Looking for a Model of Scientific Integrity?

Stuff happens. Errors creep into research papers. We gripe about it here from time to time. But today we have a great example of how scientific integrity works.

Back in July, a group of researchers, led by Yulyu Yeh, published a paper about nutrition education for African-American preschoolers. Their analysis found a benefit for the children with excess weight and obesity at baseline.

Unfortunately, after publication, they found a serious error in their analysis. So last week, without mincing words, they published a correction. They said it simply:

We did not take into consideration the phenomenon called regression to the mean. As a result, we cannot make any affirmative statements about the effectiveness of our interventions.

Their conclusion was wrong. They swiftly made it right.

Regression to the Mean (RTM)

RTM is a common statistical problem. It crops up when you have a study sample that differs from the average of a population. Any time you have a sample that differs from the mean of  the population, it’s likely that the next time you take a measurement on that sample, it will regress to the mean.

Let’s say you have non-random group of people with an unusually high BMI at one point in time. They are likely to have a lower BMI the next time you look. That’s RTM in a nutshell.

Studying people with high BMIs is a standard feature of obesity research. So RTM has a way of sneaking into scientific publications about obesity interventions. And as Asheley Skinner et al recently pointed out, it’s an especially common problem in pediatric obesity research. The result is many studies that falsely claim something is effective based on a reduction in BMI from a high baseline.

The solution is careful study design up front. A randomized, controlled study is best. But when that is not possible, other designs can help to avoid the issue. But after the fact, it’s often too late.

No Spin, Varnish, or Denial

By correcting the record for their study with integrity, Yeh et al are providing an important example, says Indiana University Public Health Dean David Allison:

I admire the authors for simply admitting they made an error. They stated clearly and without equivocation that their original conclusions were not substantiated.

More attention to the confusing effects of regression to the mean is warranted. And likewise, people deserve more praise when they come forward to say “I made a mistake and my original conclusions were incorrect,” without spin, varnish, or denial.

Confronted with a statistical error in a published paper, authors face an ethical dilemma. Rationalization can be tempting, as we’ve seen before. We’re delighted to see a better model for scientific integrity.

Click here for the study by Yeh et al and here for the correction. For more on RTM, click here.

Got Integrity? Photograph © Misty Johnson / flickr

Subscribe by email to follow the accumulating evidence and observations that shape our view of health, obesity, and policy.

October 29, 2017