James Schneider  

Where Does All the Booze Go?

Feminizing Famine... What If Firms Could Opt Out of...

In a recent post, Bryan wrote about how people do not accurately report how much tobacco they purchase. If they did, then the survey data would be consistent with tobacco sales. However, surveys report less than half of the actual tobacco sales. People are reluctant to admit to socially frowned upon activities. This pattern of biased responses occurs with other health issues as well.

Based on sales figures, self-reported drinking in Great Britain is only 60 percent of the actual alcohol consumed. Biased answers present a serious challenge to medical researchers because many studies on the health dangers of drinking rely on self-reported drinking levels. Bias is especially problematic with alcohol because drinking likely follows a U-shaped curve -- even though heavy drinking is dangerous, moderate drinking might have health benefits. If drinking in moderation is good for you, then finding moderation's sweet spot requires that researchers have accurate measures of individual alcohol consumption. Not surprisingly, researchers have spent a lot of energy studying the best ways to get people to report accurate measures of alcohol consumption.

Some studies ask people to write down their alcohol consumption retrospectively whereas other studies ask people to keep track of their consumption on a go-forward basis. When the former approach is used, people report approximately 20 percent less alcohol consumption. This probably isn't caused by people increasing their drinking to look cool on the go-forward questionnaires. It is more likely that people have imperfect recall about past actions. Further evidence of imperfect recall is the fact that asking people about their drinking in greater detail yields higher self-reported drinking. When beer consumption, wine consumption, and spirit consumption are requested separately, people report approximately 20 percent higher alcohol consumption. Asking separate questions forces people to think more carefully about the response. This pattern is also consistent with social desirability bias. Perhaps people default to socially desirable answers, but they will give less biased responses when they are forced to carefully consider their behavior. (Investigators don't "break" suspects by asking a single question about their guilt.) In general, using better methods to ask about drinking yields higher reported drinking. Of course, if heavy drinkers are reluctant to fess up, then any method based on surveys will underreport consumption.

Social desirability bias would be a particular problem when surveying pregnant women about how much alcohol they consume. While most heavy drinkers would be merely embarrassed to report their actual drinking, pre-natal drinking comes with a much higher level of shame. One study received more truthful responses from mothers by waiting 14 years after the pregnancy and asking about pre-natal drinking a second time. Normally, waiting so long might bias drinking results downwards since people are more likely to forget than to hallucinate. However, it seems that the shame of pre-natal drinking ebbs with time. When mothers were asked about their drinking 14 years later, they gave higher estimates of their drinking than when they were initially questioned. These higher estimates seem to be more accurate; they did a better job of predicting problems that developed during the child's teenage years.

Self-reports are also an issue when studying obesity. People have a tendency to underestimate their weight and over-estimate their height. Again, not all survey techniques are created equal.

It should be noted that the under-reporting of weight and BMI tends to be smaller in the NHANES than in other studies, possibly because participants are aware that the questionnaire survey is followed up by the examination.

Honesty and reliability are socially desirable traits. Maybe the desire to show these traits moderates the bias shown in self-reported height and weight.

This is my last post as a guest blogger. I would like to thank Amy, Lauren, Bryan, and David for giving me this opportunity. It's been a pleasure being associated with such a great group of people.

Comments and Sharing

COMMENTS (7 to date)
Tracy W writes:

House: everyone lies.
Thank you for a very interesting final post.

Greg Heslop writes:

Those are quite large effects from different ways of asking. I wonder if paper surveys and computer surveys would also yield different responses, but I don't know if there are any studies on this?

I am thinking that respondents will be more likely to answer in untruthful but socially desirable ways if they fill out a paper form, because when they are done, the paper is handed to someone, or put on a pile. Either way someone might see who handed in the paper. This problem would not exist if the survey were done using a computer.

Thank you for this post and for all your others, Professor Schneider. I have enjoyed reading them.

JJ writes:

Have you hosted a party recently? Go around afterwards, and you'll see that 75% of the beer that was opened is wasted. I'm a 24-year old in NY. If it happens by me, I can imagine the effect isn't negligible.

KLO writes:

One interesting possibility is that one's ability to accurately estimate personal consumption affects how much one consumes. Fat people very frequently claim to eat very little, while simultaneously claiming that others who are much lighter eat much more. The general assumption is that these fat people are simply liars. However, it is possible that fat people are simply poor at estimating consumption, and really do not know that they are eating as much as they are. While I have doubts that fat people aren't simply lying, in principle it makes sense that some people are better at estimating consumption and those who are worse at it would be fatter.

Jay writes:


While I agree with you that consumption isn't 100% a perfect measure, I don't think England's 40% of alcohol numbers are left over in solo cups sitting on a porch railing.

MingoV writes:
Biased answers present a serious challenge to medical researchers...
Any person who relies on self-reported information on behavior should automatically be disqualified from being a medical researcher. The unreliability of survey-obtained data is why most physicians are skeptical of reports from NHANES and other surveys.
Shane L writes:

In some circumstances I would expect social desirability bias to cause respondents to exaggerate their drinking, not under-rate it. There are macho drinking subcultures, such as that prevalent in my college years. Drinking heavily was a part of young people's identities and self-image. People might boast about how much they drank and how drunk they became, all part of the youthful machismo.

Comments for this entry have been closed
Return to top