correctly.  I particularly dislike stacked bars: unless you have only two categories, it’s really difficult to compare proportions between bars.  If each stacked bar charts represents 100% and there are something like four or five categories, then it’s almost impossible to read. If your data are that complicated, maybe you’d be better off with a table.
• Although PowerPoint seems to like 3-D bars, avoid them, as they can make it difficult to determine the true height or value of the bars.  It might be OK if there are large differences between the bars, but if they are close together it can be hard for the eye to estimate the gaps.
• For line graphs, avoid broken lines because they require the brain to try and link the pieces together, or colours that are hard to see or to distinguish from one another.  If you’re resorting to Kelly green and lime green, maybe you have too much information on the chart.  Information overload is bad for clear communications.
• The “aspect ratio” refers to the comparison of the width to the height of the chart.  Apparently, the optimal aspect ratio is 1 (height) to 1.3 (width).  If you fiddle with the aspect ratio too much, you can create a misleading impression of the data (e.g., overly tall, skinny bars or short, fat bars).  Make sure the scale starts at zero so you’re accurately portraying the data and that all bars are the same width.
• The rule of thumb is that categorical data should be portrayed with bars and only continuous data can be shown using line graphs.  But if you have a lot of data points (e.g., number of Baptists in the population at the time of some fictitious census, by year between 1901 and 2008) then you may use a line graph.  Theoretically, each data point should be marked with a symbol such as box or triangle, which gives the reader a hint that it’s actually a point estimate.  In other words, the line between points does not represent change in the number of Baptists throughout the year, but change between one year and the next.
• It’s tempting and may make you feel clever, but using a chart with one scale on the left-hand side and another on the right-hand side can confuse people.  Aw shucks, Time magazine does that thing all the time, you say.  Yes, but Time magazine has a whole graphics department to make sure people aren’t confused by which scale is being applied to which variable.
• Label everything very carefully so your audience can see at a glance what you are illustrating or comparing.  Don’t assume they have actually read your article or are listening to your presentation.  The same applies to tables.

Take-away message/bottom line Good graphs that are accurate and easy to read take time and effort but are worth it.

 

Favs from my bookshelf — statistics with Andy Field

A few years ago, I discovered Andy Field’s book Discovering Statistics Using SPSS, the second edition (2005).  I absolutely loved it.  Field is a psychologist at the University of Sussex and his book did three important things.  First, it used humour to explain statistics so they were not only understandable, but – dare we say? – interesting.  That’s no mean feat (statistical pun not intended but pretty nifty).  Second, it explained all those numbers and stats that SPSS spits out at you, giving you practical advice on what options to choose and what the output means.  That’s also very helpful.  I haven’t found a publication that helpful since that purple book we all used to have back in the olden days, when SPSS was still a mainframe package.  Third, Field also explained how to use SPSS output to calculate additional stats, such as effect size.  Yeah, I love effect size!  (Which suggests that I’m in need of psychiatric help, but that’s another story.)

Times change.  SPSS (Statistical Package for the Social Sciences) is now calling itself PASW (Predictive Analytics SoftWare).  I have no idea why the name change occurred.  Field has written a third edition, although it predated the software name change and continues to be Discovering Statistics Using SPSS, just 3^rd edition (2009).  The third edition is even heavier than the second, and he’s added a chapter on multilevel linear models (hierarchical modeling), the new hot babe of statistics.  But to tell you the truth, I’m a bit disappointed in the 3^rd edition.  In the 2^nd edition, he used personal anecdotes to lighten the mood, but in the 3^rd he’s gone somewhat overboard on the whole thing.  I really don’t need to see his baby picture, his first guitar, his first girlfriend, etc., throughout the years.  He needs to go back and strike a better balance between being funny and personable and giving us Too Much Information. There’s an accompanying website for the 3^rd edition (available through the publisher at http://www.uk.sagepub.com/field3e/default.htm) which has a number of complementary materials, including podcasts corresponding to some (but not all) of the chapters.  Field’s own website is Statistics Hell (www.statisticshell.com <http://www.statisticshell.com) where he also posts podcasts and class handouts.  It’s probably the only statistics site I’ve ever seen that posts a warning about its language and sexually explicit content.  (Maybe I should change my blog to Statistical Dominatrix?  Statistics S&M?)  If you have enough self-discipline, you could use the book and websites to teach yourself statistics, although the sound quality of the podcasts is highly variable and sometimes disappointing. FYI, if you’re a SAS user, in 2010 Field and someone named Jeremy Miles of the RAND Corporation published Discovering Statistics Using SAS.

Take-away message/bottom line There are some good books out there is you want to learn more about statistics or statistics software.

 

Can RCTs be used to evaluate websites?

In the world of medical and health research, the “gold standard” of proof is the randomized controlled trial (RCT).  The heart of the RCT is that it’s a trial or study that:

• compares a control group — a group that gets a placebo or existing standard intervention — to an experimental group that gets the new, experimental treatment (hence, it’s controlled)
• doesn’t consciously select people to go into one group or the other but does it using some sort of procedure that supposedly leaves it to chance (hence, it’s randomized)
• typically doesn’t disclose who is getting which treatment until the trial is completed, so patients, physicians or researchers cannot consciously or unconsciously skew the results (blinded if only patients are unaware and double-blinded if both patients and physicians are unaware)

RCTs are incredibly powerful but they have one serious limitation: they can only tell you what works within the carefully-controlled, somewhat artificial environment of the trial.  They aren’t so good at telling you what will happen in the messy real world, where people don’t have research assistants bugging them to take their medication every day, they aren’t being examined every couple of months, and they know they are participating in a trial and being watched, and they are told their participation is important.

That’s why in medical research a distinction is made between efficacy and effectiveness, two words that most of us consider synonyms.  Efficacy is used to refer to how well something works in a RCT whereas effectiveness is how well it works out there in the real world.  So a drug that has high efficacy in a clinical trial conducted at a major medical centre with motivated patients and specialist physicians may not have the same level of effectiveness when its given to everyday patients by family docs who don’t have the time or energy to preach adherence and may not be as skilled in titrating doses.

As websites are being used more and more to deliver health promotion programs, there’s a growing literature in which social scientists have used the RCT model to test them.  RCTs can give us some important information but participants need to be deliberately recruited and so are aware they are in a trial.  As well, research assistants spend a lot of energy trying to keep them involved.  As a result, the adherence rates of websites in RCTs are typically waaaaay higher than you would ever see in real life.  And I suspect the way people interact with a website is also different in real life than in RCTs, where tools are typically explained to participants.  In real life, people skip through sites, picking and choosing where they will go, scan – not read – text, and (like most men) ignoring instructions.  They use sites the way they want to, not how the designers or social scientists think it should be used. More importantly, not all users are the same.  There are actually naturally-occurring groups of users  — something you’re likely to miss within the limited number of people participating in a RCT.  For a couple of different takes on internet users, check out http://www.youtube.com/watch?v=uZa6CvC500k (a summary of six groups identified by Forrester Research) or http://www.youtube.com/watch?v=hP_5gC7aEM8.  Looking at these naturally-occurring groups within the natural setting of the real internet is the subject of my own doctoral research.  

Take-away message/bottom line Experimental settings such as randomized controlled trials can be helpful in studying the efficacy of websites but results may not be generalizable to the real Internet.

 

Clusters and classes and groups – oh my!

Dividing people into groups is a common practice in research, polling, marketing and health.  Some examples include:

• in the Diffusion of Innovations, Everett Rogers defined five groups according to how they respond to new technology: innovators, early adopters, early majority late majority and laggards (a classic paradigm that is still frequently used)
• in The Tipping Point, Malcolm Gladwell argues that change, fads and epidemics are shaped by the involvement of three key sub-groups: Connectors who link people, Mavens who are information brokers, and Salesmen (and presumably also saleswomen) who are charismatic “persuaders.”
• In Sex in the Snow, pollster Michael Adams argued that Canadians can be divided into three demographic groups (Elders, Boomers, and Gen-Xers), each of which has various psychographic “value tribes” (for Elders, Rational Traditionalists, Extroverted Traditionalists, and Cosmopolitan Modernists; for Boomers, Autonomous Rebels, Anxious Communitarians, Connected Enthusiasts, and Disengaged Darwinists; and for Gen-Xers, Aimless Dependents, New Aquarians, Autonomous Post-Materialists, Social Hedonists, and Thrill-Seeking Materialists; if nothing else, you’ve got to love the imaginative names).

Typically, sub-groups are formed by using segmentation statistics, such as cluster analysis or latent class analysis.  Segmentation is another area of quantitative statistics where the results can look deceptively definitive.  It’s important to keep in mind that these are statistical constructs.  We hope they reflect what is happening out there in the real world but in most cases we can’t be certain. Most standard statistical packages such as SPSS/PASW or Systat will give you cluster analysis procedures such as hierarchical and K-means, both of which require continuous variables.  Hierarchical analysis will give you a cluster tree and leave it up to you to decide where you think you should cut off the number of groups.  K-means asks that you specify the number of groups you think there should be and then will spit out a solution for you.  The 2-step procedure in SPSS/PASW has an option by which it will choose the number of groups for you, but is not as widely used as the other procedures. How many groups is the key issue in clustering.  Theoretically, your choice should be based on theory or previous research.  But what if there is no previous research to guide you?  Then you’re pretty well on your own.  I know of one polling firm that as a standard chooses four groups, on the grounds that “four is a good number for policy making.”  But if in the real world there are actually three groups, or five groups or even six (I’ve seen cluster analysis with six groups, and as they were clustering the entire population of Australia, that is reasonable), using an arbitrary four is not going to give you an accurate picture of realty. In another blog, I’ll give you a short summary of the methods commonly used to validate clusters.  And I haven’t even gotten to latent class analysis – if you’re dealing with people giving you psychodemograpics or things like this you should be aware of this procedure as well

Take-away message/bottom line If you have a polling or marketing firm presenting cluster analysis to you, ask how they decided on that number of groups.  Did they test other options to see if they might fit the data better?  

 

The computer persona – to be replaced by segmentation?

In November, Forrester Research (www.forrester.com <http://www.forrester.com>) started a discussion regarded the future of design personas, which it defines as “models of target customers based on ethnographic research.”   I’ve been involved in projects where personas (I think it should actually be personae, but what the heck) were developed to illustrate the sort of “typical” user the website would (in the estimation of developers) attract.  Personas can be helpful in getting developers focused.  But as with everything involving the Net, things are changing. One commentator on the Forrester discussion board described his company’s use of personas as shifting to a form of storytelling and using actors to act it out with stakeholders.  Another commentator stated that in his company, personas are tied to quantitative research and data. You’ve probably already figured out where I’m heading.  Segmentation procedures such as cluster analysis appear to be ready-made for identifying potential personas by dividing users into groups.  There are problems, of course.

1. You need data to segment.

If you’re designing a new site, you’ll naturally lack data.  Is there any you can beg or borrow from similar websites or studies that might act as a reasonable proxy?

2. You need to know what you’re doing when you use packaged segmentation procedures such as cluster analysis.

Cluster analysis is designed to create groups based on similarities.  But before you start churning out solutions, you need to have a reasonable idea of:

• how many groups you should have.

I talked about this in an earlier post, if you’re interested.  To recap briefly, don’t go for a canned solution (e.g., always four groups) – there should be some logical, convincing or at least plausible reason for the number of groups formed.

• what factors are important in forming your groups.

The most egregious cluster analysis I ever saw was one by a polling firm that used 21 variables to form four clusters.  Twenty-one??!  I’m surprised they missed the kitchen sink.  Clusters are groups based on similarities and should be based on important or meaningful similarities.  How similar or dissimilar can four groups be when you’re looking at 21 traits or variables?

3. As cluster analysis software is vulnerable to the GIGO syndrome, you need to validate your solution.

I’ll start by admitting that I’ve only been privy to a few polling firm cluster analyses.  So my sample is small and could easily be biased.  But in those cases, not once was there a section or discussion of how the cluster analysis was validated.  I think that’s too important a step to just skip.  I’ll follow up on this discussion in another blog, giving you a run-down on the strategies academic researchers use in validating cluster analysis.  It’s not too much to ask polling or marketing firms to use at least one of them so clients and users can have more confidence in the results.

Take-away message/bottom line Segmentation procedures may be helpful when developing or validating website development personas.

 

I have groups – but are they the right groups?  Validating clusters.

In research, validity refers to the accuracy of your observations or solution. Validity can be divided into different types (content, face and quantitative) but that’s probably another discussion.  Rather, in followup to my last blog, I wanted to give you an overview of how groupings created by clustering procedures can be validated.  My source is that scintillating page-turner, Cluster Analysis for Researchers by H. Charles Romesburg (Lulu Press, 2004).  (If nothing else, by summarizing some of the key points, I’m trying to spare you the experience of reading books like this.)  Romesburg lists eight ways to validate cluster solutions.  I wouldn’t expect a marketing or polling firm to use all of these, but if you get a report with a cluster analysis, it’s entirely kosher to ask if they used at least one.

1. Do more than one solution and pick the best and most well-structured one.

He goes on to talk about “average Euclidean distance coefficient” but you don’t need to go there.  The point is just don’t pick the first solution the computer spits out at out, unless it’s both perfect and angels sang as you read it.

2. If there are existing cluster solutions out there to compare to, pick the one that best agrees with what you already know.

Good advice but there may not always existing literature out there. Or polling or marketing firms may not be set up to scan the research literature for previous solutions.

3. Go with expert intuition.

A variation of my previous advice not to let the statistical tail wag the research dog.  Statistical procedures such as cluster analyses are unthinking tools.  Sometimes we need to step back and listen to what real experts with a lot of previous experience and training have to say.

4. See if the solution holds up when you apply it to a different (but presumably similar) database.

Nifty advice, although I don’t know how often you’ll have another database just sitting around for you to use.

5. Compare results when you use different multivariate analyses on the database, such as principle component analysis, multidimensional scaling and factor analysis.

This is getting technical.  My advice if you’re the user of a cluster analysis is simply to ask the creators if they used other statistical procedures to see if they produced findings that support the clusters.  If the clusters are reflecting reality, then other procedures should point in the same direction.

6. Split your sample in two and run each independently to see if you get the same solution.

This is a common option, although it has also been argued that it tests for replicability of the solution rather than validity.  But it’s worth a try and it’s easy to do.

7. See if the solution remains the same even if you add more information or alter a small number of the data.

If you add more information and you get the same solution, the clustering is referred to as stable; if it resists changes to a small number of the data points it’s referred to as robust.  I don’t know how frequently polling or marketing firms would devote this amount of time to validating a cluster solution.  Whether it’s worth your while would depend upon how critical the clusters are to your organization.

8. Does the solution agree with what you thought might happen before you run the analysis?

The reasoning here is that your a priori assumptions are based on science or previous experience, whereas conclusions you make after you do your runs (a posteriori) are more likely to be rationalizations or justifications for your lovely solution.

Take-away message/bottom line If your polling firm or market research house presents you with cluster analysis, ask how it was validated.  At least one method should have been applied to determine if the cluster solution is reasonable.  

 

Can you bribe people to be healthier or more productive?

Over the past year and a half, I’ve been doing a lot of reading on motivation as part of my doctoral studies, with a focus on health behaviour change. Most people would like to be healthier but find it difficult to rustle up or maintain enough motivation to stick with it over the long term. It’s not that people are stupid or don’t care – the reality is that we live in an obesogenic society where it’s typically easy to be make unhealthy choices and hard to make healthy ones. One approach that is being tried is using a reward system to reinforce healthier choices.   In the UK, for example, the Points4life program has been launched in Manchester to test a card loyalty system. Loyalty programs have also been used by a number of insurance companies and heath maintenance organizations (HMOs) in the US to try and get employees or patients to lose weight, eat healthy, etc.

To date, the literature hasn’t shown a lot of success. Some small studies have shown the incentives or rewards can change simple behaviours such as keeping doctors’ appointments, but there’s been no evidence of substantive or lasting effect for complex behaviours such as losing weight or quitting smoking. In fact, in one study, incentives to quit smoking were actually associated with a higher relapse rate. Yikes, so you’re going backwards instead of forward. The finding that incentives may be associated with short-term benefit but long-term harm is actually in keeping with research on self-determination theory. Psychologists who study this theory (such as Edward Deci and Richard Ryan at the University of Rochester) believe that external pressures to change (such as rewards) actually undermine people’s intrinsic (just think of it as internal) motivation.  In the classic experiment of this theory, subjects were asked to complete a puzzle but were also filmed after the researcher left the room. As a rule, people tended to keep playing with the puzzle when on their own – after all, puzzles are kind of interesting and there probably wasn’t much else to do. But when people were paid for solving the puzzle, the amount of time they spent playing with it on their own fell precipitously. A reward shifted the puzzle from something you do because it’s interesting and fun, to something you do because you’re paid to do it. There’s a tremendous potential to apply self-determination theory in a number of fields.  I’ll give you a link to a real nifty You Tube video by RSAnimate that neatly encapsulates self-determination theory and shows how it applies to employee motivation: <http://www.youtube.com/watch?v=u6XAPnuFjJc>. As the video points out, rewards are only beneficial with simple, mechanical tasks; when something involves cognitive tasks or complex behaviours, they can be detrimental. This really brings into question the whole concept of using bonuses to motivate people to be more productive – or healthier. Closer to home, the Toronto Board of Education has been reported to be in the process of considering a reward system to motivate under-performing students.  Hmmmm, if you watch the video you might come to question that approach.  

Take-away message/bottom line Money may make the world go round but it may not be the best way to motivate people for the long term or for complex behaviours.

 

Making change – battle of the goals?

When we think of goals, we tend to think of ones that we set up deliberately and consciously, such as New Year’s resolutions.  But goals can also be unconscious (implicit).  As well, goals can conflict, particularly when they address short- vs long-term benefits.   Psychologists have determined that people can unconsciously manipulate their interpretation of clues to increase the odds of achieving their goals, and even (self) impose penalties or rewards to keep themselves focused (Counteractive Control Theory).  But there’s a catch: how important is the target goal compared to other goals or competing interests? Understanding goal conflict is important if you’re trying to pursue your own goals (e.g., losing weight) or supporting someone in their efforts (e.g., mentoring a colleague or employee).

In his book Dieting, Overweight, and Obesity, Self-Regulation in a Food-Rich Environment (APA, 2008), social psychologist Wolfgang Stroebe focuses upon eating, but I think you can probably draw analogies to other areas of life. The basic premise of goal conflict theory is that your behavour is shaped by the conflict between incompatible goals.  In eating, it’s the conflict between eating enjoyment (which for some people may also have emotional connotations) and your desire to look buff.  One goal is immediate and sensual, while the other is long-term and, let’s face it, often not a lot of fun.  Complicating the situation is that we live in an environment rife with external clues to induce eating enjoyment and overeating, ranging from the smell of Cinnabons at the mall to fast food ads on TV (food pornography). Coping with frequent or strong triggers requires mental resources (Stroebe describes them as attentional).  Emotions, particularly strong emotions, can reduce the attentional resources needed to resist temptation.  Hence, the stereotype of the couple who has just broken up: he heads to the bar to drink and she heads to the frig for ice cream. Stroebe’s work focuses on dieting and eating but you can see where the basic premise can apply to other areas of life.  So you have a student who says he wants to do well at school so he can get into med school/law school or whatever, but somehow can’t resist the temptation to party instead of studying.  Or an employee that you think could go far in the company but who consistently procrastinates and is late with projects.  It’s the same basic issue of goal conflict.  There’s no simple answer. Another thing to consider is that people differ in their inherent ability to delay gratification and their time orientation.  For a nifty summary of Philip Zimbardo’s research on people’s use of time, see the RSAAnimate video: http://www.youtube.com/watch?v=A3oIiH7BLmg.   (I really recommend this video.)  Manipulating the environment (e.g., removing distractions) and priming for goal achievement may help.  But trying to get a present-oriented teen or employee to focus on long-term goals can be extremely difficult. What I like about the goal conflict theory is that it helps to dispel the myth that people who aren’t successful in something, such as dieting, are stupid, lazy or unmotivated.  I don’t know many people who deliberately choose to be overweight, unhealthy or unsuccessful.

Take-away message/bottom line It’s not surprising that people’s goals can come into conflict with one another, particularly if goals operate on different time frames (e.g., short-term gratification vs. long-term benefit).  

 

If bribery doesn’t help people change, what about penalties?

In an earlier blog, I talked about the debate over whether rewarding people is an effective in changing behaviour.  As we saw, the evidence is not promising for this approach, even though there’s increasing interest in it.  A lot of buzz is being generated by HealthyWage (www.healthywage.com), a website that promises to pay you if you meet your weight loss goal.  The trick about HealthyWage is that it’s not based on self-report: to get your money you need to be able to produce a special verification form signed by either your doctor or a participating health club. So if there is benefit, is it because you’re being rewarded or because you’re accountable to someone else? But what about turning the tables – creating a situation whereby people are penalized if they don’t perform? An etool called StickK (www.stickk.com) is also generating buzz, in part because it’s origins.  It’s the brain child of four academics at Yale University from the departments of Economics, School of Management, and School of Law (but, alas, none from health or medicine).  Being an academic from Yale doesn’t mean that you’re right but it does mean you probably have a certain amount of brain power.  In fact, one of the founders wrote a book entitled Carrots and Sticks: Unlock the Power of Incentives to Get Things Done.  They may be doing something right – apparently the American Cancer Society is using StickK and there are corporate clients such as Staples.  One article states that it has more than 80,000 users. At the heart of StickK is the concept of the Commitment Contract. The Contract is your public pledge to make changes that you back up with money (if you want – it’s not mandatory and apparently only 30% do it).  If, for example, you want to exercise and you fail, your credit card will be charged whatever you pledged (for example $100).  You stipulate who the money goes to at the end of your contract, such as a friend or favourite charity.  You can have the money go back to yourself but the program doesn’t recommend it: the idea is to create accountability to someone else for your failure to act.  In some cases, users designate an “anti-charity” or an organization that they absolutely don’t want their money to go to (some examples being things like the National Rifle Association). StickK claims that the combination of accountability and financial incentives results in a 74% success rate, with the rate rising to 80% when you designate an anti-charity.  I don’t know whether we should take those numbers too seriously: they are short-term results based on self-report.  Stories have already emerged of people who have cheated in order to avoid paying up: honesty is the Achilles heel of systems based on self-report.  More importantly, we don’t know what happens over the long term.  There is research evidence showing that financial incentives may work in the short term but that the weight loss is not maintained over the long term.  Does the same happen here?  After all, in his book Punished by Rewards, The Trouble with Gold Stars, Incentive Plans, A’s, Praise, and Other Bribes Hoghton Mifflen,1999), Alfie Kohn states: The troubling truth is that rewards and punishments are not opposites at all; they are two sides of the same coin. And it is a coin that does not buy very much. (pg. 50) Although either penalties or rewards may seem exciting at first, the novelty may quickly wear off – especially as the reality of what losing weight requires sinks in.  Losing weight and keeping it off requires ongoing effort: read Gina Kolata’s Rethinking Thin if you’re tempted to think it’s easy. Dr. Kelly Brownell, director of Yale University’s Rudd Center for Food Policy and Obesity, has been quoted as describing incentive programs as “a waste of time.”  His apparent pessimism is rooted in the fact that such toys quickly lose their effectiveness when you live in an obesogenic environment.  Others are concerned that incentive programs turn weight loss into a contest that promotes risky and unhealthy behaviours. The focus may become on dropping the weight to meet your deadline, rather than safely or permanently.

Take-away message/bottom line Both rewards and penalties are being tried to help people make behaviour change but there is no evidence they are effective in the long term – and good reasons to suspect they may be counterproductive.

 

Getting better poll responses

Getting people to honestly and thoughtfully answer survey or polling questions can be a challenge.  There’s a lot of factors involved, but let’s start by focusing on those involving the respondent.  Things to consider include:

• the urge to be agreeable (acquiescence bias) or as Smith and Fletcher call it, “politeness bias” (Inside Information, Making Sense of Marketing Data, John Wiley & Sons, 2001)
• the urge to be seen as responsible, kind, mature, successful, etc. (social desirability, such as claiming you watch public television when actually you’re a Three’s Company rerun junkie)
• the fuhgddaboutdit syndrome: answering the question correctly is not worth the time, energy or mental effort so you fuhgeddaboudit and pick whatever so you can finish the damn thing as quickly as possible
• the tendency to score around the middle and to avoid extreme responses (think Prozac), or to score at the extremes (think drama queens)

There are individual and socioeconomic differences in how people respond to survey/polling questions.  For example, some studies have shown that those with less education are more likely to score on the extremes. The tendency to give positive or extreme answers also varies by ethnicity; in one study, for example, Chinese and Japanese students were more likely to select midpoints compared to American and Canadian students.  (If you want to read more about the effect of ethnicity, a study comparing response styles in 26 countries can be accessed here: <http://www.harzing.com/download/respstyles.pdf>.) How do you get around this issue?  An important first step is looking at how you ask your questions.  A 2010 article in Survey Research Methods by Saris et al. (vol 4, pp 61-79) found rewording questions could help to reduce acquiescence bias. Avoid questions that set people up to agree, such as:  To what extent to you strongly agree, agree, disagree or strongly disagree with the following statements: 1)   My health is excellent. 2)   The issue of abortion is very important to me personally. 3)   I rarely feel sad. Instead, score each question separately and avoid using agreement words.  Like this:

1. How would you rate your health overall: excellent, very good, fair, bad or very bad?
2. How important is the issue of abortion to you personally?  Is it extremely important, very important, somewhat important, not to important, or not important at all?
3. How often do you feel sad?  Constantly, very often, somewhat often, rarely, or never?

The first method makes it a lot easier to combine scores into a scale and may appear easier to program into a CATI script.  But as you can probably see, asking questions the same way reduces the need to think about the answers you give and increases the danger of acquiescence.

Take-away message/bottom line When designing or interpreting polling or market research questions, try to avoid writing questions that increase the risk of acquiescence bias or bias due to social desirability.  

 

The Ministry of Silly (Survey/Poll) Questions

In their book on the practical use of survey and polling data, Inside Information, Making Sense of Marketing Data (John Wiley & Sons, 2001), British marketing executives David Smith and Jonathan Fletcher give a 24-point “ask a silly question” checklist.  The title is a misnomer: it’s actually advice on how to avoid silly questions.  So I thought that I might rework their list to more accurately reflect its title.  I also combined some of the points so it would be a somewhat shorter list.

Silly Questions Checklist

1)   Make sure questions are irrelevant to your sample population (e.g., ask people in Saskatchewan about avalanche control).

2)   Ask people questions about things they have no experience in (e.g., ask someone who has never flown about in-flight conveniences).

3)   Make sure people feel that answering honestly could undermine their vision of themselves – either in their own eyes, or in the eyes of others (e.g., asking married presidential candidates if they have ever found anyone other than their spouse sexually attractive).

4)   Pose hypothetical questions that are so unlikely that the respondent has never thought about it (e.g., if you knew little green men were landing on the earth in ten years, would you buy more life insurance?)

5)   Create a situation whereby people feel rude if they don’t answer you a certain way (e.g., I’m a parent of a child with cancer. Do you think more money should be spent on cancer care for children?)

6)   Make sure you get the answer you want by creating a leading question (e.g., Doctors have confirmed that Stephen Harper has no heart.  Are you still planning to vote for this heartless politician the next time he forces an election?)

7)   Ask sensitive questions without any tack, assurances of confidentiality, or too early in the questionnaire, before people are more comfortable with the interviewer (e.g., How many times a week do you beat your wife and/or children?).

8)   Ask questions based on an underlying assumption that might not be true (e.g., asking every respondent with a darker skin tone “which country do you come from?”).

9)   Ask questions that are vague and abstract (e.g., asking which disease worries the respondent the most but not specifying whether “worrying” refers to the risk of death, suffering, or disability).

10) Use words that have connotations that can bias responses (e.g., should all Middle East terrorists be arrested? Should sexual perverts have access to counseling?).

11) Use unfamiliar words just to make sure respondents can’t understand the question (e.g., do you believe that a more sanguine demeanor on the part of Stephen Harper will assuage tensions in the House of Commons?).

12) Put two concepts in the questions or essentially ask two questions in one (e.g., are your two major suppliers Canadian-based?  How important to you is it to stop smoking and lose weight?).

13) Damn the grammar police – use double negatives (e.g., do you agree or disagree with the following statement: offenders under the age of 18 should not be allowed not to do community service?).

14) Ask respondents to remember things from the past – the further back, the better (e.g., how many times in 2009 did you clean your bathroom?).

15) Ask people to do feats of mental arithmetic (e.g., what proportion of your weekly pay goes to housing?)

Take-away message/bottom line There are lots of ways to write bad or silly survey questions.

 

Take two placebo and call me in the morning ….

In the November/December (2010) issue of The Skeptical Inquirer, there’s an interesting article by Yale University neurologist Steven Novella on the placebo effect.  (Yes, I confess, I subscribe to The Skeptical Inquirer.  But hey, I thought Big Bang Theory was making nerdy the new cool.)  Novella’s point is that we tend to think of the placebo effect as a unidimensional or simple phenomenon of mind over matter: you think the medication or treatment will help you, by gosh, it does – even if there is no active ingredient.  Novella’s point is that the placebo effect is actually much more complex; in fact, a number of factors may contribute.

• Bias or expectations of the subjects: you interpret symptoms or feelings to fit the scenario described – or at least implied – by the researchers
• Bias or expectations of the researchers: they find what they’re looking for
• Observer or Hawthorne effect: being observed or studied changes the behaviour or reports of subjects and researchers
• Cheerleader effect: subjects’ reactions may be influenced by their feelings about being in a study, such as relief that doctors are finally taking your concerns seriously and you’re receiving treatment, the social benefits of meeting regularly with research staff, or a renewed dedication to eating healthy or being active
• Physiologic changes: as Novella describes, participating in a trial may reduce stress, which in turn can help trigger the release of brain chemicals such as endorphins or dopamine.

Novella also talks about the importance of regression to the mean in understanding the placebo effects.  Statisticians frequently refer to regression to the mean so it’s a concept that is worth understanding.  It’s not rocket science: all it means is that if you get an extreme result, chances are that future results will tend to move (i.e., regress) closer to the average (i.e., the mean). To understand regression to the mean, let’s take an example some of my friends may appreciate.  You’re out shopping for boots and you find the pair that you’ve been eyeing for weeks not only on sale – but on sale for 80% off.  And they have them in your size, and they have the colour you want, and you get double airmiles, and they’ll give you a free leather protection spray, and the salesman has the cutest smile in the entire world.  Wow, what a shopping coup.  Now, what are the odds that all of these elements will ever align again in such a perfect combination?  Not huge.  Rather, your next boot-shopping expedition will probably regress to the mean, whereby you’ll get some of these elements but not necessarily all. The good thing about regression to the mean is that it also works in reverse.  Let’s say you go shopping and it’s the trip from hell: the stores don’t have anything in your size, everything is overpriced, the salespeople are snarky, the other shoppers are blatantly rude, your feet hurt, someone dings your car in the parking lot and doesn’t leave a note – so some kid thought it might be a good time to key the other side as well.  At least you know that according to regression to the mean, it’s unlikely your next shopping trip will be quite this bad.  However, I should add that regression to the mean is not necessarily immediate nor universal.  For some reason, it doesn’t seem to work well in restaurants – a bad restaurant typically remains a bad restaurant. If you’d like to see Novella’s blog, NeuroLogica, which he bills as “Your daily fix of neuroscience, skepticism, and critical thinking” you can find it at <http://theness.com/neurologicablog/>.  The Skeptical Inquirer website is www.ciscop.org <http://www.ciscop.org>.

Take-away message/bottom line The placebo effect is actually quite complex and can be influenced by psychological, study or statistical factors.

 

Is being overweight a death sentence?

A December, 2010, publication in the NEJM on all-cause mortality and weight has attracted a fair bit of media attention.⁽¹⁾  Most of the media pickup has interpreted the study to say that being overweight increases your risk of dying.  But how great a risk?  Ah, there’s the rub. Here’s a table derived from the article showing the results for women (the data in the abstract and the numbers most frequently cited in the media reports, although results were similar for men).  Hazard ratios were adjusted to account for age, alcohol intake, education, marital status and overall physical activity. If you’re curious, a hazard ratio is the statistics used when describing survival data. It tells you the odds that, compared to a reference group, how many people in the other groups will die during a specified period — the study period, which in this case was about ten years. It can’t tell you what will happen over a lifetime – just within the study followup period. Just so you know, the hazard ratio is also based on the assumption that for any fixed point in time, the risk of death for people in a group remains the same. (If you want more information about hazard ratios, there’s a nice if somewhat technical summary from Oxford that’s available at: http://www.medicine.ox.ac.uk/bandolier/painres/download/whatis/What_are_haz_ratios.pdf.)

BMI range	This falls into the Canadian weight category of …	Hazard Ratio	95% Confidence Interval
15.0-18.4	Underweight (<18.5)	1.47*	1.33, 1.62
18.5-19.9	Normal weight (18.5-24.9)	1.14*	1.07, 1.22
20.0-22.4		1.00	0.96, 1.04
22.5-24.9		Reference category
25.0-29.9	Overweight	1.13*	1.09, 1.17
30.0-34.9	Obese Class I	1.44*	1.38, 1.50
35.0-39.9	Obese Class II	1.88*	1.77, 2.00
40.0-49.9	Obese Class III (>40.0)	2.51*	2.30-2.73

* significantly different from the Reference Category

Note: In reading 95% CIs, the rule of thumb is that if it cross 1.00, such as 0.96-1.04, then the RR is not statistically significant from the reference category.

You’ll see that being at the low end of what Health Canada calls the normal weight range (BMI 18.5-19.9) has the same risk of death (14%) as being overweight (13%).   Being underweight (BMI 15.0-18.4) carries the same risk of death (47% higher than the reference category) than class I obesity (44%).  But I don’t hear much talk about these facts. One of the common explanations for the increased risk of death among the underweight category is that they may be already sick, i.e., that their thinness is due to diagnosed or nondiagnosed disease. But to my mind, it’s just as likely that people who are overweight or obese may be so because of a condition or problem. For example, someone with arthritis may find it difficult to exercise and inactivity contributes to weight gain; people who are depressed are also vulnerable to weight gain. In other words, if it’s good for the goose (explaining increased mortality for the underweight) I don’t understand why it’s may not also be good for the gander (explaining excess mortality among the overweight or obese). I have no problem with the idea that obesity is associated with increased risk. But I’m not so sure about overweight. After all, a 13% increased risk is pretty small. According to the Bradford Hill criteria, the stronger the relationship between two variables, the less likely it is due to extraneous factors. How big a relationship is big enough? That’s hard to tell. Studies on smoking have reported risk and hazard ratios between 10.00 and 30.00. Next to that, a hazard ratio of 1.13 seems puny. I remember years ago reading an opinion piece by an epidemiologist stating that a doubling of risk should be required before people get excited. Another problem is that there are studies showing that people in the overweight category may actually have a reduced risk of death compared to the normal weight class. Using a sample of 11,326 respondents to the 1994/95 National Population Health Survey followed for 12 years, Orpana et al.⁽²⁾ found that compared to the healthy weight class (BMI between 18.5 to less than 25), being underweight was associated with a 73% increased risk of mortality and severe obesity (BMI greater than 35) a 36% increased risk. But being overweight was associated with a significantly decreased risk of death (RR=0.83, which translates into a 17% lower risk) and there was no difference for those in obesity class I (for those with a BMI between 30 and 35 the RR was 0.95 with a p value greater than 0.05).  When they controlled for sociodemographic factors and health behaviours, being overweight remained protective (RR=0.75. 95% CI=0.65-0.86), but the effect of being underweight, severely obese or mildly obese were no longer significant.  For those who like numbers, they were: for underweight, RR=1.51, 95% CI=0.99-2.30; for severely obese, RR=1.09, 95% CI=0.86-1.39; for mildly obese, RR=0.84, 95% CI=0.67-1.05. Moreover, as Orpana et al pointed out, this wasn’t an isolated finding: other studies have also reported that overweight may actually be protective against mortality. That’s important, as another of the Bradford Hill criteria for evaluating a relationship concerns consistency: the more consistent findings are, the more confidence you can have in it. We tend to toss overweight in with obesity and it does plump up your numbers if you’re talking about the “obesity epidemic.” According to the latest figures from the 2007-2009 Canadians Health Measures Survey, 24% of Canadians 20-69 are obese and 37% overweight.  Add them together and you get an impressive stat of 61%. But if we’re talking about health risks, then it’s best to keep the two categories distinct.

Take-away message/bottom line The health risks of being overweight – as opposed to obese – are still unclear.

(1)  de Gonzalez AM, et al Body-mass index and mortality among 1.46 million white adults. NEJM 2010;363:2211-19

(2) Orpana HM, et al. BMI and mortality: results from a national longitudinal study of Canadian adults. Obesity 2009; doi:10.1038/oby.2009.191.

 

Is taxing sugar-sweetened drinks the solution to our obesity dilemma?

In late 2010, there was a fair amount of media pick-up of a study claiming that a tax on sugar-sweetened drinks could reduce people’s caloric intake and lead to a weight loss of ten pounds over ten years. Sounds great and isn’t the first time this sort of modeling has been performed – Kelly Brownell et al already published similar findings in NEJM in 2009. Let’s look behind the curtain, however. First of all, this wasn’t a study of what actually happened but computer modeling of what might happen if you imposed a sugar-sweetened beverage tax. But as the modeling showed, the entire population wouldn’t respond in the same way to such a tax. High income familiar would ignore it – as would low income families.  The only people who would really be affected would be those in the middle class. But which income group has the highest rate of obesity and thus is at higher risk of obesity-related diseases?  It’s not the middle class. I’m not a big fan of interventions that widen health disparities. The authors estimated that if you imposed a 40% tax (three times higher than our current HST) it would cut 12.5 calories per day out of the average diet and theoretically result in a 1.3 pound weight loss per year. A 20% tax would be associated with only a 6.9 calorie reduction, for an annual theoretical weight loss of 0.7 pounds per year. There are a lot of problems with that optimistic calculation. First, it’s based on the assumption that people who stop drinking sugar-sweetened soft drinks don’t replace the calories with something else. If they replace the drink with water or a non-caloric alternative, then the calculation works (but diet sodas are coming under fire too, so that may be off the table). But what if they replace their drink with something they think might be healthier? Here’s the calorie count of the same amount (100 grams) of different options:

• Sugar-sweetened cola = 37 calories
• Orange juice from concentrate = 49 calories
• Apple juice = 46 calories
• Whole milk = 61 calories
• 1% milk = 42 calories
• Chocolate milk = 83 calories

Coffee has no calories but that Canadian standard, the double-double (two servings of cream and sugar) has 150 calories and even a caffe latte made with nonfat milk (16 oz size) has 168 calories. It’s rare that people just stop doing something – usually they substitute something else. So what are the odds that you’d really end up with calorie savings? I’m no fan of sugar-sweetened cola (although I must admit to an addiction to diet Dr. Pepper – my mid-day caffeine boost).  I agree that most people would be healthier if they stopped drinking them.  But I also don’t like it when things are promoted as “fact” when they are based on hypothetical estimates based on what may be unrealistic assumptions (e.g., that people won’t switch to some other beverage).

Take-away message/bottom line Public health policies should narrow rather than widen health disparities. What works on paper may not work in reality

 

What in the world is Bayesian statistics?

There was a CP (Canadian Press) article in the paper the other day and it referred to an academic using “a rare system known as Bayesian statistics.”  Give me a break.  Bayesian statistics is not rare.  It’s not as commonly used as inferential statistics (your standard “testing for statistical significance”) but it’s not rare. In fact,  just recently Archives of Internal Medicine published a study in which Bayesian statistics was used to look at the ability of different classes of anti-hypertensive medications to reduce the odds of heart failure in people with high blood pressure. So it can be used to address real-life issues. After reading Nassim Nicholas Taleb’s The Black Swan, The Impact of the Highly Improbable (Random House), I got interested in non-inferential statistics.  Among other things, that led to Donald Berry’s Statistics, A Bayesian Perspective (Duxbury Press,1996).  Berry’s book is a little older (it comes with a disc, which is useful only as a bookmark) but is surprisingly readable.  Mind you, I don’t think it will ever make it to the Heather’s Picks list. Bayesian statistics deals with conditional probabilities.  Most introductory stat courses make you go through the ABCs of probabilities, although many students don’t enjoy it – they want to get to more meaty topics like chi squares and t-tests (statistics courses can be mentally unhealthy environments).  After reading Berry, I still don’t love probabilities (they can be frustrating little dickens) but I have a greater appreciation of what they do.  Moreover, it’s given me some ideas of how I might be able to apply Bayesian statistics to some data with which I’m working.. If you’d like to learn more about Bayesian statistics (yeh, sure), there’s some YouTube videos to help you:

• Math: Conditional Probability by Philip Brocum – http://www.youtube.com/watch?v=4PwnvqGEHoU&NR=1&feature=fvwp.  Does a nice job explaining conditional probability; starts by talking about OJ Simpson and then goes on to talk about HIV testing.
• The Monty Hall Problem Explained by Philip Brocum – http://www.youtube.com/watch?v=koPBkK_Ra-k&feature=channel  The classic three-door dilemma but be warned: the answer is not what you’d expect.
• Bayes’ Theorem (Formula) by bionicturtledotcom – http://www.youtube.com/watch?v=pPTLK5hFGnQ&feature=fvwp  There’s an error in the formula at one point (it’s noted) but it’s very good at showing the sort of formulas involved.

As Philip Brocum explains in his video, Bayesian statistics is not easy – quite frankly, he admits it’s hard.  Probabilities are often counter-intuitive.  Although I can intellectually follow the solution to the Monty Hall Problem, my gut still tries to tell me that since there are two doors left, the odds of winning are 1 out of 2.  Needless to say, I’m not exactly sailing through Berry’s book.

Take-away message/bottom line Bayesian statistics is a field of statistics that focuses on conditional probabilities.

 

Betting for dummies (which includes me!)

In a previous post, I told you how I was working my way through Donald Berry’s Statistics, A Bayesian Perspective (Duxbury Press,1996).  Still working on it.  It may take a while, as I’m actually trying to work through the problems. Berry likes real-world examples and in the process I learned things about betting that I didn’t know (guess my life has been too sheltered).  The odds for a race, for example (called track odds) depend upon the amount of money wagered adjusted for the track take (the fixed percent of every dollar wagered that is taken by the track).  The racetrack sets the final odds after seeing how much money is bet, so the bettor doesn’t know the odds until all betting is completed.  And as you can see, what is distributed to the winners is not the entire pot of all bets, but the pot minus the track take.  Also the final payouts are not based on the horse racing odds displayed when you place your bet, but on the tote board as it reads upon the close of betting. Bookies, on the other hand, don’t have the luxury of waiting until the wickets close to set their odds (e.g., for a football game).  To set their odds they have to guess how much money is going to be bet on the two teams and set the odds so it will be approximately balanced.  Too many bets on one side could enrich a bookie or bankrupt him. Berry writes that the day before the 1989 World Series the bookmakers advertised the odds as “9-11” in favour of Oakland.  He explains that what is missing are two implied 5’s.  The odds were actually9:5 against San Francisco and 5:11 against Oakland.  If you bet $5 on San Francisco and it won, then you would win $9 and get your $5 back, for a total of $14; of course, if the team lost you lost your $5.  But if you were betting on Oakland (because they were the favourite), you’d have to bet $11 in order to win an extra $5. When I look it up in Wikipedia, it states that the US uses format odds, which is the amount won on a 100 stake when positive, and the stake needed to win 100 when negative.  In an example on Understanding Money Lines, Allen Moody explains that if the odds are for the two teams in a game are written as -130 and +120, it means:

• the minus indicates that this club is favoured to win, so you’d need to risk $130 to win $100
• the plus sign indicates the underdog: if you risk $100 you could win $120.

Of course, this is just the beginning.  In sports you can bet to cover the spread, over/under, etc., etc. Not being someone who plays the ponies or sports teams, I didn’t know any of this.  So you see, people, there is value in reading statistics books.  Now, if they could just help me learn how to shoot pool.

Take-away message/bottom line If nothing else, Bayesian statistics can help you understand real-life processes such as betting.

 

Is obesity responsible for slowing life expectancy growth in the US?

Since 1980, the growth in the average life expectancy in the US has lagged behind some of the other western countries, says the US policy and research nonprofit, the National Research Council.  And the culprits, according to the NRC, are smoking and obesity.  (To read one media interpretation of the NRC release see http://www.canada.com/health/Study+says+smoking+obesity+blame+divergent+life+expectancy+rates/4164058/story.html . The NRC’s own release is available at http://www8.nationalacademies.org/onpinews/newsitem.aspx?RecordID=01252011.) The first news report I saw sort of mixed up two different measures of life expectancy.  Let’s separate them out.

1. Average life expectancy from birth – the number of years you estimate the typical person in the population will live.  As its name implies, it’s based on an average (i.e., a mean) and includes everyone who is born alive.

1. Life expectancy from age X – in the NRC’s original release, they emphasized that they were talking about average life expectancy from age 50. That means you calculate number of additional years most 50-year-olds will live. To be included, you have to make it to age 50.

It’s important to understand the difference between the two numbers.  Average life expectancy from birth is influenced by the infant mortality rate and childhood/youth mortality. As someone wrote in Wikipedia, “Because of this sensitivity to infant mortality, simple life expectancy at age zero can be subject to gross misinterpretation, leading one to believe that a population with a low overall life expectancy will necessarily have a small proportion of older people.” Let’s give an example (courtesy of Wikipedia). The average life expectancy from birth during the Roman empire was about 28 years. But that didn’t mean there were no old Romans – there were. They may not have looked good in their togas, but there were old people running around, probably muttering about how civilization was going to hell in a hand basket because of those crazy kids. Also, if you made it to age 15 (which means you survived all the common causes of infant and child death), the “typical” Roman could look forward to living another 37 years, to about age 52.  Vespasian was 60 were he became emperor and he reigned for ten years; Tiberius was about 56 when he took over from Augustus and he reigned for 23 years. In interpreting American life expectancy from birth, it’s important to remember that infant mortality in the US is not great.  According to the OECD statistics portal, the infant mortality rate in the US is 6.7 per 1,000 live births.  In Canada, it’s 5.1, while in the UK it’s 4.7 and in Japan it’s 2.6.  However, take the Japanese number with a grain of salt replacement – apparently Japan’s high average life expectancy may be slightly inflated by the fact that they count many infant deaths as stillborns (so they aren’t include in the calculations). The NRC wanted to look at the effect of lifestyle factors on the life expectancy of middle-aged adults, which is why they looked at life expectancy from age 50. That means accidents and injuries (the most common causes of death among infants, children and youth) have only a moderate influence on the average and chronic diseases (which are more common after age 50) make a greater contribution. But if chronic diseases are more common, it also makes sense that other factors – such as access to health care and income – become more important. And as we know, access to health care in the US is very much tied to economics and, ultimately, race. It’s complex and I don’t know if we can actually tease out the effects of individual lifestyle factors.

Take-away message/bottom line If you’re reading something involving life expectancy, be sure you understand whether the authors are talking about average life expectancy from birth or from a specific age. Different factors may influence different life expectancy calculations. Average life expectancy is not a good indicator of life span or the number of old people in a population.

 

You too can be a statistician and play with chi squares!

Ok, so it’s not as much fun as having your teeth cleaned.  But if you’re looking at rates or proportions and want to find out if there is a significant difference, the first test to consider is probably a chi square. I’m going to explain here how you calculate a chi square.  The point is not necessarily that you want to start calculating chi squares (there are lots of nifty calculators on the web and I’ll give you some links at the end) but I want you to see that statistics isn’t all that mysterious.  Pretty well anyone can do stats if they want to (the issue is more, how many people want to?).  In fact, the biggest challenge in stats isn’t how to do them (we have calculators and software packages to do that) but deciding what stat is appropriate for your data. Let’s take an example.  You find that 24% of people who watch What Not to Wear are Tea Party supporters while 27% of people in a survey you conducted are Tea Party supporters. The percentages are pretty close but your sample sizes are different. Are WNTW viewers less likely than the general population to be Tea Party supporters? First, start by organizing your data into a table using the actual numbers, rather than the percentages.

		Poll
		What Not to Wear Viewer poll	My poll	Totals
Tea-Party Supporter	Yes	120	324	462
	No	380	858	1238
Totals		500	1,200	1,700

If you are desperate or have done as many chi square calculations as I have, then you can do it by hand.  The formula is based on looking at the difference between the distributions observed and the distributions you would expect if there was no relationship between population (poll) and Tea Party membership.  It’s actually not hard to do at hand, but don’t worry, you don’t need to haul out your calculator.  There are plenty of free chi square calculators on the Internet you can use. Key points to remember are:

1. The chi square is only appropriate when you are comparing proportions in categories that are nominal (e.g., Tea Party supporter vs. not supporter) and are mutually exclusive (i.e., Tea Party membership is not a requirement of getting on WNTW).

2. In a nutshell, the chi square is a test of the association between two variables. The larger the sum of the chi square, the more the numbers vary from what we would expect if there was no relationship (association).  But remember – this is a test of statistical significance and so is affected by sample size and the accuracy of your data.  It can’t tell you anything about the size of an effect.

3. The chi square is unreliable if any of the four expected observations is less than 5.  If that happens, you’re supposed to use another test, called the Fisher’s Exact Test (although as far as I know, there’s no Fisher Inexact Test).

4. If you read that the chi square is Yate’s corrected, don’t panic.  It’s just an attempt to make it harder to find things statistically significant (i.e., to make the test more conservative). It typically doesn’t make a huge difference in the results.

What if you read that a test is 1-tailed or 2-tailed?  What is this – some sort of lab experiment gone wrong?  Actually, a 1-tailed test means that you decided ahead of time that you were only interested in whether group A was greater than group B (or less than, if that’s your hypothesis). Typically, you don’t know for certain the direction of the difference so you go for a 2-tailed test, which sets the bar a little higher.  Most of the time, it’s recommended (by those powers that be) to default to the 2-tailed. Oh, and in case you really need to know, there was no difference between the two hypothetical polls.  The sum of the chi squares was 2.104 with 1 degree of freedom, which translates to a 2-tailed p value of 0.147.  In other words, p was greater than .05 so do not pass Go, do not collect $200. In fact, the hardest thing about the chi square is that you need access to a Chi square value chart to determine your p value. Most introductory stats books have the table at the back; it’s also easy to find on the web. Of course, you don’t need to do the calculations by hand.  Below are some 2×2 chi square calculators on the web.  Again, I have no ties to them but they’re free and seem to be good. http://faculty.vassar.edu/lowry/odds2x2.html  This is the odds ratio calculator from Vassar that I gave you in an earlier blog.  As well as giving you the OR for a 2×2 table, it calculates the chi square or, if appropriate, the Fisher’s exact test.  Pretty nifty. http://www.graphpad.com/quickcalcs/Contingency1.cfm  This calculator from GraphPad Software enables you to select chi square with or without Yates’ correction, Fisher’s exact test, and between 1- and 2-tailed p values.

Take away message/bottom line The 2×2 chi square is a handy statistic for looking at whether there is a statistically significant difference in the proportions between two groups.

 

Are chi squares only for two groups?

In a previous post, I talked about chi squares when you had two groups and wanted to look at differences in two outcomes (Tea Party supporter vs. not Tea Party supporter).  This is what is referred to as a 2×2 table. But what if you have more than two outcomes?  What if your choices were something like Liberal, Conservative, NDP, Green or Totally Turned Off?  Can you still use the chi square? The answer is yes.  You can use the chi square to compare proportions between two groups when there are more than two variables.  Again, let’s start with a table.  In this table, you have two categories (c) and five rows (r).

Political affiliation	Hip urban professionals	Old fogies	Total
Liberal	80	90	170
Conservative	60	150	210
NDP	80	70	150
Green	100	70	170
Totally Turned Off	180	120	300
Totals:	500	500	1,000

You can still do this by hand if you love math but again I recommend you use one of the free online calculators. If you do that, you’ll find that the sum of the chi squares is 57.1 and the degrees of freedom=4. (If you’re interested, degrees of freedom = number of rows (r, in this case 5) minus 1. You need to know the degrees of freedom for looking up the p value of your chi square but if you’re using an online calculator, don’t fuss over it.)  With 4 degrees of freedom, the p value of 57.1 is less than .001. So you can say there is a statistically significant difference between hip urban professionals and old fogies in their political affiliations (p<.001). Ah, but here’s a key thing to remember. A chi square like this can tell you whether or not there is a difference between the two groups. But it can’t tell you which row is responsible for that difference. And now for online calculators I’ve stumbled across: http://www.physics.csbsju.edu/stats/contingency_NROW_NCOLUMN_form.html There’s a whole range of nifty calculators. http://faculty.vassar.edu/lowry/newcs.html Vassar also has a chi square calculator for rows by columns.  It also gives you some other stats that you may or may not be interested in (Cramer’s V and lambda).

Take-away message/bottom line The chi square statistics can also be used when looking at more than two groups. It will tell you whether there is a statistically significant difference between the two groups but can’t tell you which row or category is responsible for the difference.

 

You too can be a polling maven

An earlier post mentioned the fact that the margin of error reported for polls is typically an overall confidence interval and is seldom calculate for each, individual question in a poll. As a rule, if you have a reasonably-sized sample, the margin of error will change very little between different questions.  But I also advised that if something is really, really important to you, you might want to calculate its specific margin or error. So how do you do that?  It’s really not that hard.  First, you need to know the proportion (percent) you want to test.  Let’s say 61% of 600 people you polled said Justin Trudeau is a sweetie.  Make that 61% into .61 and you have your sample proportion.  From that you can calculate the anti-proportion: 1-p, which is 1.00-.61=.39. The other number you need to know is your sample size (n) or the number of people who answered that question. Finally, you need to decide what size margin of error you want.  If you want to have a 95% margin of error, the key number for you is 1.96.  If you want a 99% margin of error, you need to use the number 2.575.

Step	What you do	Our example
1	Multiple p x (1-p)	(.61)(.39) = .2379
2	Divide the result of step 1 by your number of respondents (n)	.2379/600 = .0003965 (some calculators may show it as 3.954-04)
3	Find the square root of the result of step 2	The square root of .0003965 = .0199
	To calculate the 95% margin of error, multiple the result of step 3 by 1.96.	.0199 x 1.95 = .039
	To translate this back into percentages, multiply by 100.	Your p (.61) is now 61%.Your margin of error (.039) is now 3.9%.  In other words, 61% of respondents thought Justin Trudeau was a sweetie, with a margin of error +3.9%, 19 times out of 20.
	To calculate the endpoints of your 95% margin of error, add and subtract 3.9% from 61%.	Range:a)    61% – 3.9% = 57.1% b)    61% + 3.9% = 64.9% In other words, the proportion of respondents who thought Justin Trudeau was a sweetie ranged from 57% to 65%.

As you can see, it’s not that hard.

Take-away message/bottom line It’s not hard to calculate your own polling margin of error.

 

Chemophobia – let’s get real!

A news release claiming that a link had been found between diet soda consumption and stroke got me thinking about the widespread existence of chemophobia – the irrational fear of “chemicals.” Part of the problem lies in the fact that there seems to be the assumption that naturally-occurring chemical are all good while manufactured chemicals are inherently bad. Let me give you a true example. Years ago, a friend of mine told me that she was thinking of going off The Pill because she didn’t like the thought of exposing her body to all those chemicals.  Except she was rolling a joint at the time. Okay. The Wikipedia entry states: “General chemophobia derives from incomplete knowledge of science, or a misunderstanding of science, and is a form of technophobia and fear of the unknown.” Chemophobia is rooted is both science illiteracy (not appreciating that all substances – natural or man-made – are composed of chemical) and mistrust of industry. In some cases, the mistrust is well-founded but it’s also fairly common for fears to be manipulated by the media or other self-interested parties. It’s important that we look at claims individually and carefully. A good place to start is testing a claim against the Bradford Hill’s criteria of causation. Hill argued that the minimal conditions needed to establish a causal relationship between two items were:

1. Temporal Relationship: If A is believed to cause disease B, then A must always precede the occurrence of B. Occurring at the same time suggests – but does not necessarily prove – correlation but not causation. If A occurs after B, then there’s clearly no causal relationship.
2. Strength: The stronger the association, the more likely the relationship between A and B is causal (this is where the odds ratio comes in handy).
3. Dose-Response Relationship: An increasing amount of exposure to factor A increases the risk of developing condition B.  This is also referred to as the biological gradient.
4. Consistency: Results are replicated in studies in different settings using different methods. This is an important but often over-looked criteria.
5. Plausibility: There should be some currently-accepted or reasonable biological process that explains the relationship.
6. Consideration of Alternate Explanations: Other possible explanations have been considered, tested and ruled out.
7. Experiment: By manipulating factor A in an experimental setting, you can prevent or ameliorate disease B. Also referred to as reversibility.
8. Specificity: The factor A is linked to a specific effect (disease B). This is arguably the weakest of all the criteria.
9. Coherence: The relationship should be compatible with existing theory and knowledge (e.g., is not based on something like UFO technology).

Wikipedia has an entry on the Bradford-Hill criteria, if you’d like to learn more.  As with all things in science, there’s some healthy debate about it.  But it’s a great place to start when trying to separate irrational chemophobia from rational concern.

Take-away message/bottom line First, all substances – whether natural or man-made – are composed of chemicals. Second, when assessing claims about the effects of chemicals on health, try applying the Bradford Hill criteria.  

 

We could all use a little quirkology

During my last trip to England, I picked up Quirkology, The Curious Science of Everyday Lives by Richard Wiseman (Pan Books, 2007).  Wiseman is described as a professor of Public Understanding of Psychology and is based at the University of Hertfordshire.  His book looks both at scientific investigations into offbeat subjects as well as investigations into urban myths and misperceptions.  You may know of Wiseman through his work on the LaughLab, the search for the funniest international joke.  In the book he admits that the more highly-rated jokes weren’t necessarily the funniest because the project had to be G-rated.  A lot of the funniest jokes were just too salacious to post publically. What gives Wiseman’s (little and easy-to-read) book weight is that – unlike the sensational use of research by the media – he is willing to admit that there can be multiple explanations for observed effects and sometimes both positive and negative research findings.  For example, one team of researcher found that men with initials that form a positive word (such as JOY, HUB or ACE) lived four or more years longer than those with initials forming a negative word (example being PIG, BUM or DIE).   (Makes me wonder if the idea for this research wasn’t cooked up after one too many cocktail at the faculty club.)  The effect was somewhat different with women: having initials that formed a positive word was associated with three more years of life but there was no effect for initials forming a negative word.  But the key is that Wiseman doesn’t stop there: he also describes the work of another team that, using what it felt were more sophisticated statistical techniques, failed to find any effect.  So is your lifespan dictated by your initials?  It’s not clear. It’s nice to have simple and clear conclusions and certainly this is what media reports of science and research often promote.  But in reality, one positive experiment doesn’t necessarily mean that something is “proven.”  Chrétien was wrong when he tried to infer that something was proven because there is some sort of “proof.”  Sometimes what we think is “proven” may have alternative explanations. Incidentally, Wiseman is a skilled magician and he uses this and his training in psychology to help explain how we are biologically set up to sometimes fool ourselves.  If you’ve got kids hooked on the silliness of Ghost Hunters or shows of that ilk, you may want to get them to read this book or Wiseman’s new book, Paranormality, Why We See What Isn’t There.

Take-away message/bottom-line Science can be full of interesting surprises and one experiment or study is typically insufficient to claim something is proven beyond any doubt.  Keep an open mind when reading research findings.

 

Who’s leading your focus group?

John Rossiter is an Australian professor in the Institute for Innovation in Business and Social Research and his latest book has the unwieldy title of Measurement for the Social Sciences, The C-OAR-SE Method and Why It Must Replace Psychometrics (Springer 2011).  Rossiter revels in the image of being an iconoclast but there’s also some good, common-sense ideas in this book, some gleaned from his years of doing market research in the US and Australia. Rossiter has been involved in more than a few projects involving focus groups.  Here are some point that he makes that may be helpful for those commissioning this type of research.

1. Sample ideas, not necessarily people.

Rossiter makes the point that you don’t need or want a random sample of the population when you form focus groups.  Rather, you want to sample the range of ideas relevant to your research topic.  If your research topic is a consumer product, then you want people who represent a range of ideas or attitudes: from average people to extremes such as heavy users, what he calls “averse non-users,” users of niche brands, and so forth.  The use of focus group “groupies” doesn’t bother him as long as they represent a range of ideas. So you don’t need to worry about getting a representative sample of the population as much as a purposive sample relevant to your product or service.  At the same time, he warns that research companies have very different lists or strategies for recruiting focus group participants. As a result, you shouldn’t expect different companies to produce similar results.

2. The person analyzing or leading the focus group can contribute up to half of the variance in the findings.

In quantitative research, if you give different people the same set of questions up to 90% of the variance in results are attributable to differences between respondents. In qualitative research such as focus groups, Rossiter figures that about 50% of the variance is explained by those who lead the focus group (and thus set the tone and determine what topics are and are not raised) and analyze participants’ responses. Rossiter identified the variables that affect the quality of a focus group analyst as:

• ability to understand verbal and nonverbal communication (i.e., emotional intelligence)
• knowledge of psychological concepts and processes
• knowledge of causal mechanisms in social science
• personal values
• mental and physical factors, such as his/her attention span, state of mind, physical health and other situational factors.

Take-away message/bottom-line When hiring a company to do focus groups, ask how it will ensure a purposive sample, the training and experience of the focus group facilitator, and the training, experience and capacities of whoever writes the final report.  What you hear is raw data but the report you read will be heavily influenced by the analyst(s).

 

What is the meaning of “100-year storm”?

Like most people, I’ve heard about the concept of the “100-year storm” or its variants. Several years ago, we had a rainstorm early in the spring when the ground was still frozen and the stream behind our house flooded. Several neighbours on our street ended up with inches of water in their basements.  Our basement remained dry but the pool in the backyard was reduced to a mud and branch-filled swamp.  It took the entire summer to clean it. What we experienced was described by some locals as a “60-year storm,” which raises the comforting expectation that it won’t happen again for another 60 years (at which point, I won’t care one way or another).  But does the weather know that?  And what about the 100-year storm that the city claims it is planning for – just when is it due?  In this part of Southern Ontario, Hurricane Hazel has been described as the last 100-year storm. Since that happened in 1954, is it safe to assume that the next one should be expected some time between 1954 and 2054? In his book, Numbers Rule Your World (McGraw Hill, 2010) Kaiser Fung explains that while it’s common to define a “100-year” storm as a storm of such power that it will appear only once every 100 years, it’s wrong.  Statisticians working in the insurance industry define a 100-year storm as a storm that produces damage greater than 99% of all storms. The US Geological Survey defines the 100-year part as referring to “the recurrence interval … based on the probability that the given event will be equaled or exceeded in any given year.”  In other words, the term refers to the probability of an event, rather than its predicted frequency. A 100-year storm means that the probability of a storm greater than the previous biggest storm is approximately 1 out of 100 for any given year. It is possible to have 100-year flood two years in a row – ask the people living in the Prairies. The USGS also points out that a 100-year storm doesn’t necessarily cause a 100-year flood, and that the baseline for calculating a 100-year storm or flood can change over time as data changes. As Fung points out, the term “100-year” storm is a misnomer and gives people a false sense of security.  Especially, I might add, in this era of climate change.

Take-way message/bottom line The term 100-year storm or flood refers to its probability of occurring in any one year and not its predicted frequency.      

 

Who are “health conscious” and how do we measure it?

One of my more recent reads was John Rossiter’s Measurement for the Social Sciences, The C-OAR-SE Method and Why It Must Replace Psychometrics (Springer 2011).  Rossiter is an Australian professor in the Institute for Innovation in Business and Social Research and likes to portray himself as a rebel with a cause. His book is worth a read if you do a lot of market research.  You may not agree with everything he says but you’ll have to admit that he raises some good points. What I liked about Rossiter’s book is his insistence that we be more rigorous about defining the concepts we’re talking about or measuring. For example, the concept of “service quality” is frequently a focus of marketing research.  But what exactly does that mean? When you’re trying to measure it (e.g., a customer survey), is it being defined by the customer or by the researcher?

Rossiter argues that if we are trying to measure people’s actual perceptions (what he refers to as concrete perceptual attributes, which is a mouthful) that we need is a concept (attribute) with one definition that the person can self-rate using one clear and specific question. In other words, “service quality” may be too wide a term to encompass what different consumers experience. Experts use multiple questions and measures but Rossiter argues that this means we’re not measuring actual customer experiences/perceptions but rather psychological concepts as defined by the expert. The attributes of “service quality” picked by the experts may not accurately reflect how customers experience service quality. This is interesting for me because I’ve started reading about the concept of “health consciousness.”  In health promotion and education, we frequently talk about “health conscious consumers” – those are the folks that we think come to health websites, buy vitamins or go for their annual checkups.  One article I read defined the health conscious consumer as individuals who lead a “wellness-oriented lifestyle” and are concerned with nutrition, fitness, stress and the environment. The questionnaire developed to measure this definition of “health conscious” includes questions on attitudes (towards the environment, exercise and health) as well as behaviours (information-seeking, nutrition, stress and exercise).  Although people score themselves, this is clearly a tool measuring “health conscious” as arbitrarily defined by the experts. But does this really reflect the health conscious consumer? We know, for example, that women are often the health information seekers not only for themselves but for others in their family. So is being “health conscious” slightly different for women than men? And wouldn’t the nature of being health conscious be different for someone who is healthy than for someone who has been diagnosed with a chronic condition?  Can someone who is afflicted with a chronic disease so they can’t exercise still be health conscious? In short, I have a lot of work cut out for me in understanding what is meant by this concept before I can use in it my research.  And this is the way it should be. Concepts or categories used in research should not be assumed or plucked out of the air but carefully thought through in detail. When reading marketing or polling results, look to see whether key concepts have been carefully developed, defined and measured.

Take-away message/bottom line It is important to distinguish between how people themselves would define or measure a concept such as “service quality” or “health consciousness” and how experts may define or measure it.

 

Is “learning style” an example of junk science?

OK, it’s a bit dorky but one of my favourite books has the unwieldy name of 50 Great Myths of Popular Psychology, Shattering Widespread Misconceptions About Human Behavior (Wiley-Blackwell, 2010).  The book is written by four established professors of psychology; four in the US (Scott Lilienfeld at Emory, Steven Lynn at the State University of New York, and John Ruscio, College of New Jersey) and one Canadian (the late Barry Beyerstein who was at Simon Fraser University). The point of this book is that when psychology is reduced to sound bites and self-help axioms, ideas can be misinterpreted, taken way out of context, or erroneously reified.  As a result, misinformation about psychology is common.  To illustrate their point, they describe 50 beliefs that are actually woefully lacking in scientific evidence. One of the 50 is the idea that students learn best when teaching style is matched to their “learning style.”  This was of particular interest to me because when I started my long-distance doctoral program, students were asked to fill out a questionnaire to identify their learning style.  So if a university is subscribing to this theory, it must be scientific? Well, not really, according to Lilienfeld et al.  Their criticism focus on the following points.

1. Despite being around for decades, there is no clear agreement on what learning style means.  One of the most popular models is based on preferred sensory modality (you are a visual, auditory, or kinesthetic/tactile learner).  Another divides people into activists, reflectors, theorists and pragmatists, while yet another uses the categories of converger, diverger, assimilator and accomodator.  OK, so what is it?
2. There’s no reliable and valid way to assess learning style.  Most of the questions lack any sort of context, whereas how you prefer to learn may be shaped by what you’re trying to learn. How you study a foreign language can be very different from how you learn a skill such as swimming or archery.
3. So far, there’s little or no evidence that matching instructors’ teaching style to students’ learning style is helpful or effective.  In fact, most of the studies that have been conducted have found no effect.
4. There’s little or no evidence that people can be trained to adapt teaching styles to students’ learning styles.

The authors end the discussion by pointing out that it’s limiting to infer that people only have one learning style.  In fact, evolution has given us multiple ways of learning and which ones we use will depend in large part upon what we’re trying to do. If you’d like to read more about this issue, Wikipedia has a nice entry on learning styles, including a description of the 2009 panel that reviewed the evidence for the Association for Psychological Science.  The panel concluded there were only a few studies that used an adequate methodology for studying learning styles.  Moreover, of those few that were methodologically sound, all but one had negative findings.  Not everyone is willing to give up on the learning styles theory, but I think that finding is pretty telling. 

Take-away message/bottom line There’s a lot of popular pseudo-science and because it lacks a consistent definition, verified means of measurement, and supportive scientific evidence, learning styles may be one example.

 

When health promotion claims are exaggerated: preventing death

Recently, I saw a claim that “80% of deaths due to heart disease can be prevented by changing behaviour.”  Yikes, that’s just so …. wrong. To be fair, I don’t think people or organizations that make such claims are being deliberately misleading. Claims like this are more a matter of fuzzy logic. Okay, let’s start about the claim that you can prevent death. Anyone who has ever worked with me knows my standard reaction to this sort of statement is: “You can’t prevent death.  Everyone has to die eventually.” It is possible to try and prevent premature death, typically defined as death prior to the average life expectancy, which is current about 81. Or premature death may be defined as death prior to age 75. Either way, you can see what you’re talking about isn’t preventing death but rather postponing it. The oldest verified human is Besse Cooper of the US who died at 114. A French woman, Jeanne Calment, is thought to have lived to 122. But eventually they died – death wasn’t prevented.

If you’re looking at preventing premature death, then the impact is less dramatic than overall figures might lead you to believe. Let’s take heart disease, just because it’s an area with which I’m familiar. In 2007, there were 50,499 deaths in Canada from heart disease (21.5% of all deaths). Seventy percent (35,324) of these deaths occurred in people 75 and over.  Even if you up the ante by increasing the age cut-off to 80, 57% of all heart disease (28,713/50,499) occurred among people aged 80 and over. In other words, in terms of postponing premature death, you’re actually dealing with at most 30% of all heart disease deaths: 15,175 if you use the age 75 cut-off.  Eighty percent of 15,175 is 12,140. So is what we’re saying is that we can postpone approximately 12,000 deaths per year? That would account for 13% of all premature deaths in Canada (12,140/92,343). So the best we may be able to do is postpone death from one cause. But when you do that, you’ve got to accept that you open up the door to death from other causes. The two more significant causes of death in Canada are cancer (69,595 in 2007 or 29.6% of all deaths) and heart disease (50,499 or 21.5%). If you’re 75 or over, chances are that you’re going to die from one or the other. For example, in 2007, 69,595 Canadians died from all forms of cancer (a sort of catch-all, as cancer is not one, homogeneous disease), of which 47% (32,467/69.595) were age 75 and over. The fact of the matter is that young people tend to die from accidents whereas older people tend to die from diseases traditionally referred to as “diseases of aging” (a term now out of favour in health promotion). Life can be a trade-off. The person who avoids a heart attack in their sixties or seventies may live to present with cancer, or vice versa. It is commonly accepted that prostate cancer tends to develop in men 50 and over but many (one estimate is about two-thirds) have a slow-growing form. An unknown proportion of these men never have symptoms, aren’t diagnosed or treated, and eventually die of other causes, such as heart disease. Even among those who are diagnosed and treated, a proportion will die from causes unrelated to their prostate cancer. As you can see, you’re not preventing death — you’re simply substituting one cause of death for another. Using messages about preventing death may be catchy but I think it misses the most important reason for healthy living: you’re going to feel a lot better.  People who maintain a healthy weight, exercise and eat a healthy diet have more energy, are less prone to depression and have a better chance of successfully managing any chronic conditions they may have (diabetes, arthritis, high blood pressure, etc.).  Rather than focus on a long-term, intangible and problematic benefit of “preventing death” I believe health promotion should focus more on the immediate, tangible and known impact of healthy living on health status and quality of life.  But that’s just me.

Take-away message/bottom line Take care when reading messages about “preventing death.”  Death is hard to avoid – the best you can do is postpone it or substitute one cause for another.

 

When health promotion claims are exaggerated: risk factor x causes y number of deaths

On WebMD you can find the following statements:

• about 20% of all deaths from heart disease in the US are related to smoking
• 35,000 nonsmokers will die from heart disease each year as a result of exposure to environmental tobacco smoke.

This example cites smoking stats but similar claims have been made for other risk factors.  For example, it has been estimated that 1.6% of all deaths worldwide are caused by high blood pressure and obesity causes 112,000 excess deaths from cardiovascular disease in the US.  I could go on and on. I’m not arguing that risk factors such as smoking, hypertension or obesity don’t increase the risk of various chronic diseases – they do. My point is that we need to keep in mind that these numbers are estimates and not actual counts. In reality, if you die of heart disease, stroke, cancer, respiratory disease, etc., your death certificate does not list whether you smoked, were overweight, had high blood pressure, etc.  Even if we had that information, how would you apportion responsibility?  Most people have multiple risk factors. So was Joe’s death 50% due to his smoking, 30% due to his diabetes and 20% due to his high cholesterol? How would you make that sort of call? The numbers of deaths cited in many reports are estimates calculated from the difference associated with the incidence of a diseases observed in populations with or without a specific risk factor. Typically, these incidence rates are drawn from large studies that counted the number of deaths among, for example, smokers vs. nonsmokers. The calculations are based upon the theory that the death rate among the unexposed (e.g., nonsmokers) is the baseline and the death rate among the exposed represented excessive deaths linked to this factor. Different studies can produce different estimates – which affects your attributable risk calculations. There is simply no one “true” set of numbers. In a 1998 article in the American Journal of Public Health, Rockhill, Newman and Weinberg discuss how population attributable fractions can be misused or misinterpreted.  For example, some studies have inappropriately added up single risk factor fractions in an attempt to derive the total proportion of disease risk that can be attributed to multiple risk factors (e.g., smoking, obesity and hypertension). But you really can’t do that – even though in reality risk factors typically cluster (e.g., the person who is overweight is more likely to have hypertension). Rockhill et al also object to the practice of equating the population attributable fraction with the proportion of disease cases “explained” by the risk factor, such as number of heart disease deaths “caused” or “explained” by smoking. The reality is that exposure does not automatically mean that you will get the disease in question.  Believe it or not, not everyone who smokes, is obese or is hypertensive develops heart disease. So stats like the 35,000 nonsmokers dying from second-hand smoke shouldn’t be taken as real deaths. They aren’t – they are illustrations to show us the potential scope of the issue. They are statistical tools to help guide health care and promotion.

Take-away message/bottom line If you read that risk factor x causes y number of deaths from a disease, be aware that these are statistical estimates based on existing incidence or mortality rates. They are tools to help us understand the impact of risk factors, not counts based on actual deaths.

 

Leadership – anti-lessons learned when taking aikido

This is a little off-topic for this blog, but it concerns something I’ve been thinking about lately.  Years ago when I was younger and much more flexible, I took aikido. I was never a star but for about four years I faithfully went twice a week to classes and once a week to practice.  So I was spending a fair bit of time on it. But although I enjoyed the physical part of the class, overall the experience was not very positive. Why was the experience so unsatisfactory? Like many situations in life, a lot of the responsibility lay with the person at the top: in this case the sensei or teacher but in other situations it may be your boss, supervisor, teacher or parent. What made this sensei such a poor leader? Let me count some of the ways. An ineffective leader:

1) Is not committed or invested in the organization:  In this case, the sensei didn’t run his own dojo but rather had a teaching contract with a recreation facility. So it really didn’t matter to him how many people did or didn’t sign up for his classes, retention rate, student satisfaction, etc.: he’d be paid the same no matter what happened. In the business world, it’s not unusual to find CEOs who come into a company in order to advance their own careers and consider the organization and position as a stepping stone to something better. Like the sensei they are not invested or committed to the organization and therefore to the people who work there. That doesn’t make for sensitive, inspiring or committed leadership.

2) Fails to motivate good people so they stick around:  One thing I noticed early on was the lack of senior students: there was only one black belt and no brown belts.  Most senseis have a cadre of black belts that have studied with them for years, as well as a steady supply of people coming up through the ranks. Senior belts are important, not only because they give dojos a cheap supply of extra teachers but because they provide aspirational models for new students. I worked in an organization that had a similar issue. Rather than promote from within, the CEO preferred to recruit from outside. As a result, there was little or no path for advancement for most of the people who worked there – and they knew it. The good people stuck around long enough to get what they wanted for their CVs and then left. Not a prescription for building a high-performance workforce.

3) Is resistant to recognizing individuals’ growth: The story I heard about this sensei is that his lone black belt was a brown belt when the two went to Japan to train at the school’s home dojo. At one point – much to the sensei’s surprise – the brown belt was tested and awarded his black belt. In other words, the teachers in Japan recognized that the student was ready for his black belt but his own teacher did not. I had a similar experience in one place where I worked. After eight years, it was obvious that the CEO still saw me as the person I was when I was first hired. He couldn’t see that over time I had changed, grown and matured or that I could continue to grow. And I wasn’t the only one in that organization with that problem. The possibility that people could grow — or be mentored to grow — was foreign to him. Guess what – this is the organization where there was no internal advancement and good people inevitably left.

4) Doesn’t believe in or bother with explanations:  My sensei was a man of few words who would demonstrate moves a few times but gave little or no verbal explanations. I was told he was emulating the traditional Japanese system in which students learn by watching. That may work in Japan but I’m not so certain it works here. In fact, a few months ago I stumbled across some YouTube videos in which an American sensei was explaining moves I supposedly “learned” years ago and I was amazed at what I hadn’t known. How many CEOs make the same mistake: keep the people who work for them, particularly those outside of the inner sanctum, in the dark as to where he is trying to head the company, why changes are being made, why policies are being revised, etc.? Without explanations, two things happen. First, people underperform because they don’t clearly understand what they are supposed to do. Second, it’s a fact of life that whenever there is an information vacuum, something will rush in to fill the void. In the case of my aikido class, students tried to help each other by giving what they thought were helpful tips or explanations. The problem is that some of this information was wrong or based on other martial arts (because some of the green belts had studied jujitsu, some of my aikido moves ended up being a sort of strange hybrid). In organizations, if leadership does not provide full and convincing explanations, you can be sure that rumours and speculation will quickly fill the void.

5) Uses mind games to exert control: In my class, I think belt testing was conducted once a year and the sensei picked who would be tested. One year, an eager white belt who wasn’t chosen with the rest of his cohort went to the sensei after class and asked why. Instead of explaining his decision or offering constructive criticism that the student could have used to improve his performance, he was told, “The fact that you ask that question shows you aren’t ready.” Yikes, talk about mind games. To add to the madness, although the sensei chose who was tested, it was common for at least one person to fail. It made the point that the sensei was in control but it made it confusing for the students. Unfortunately, mind games can be common in the corporate world. For several years I worked with a woman who had two favourite games. The first was creating crises, sometimes out of virtually nothing, that she would fan into panic situations. Of course, then she would paint herself as the only person who could save the organization from imminent and total diaster. Her second game was “That’s not what I told you to do.” This game was played by giving vague and ambiguous instructions to staff, who – because they were in the midst of a crisis, mind you – would rush out and try to do as she instructed. But inevitably, when they came back to her they would be told “That’s not what I told you to do.” It’s a wonderful way to drive staff to mass suicide. I should point out that the problems I had with aikido are not in any way representative of the discipline itself.  Aikido is considered one of the most cerebral of the martial arts and incorporates a rich and beautiful philosophy.  No, what I’m describing are the leadership limitations of a specific individual. Blame the individual, not the discipline.

Take-away message/bottom line No matter what field they are in, good leaders are people who are committed, motivate loyalty, recognize and support individuals’ growth, explain what they are doing and why so there are no information vacuums, and avoid using mind games.

 

Ten Tips for Increasing Self-Control

In the May/June (2011) edition of Scientific American Mind, Wilhelm Hofmann and Malte Friese talk about the science of self-control.  They focus on the conflict between impulses aimed at immediate gratification (aka cravings) and our reasoning of what we should do in order to achieve long-term objectives (aka convictions, such as to be healthy, lose weight, etc.).  When our convictions are high, we can be successful in resisting immediate temptations or cravings.  But when stress, emotional strain, or alcohol get in the way, our convictions can crumble and we give into temptation.  Hence, the proverbial “falling off the wagon” – whether it be alcohol or food – after a fight with a significant other or a bad day at work. In his book, Dieting, Overweight and Obesity, Self-Regulation in a Food-Rich Environment (American Psychological Association, 2008), Wolfgang Stroebe of Utrecht University applies this theory to the issue of weight.  He uses slightly different terminology, referring to goal conflict theory, but the basic premise is the same. His conclusions are a little pessimistic.  After reviewing the evidence he thinks that behavioural program may help people to improve their dietary choices or become more active but are unlikely to “cure” obesity.  Part of the problem may lie in the fact that the people who enroll in weight-loss programs tend to be the more severely obese – the very people who may be more resistant to treatment.  But he does give that, as behaviour change takes time, whereas any single attempt has a low probability of success, the cumulative probability will add up if you keep trying. Oh my, that sounds discouraging.  At least Hofmann and Friese give some advice on how to increase your self-control.  So for those of you struggling with behaviour change, here are their tips (somewhat rewritten) on increasing your level of self-control so you can resist temptation.

1. Educate yourself and keep reminding yourself of the risks and long-term negative consequences of the behaviour (e.g., overeating) that you’re trying to avoid.  Some people like to post a thin picture on the frig as a reminder of what they want to look like but I’ve always found a picture of a fat person more effective.
2. Increase your sense of personal engagement and accountability by telling friends about your goal.
3. Long-term goals can sometimes be abstract so shape them into intermediate steps or milestones that are more concrete and measurable.  This is very much in the vein of the SMART goal approach (SMART standing for specific, measurable, attainable, realistic and time-based).
4. Pat yourself on the back and take pleasuring in achieving partial successes and reaching your intermediate milestones.
5. Plan for problems by formulating “if..then” resolutions.  Think in terms of “if this happens [e.g., someone brings Timbits to the office] then I’ll [e.g., chew on some gum instead of eating them].”
6. Replace old bad habits with new good ones. This is a common tip, although I’ll admit that it’s easier said than done. Substituting carrot sticks for cupcakes sounds like a great idea but is only really feasible when your mental resources (i.e., your convictions) are up to the challenge.
7. Change your impulses or cravings by learning to associate the mere sight of temptations with negative stimuli. The classic example is to create a mental image of something you love, such as chocolate brownies, but to add cockroaches to the image.  The goal is to create a yuck factor.
8. Identify situations that pose a particular risk for you and avoid them as much as possible.  Some of these “high risk situations” are relatively easy to steer clear of, such as party hours in bars. Others, such as family dinners, are a lot harder to avoid..
9. Train your working memory. By this, the authors mean you need to train your brain so you don’t lose your focus on your behaviour change goal and can inhibit counter-productive behaviour.
10. Plan breaks and relaxation periods so you don’t run out of the mental resources needed to resist temptation. This is particularly important in our obesogenic environment, where temptations lurk around every corner.  In other words, try to avoid getting so stressed and worn out that your guard is down and your susceptible to cravings.

Take-away message/bottom line When trying to change, knowing what to do is not enough – people need strategies to bolster their self-control so they can successfully resist temptations or cravings.

 

Eat a salad and you eat alone – teens’ attitudes about healthy eating

We spend a lot of time and money on educating people – particularly youth – about what is healthy for them. And that’s okay, as education is a necessary first step in helping people lead healthier lives. For example, look at what education has achieved in changing the images of smoking (once advertised as cool, normal and even approaching healthy) and wearing seat belts (something only nerds with Volvos used to do).  Changing social attitudes was a huge first step in instigating change and that change would not have happened without a lot of education through social marketing. But education is not enough to change behaviour.  Again, look at the examples of smoking and seat belts.  In North America, you’d have to have lived under a rock for the past 30 years not to know that smoking is unhealthy for you.  But even so, a minority (about 18%) continues to smoke.  Even more frustrating for health promoters is the fact that youth — who have not only grown up in  society that has denormalized smoking but in most cases have been taught in schools that smoking is bad for you — continue to take up the habit.  And although seat belts are now standard in all cars and use is required by law, some people continue to resist.

Which brings me to the recent study by a group in the UK (Stead et al 2011)¹ on eating choices of adolescents aged 13-15.  By conducting focus groups, the researchers came to the rather startling conclusion that making healthy food choices is actually bad for one’s social health.  In a nutshell, although the kids knew what foods are healthy (no-one mistook potato chips as a healthy) and could recite the benefits of eating healthy foods (e.g., it allows you to keep fit and be more energetic), they also knew that showing up with a “healthy lunch” could land you in the category of social geek. Ironically, although being fat was social leprosy, being seen as trying to lose weight by making healthy food choices was equally detrimental.  Cool kids are those who can eat junk but not get fat. This study is not the only one to suggest that the social aspects of food and eating may play important roles in our current obesity crisis. For example, studies have shown that youth from minority ethnic backgrounds often choose to eat “western” foods when with their peers as part of their efforts to “blend in” and feel included.  Share your pizza or French fries with others and you’re part of the group; eat a salad, and you eat alone. As any anthropologist or sociologist will tell you, food is an important and powerful cultural and social symbol; what you eat defines what you are as a social being (remember Levi-Strauss’ The Raw and the Cooked?). So youth don’t make what nutritionists might define as unhealthy food choices because they are simply ignorant or can’t afford nutritionally better foods – but because of cultural barriers.  Healthy eating is associated with being marginalized, being a “health nut,” “goody-goody” or “fitness freak.” There is, in the words of Stead et al, a “cultural ‘non-ownership’ of healthy eating” that dooms most health promotion programs to failure.

Take-away message/bottom line Eating is another way in which youth try to establish and maintain social identities. Currently, making nutritionally healthy food choices may be bad for a youth’s social health.

(1)  Stead M, McDermott L, MacKintosh AM, Adamson A. Why healthy eating is bad for young people’s health: identify, belonging and food. Social Sci & Med 2011;72:1131-9

 

How reliable are news reports of research findings?

An increasing number of studies have shown what many previously suspected: news reports of research findings or medical developments are heavily dependent upon the information contained in media releases.  Some of these media releases are written by journal editors or university or hospital PR departments, but others may be written by pharmaceutical companies. In fact, if a journal article isn’t promoted by a media release, odds are that you’ll never hear it reported in the news. Scientific American guest blogger Hadas Shema argues that what we’re coming to is a reliance on “canned news reports” heavily influenced by PR departments – and by the forces that bias journal publishing.¹  For example, she cites a review in PLos ONE of 61 studies looking at the relationship between antidepressants and breast cancer. None of the 15 researchers who had industry ties reported a positive link between the drugs and breast cancer, compared with 43% of those without industry ties. In other words, there was a statistically significant (p=.001) relationship between industry ties and the conclusion the researcher came to regarding antidepressants and breast cancer.² Does this mean that research is hopelessly contaminated by financial considerations and we shouldn’t trust any news report we read, hear or watch on TV? Not necessarily. There are a number of reasons why research results can sometimes be suspect. Although there may be a few unethical apples in the barrel, in most cases researchers’ bias may be unintentional (as in, you find what you’re looking for). Or the problem may reflect the fact that it’s difficult to publish studies with negative findings (an issue science is trying to address). The good news is that you’re not alone in trying to decipher media reports.  Originating in Australia, a loose network of sites referred to as Media Doctor has been established in which teams of medical doctors and researchers review health stories and rate them. In Canada, the site can be found at www.mediadoctor.ca.  In it, for example, a CBC story suggesting the growth of prostate cancer tumours could be slowed by a drug was rated 2.5 out of a possible five stars.  As well, the article was rated as satisfactory for on four out of nine criteria and a short explanation and discussion was posted. The US version of the site is found at www.healthnewsreview.org.  Again, there’s an overall score for a news report, as well as more detailed analysis and discussion.  On the day I visited, an article in Time on two prostate cancer treatments rated only one out of five stars, while a story in the Wall Street Journal on heart disease screening got four stars.  Talk about a quick and easy way to separate the health news wheat from the chaff! The British version is called Behind the Headlines and is sponsored by the National Health Service to help the general public (www.nhs.uk/News/Pages/NewsIndex.aspx). It’s less concerned with scoring news reports than explaining them – and the health issues they represent – in a fair and scientifically-accurate manner.  It also posts special reports, such as Miracle Foods, Myths and the Media.

Take-away message/bottom line Too many news reports on medical developments are based largely, if not solely, on media releases distributed by interested parties. The Media Doctor network of websites can help by providing expert commentary on the quality of media reports.

¹ Shema H. Health Reporting and Its Sources. Scientific American online, May 31, 2011.  http://www.scientificamerican.com/blog/post.cfm?id=health-reporting-and-its-sources-2011-05-31&WT.mc_id=SA_CAT_HLTH_20110531

² Cosgrove L, et al. Antidepressants and breast and ovarian cancer risk: a review of the literature and researchers’ financial associations with industry. PLos One 2011; 6(4):e18210.doi:10.1371/journal.pone.0018210

 

Manipulating polling results: aided vs. unaided awareness

A recent media release by the Heart and Stroke Foundation reported “only 28% [of Canadian women] recognized high blood pressure as a risk factor” for stroke.  This sort of stat is also a nice illustration of an important issue when you’re doing public opinion polling: whether you use aided or unaided awareness. Aided awareness is when you give people a list and you ask them to tell you whether they recognize something. It could be the risk factors for stroke or names of politicians, TV shows or toothpaste brands. When people have a list to prompt them, rates are typically high. In fact, sometimes people will say “yes” to something they’ve actually never used or aren’t aware of – or even exists. In asking aided questions, you need to keep in mind that responses may be influenced by acquiescence bias, social desirability or the fuhgddaboutdit syndromes (see my April 13 post Getting better poll responses for more information on what these mean). When there’s no list or prompt (i.e., unaided awareness), rates of recall are usually much lower because you’re relying on people to pull the information from their memory. Unaided awareness is commonly referred to as “top of mind” because that’s essentially what you’re doing – you’re asking people to give you the first few answers that occur to them. In 1999, for example, I was the consultant for a poll that measured both aided and unaided awareness of stroke risk factors. Needless to say, you have to ask the unaided question first. When asked the unaided question (i.e., to spontaneously name the risk factors for stroke), about 30% gave high blood pressure as one of their five responses.  But when we gave them a list of risk factors, 85% agreed that high blood pressure was a risk factor. So what’s the truth? The danger with aided awareness is that it’s easy for people to just agree with everything on the list (although there was some discrimination in this particular survey, as not many people agreed with ringers such as aluminum cookware).On the other hand, the danger of unaided awareness is that it’s limited to what people think of first – rather than what they actually know. So if you want to fudge your poll, you can get the kind of results you want by choosing how you ask the question. You want a big number, such as 95% of Canadians agree that second-hand smoke is unhealthy for nonsmokers? Then measure aided awareness. You want a small number, such as only 5% of Ontarians realize that diabetes is a risk factor for stroke? Then measure unaided awareness. Unfortunately, when reading a news story that quotes polling results, we’re seldom told whether the stat is based on aided or unaided awareness. That’s a shame, as it could really help in interpreting the numbers

Take-away message/bottom line Polling questions that measure aided awareness typically produce higher numbers than those that asked people to spontaneously provide responses/information (unaided or top-of-mind awareness).

 

Why don’t heart attack survivors change their evil ways?

I’ve talked with a lot of cardiologists over the years for my work and a common theme is their frustration at getting patients to adopt and maintain healthier lifestyles.  You would think if someone has survived a heart attack or bypass surgery they would see the light and quit smoking, start exercising, eat healthy, etc.  But time and time again I’ve heard cardiologists complain that their patients don’t seem to listen or are incapable of staying on track.  Some start out with good intentions, mind you, but only a minority seem to be able to maintain it. A couple of articles I read recently may give us some clues as to what is going on. To begin with, let’s establish that heart attack survivors aren’t acting irrationally or are necessarily in denial: their behaviour is actually quite rational and sensible if you understand what they are thinking. Irish researchers Carol Condon and Geraldine McCarthy⁽¹⁾ found that during the early recovery period, patients were indeed motivated to make lifestyle changes.  They were willing to admit that lifestyle risk factors may have contributed to their event and were actually quite ambitious in their plans to make changes.  In fact, the authors felt people tended to be too ambitious and to take on too many changes at the same time. But patients’ enthusiasm flagged when they found that making healthier choices often clashed with social and family activities. What they wanted after their hospitalization was to “get back to normal” – but making lifestyle changes could make some of their old “normal” verboten. Hmmm, a conflict.

Additional insights can be gleaned from an earlier (1998) study by Rose Wiles.⁽²⁾ Rose conducted in-depth interviews with heart attack survivors in order to probe their understanding and interpretation of their disease and recovery.  She found that people didn’t see a heart attack as a form of chronic disease but rather as an acute event that was usually fatal.  Somewhat surprised (and grateful) at being alive, most eagerly accepted the line from doctors and nurses that a complete recovery was probable within a relatively short time period (about three months). But if you are “recovered” then any subsequent heart attack would be a separate, unrelated event, rather than part of an ongoing condition. So why this talk about prevention? After all, if you recover from an infectious disease, it’s over – you don’t need to worry about it any more. Because what people interpreted “mild” heart attacks (which they defined as anything that didn’t require surgery or result in death or severe disability) as something you recover from, they saw lifestyle changes as part of the recovery process. So although they could be quite good during those three or so months of recovery, there was no real motivation to sustain them for the purpose of prevention. Complicating the process is what I once heard referred to as the “Uncle Harry” syndrome.  Everyone knows at least one Uncle Harry – someone who does all the wrong things health-wise (overweight, drinks, smokes up a storm, eats like a pig, allergic to exercise) but lives to a ripe old age, hale and hearty almost to the end. Come to think of it, it should be referred to as the Winston Churchill syndrome.  At the same time, everyone also knows at least one “Aunt Harriet” – a health paragon of virtue who did everything “right” but was struck down by a heart attack early in life. Maybe that should be the James Fixx syndrome? So if good (health-habits-wise) people can die young and bad people live long and prosper, then what causes heart attack? For many people, the answer to the “why me?” question is fate. Unforeseen forces must the ones holding the trump cards; as a result, what you do may not in the end determine whether your heart attack number comes up. Which in some respects is true.  Odds are probabilities that apply to populations and though you can reduce your odds of disease by making lifestyle changes, there’s no guarantee what any individual’s outcome will be. Why is this type of research important?  It’s important because it helps us to understand why people are having trouble following the sort of health promotion advice routinely (and often briefly) handed out by physicians or other healthcare providers. In my mind, it’s not helpful if we just dismiss people as “difficult” or “stupid” because they fail to comply. What we need to do is to understand why they don’t comply. Understanding is the first step in developing approaches that better meet their needs.

Take-away message/bottom line Understanding how people interpret the relationship between lifestyle and heart attacks and the nature of heart attack may give us insights into why lifestyle changes are not maintained.

⁽¹⁾Condon C, McCarthy G. Lifestyle changes following acute myocardial infarction: patients perspectives. Eur J Cardiovas Nurs 2006;5:37-44

⁽²⁾Wiles R. Patients’ perceptions of their heart attack and recovery: the influence of epidemiological “evidence” and personal experience. Soc Sci Med 1998;46:1477-86

 

Making assumptions about depression in heart disease and seniors

It’s been estimated that about a third of patients with heart failure are depressed or at least have depressive symptoms. Of course, you could argue that it’s surprising the proportion isn’t greater: heart failure is a miserable disease with symptoms as bad as, and a mortality rate worse than, many forms of cancer. With this diagnosis and its associated poor quality of life, there are plenty of reasons to be depressed. But a small, qualitative study published in 2009 suggests that our assumption (you become depressed because you are dealing with heart failure) may be wrong.  In interviewing ten heart failure patients, Rebecca Dekker and her colleagues found that seven reported they were depressed before they were diagnosed.⁽¹⁾ In other words, disease followed the depression, not the other way around. The authors comment that this makes sense, in that we know depression is a risk factor for developing cardiovascular disease. Knowing what comes first has clinical implications. It suggests, for example, that people with depression might benefit from careful screening for heart disease so you can prevent or at least postpone the development of heart failure. As well, it might be helpful to screen for depression in newly-diagnosed heart failure patients, so you can treat the depression early rather than waiting until it worsens. I suspect heart failure may not be the only disease in which we’re making assumptions about depression. Research indicates, for example, that depression is more common among people with Parkinson’s disease, stroke, diabetes, lupus, multiple sclerosis, and dementia. For example, the Parkinson’ Disease Foundation in the US estimates that up to 60% of people with the disease experience mild or moderate depressive symptoms.  But are people getting depressed because:

• they’ve been told they have the disease?
• there’s some biological or neurological process associated with the disease that affects mood? (e.g., there is evidence that Parkinson’s causes chemical changes in the brain that may lead to depression)
• they’ve been prescribed medications that can cause depression (e.g., some blood pressure and beta-blocker medications, sleeping pills, or tranquilizers)?
• before their diagnosis, they were already depressed or at risk of depression?

How you treat depression could vary according to what causes it. Another common assumption about depression is that it’s something that naturally occurs as you get older. After all, the reasoning goes, as you enter your 70s, 80s and 90s, your independence and ability to do the activities of everyday life declines (e.g., you have to give up driving, you need help doing chores you used to do by yourself), you begin to lose loved ones and friends, you face new health challenges, and you may need to cope with financial issues (e.g., that fixed pension doesn’t go as far as it used to). But humans have an amazing ability to cope and depression is not a normal part of the aging process. The paradox is though we think that seniors have a lot of reasons to be depressed, the condition is seriously undertreated. I don’t know what drives that. Is it thinking that because it’s “normal” we shouldn’t interfere? That treatment won’t be effective in seniors? That’s it’s not worth the time and effort to treat it? It can sometimes be difficult to spot the signs of depression in seniors – even more difficult if you’re not even looking for them. Irritability may be dismissed as being “grumpy,” forgetfulness or lack of concentration as an early signs of dementia, changes in diet as driven by income or lack of motivation to cook, or changes in sleep patterns as the result of medications. Some of the signs of depression are similar to, and can be confused with, other medical conditions. The good news is that where depression is recognized, diagnosed and treated (by talk therapy, anti-depressant medications or a combination of both), the outcomes are typically good. This can make life a lot happier and healthier for the senior, his/her family and caregivers.

Take-away message/bottom line Too often, we make assumptions about what causes depression. It’s important to look at the actual causes of depression if we are to effectively treat this important medical condition.

⁽¹⁾  Dekker RL et al. Living with depressive symptoms: patients with heat failure. Am J Crit Care 2009;18(4):310-8

Depression. Parkinson’s Disease Foundation. http://www.pdf.org/en/depression_pd Seniors and Depression. Canadian Mental Health Association. http://www.ontario.cmha.ca/seniors.asp?cID=5800

Depression in Older Adults and the Elderly. Recognizing the Signs and Getting Help. http://www.helpguide.org/mental/depression_elderly.htm

 

Do medical centres exaggerate research findings? 

Press releases increase the chances that findings from a research study will be picked up by the media.  In fact, a 2008 study found that about a third of American health news stories rely solely or largely on press releases.⁽¹⁾ There’s a tendency for press releases from pharmaceutical companies and medical journal to overstate the important of the research findings and to downplay or ignore their limitations.^(2,3)  But what about news releases from academic health science or medical centres – institutions that are supposed to be bastions of objective research? A study by Steven Woloshin and colleagues suggests that academic research centres aren’t doing any better when it comes to accurately positioning the clinical significance of research findings.⁽⁴⁾  Analysis of a random selection of 200 press releases from ten American medical centres found:

• almost half (44%) were describing animal or laboratory research, of which three-quarters explicitly claimed the results were relevant to human health.  However, 90% lacked disclaimers or warnings about the dangers of extrapolating laboratory or animal results to people.
• 29% (or three out of ten) releases were rated as exaggerating the importance of the findings. Exaggeration was frequently found in quotes attributed to the researchers.
• a third (34%) of studies involving humans failed to quantify results (such as, for example, giving a risk ratio or an effect size) and a quarter (23%) didn’t tell how many subjects were involved (study size),

In summary, Woloshin et al concluded that 40% (four out of every ten) media releases reported on studies that have serious research limitations, such as small samples (less than 30 participants), no comparison or control group, surrogate or indirect measures of the primary outcome, or unpublished data.  Yet over half (58%) of these studies failed to include disclaimers or qualifiers cautioning the reader about study limitations. Ouch, these are pretty damming findings.  As someone who has been involved in writing media releases, I feel pretty confident in saying this doesn’t happen because evil people are trying to deceive people. Rather, the problem is that media releases on research have to compete to get the attention of reporters and editors. Most research doesn’t appear sexy or interesting to outsiders or the lay public. To get in the news and build profile for an institution or organization, publicists have to try and make the case that the research is new, innovative and a step forward for human health. So the study in mice is reported as though it will be of immediate benefit in the treatment of humans, or tentative findings are presented as solid evidence. And those quotes? Sometimes, the publicist will take what is said by a researcher (who is – quite understandably — trying to promote his/her research for a variety of research and career reasons) but often they are just written by the publicist. There are a couple of other things to keep in mind regarding media releases. First, sometimes qualifiers are present in release itself but are cut out by reporters or editors. In the need for brevity, disclaimers are often seen as unnecessary. Second, sometimes the biggest offender is not what is said in the story itself but its headline. The key here is that the headline is typically written by a copy editor, not the reporter. The headline is supposed to draw the attention of the reader, so it’s not too surprising if it distorts the science.

Take-away message/bottom line Even media releases from academic health sciences may exaggerate the importance or relevance of research findings. Readers need to learn how to read about research if they want to separate the wheat from the chaff.

⁽¹⁾      Schwitzer G. How do US journalists cover treatments, tests, products, and procedures? An evaluation of 500 stories. PLoS Med 2008;5:e95 http://www.plosmedicine.org/article/info:doi/10.1371/journal.pmed.0050095

⁽²⁾      Kuriya B et al. Quality of pharmaceutical industry press releases based on original research. PLoS ONE 2008;e2828 http://www.plosone.org/article/info:doi/10.1371/journal.pone.0002828 ⁽³⁾      Woloshin S, Schwatz LM. Press releases: translating research into news. JAMA 2002;287:2856-8

⁽⁴⁾      Woloshin S et al. Press releases by academic medical centers: not so academic? Ann Intern Med 2009;150:613-8

 

On depression and “thinking positive”

Psychiatric genetics is an emerging and important area of research.  As its name implies, it’s the study of how genetics contribute to psychiatric issues such as schizophrenia, autism, ADHD and depression.  Not that it’s a simple relationship. For example, studies on twins in Sweden have shown that genetics explain about 40% of a woman’s risk of depression and about 30% of a man’s. ¹That means even if you have the “depression” genes, you still have a 60% to 70% chance of not becoming depressed. That’s not so bad when you think about it. The important thing about this sort of research is that it helps us to see just how complex mental health issues are.  Like other diseases, such as heart problems, asthma and diabetes, both genetic and environmental factors may be involved. If you’re genetically susceptible, you may be more vulnerable to the effects of a bad environment. But nobody’s totally immune: even if you’re genetically resistant, a bad environment may still get you. It also illustrates how much mental health issues are real diseases. Thinking that people can just “get over” things such as depression, anxiety, or phobias makes as much sense as asking them to “get over” diabetes or heart disease. It just ain’t that simple. Somehow, you’ve got to get body chemistry, genes and environment supporting one another. And the sort of environment that’s helpful?  Well, that may vary by culture. For example, in studies of depression, some factors such as stressful life events and childhood abuse are pretty well universal. But in Europe and the US, a higher level of education is associated with a lower risk of depression, while in China and Japan, it’s the opposite: the more educated you are, the greater the risk of depression.¹ In today’s media environment, it’s easy to forget these factors. Books like Oprah’s favourite, The Secret, are based on the assumption that if you just want something badly enough and think positively enough, anything is possible. You’re depressed?  Think positively and the depression will go away. You’re sick? Think positively and you’ll get well. You’re poor?  Think positively and you’ll get rich.  As Jamie Cundy writes in a Psychology Today blog: While the idea of positive thinking is a good one, the concept is presented in a way that lays blame on the individual if these good things don’t happen.  Even worse it places the blame on the individual if BAD things happen. ²

Take-away message/bottom line New research is giving us insights into the complex relationship between nature and nurture when it comes to mental health.

¹Cyranoki, D. Chinese depression survey holds surprises. Nature. Published online August 1, 2011. http://www.nature.com/news/2011/110801/full/news.2011.450.html?WT.ec_id=NEWS-20110802

² Cundy J. What is the ‘Secret’ to happiness? The Beauty in the Beast blog. Psychology Today. Published June 6, 2011.  http://www.psychologytoday.com/blog/the-beauty-in-the-beast/201106/what-is-the-secret-happiness  

 

Do you know what your chiropractor believes?

Many, many decades ago, my father was sent to a chiropractor by his family doctor.  Well, “sent” is a little strong.  Actually, the family doctor said there wasn’t much he could do for my father’s back pain and so he might consider to go to the chiropractor in town, but to never tell anyone that he had heard those words spoken by a M.D. Times have certainly changed. Currently, chiropractors are itching to get into physician-led Family Health Teams. Here’s my disclaimer: I’ve been to chiropractors in my time and generally I tend to think they may be beneficial for back problems. Or at least they appear to be beneficial: the reality is that most back pain goes away within 6 to 8 weeks anyways. In fact, a 2011 review of the research evidence by the Cochrane Collaboration found spinal manipulation wasn’t any better than other common therapies, such as exercise or physiotherapy, in relieving back pain or improving function.¹ Perhaps the only good news is that it’s generally considered safe – provided you stay away from neck manipulations. What is scary about chiropractic is the vehemence of its supporters.

A friend of my husband, Paul Benedetti, collaborated with a friend several years ago to write Spin Doctors: The Chiropractic Industry Under Examination. The wave of criticism that descended on them wasn’t pretty. Part of the problem is that the chiropractic industry didn’t appreciate the fact that its rather dubious origins were discussed. The founder, D.D. Palmer, was an ex-pat Canadian who had practiced magnetic healing before receiving messages on the principles of chiropractic from a deceased “doctor” at spiritual camp meetings in Iowa ² Writing in 1914, he referred to chiropractic as a form of religion, although he later backed off from that track. Palmer’s theory was that misalignment of the vertebra interfered with the body’s functioning and innate ability to health itself. He called those misalignments vertebral subluxations. The problem is that subluxations don’t show up on x-rays. As described in Wikipedia, the regulatory body for chiropractors in the UK has stated that a vertebral subluxation “is not supported by any clinical research evidence that would allow claims to be made that it is the cause of disease.”³  Despite this, a 2003 survey of North American chiropractors found that 88% wanted to continue using the term. In contrast, when mainstream medicine uncovers new facts, it typically accepts the need to change. So we now accept that ulcers are caused by bacteria, malaria is a virus carried by mosquitoes and not by “bad air,” and most children don’t require tonsillectomies. Let’s be fair: it’s not a big deal that the original founder of chiropractic was what we would now consider a flake. What is now mainstream medicine used to believe in leaches and bleeding people. Over time, science has changed the practice of medicine radically and it will continue to change. Physicians have learned to lay bare what they do to the rigours of science and to stop doing things that are shown to be ineffective (granted, not all physicians are equally receptive to change). Chiropractic must learn to be just as open to science and scrutiny. Attacking anyone who asks questions is only a means of ensuring the practice never moves forward.

Take-away message/bottom line If chiropractic wants to be considered a medical science, it must walk the talk and truly embrace the scientific method – including being open to criticism.

¹Rubinstein SM, van Middelkoop M, Assendelft WJJ, de Boer MR, van Tulder MW. Spinal manipulative therapy for chronic low-back pain. Cochrane Database of Systematic Reviews 2011, Issue 2. Art. No.: CD008112. DOI: 10.1002/14651858.CD008112.pub2

²  Wikipedia. D.D. Palmer. http://en.wikipedia.org/wiki/Daniel_David_Palmer

³ Wikipedia. Chiropractic. http://en.wikipedia.org/wiki/Chiropractic

 

Motivation – time to accept the new paradigm?

A number of books are now out there showing that the idea that positive reinforcement (e.g., a bonus or commission) is actually a bad idea.  But unfortunately, as predicted in Thomas Kuhn’s The Structure of Scientific Revolutions, it’s been hard to get rid of the old paradigm.  Schools, businesses and families around the world continue to operate under the old and misguided concept that you can bribe or reward employees or kids into acting the way you want.  Well, you can in the short term but it’s unlikely you’ll be happy with the ultimate results. Why? Because what you end up with are individuals who lack internal (intrinsic) motivation and so are going through the motions in order to get the reward but apt to take short-cuts to get it. In his book Drive (Riverhead Books, 2009) Daniel H. Pink quotes research showing that rewards not only reduce intrinsic motivation – but are addictive. Like drugs, rewards give an initial buzz – but one that fades. A $5 reward is great the first time you get it, but after the thrill dies away you want another – and this time it has to be even bigger to give you the same sensation. Pink describes research using MRI technology in which the brain functioning of healthy volunteers was monitored during a game involving the prospect of either winning or losing money. When participants knew there was a chance of winning money, a part of the brain called the nucleus accumbens received a burst of dopamine. This is the same physiologic process that occurs in drug addiction. But as the dopamine dissipates, so does the feeling. Pink argues that the fact that rewards mimic the physiology of addiction is disturbing if what we really want are healthy, motivated employees or kids. Furthermore, he goes on to ask whether this process can perhaps explain why so many reward programs for health behaviours backfire. For example, a program that pays people to not smoke may, Pink suggests, simply replace one dopamine-fueled addictive behaviour (smoking) with another (reward seeking). That works in the short term but when the reward is no longer sufficient to trigger the dopamine release or is even stopped, then you’ve got an addict in search of a drug of choice. Which for many people means going back to smoking, while some may replace it with food. The data on using rewards for health behaviours appears to support Pink’s predictions. So far, studies have shown that rewards can trigger people to make changes like quitting smoking, exercising, or losing weight – but only in the short term.  When the rewards end, the relapse rate can be high.

Take-away message/bottom line It’s time to shift from the old paradigm of motivation based on rewards and punishments, to one rooted in the psychology of self-determination theory.

 

Nail in the coffin of eugenics?

The idea that people are born with innate abilities such as intelligence dates back to the ancient Greeks.  In western society, it really picked up steam in the 19^th century with the publication of Galton’s Hereditary Genius. According to psychologist Victoria Plaut and Hazel Markus¹, the idea spread widely through American academics and politics during the period between the 1890s to the 1920s.  Why then?  In part, it was motivated by waves of immigrants to the US from southern and eastern Europe. The good people of the establishment feared they were going to be swamped by hoards of messy, smelly people who, by Lord, didn’t even speak English! The basic premise of eugenics is summed up by Popeye: “I yam what I yam.” In other words, people are born with a certain amount of brain power and that’s all they’re ever going to have. The original intent of the Binet-Simon Intelligence Scale (which was actually commissioned by the French government) was to measure current performance so slow learners could be identified and given special help. But in 1916, a Stanford professor known to believe in eugenics adapted the Binet-Simon for Americans (the Stanford-Binet Intelligence Scale) and presented it as a measure of innate intelligence. The fact that intelligence was described as innate fed into the notion that intelligence (or even other human traits such as honesty or compassion) is genetic. Nature, as opposed to nurture.Of course, no actual genes had been identified but that was just a technicality according to believers.  And to be fair, it wasn’t until fairly recently that genetics had reached the level of development that it had the capacity to look at the genetics of complex characteristics. But genetics is beginning to reach that stage – and the news for eugenics is bad. A publication in the August issue of Molecular Psychiatry (which strangely enough is not on my “must-read” list) reports that a study of 3,500 adults 19-90 from the UK and Norway was unable to find any specific genes liked to higher cognitive abilities.² Then, using advanced statistical techniques looking at 500,000 SNPs (which stands for single nucleotide polymorphism or variation in a gene of a single nucleotide) they found that roughly half of the variation in intelligence between individuals could be attributed to underlying genes. But there were some important catches in this research: 1:  Genetic variation explained only about half of the variance in IQ. 2.  The results suggested that hundred or even thousands of genes may be involved, with each gene contributing its own little piece to the puzzle. 3.  The specific genes involved have not been identified. Probably the most important implication of this research is that it shows the eugenic vision (turned nightmare) of “breeding” for a specific type of person is ridiculously simplistic and just downright unfeasible. Far too many genes are involved to be able to select for intelligence. And even if you somehow puzzled your way through the right combination of hundreds or even thousands of genes, you’d still account for only about half of a person’s ultimate IQ. What accounts for the other half? Who knows? Its probably shaped by environment but “environment” is a pretty broad category, ranging from the type of foods you ate as a kid, air and water quality, parenting, childhood experiences, friends, schooling, extracurricular activities, health, personality, role models, motivation, etc., etc.  You get the point. Moreover, other research suggests that the ability of genes to predict intelligence is even smaller.  As reported in a post on the online The Scientist, another researcher thinks genes can predict only about 1% of variance in intelligence. ³ Eugenics has been responsible for a lot of suffering in this world, ranging from forced sterilizations of people with Down’s Syndrome or mental illness to the horrors of Nazi Germany.  Skin-heads, racists and neo-Nazis be advised: you don’t have a (scientific) leg to stand on.

Take-away message/bottom line New genetic research is establishing that eugenics is bunk and you can’t “selectively breed” for human intelligence.

¹  Pault VC, Markus HR. “The “Inside” Story. A Cultural-Historical Analysis of Being Smart and Motivated, American Style” IN Handbook of Competence and Motivation, Elliot AJ and Dweck CS, eds. London; Guildford Press; 2005. pp 457-488

² Davies, G., et. al, Genome-wide association studies establish that human intelligence is highly heritable and polygenic Molecular Psychiatry, epublished August 9, 2011 doi: 10.1038/mp.2011.85, 2011.

³ Ghose T. Heritability of intelligence. The Scientist August 9, 2011. http://the-scientist.com/2011/08/09/heritability-of-intelligence/

 

Off Topic: Of Mad Men, the FLDS and My Big Fat Gypsy Wedding

Not long ago, my friend, the fab media maven Elissa Freeman, wrote a guest blog concerning money and relations between the sexes (“Contemporary marriage unions would SHOCK Carol Brady and my mom” http://www.marriedmysugardaddy.com/contemporary-marriage-unions-would-shock-carol-brady-and-my-mom?utm_source=twitterfeed&utm_medium=twitter).  What I found interesting were some of the comments. One woman wrote about remembering her father telling her mother that she didn’t need a new winter coat, meaning forget it, babe, you’re not spending “my” money on that. I found that hard to comprehend. I came from what on the surface would appear to be the quintessential 50’s family: Dad was the wage earner, Mom was the homemaker, and we kids – well, we were told our job was to do well at school. But my father was also a reasonable man and my mother came from a long line of strong women. I can remember my mother being shocked and appalled when fellow ladies of the UCW (United Church Women, for those non-WASPs among you) said they had to ask their husbands for extra money for this or that. My mother operated on the assumption that it wasn’t solely my father’s money because she contributed equally to the relationship, albeit in a different way (such as keeping house, sewing our clothes, canning endless but economical jars of tomatoes, peaches, jam, and whatever else could be canned). So she felt – and obviously my father agreed – that it was their money.

In commenting on Elissa’s blog, I attributed my parents’ attitude to the fact that both came originally from family farms. On the traditional family farm, labour was clearly gender-specific but both sexes contributed to the success of the operation. In my experience, a farmer’s wife worked just as hard (just in a different way) as her husband and garnered equal respect in the community. I speculated that perhaps the attitude of women’s dependency on men that we identify with the 50’s (and thus with Mad Men) was a function of the growth of the suburbs. When you have TV dinners, well-stocked grocery stores and cleaning services, the value of the services performed by women (or, at least, the perception of their value) plummeted. Guess my mother missed that memo – she was probably too busy canning and sewing and stuff to pay attention. Of course, this is a vast over-simplification and applies only to mainstream culture. Look at the Fundamentalist Church of Jesus Christ of the Latter-Day Saints, better known as the FLDS, and its ilk. The conviction of Warren Jeffs has lifted the lid on a systematic and sick culture of de-valuing women. If it’s any consolation, the FLDS doesn’t limit its sickness to only women: Warren Jeffs has been quoted saying “the black race is the people through which the devil has always been able to bring evil unto the earth.”  (Jon Krakauer’s 2004 book, Under the Banner of Heaven, is an interesting read if you want to learn more. I don’t know about the accuracy of all of his writing – his book Into Thin Air is pretty controversial – but darn, the man can write.) Women unfortunate enough to be raised within the FLDS are so physically isolated from mainstream society that it’s not surprising that they grow up indoctrinated. Few are able to see alternatives and able toUKgraph get out. Power to those who make the leap! Which brings me to Irish Travellers, as depicted in My Big Fat Gypsy Wedding (apparently, most of the episodes show Irish Travellers rather than Roma Gypsies). Like the FLDS, Traveller culture as portrayed in the show appears to be appallingly backward in its view of women. In the episode I saw, a 15-year-old girl who left school at about 13 admits that she can’t read very well but argues that schooling is unnecessary since no Traveller girl will ever become a doctor or lawyer or even hold a job. Nope, the best a Traveller girl can look forward to is a big fancy wedding and then it’s all downhill. I think the interesting thing is the show’s illustration of the power of culture. Unlike women in the FLDS, who are physically isolated on compounds or communities like Bountiful, girls portrayed in MBFGW are surrounded by mainstream culture. For a while at least, they go to school with non-Travellers. Yet the power of socialization is such that the majority apparently buys into the notion that because they are female, they lack rights and value. Scary stuff. Of course, when it comes to the portrayal of women – and role models for youth — there’s a lot of scary stuff out there in reality TV land. Like Toddlers & Tiaras and anything involving a Kardashian – any Kardashian.

 

What is the Edmonton Obesity Staging System and why does it matter?

A fair bit of media buzz was generated by the recent publication of the Edmonton Obesity Staging System.¹ The EOSS was created in large part because of the limitations posed by the Body Mass Index (BMI) and even waist circumference. Both may be helpful in dividing large numbers of people into risk groups but are pretty poor at predicting health consequences for individuals. As I’ve discussed on this blog, the relationship between weight and health is complex. My particular complaint is that in health promotion messaging we tend to lump those who fall into the overweight category with those you are obese. Scientifically, I don’t think that’s justified, as I discussed in my April 21 post, Is being overweight a death sentence? If you look at the fine print, we’ve always known that BMI has a lot of limitations. It’s only supposed to be an indicator of health risk if you’re:

• Between 18 and 64 years of age
• Not pregnant
• Not highly muscular or athletic (football linebackers and Iron Man participants need not apply)

Moreover, as discussed in Wikipedia, the BMI doesn’t account for body frame.²  A large-framed (what we colloquially call “big-boned) individual could have a low body fat percentage but still be classified as “overweight.” Conversely, a small-framed individual could squeak by with a “normal” BMI but actually be carrying a heck of a lot of body fat (as opposed to lean body mass). Finally, the BMI doesn’t address the problem that you can lose height as you age. As you do, even if you don’t gain an ounce, your BMI will increase, which hardly seems fair. This is one of the reasons why it’s generally acknowledged that BMIs aren’t accurate indicators of risk for people age 65+. So I think the EOSS is a great idea. It is recognizes that health is more than just a number on a scale, that there’s such things as “healthy fat” and “unhealthy thin.” Just one thing you should be aware of, though – this system has yet to be tested. As the authors point out, this is a proposed system. The system has yet to be tested to see if it actually can predict health consequences. If you’d like to read more about the EOSS, Ayra Sharma runs a very good blog (http://www.drsharma.ca/). If you go there, you can also download/print off a chart outlining the EOSS stages.

Take-away message/bottom line A proposed (but yet tested) system (the Edmonton Obesity Staging System) has been developed that may address the limitations of the Body Mass Index (BMI) in predicting the health risk associated with obesity.

¹ Sharma AM, Kushner RF. A proposed clinical staging system for obesity. International Journal of Obesity (London) 2009;33:3:289-95

² Body Mass Index. Wikipedia http://en.wikipedia.org/wiki/Body_mass_index

 

Is it time to move beyond ‘fight club’ analogies?

In light of Jack Layton’s death from cancer, there was an op-ed article in the Globe and Mail about the problems and limitations of the common practice of using fight analogies when talking about disease.¹  The author, Carly Weeks, makes the point that equating illness with a war, battle or fight with an enemy diminishes our understanding of the challenges and complexities of living with a serious illness. By talking about “a battle against cancer” (or ALS or MS, or whatever), there’s the subtle inference that people who “lose” didn’t fight “hard enough.” People are artificially divided into “winners” – “survivors” being a common lexicon – or “losers.” She also quotes a British writer who has multiple myeloma (a cancer of the bone marrow) who feels the emphasis on the patient’s “bravery” and “courage” implies that if you die – if you don’t “conquer your cancer” – “there’s something wrong with you, some weakness or flaw.”  Christopher Hitchens, who has esophageal cancer, has also gone public with his dislike of the “battle” analogy. It’s a problem that we as a society need to recognize.  Sometimes, no matter what you do, you can’t stop a disease process.  A positive attitude is great and will undoubtedly support a better quality of life, but we shouldn’t act like patients are responsible for their own prognosis. They aren’t. They can do everything they are capable of to help themselves and to live as well and fully as possible, but the responsibility – and the blame – doesn’t lie with them. The use of the ‘fight club’ analogies extends far beyond cancer. We talk about people ‘battling’ mental illness, heart disease, MS, AIDS, etc., etc. But it’s not like people have a choice or there’s any sort of level playing field. It’s not a test of anyone’s courage or determination – it’s a disease. About 30 years ago, my father had a severe stroke that left him paralyzed on one side and aphasic. Around the same time, a TV movie was made about Patricia Neal’s stroke and recovery. I remember my mother hating that movie. She hated it because it inferred that Neal recovered because of her hard work and determination. It inferred that if you just worked hard enough, you too could make a full recovery. But the fact of the matter is that although my father worked as hard as possible at his rehabilitation (my mother attended some of his rehab sessions and was impressed and saddened by how hard he tried), his brain damage was so extensive that he could never hope to make a recovery like Neal’s. The “fight hard enough and you can triumph” scenario can be a cruel mirage. It’s interesting to see the comments on Weeks’ article on the Globe and Mail website. Overall, they are positive – particularly among those who have been touched by cancer. A small minority thinks it’s nitpicking, political correctness run amok or even offensive. Needless to say, I don’t agree with the minority. Language matters – it influences our interpretation of the world and thus shapes our reality.

Take-away message/bottom line Current common analogies or clichés can have unfortunate influences on our perceptions and expectations.

¹ Weeks, C. Jack Layton didn’t lose a fight: he died of cancer. The Globe and Mail. August 22, 2011.  http://www.theglobeandmail.com/life/health/new-health/conditions/cancer/jack-layton-didnt-lose-a-fight-he-died-of-cancer/article2137736/

 

Can we “end” the two primary causes of mortality – heart disease and cancer?

Recently, there’s been a spate of campaigns promising to “end” cancer, or at least certain types of cancer.  I’m not naming any names here. I mean, I understand why organizations are making these sorts of claims – they want and need some sort of dramatic claim to galvanize donors and volunteers. The more flamboyant the better, some PR firm probably told them. But the fact of the matter is that claims of this type are just – well, to be honest – silly. First of all, diseases such as cancer and heart disease have multiple causes and risk factors, including genetics, ethnicity, socioeconomic status and spontaneous gene mutations. In the case of coronary heart disease (just one form of heart disease), the consensus is that there is “no clear etiology” and numerous risk factors. ¹ How in the world can you control or prevent all possible causes of such diverse and complex diseases? Well, what about lifestyle changes to prevent diseases?  Wikipedia quotes a report by the World Cancer Research Fund that estimated that 38% of breast cancer cases in the US are preventable through lifestyle modification (reducing alcohol intake, increasing physical activity levels and maintaining a healthy weight).²  That’s good but leaves 62% up for grabs. The fact of the matter is that breast cancer is, to quote Wikipedia, “strongly related to age with only 5% of all breast cancers occurring in women under 40 years.” ² And heart diseases? As the World Heart Federation puts it, “Simply getting old is a risk factor for cardiovascular disease; risk of stroke doubles every decade after age 55.”³ So unless we’re going to resort to some sort of Logan’s Run future, I think we’re stuck with a world with both cancer and cardiovascular disease as the major causes of mortality. The only way it could change would be if we were to have huge epidemics so the major causes of death would become (once again) communicable diseases. But is that an alternative anyone would like to see? Given this PR-driven hyperbole, it was refreshing to see the announcement of the US government’s Million Heart initiative. ⁴ The campaign is based on the calculation that more than 2 million American die of heart attack or stroke each year, and the goal is to prevent 1 million of these deaths over the next five years. It’s not claiming to be able to wipe out heart disease or stroke – just to prevent 250,000 deaths per year.  So it’s reasonable but the 1 million figure still gives it PR flash.

Take-away message/bottom line Unless we want to go back to the good ol’ days of epidemics of typhoid, scarlet fever, smallpox and bubonic plaque, complex diseases that increase with age, such as most forms of cardiovascular disease and cancer, are probably going to stay with the human race.

¹  Coronary disease.. Wikipedia. http://en.wikipedia.org/wiki/Coronary_heart_disease

²  Breast cancer. Wikipedia. http://en.wikipedia.org/wiki/Breast_cancer

³ Cardiovascular disease risk factors. World Heart Federation. http://www.world-heart-federation.org/cardiovascular-health/cardiovascular-disease-risk-factors/

⁴ Frieden TR, Berwick DM. The “Million Hearts” initiative – preventing heart attacks and strokes. NEJM September 13, 2011. http://www.nejm.org/doi/full/10.1056/NEJMp1110421?query=featured_home

 

As Kenny Rogers said, you’ve got to know when to hold ’em, know when to fold ‘em

I’ve been writing a fair bit about motivation and different strategies that have been proposed for helping people stay on track. But there’s another side to the issue that is often overlooked: knowing when to give up. As Kenny Rogers sang, you’ve got to know when to hold ‘em and keep trying but you also have to know when it’s a lost cause and you should disengage. As described by psychologist Charles Carver and Michel Sheier,¹ there’s no simple, linear relationship between goals and effort. Intentions can tell us where to direct our efforts (whether we want to approach a positive goal or avoid a negative goal) but emotions shape the intensity of our motivation. Those emotions can be influenced by our basic personality (whether we’re an optimist or pessimist), past experiences (perceived self-efficacy) and how things are going. If it’s an approach goal, such as losing weight, then when we see we’re making progress we feel happy, eager and excited. But when things start to go bad, we may experience anger. That anger may help us stay on track and maybe even intensify our efforts. But too much failure and we’re likely to end up feel depressed, sad or beaten. Those feelings are associated with giving up.

In our society, we tend to think of giving up as a bad thing. But there are many situations in which giving up is actually a good and healthy thing. Not every kid in Timbits Hockey has the natural ability to make it to the NHL, just as not every woman can look like Angelina Jolie. We’ve all known people who have pursued goals to the point of damaging their families (e.g., the guy who persists in an unrealistic dream of becoming a golf pro) or well-being (e.g., someone addicted to booze, alcohol, gambling or sex). Sometimes, walking away is actually the smart and sane thing to do. What shapes our ability to let go? Carver and Sheier point out that goals such as perfecting your golf game are often inputs into how you define yourself (e.g., as being athletic or successful). The key to disengagement without despair is finding some other way to support that definition or vision of yourself.  Sometimes, this can be accomplishing by scaling back from a lofty goal to something more realistic (e.g., abandoning dreams of competing on the PGA and focusing instead on winning local amateur tournaments). Other times, it may involve shifting from one activity to another. That could mean shifting your attention to something you already do (e.g., from golf to curling); other times it may mean stepping outside of your existing framework and developing new goals. We live in a society that glorifies stories of success through perseverance and vilifies giving up. It’s especially acute when it comes to behaviours that we see as socially unacceptable, such as addictions, being overweight or, for the Tea Party types, poverty. “Try, try again,” we admonish, “Never say never,” “Quitters are never winners and winners never quit,” and so on and so on. But not only is it important to “know when to fold ‘em” but to “quit when you’re ahead,” and to “stop beating a dead horse.”

Take-away message/bottom line To live a healthy life, we need to know when to persist at something and when to disengage and direct our energies elsewhere.

¹ Carver CS, Scheier MF. Engagement, Disengagement, Coping, and Catastrophe. IN Handbook of Competence and Motivation, Elliot AJ, Dweck CS, eds. 2005; New York: Guilford Press. pp 527-547

 

The lay epidemiology of heart disease or why bad things happen to good people

Recently, I was reading some interesting articles on “lay epidemiology” – how the general public understands the incidence of disease. ^{1 2}  In this case, the disease was heart disease, specifically cardiac deaths.  As the authors point out, the health community has spent a lot of time, money and effort educating people that cardiovascular mortality is determined by modifiable behaviours such as smoking, inactivity, and poor diets.  And to a large extent, the education has been effective.  Ask people and they can readily recite the modifiable risk factors for cardiovascular disease. And yet, there’s still a gap between knowledge and actual behaviour. So what gives?  Are people in denial, foolhardy, or just plain stupid? As it turns out, when you sit down and listen to people, it emerges that they are actually quite rational. They understand the idea of behavioural risk factors and can easily identify people who fall into the “high risk” category (typically, those who are overweight, out of shape, or highly stressed). But their own experiences have shown them that risk factors aren’t the only forces at work. Almost everyone knows of, or knows someone who knows, the seemingly “healthy” person who did “everything right” and yet died at a young age from heart disease. They also know lots of people who have done “everything wrong” and yet lived to a ripe old age. The only way of reconciling the concept of risk factors with the reality of what they’ve seen is to add another concept: that of luck, fate or destiny. In other words, although people may increase or decrease their risk by their behaviours, their ability to change their fate is limited. If “your number is up” then even the best of behaviours won’t save you. The tenants of prevention are fallible; they can’t explain the “unwarranted longevity” of people who routinely make unhealthy choices or the “unwarranted deaths” of those who make healthy ones. Risks work only up to a point, and then random fate takes over. The lay epidemiology of heart disease is dichotomous. As described by Davison et al, this dichotomous lay epidemiology is in many respects a reflection of our health promotion strategies. The essence of the population approach to heart disease prevention is to reduce the risk of as many people as possible, if not all people. But risk is determined by identifying characteristics (risk factors) that can distinguish between low and high risk of a disease. Risk is a concept rooted in the group and not the individual. Only a relatively small proportion of people fall into the highest risk groups; if everyone fell into the high risk group, the factor would not be helpful in determining risk.  As a result, the majority of people fall into the medium or even low risk groups. In the case of heart disease, the outcome is that most fatal heart attacks happen to people outside the high risk group. Wow, you say, how does that happen? Let’s work through a hypothetical situation and I’ll keep it as simple as possible.  You screen 100 people and 25 fall into your high-risk category of having high blood pressure. Of the high risk group, 60% have a heart attack in the following five years (an unrealistically high rate, but this is just an illustration). That works out to about 15 deaths. Of the 75 people in your medium- to low-risk group (i.e., those without high blood pressure), the death rate is only a measly 25% — less than half that of the high risk group. But 25% of 75 works out to roughly 19 people, 4 more than in the high risk group. Can you now see the problem? In a classic article published in 1981, Gregory Rose called this the prevention paradox. You ask people to change their behaviour to theoretically reduce their heart health risk but many of them wouldn’t have had a heart attack anyway (or at least a premature heart attack). You can’t predict which individuals may benefit from making changes. So you ask entire population to change their evil ways even though you can’t predict who will benefit and who won’t. Given this reality, it’s not surprising that lay epidemiology views death as a bit of a crap shoot. In many respects, it is.

Take-away message/bottom line The general public is not being irrational when its explanation of cardiovascular death includes an element of luck or fate; the prevention paradox means it’s impossible to tell who may benefit from health promotion interventions.

¹  Davison C, Smith GD, Frankel S. Lay epidemiology and the prevention paradox: the implications of coronary candidacy for health education. Sociology of Health & Illness. 1991;13(1):1-19.

²Davison C, Frankel S, Smith GD. The limits of lifestyle: re-assessing ‘fatalism’ in the popular culture of illness prevention. Soc Sci Med. 1992;34(6):675-85.

 

Women and heart disease – time to stop blaming menopause?

For a long time, it’s been argued, women’s risk of heart disease is relatively low until menopause, at which point it begins to increase dramatically. However, new analysis is challenging this paradigm. In an article published in BMJ (British Medical Journal),¹ researchers at John Hopkins and the University of Alabama looked at ischemic heart disease mortality data from England, Wales and the US. They concluded that in women, heart disease mortality increased exponentially with age – period. They found no evidence of any acceleration or jump in mortality at menopause. Among men, risk increased rapidly during young adulthood and then slowed down somewhat. But at no point does the mortality rate among women match that of men; even after menopause mortality rates for men are much larger than they are for women. The authors compared the pattern of mortality of three different birth cohorts: people born between 1914-25, 1926-35 and 1936-45. Presumably, factors that could impact on CVD mortality, such as medical treatment and lifestyle and environmental risk factors (including smoking rates, prenatal nutrition, diet, etc.) would vary between the three cohorts. And yet, the same mortality patterns emerged across all three. That’s pretty convincing. The authors conclude that the early and rapid acceleration in male heart disease mortality may have misled people to thinking that women’s heart disease risk was linked to the effects of menopause. In other words, because acceleration of risk of women occurs later than it does in men, it happens to coincide with the time that most women become menopausal. As a result, we assumed the two were linked. You’re not alone if you’ve heard that “after menopause” there are increased rates in women of many CVD risk factors, such as high blood pressure, blood glucose, and high cholesterol. But in actual fact, there is some evidence that at least some of these risk factors may increase because of age rather than menopause. For others, such as cholesterol, the relationships between age, lipids and menopause appears to be complex.² Why is this important? For one thing, the authors of the BMJ article point out that it means we’ve been sending out a misguided health promotion message. We’ve been saying that women pretty well don’t have to worry until after menopause, whereas what we should be focusing upon is lifetime risk. Second, I think anything that helps to relieve the stigma of menopause is a good thing. Right now, women are viewing menopause as something akin to the Black Death. It’s not the end of good health, youth or sexuality – it’s just a part of the natural aging process. In case you’re interested, the authors of the BMJ article speculate that the root cause of the patterns they observed in CVD mortality involves telomeres. A telomere is like that plastic tip on the end of your shoelaces, except it’s found at the ends of a chromosome. Its job is to protect the end of the chromosome from deteriorating or from fusing with neighbouring chromosomes. Each time a chromosome reproduces itself, the telomere is shortened a bit. Over time, the telomere becomes too short to do its job and the chromosome loses its ability to reproduce. The good news is that this is a mechanism that helps to prevent cancer (which is a form of uncontrolled cell reproduction). The bad news is that when cells stop reproducing, we start to age. In the case of heart disease, what we may be seeing is the result of cumulative insults to the blood vessels. Some of these insults may be from things we do (e.g., smoking), whereas others may be genetic or just part of the natural physiology of the body (e.g., turbulent blood flow at places where the blood vessels turn or twist). When we’re young, the body’s natural defenses are able to respond to these insults and repair any damage. But over time, as the telomeres wear down, the body loses its ability to respond and vascular injuries and damage begin to take their toll. According to the authors, there’s evidence that the telomere loss per age-year is slightly (but not significantly) greater in men than women but age-adjusted telomere length substantially shorter in men. This difference, they say, is equivalent to the telomere-shortening process occurring 7.6 age-years sooner in men compared to women. In other words, men get the short end of the stick, telomere-speaking. As a result, they develop heart disease earlier than women. Warning, Will Robinson! Just because there’s a correlation between the time lines of telomere shortening and CVD mortality, doesn’t mean it’s been proven that one causes the other. The authors only analyzed mortality data: they didn’t measure any telomeres. So their explanation that telomere length, rather than menopause, is the reason for the mortality curve they found in women is a hypothesis, not a fact.

Take-away message/bottom line New research suggests that age, not menopause, is responsible for the increase in heart disease witnessed in women age 55+.

¹  Vaidya D et al. Ageing, menopause, and ischaemic heart disease mortality in England, Wales, and the United States: modeling study of national mortality data. BMJ 2001;343.d5170

² Bittner V. Menopause, age, and cardiovascular risk, a complex relationship. J Am Coll Cardiol 2009;54:2374-5

 

Recognize yourself – self-handicapping, beneffectance or defensive pessimism?

When in school, did you ever know someone who went out and got hammered the night before a big exam? Someone who blows a job interview by making unreasonable demands, telling an off-colour joke or wearing the wrong clothes? According to Frederick Rhodewalt and Kathleen Vohs,¹ these are examples of self-handicapping, just one of a number of defensive mechanisms people use to protect their image of themselves. As the authors put it, “People’s self-protective cognitive gymnastics can be quite remarkable. They have an uncanny ability to take self-relevant but threatening information about the self (or the anticipation thereof) and turn it into something more benign.” In other words, most of us are pretty good at twisting our interpretation of reality in order to see ourselves in a more flattering light. Self-protection, let me count the ways (according to Rhodewalt and Vohs):

• Self-handicapping: creating or claiming there are obstacles that prevent you from performing well, which helps to discount the idea that you lack ability (“It’s not that I’m stupid – it’s just that I was too hung over to concentrate”)
• Beneffectance: referred to by the authors as part of the “duct tape” in the defensive strategy toolbox (at the time this was published, Voh was at UBC, so hence the duct tape analogy?), it refers to the tendency of people to view their intentions and actions as good and to attribute bad outcomes to the effect of external circumstances (e.g., “I’m a good golfer but my caddy sneezed when I wound up for my shot, causing the ball to verve to the left”)
• Putting down others: there doesn’t seem to be a scientific name for this, but we’ve all seen it in action at one time or another: the person who when threatened, tries to puff him/herself up by denigrating or putting down someone else. Also known to most people as “the shit rolls downhill” syndrome.
• Defensive pessimism: people who even though they have a history of achievement, expect nothing but failure in the future (e.g., “Well, I’ll try to manage the project but you should keep your expectations low because you know how inexperienced I am”)
• Rejection-sensitivity: the person who is chronically anxious and expects to be rejected by their significant other – to the extent that he/she often acts in a way (withdrawing, being suspicious, or being clingy) that creates a self-fulfilling prophecy.

All of these defensive strategies are common and in some circumstances they’re helpful by propping up our self-image as competent and likeable, despite life’s injustices and cruelties. But when used too frequently or too intensely, they can also become harmful crutches. The key may be self-awareness: are you using defensive strategies so frequently that you’re not listening to what life is trying to tell you? These defensive strategies remind me of the physiologic process of stroke. Years ago, Dr. Antoine Hakim (a great neurologist and general nice guy) explained to me that when the blood supply to the brain is interrupted, chemicals are released that are designed to provide short-term protection for the brain cells. If blood flow is restored quickly, the strategy works and damage is minimized. But if blood flow is not restored quickly, the chemicals released can damage other, nearby cells, creating a cascade of destruction. I think defensive psychological processes are similar: they work in the short term but you have to be careful: over the long term, they could be harmful.

Take-away message/bottom line When we face failure or feel psychologically attacked, we resort to self-protective strategies such as self-handicapping, beneffectance, bullying, defensive pessimism or reactive rejection-sensitivity. The key is to avoid using short-term strategies as long-term crutches.

¹Rhodewalt F, Vohs KD. Defensive strategies, motivation and the self, a self-regulatory process view. IN Handbook of Competence and Motivation, Elliot AJ, Dweck CS, eds. 2005; New York: Guilford Press. Pp 548-565

 

It’s not what you think – when health stats are twisted

Years ago, I worked for a health charity that sent out fundraising letters with a message on the envelope proclaiming “One out of every two people who open this envelope with die from [insert disease]!”  The folklore was that it even became a joke on late-night US television, as in “who would want to open the envelope if this is what will happen?”   At the time, I tried to get through to the TPTB at the organization that this was a distortion of the facts. At that time, about 48% of Canadian deaths had the disease in question listed on the death certificate. That means that almost half of Canadians who died in any one year were considered to have died from that disease. That doesn’t mean that half of the entire Canadian population of that time would die from the disease – there’s a difference between all Canadians and those Canadians who die in any one year.   Two decades later and the organization in question is still making this sort of claim (although the numbers have shifted). And I still see it as inaccurate and a personal pet peeve.   The problem with the claim is that it confuses two different populations: all Canadians and that sub-set of the population that dies in any one year. There are not the same, primarily because not all people have the same risk of dying in any one year. Children have a much lower risk of dying than adults and men typically have a higher risk than women. The following chart is taken from an article published by Bandolier (a really great site) and shows annual risk of death by age and gender in the UK. The pattern is similar in most western countries. ¹

So you can see the lovely J-shaped curve and the epidemiologic issue. In any one year, a male between the ages of 25-34 has a risk of 1 in 1,215 of dying, while a female in that age group has a risk of 1 in 2,488. In comparison, a male between the ages of 75 to 84 has a 1 in 15 risk of dying and a female 1 in 21. Big difference. So the people dying in any one year are not representative of the general population: they are a sub-set of the population that is strongly skewed to the older age groups. The other thing to consider is cause of death. Let’s look here at the top three causes of death for Canadians at different ages: ² Under 1 years of age: congenital abnormalities, short gestation, maternal complications of pregnancy

• 1 to 14 years: accidents, cancer, congenital abnormalities
• 15 to 24 years years: accidents, suicide, and a tie for third place between cancer and homicide
• 25 to 34 years: accidents, suicide, cancer
• 35 to 44 years: cancer, accidents, suicide
• 45-54 years: cancer, heart disease, accidents
• 55 to 64 years: cancer, heart disease, accidents
• 65 to 74 ears: cancer, heart disease, chronic lower respiratory disease
• 75 to 84 years: cancer, heart disease, stroke
• 85 and over: heart disease, cancer, stroke

The point of this is that younger people have a much lower risk of dying and, more importantly, when they do die, they typically die from different things than older people. Again, a stat developed by looking at deaths (“1 in 3 deaths are due to x”) is not transferrable to the general population (“1 in 3 Canadians will die from x”). Of course, what we’re actually talking about here is the gap between statistics and marketing. For marketing and fundraising purposes, health charities and other organizations need simple messages that are supposed to be strong and – let’s face it – scary. A message like “1 in 3 Canadians who die will die from our disease” is considered too complicated and obtuse for mass communication. I get that – I don’t like it, but I get it.

Take-away message/bottom line In trying to create marketing messages, health statistics can sometimes be mis-interpreted. The risk of dying from any one cause differs widely among the general population by age and gender.

¹ Risk of death by age and sex. Bandolier “Evidence based thinking about health care” http://www.medicine.ox.ac.uk/bandolier/booth/Risk/dyingage.html

² Statistics Canada. Leading Causes of Death in Canada, 2007. Table 3 Ranking and number of deaths for the 10 leading causes by age group, Canada, 2007.  http://www.statcan.gc.ca/pub/84-215-x/2010001/table-tableau/tbl003-eng.htm

 

Health messaging & the media – walking a thin line

Trying to communicate health messaging through the mass media is a balancing act. You need to find some sort of acceptable compromise between the epidemiologic facts and stats (which tend to be complicated, murky and often not very exciting) and the sort of short, sensational sound bites needed for reporters and marketing types. It’s a thin line and sometimes organizations struggle. Take the case of the Imperial Cancer Research Fund in the UK. In 2000, the charity ran a major advertising campaign that showed three little girls sitting on a wall. Above each head was a label: lawyer, teacher, cancer. But as a critic pointed out, while it’s true that “one in three people will get cancer if they live long enough….it does not mean that one person in three dies of the disease.”¹ And this made it to the august pages of BMJ (British Medical Journal). ¹ Ouch! The issue is that there’s a big difference between the lifetime risk of developing a disease and the risk of dying from it.  In the US, for example, a woman’s risk of developing cancer is 1 in 3 – as suggested by the UK campaign – but her risk of dying from cancer is 1 in 5.² Likewise, a man’s risk of developing cancer is 1 in 2 but his risk of dying from cancer is actually 1 in 4. The BMJ article also criticized a story line in a popular soap opera in which a 32-year-old woman is diagnosed with breast cancer and undergoes a mastectomy. According to the BMJ, the story line distorts the reality that “such young women have a 100^th the risk of beast cancer of older women, that multifocal breast cancer is found in only 5% of cases, and that mastectomy is seldom performed.”¹ Which brings me to the oft-quoted and scary stat that a woman’s risk of developing breast cancer is one in eight (US) or one in nine (Canada). As Sarah Boseley writes in The Guardian, most people hear that stat and immediately start to apply it to any group of women they see, whether it be girls playing on a schoolyard, a class of university students, or a gathering of young mothers.  But in reality

… the picture is not as bleak as the headlines might suggest. For a start, it is not the little girl in primary school who has a one-in-eight risk of breast cancer – it is her grandmother, who has turned 70. Second, we can all do something to reduce our individual risk of getting the disease. And third, breast cancer is now a disease that most women survive, thanks to earlier diagnosis and better treatment. Cancer is largely a disease of old age. Young women such as Kylie Minogue are the exception, not the norm.³

To show you the issue, I pulled data on breast cancer risk by age from the Centers for Disease Control website.⁴ The site provides info on the percent of US women who develop breast cancer over the next 10, 20 and 30 years, depending upon their current age. To make it easier to see what is happening, I added the “1 out of x” equivalents.

Woman’s Current Age	Percent of women who are expected to develop breast cancer during: 3
	Next 10 years	Next 20 years	Next 30 years
30	0.43%: 1 out of 233	1.86%: 1 out of 54	4.13%: 1 out of 24
40	1.45%: 1 out of 69	3.75%: 1 out of 27	6.87%: 1 out of 15
50	2.38%: 1 out of 42	5.60%: 1 out of 18	8.66%: 1 out of 12
60	3.45%: 1 out of 29	6.71%: 1 out of 15	8.65%: 1 out of 12

So, you say, where does 1 in 8 come from?   It comes from the population estimate that among women who live until age 85, one in eight develops breast cancer sometime during her lifetime.⁵  And remember, the average life expectancy of a woman in Canada is about 83. In other words, appreciable numbers of women are living long enough to be in the highest-risk age group for breast cancer (i.e., age 70 and older). This doesn’t mean that younger women can’t get breast cancer; in fact, I’ve had personal experience of two women who developed breast cancer in their 30s and subsequently died. The point is that these women are exceptional (i.e., they fall into the 0.43% of that age group) and not the norm. If younger women can develop breast cancer, then what difference does it make if current messaging is somewhat misleading? The problem with current messaging is two-fold. First, it may unnecessarily scare younger women. Second – and more importantly, in my opinion – is that it may lead to complacency on the part of older women. A woman in her 70s may figure she doesn’t need to worry about breast cancer (“if I haven’t gotten it yet, I’ll probably never will”) when this is precisely the time when she needs to be most vigilant. A lot of the problem and confusion stems from the difference between population-based probabilities and individual risk. As I’ve explained in the past, risk statistics are population-based probabilities: they tell us what will happen for a group as a whole. Population or group-based stats are woefully bad at telling us what will happen to any one individual within that group. Each individual’s risk will vary, according to any number of genetic, personal and environmental factors. Accurately communicating health statistics is difficult. Organizations struggle to get the messaging correct, especially when competing for the attention of people in a media- and message-saturated environment. I’m sure that the Imperial Cancer Research Fund and various breast cancer charities do excellent, valuable and important work. It’s just a shame that they need to resort to sensational messaging in order to draw attention to their work and issue. But if they have to do it, then they need to take care. No-one wants to be called out on the pages of publications such as BMJ.

Take-away message/bottom line Accurately communicating and interpreting population-based probabilities can be difficult.

¹ Kent A. Raising awareness or spreading fear? BMJ 2000;321:247

²  American Cancer Society. Lifetime Risk of Developing or Dying From Cancer. http://www.cancer.org/Cancer/CancerBasics/lifetime-probability-of-developing-or-dying-from-cancer

³ Boseley S. The truth about breast cancer. The Guardian, Tuesday 8 February 2011. http://www.guardian.co.uk/society/2011/feb/08/breast-cancer-one-in-eight

⁴ Centers for Disease Control and Prevention. Breast Cancer Risk by Age. http://www.cdc.gov/cancer/breast/statistics/age.htm

⁵ University of California San Francisco (UCSF) Medical Center. Breast Cancer Risk Factors. http://www.ucsfhealth.org/education/breast_cancer_risk_factors/

 

Is healthcare biased against men? The case of PSA testing

My husband has a cadre of male friends that I refer to as his “pub buddies.” They’re great guys and range from retired professionals to tradesmen. Little moss grows on any of them. Lately, hubby came home with the message that he and his pub buddies had decided that the policy about PSA testing in Ontario is sexist and reflects a societal bias against men. After all, he says, mammography and Pap smears are paid for through public health insurance but PSA is not. Let’s look at this. PSA testing is a blood test for something called prostate specific antigen. PSA levels are typically low in men but a number of conditions – both benign and cancerous – can increase it. Both prostate cancer and benign conditions such as inflammation of the prostate (prostatitis) and enlargement of the prostate (benign prostatic hyperplasia) increase in frequency with age. It is the position of some organizations that all men over 40 have a PSA to establish a baseline value and men over 50 consider annual or semiannual PSA monitoring. PSA testing for screening purposes (i.e., in a man with no evidence or diagnosis of prostate cancer) is covered by public health insurance in most provinces and territories. But in British Columbia and Ontario, the cost is only covered if there is a diagnosis, clinical signs of prostate cancer and/or other indicators of risk such as family history, own health history or race (there are differences between the provinces). Why isn’t PSA testing routinely covered as a screening tool in all provinces? Well, the problem stems from the evidence on its effectiveness. A Cochrane Collaboration Review of the research literature found screening did not significantly reduce prostate-cancer specific mortality.¹ Moreover, PSA testing didn’t do very well when they looked at the balance between benefits and risks. Benefits of screening may take up to 10 years to occur, so screening was unlikely to be helpful for men with a life expectancy less than 10-15 years (i.e., elderly men). Meanwhile, you need to consider the draw-backs or potential risk of screening. The main problem with screening is the high rate of false-positive results. According to the Cochrane review, up to 76% of positive PSA tests are false.¹ In other words, up to three-quarters of men who get a “positive” PSA test result don’t have prostate cancer. But because of their test result, they may be exposed to the risk of additional, often invasive, tests such as biopsies. To say nothing of the worry and stress of being told you may have cancer. According to the US National Cancer Institute, “most men with an elevated PSA test result turn out not to have cancer; only 25 to 35 percent of men who have a biopsy due to an elevated PSA level actually has prostate cancer” (emphasis in the original). ² In addition, a certain proportion of tests may be false-negatives: the PSA level is in the normal range even through prostate cancer is actually present. A false-negative could lull a man into a false sense of security and encourage him to ignore warning symptoms. If you want to learn more, the October 27, 2011, issue of New England Journal of Medicine has three editorials on the topic in response to the recent US Preventive Services Task Force recommendations (www.nejm.org). FYI, the Task Force recommended against routine PSA-based screening. The issue is complicated and I think it’s the uncertainty in the science rather than overt sexism that is behind the current patchwork nature of PSA reimbursement across Canada. But I can certainly appreciate that on the surface – yes, it does look like men are being treated differently from women.  However, I think the pub buddies need to appreciate that you can’t just order a PSA – it’s supposed to be accompanied with the dreaded digital rectal exam. Obviously, better screening tools could be helpful and that’s something only research can give us.

Take-away message/bottom line There are significant issues with PSA testing which may help to explain why it is not routinely covered as a screening tool in all provinces. What is needed is something only research will be able to give us – a more accurate screening tool.

¹ Ilic D, O’Connor D, Green S. Wilt TJ. Screening for prostate cancer. Cochrane Database of Systematic Reviews. Last assessed as up-to-date: June 10, 2009. http://www2.cochrane.org/reviews/en/ab004720.html

² National Cancer Institute at the National Institute of Health. Prostate-Specific Antigen (PSA) Test. http://www.cancer.gov/cancertopics/factsheet/detection/PSA

 

A relook at the Milgram obedience experiment: are we all capable of being Nazis?

If you ever took first year psychology – or read any popular psychology – then you’re probably familiar with the obedience studies conducted by Stanley Milgram at Yale in the early 1960s. Inspired in part by the horrors of the Holocaust, Milgram wanted to assess whether some nationalities are more willing than others to obey authority figures. He designed an experiment in which participants were told by a lab-coated authority figure to administer a memory test to a learner – and to give progressively-severe electric shocks if the learner made a mistake. The learner was not only in a separate room from the “teacher” (who was actually the experimental subject) but was actually only acting to be shocked. The way I, and I think most people, remember the experiment is that it showed that ordinary people were willing to inflict severe and even lethal levels of electric shock when ordered to do so by an authority figure. It was a chilling reflection of the Holocaust: under the right conditions, any of us could be cold-hearted killers. But a review of the experiment and of Milgram’s broader body of research in this month’s Scientific American Mind suggests that this capsule summary may be misleading. ¹ To begin with, authors Stephen Reicher and S. Alexander Haslam, point out that even in the original experiment, not everyone followed orders: in fact, 35% of subjects quit or refused to continue. (Personally I don’t remember that from Psych 101 – I assumed everyone complied.) Footage of the experiment showed that many of the participants visibly agonized over what to do.  As Reicher and Haslam describe it, “They argue with the experimenter. They reflect the learner’s concerns back to him. They search for reassurance and justification.” Many felt trapped between their duties to the experimenter and the experiment – which had been described to them as a worthy attempt to advance science – and the pleas of the “learner.” Although this is the study most of us remember, in actual fact Milgram conducted a number of subsequent studies in which he looked at the influence of various factors. He found that when the subjects sat in the same room as the learner and saw the effects of the “shocks” the percentage who complied fell to 40%. It fell even further if the person had to press the learner’s hand onto an electric plate to deliver the shock. It also plummeted (to less than 20%) when two other “participants” (who were also actors) refused to participate. Finally, no-one went up to the “lethal dose” if the learner asked for the shocks, when the “authority figure” was the person being shocked, or when two authority figures gave conflicting instructions. Looking at the research, Reicher and Haslam conclude that the issue is not that followers lose their moral compasses so much as they choose particular authorities to guide them through the dilemmas of making difficult decisions. Which authority figures do we chose? Some research suggests our choices are guided by the extent to which we identify with that person and his or her goals. So we need to be careful who we choose as our “moral authority.” But even so, it doesn’t mean we necessarily turn off our own moral compass – or our responsibility.

Take-away message/bottom line Re-assessment of Milgram’s original study in 1961 suggests that the common interpretation that people are blindly obedience to authority may not be correct; the situation may be much more complex.

¹ Reicher S, Haslam SA. Culture of shock. Scientific American Brain 2011 22(5): 57-61

 

Who are Low Germans and why do they matter to public health?

The best part of my job is that I’m always learning something new. Recently, I was employed by a health department in Ontario to do some lit reviews and one of the topics concerned “Low Germans”, more formally referred to as those who are Low-German-speaking. Of course I knew a bit about Mennonites and the Amish (the Amish being found primarily in the US) but what was of interest here were Low-German families from Mexico. Mexican Mennonites? We need to start with a bit of history. Mennonites are a German-speaking people who came to Canada in the mid-1800s and because of religious beliefs prefer to remain separate from mainstream society. (In actual fact, there’s a wide diversity within Mennonite culture in the extent to which they retain traditional ways and accept or reject technology.) During and immediately after World War I, both their language and their refusal to serve in the military resulted in a lot of pressure and tension on Mennonite communities in Canada. By the 1920s, significant numbers emigrated to Mexico and other Latin American countries, particularly after the provincial government took steps to secularize education. However, for many, Mexico did not turn out to be much of a haven and many were left landless and struggling economically. To earn and save enough money to buy land or repay debts, some families have returned to parts of Canada and the US (for example, there’s appreciable numbers of them in Kansas). They typically gravitate to rural areas – parts of the country which are already facing significant public health challenges. Moreover, families are not totally settled but may make frequent trips back to Mexico. This disrupts employment, schooling, and health care. There are a lot of stressors for these families. Relatives in Mexico may feel deserted or betrayed, while Mennonite communities in Canada or the US may be less than welcoming. Education is limited: many children start helping in the fields as early as five or six years of age. Children are “God-sent” and families are large. Men are the “undisputed head of the household” and spousal abuse, alcoholism, child abuse or other family issues are typically swept under the carpet. Given all of these external and internal stressors, it’s not surprising that depression and anxiety are common; in a study conducted in Elgin Country, focus groups among the women founded that many of the men turn to alcohol, creating both financial and emotional problems for families.¹ Language barriers, transience, low education, poverty and cultural barriers make health care difficult. Women are not necessarily told about “the facts of life” or birth control, receive prenatal or post-natal care, or have the means to ensure their children receive well-baby visits, vaccinations or dental care. Even if they want or need services, how can they access them when they have no transportation, don’t speak the same language as healthcare providers, and are tied to the home by child-rearing duties? According to a report prepared by Wellington-Dufferin-Guelph Public Health Unit², girls typically finish school at Grade 8 or 9. After that some may work in a bakery or store for a while but most are expected to remain in the home helping their mothers until they marry. This remind you of anything? Like the FLDS or Irish Travellers? According to the World Health Organization, key sociocultural elements that negatively impact on women’s health include, among other things, a) an exclusive focus on women’s roles as wives and mothers and b) reduced or limited access to education and paid employment opportunities. ³ Repeatedly it has been shown that one of the best ways to improve the health of women and children is to improve women’s access to education and, as a result, information and economic opportunities. Meanwhile, public health organizations in Canada and the US are struggling to develop programs and processes for effectively reaching this Low German population. In Kansas, for example, the state Department of Health trains local Health Promoters – trusted and respected community members who establish links between health providers and the community members and provide informal health education, case management and interpretation.⁴ That’s probably only one of several options. It will be interesting to see what emerges as I continue to review the literature.

Take-away message/bottom line There are many sub-cultures around us that may be invisible to mainstream society and which require tailored approaches for health and social services.

¹ Armstrong D, Coleman B. Health Care Needs of Mennonite Women Living in Elgin County. Elgin St. Thomas Health Unit, March 2001

² Bennett J. Low-German-Speaking Mennonites from Mexico: A Review of the Cultural Impact on Health in Wellington County. Wellington-Dufferin-Guelph Public Health, December 2010.

³ Women and Health, Today’s Evidence, Tomorrow’s Agenda. 2009; World Health Organization, Geneva. http://whqlibdoc.who.int/publications/2009/9789241563857_eng.pdf

⁴ Guenther T, Treaster C. Kansas’ Low German Mennonites’: Meeting the Challenge of An Emerging Farmworker Population. Kansas Statewide Farmworker Health Program, Kansas Department of Health and Environment

 

Most chronic diseases have long, long histories

I was watching Nova the other night and the show was describing the 5,000-year-old Iceman found in the Alps in 1991.  Nicknamed Otzi, it’s been determined that he was approximately 45 at the time of his death. Analysis of his leg bones showed that his lifestyle included long walks over hilly terrain (i.e., he was physically active) and we know that at that point in history, his diet had to be purely organic (his last meal consisted of wild ibex meat and whole grains) and he was smoke free (tobacco wouldn’t make it to Europe for thousands of years). But that didn’t mean he was necessarily healthy. He had gallstones, knee problems, dental issues, and atherosclerosis (“hardening of the arteries”). In fact, during the Nova show, they said the extent of his coronary atherosclerosis was similar to what you’d expect in a modern 40-year-old male. This isn’t the only mummy to show evidence of heart disease.  Examinations of Egyptian mummies have also shown evidence of atherosclerosis of the blood vessels in the heart and kidneys, suggesting “that the stresses of modern day highly civilized life are not the sole cause of degenerative vascular diseases.” ¹ In fact, CT scanning of one Egyptian princess who died in her 40’s 3,500 years ago showed atherosclerosis so severe that one investigator said, “Today she would have needed bypass survey.” ² She wasn’t an isolated case: 45% of mummies scanned showed signed of atherosclerosis.² And one Egyptian mummy has been identified as having metastasized prostate cancer. ³ Okay, so what’s my point? My point is that it’s illogical to think we can cheat time by adopting radical diets or regimens. The human body wears out. Joints lose their lubricants and rub against one another, coronary arteries thicken and get stiff, and teeth weaken or decay. It’s what we call aging. Some people are genetically blessed and age at a slower rate or manage to avoid some diseases. Other people are not as fortunate, and may develop disease much earlier or contract a more aggressive form. Healthy living is a great idea, primarily because it makes us feel a heck of a lot better. Healthy living may be able to ameliorate chronic diseases, help us cope with them more effectively or even postpone them by a few years. But let’s not get too carried away. Any one diet, physical activity regimen, or stress management program is not going to change the biology and physics of aging.

Take-away message/bottom line Most chronic diseases have been around for a long time.

¹ Mummies; Mummies and Disease in Egypt. University of Illinois at Chicago website. http://www.uic.edu/classes/osci/osci590/6_2Mummies%20Mummies%20and%20Disease%20in%20Egypt.htm

² Lorenzi R. Mummy had earliest case of heart disease. DiscoveryNews., Friday May 20, 2011. http://news.discovery.com/history/mummy-heart-disease-110520.html

³ Luiggi C. Mummy Cancer. TheScientist. October 28, 2011 http://the-scientist.com/2011/10/28/mummy-cancer/

 

Prevalence does not necessarily equate or predict death

Unfortunately, a lot of the health messages we get in the media are written by marketing folks who may not necessarily understand or research the data. Sometimes, I wish they would talk to more epidemiologists, statisticians, demographers, and other scientists before turning on their computers. Not truly understanding statistics can lead to problems.

Error 1: Not understanding the difference between percentage of deaths, prevalence, and risk of death. Approximately 25% of deaths that occur in Canada in any year are due to x.  Does this mean that one out of every four Canadians will die from x?  Well, there’s a problem. As I discussed in my October 19 blog, the people who die in any year are not representative of the general population: they consist of a heck of a lot of people who are old and relatively few people who are young. So just because 1 out of 4 people who die succumb to disease x doesn’t mean that 1 out of 4 people in the current population will eventually die from x – especially when x is a disease that increases in prevalence with age. In that case, you have to live long enough to die from it. There’s a big difference between the odds of developing a disease and the odds of dying from it. Let’s take some examples to illustrate this.

• In the US, for example, a woman’s risk of developing cancer is 1 in 3 but her risk of dying from cancer is 1 in 5.¹ Likewise, a man’s risk of developing cancer is 1 in 2 but his risk of dying from cancer is actually 1 in 4. ¹

• According to results from the Framingham Study, if you take people who are 50 years of age (which means you have to live to be 50 for these stats to apply), the odds that they will develop CVD by age 95 is 51.7% for men (1 in 2) and 39.2% in women (almost 1 in 3).  If you look at age 75 (given that many men don’t make it to 95), the odds are about 1 in 3 for men and 1 in 5 for women. Wait a minute, you say, the odds are lower for the 75-year-olds than the 50-year-olds? Yes, because among the 75-year-olds more may die from (or have already died from) other diseases also associated with aging, such as many forms of cancer and respiratory diseases.

However, these are just the odds of developing CVD – it doesn’t tell us about the odds of dying from it: the same study found that median survival times for the 50-year-olds were 30 years for men and 36 year for women (i.e., half of the men lived beyond age 80 and half of the women beyond age 86). Okay, for what the odds of dying from CVD? I must admit to having trouble finding much published literature on this point. There’s a lot on the proportion of deaths and death rates, but not much on this crucial point.  The best I can find is the lifetime odds of death chart prepared by the (US) National Safety Council. It says that in the US in 2007 the odds of dying from heart disease was 1 in 6, compared to 1 in 7 for cancer and 1 in 28 for stroke.²

Error 2: Inferring that population stats can be applied equally across the entire population The National Safety Council’s chart on the odds of dying from various causes is sort of morbid fun. ² For example, it shows that while the odds of dying from heart disease and cancer are high (1 in 6 and 1 in 7, respectively), the odds drop a lot after that. Odds of dying in a motor vehicle accident in the US? 1 in 88.  From flying? 1 in 7,032.  From lightning? 1 in 84,079  From flood? 1 in 175,803.  But the National Safety Council is adamant that it’s not correct to compare odds or assume they apply equally to everyone. For example, the chance of dying from an earthquake is 1 in 148,756.  But that applies to the entire country – including areas where there are no earthquakes.  Obviously, your chances of dying from earthquake are a lot higher if you live in California, as opposed to Kansas. Just as your chance of dying from a flood is a lot higher if you live near a body of water. The same thing applies to diseases: prevalence or mortality stats are group estimates and not everyone in the group is necessarily equal.  For example, researchers at the Cooper Center looked at the lifetime risk of death from CVD for males.³ Overall, a 55-year-old man had about a 16% (1 in 6) chance of dying from CVD before age 80. Take a 65-year-old man, and his risk of dying before age 80 from CVD in much higher: 42% or 1 in 2.  But that’s just the overall average. In fact, the risk varied a lot according to the man’s physical activity level. Here’s the percentages from the study, with the addition of the calculation of its “1 out of” equivalent.  FYI, the authors compared their findings to previous analysis that looked at risk according to traditional risk factor burden, and found they were pretty close.

Age of Man	Lifetime Risk of Dying from CVD by Age 80
	High Fitness(24% of sample)	Moderate Fitness  (43% of sample)	Low Fitness (32% of sample)	Very Low Fitness (2% of sample)
55-year-old	4.9% (1 in 20)	9.8% (1 in 10)	19.6% (1 in 5)	28.7% (1 in 3)
65-year-old	5.6% (1 in 18)	8.8% (1 in 11)	12.2% (1 in 8)	15.2% (1 in 7)

The point here is that risk is not evenly distributed throughout the population but varies significantly with any number of factors. This study looked at the risk factor of physical activity; we need to keep in mind that not only other risk factors but also genetics play roles in determining risk. In looking at these stats you need to keep two things in mind. First, risk varies between sub-groups. Second, the number of people in the various sub-groups varies so you need to keep in mind the number of people you’re talking about. That gets us into the issue of Population Attributable Risk, something that perhaps I’ll talk about at another time.

Take-away message/bottom line The proportion of deaths due to a condition, the risk of developing a condition and the risk of dying from a condition are three distinct and different concepts.

¹Ameriican Cancer Society. Lifetime Risk of Developing or Dying From Cancer. http://www.cancer.org/Cancer/CancerBasics/lifetime-probability-of-developing-or-dying-from-cancer

² National Safety Countil.  The Odds of Dying From …. See the explanation at http://www.nsc.org/NEWS_RESOURCES/INJURY_AND_DEATH_STATISTICS/Pages/TheOddsofDyingFrom.aspx; the chart is found at  http://www.nsc.org/NSC%20Picture%20Library/News/web_graphics/Injury_Facts_37.pdf

³  Lloyd-Jones DM et al. Prediction of lifetime risk for cardiovascular disease by risk factor burden at 50 years of age. Circulation 2006;113:791-798.

 

Can you scare people into being healthy?

One of my recent tasks was looking at the literature as to what works – and, conversely, what doesn’t work – in health promotion messaging. There’s a considerable amount of experimental literature on this issue, including studies comparing gain- versus loss-based and episodic versus societal-framed messaging.  Related to this, there has also been a lot of work looking at fear-based messaging. In other words, can you scare people into being healthy? Fear-based messaging got a bad rap in the 1970s when it was used to try to scare people into quitting smoking.  It didn’t work. If anything, some ads were so scary that smokers would immediately light up to calm themselves. Oops, that’s counter-productive. So for a long time, fear-based messaging became a pariah in health promotion. The research shows that fear-based messaging actually may be effective – but there are a lot of caveats in using it. First of all, the situation is a bit like the three bears.  The message can’t be weak or it won’t make an impression.  But it can’t be too strong or it will turn people off or send them into denial (like the early anti-tobacco ads). In other words, it has to be just right in terms of the scariness factor. What can be hard, as various people – or groups of people — can interpret and react to messages in very different ways. Even more important, the message has to contain more than just fear. It is critical that the message contain information that shows people what they can do to reduce that threat and protect themselves. In other words, it only works if the message not only scares people but gives them the tools, motivation and sense of confidence that they can do something about it. ^{1 2}  After all, if you don’t think there’s anything you can do about a threat, then the natural impulse is to pull a Scarlett O’Hara – “I’ll think about it tomorrow.”  People will go into denial and the message will have no effect. Or, if you’re really unlucky, the message will – like the early anti-tobacco ads – merely drive people to adapt an attitude of, “What the heck, I’m doomed so I might as well indulge.” So it’s a delicate balancing act. You can use fear – but only if it also gives people tools and a sense of hope. Another interesting finding was a small Australian study that looked at whether fear-based messaging works with youth. ³ We typically assume that adolescents and youth adults see themselves as immortal and so fear-based messaging won’t work. After all, the threats we use with most adults – chronic diseases such as cancer — are so far in their future that we assume youth don’t worry about them. But in this study, it was found that fear-based messaging (the threat of death from smoking-caused emphysema) was equally effective among younger adults (16 to 25 years) and adults 40 to 50 years. About the only group that responded differently to fear-based messaging was older women – it responded better to non-death messaging.  Of course the study was very small and one study alone can be misleading. But it’s an intriguing thought that perhaps some of our pre-conceptions about youth may be misleading.   I hope more research is done in this area. I don’t have any experience with young adults but I was once a member of a team that polled high school students on their health concerns. Their number one worry? Acne. Number two? For boys it was height and for girls it was weight. My take-away was that teenagers are so caught up in the immediate hell of trying to survive socially that potential health consequences in the future are simply not registering. After all, there’s only so much angst you can accommodate at one time.

Take-away message/bottom line It may be possible to scare people into being healthier but messages have to be crafted with great care, provide actionable ways of dealing with the threat and support people’s sense of control and confidence.

¹  Witte .K, Allen M. (2000) A meta-analysis of fear appeals: implications for public health campaigns. Health Ed & Behavior 27(5), 591-615

²  Witte K., Berkowitz J.D., Cameron K.A., McKeon J.K. (1998) Preventing the spread of genital warts: using fear appeals to promote self-protective behaviors. Health Education & Behavior 25(5), 571-585

³ Henley N., Donovan R.J. (2003) Young people’s response to death threat appeals: do they really feel immortal? Health Ed Res 18(1), 1-14