When should you worry about recall bias?

Steve Simon

2006-08-03

Dear Professor Mean, I am studying reports of early health concerns in a group of children with a chronic disease compared to a matched group of healthy controls. I believe that the children in the disease group will have an increased rate of health problems in areas seemingly unrelated to their disease. But I am worried that there may be recall bias: the children and their parents will recall events in their past differently if they have the disease or not. What can I do about this?

Recall bias is always an issue in retrospective research, but sometimes it is a minor concern and sometimes it is a serious problem.

I would be especially concerned about recall bias if you are asking about things that most people forget. So if you’re asking about major surgery, that’s something that pretty much everyone remembers, so there is little risk for recall bias. Dietary studies are trickier because who can remember exactly what foods they ate last week (I’m not even sure I remember what I ate last night).

Subjectivity is also an important issue. If different people are likely to interpret the same event in different ways or if the event itself is difficult to pin down, then you need to worry. During an interview, Linus Pauling was extolling the virtues of Vitamin C. He pointed out that he and his family took Vitamin C supplements daily and nobody ever had trouble with colds. The interviewer pressed him on this, “No colds, ever?” Dr. Pauling replied, “No, just an occasional sniffle.”

The other thing to remember is that an inaccurate answer because some people forgot is not all that serious. You lose a bit of precision, but it doesn’t cause any serious bias. The problem occurs when one group in your study has more difficulty with remembering than another group. Frequently this happens when one of your research groups has suffered an event that stirs strong emotions--a miscarriage or a recent diagnosis of cancer, for example. There is something about tragedy that heightens people’s memory. They search through their past to see if there is any explanation for what has just happened to them. It’s not that they make up stuff, but it’s more that the control group doesn’t search as hard and will not recall events with the same level of detail. Also, the things they remember are risk factors that are described in the popular press, some of which are real risk factors, of course, and some of which are illusory.

How do you overcome recall bias? Nothing can totally remove the problem of recall bias in retrospective studies, but sometimes there are approaches you can take that will ameliorate the problem. The approaches vary greatly depending on the type of research that you do, but here are a few general suggestions:

  1. Although the person being interviewed knows whether they are in the disease group or the control group, you can still use blinding on the interviewer. If the interviewer does not know what group the patient belongs to, he/she will not subconsciously probe a bit harder for information from patients in a certain group (are you sure you never smoked cigarettes?).
  2. Measure your risk factors several different ways. Sometimes you can get information from a medical record and use it to supplement the interview or to cross check certain pieces of information. Both the interview and the medical record will have problems, of course, but the combination of the two might help avoid some problems with recall bias.
  3. Perform a quality audit of a fraction of your patients. You probably won’t have time to do this for every patient, but if you could do some additional investigation for 5 or 10% of your patients, that would give you a way of checking on the degree to which both groups had problems recalling information.

Creative CommonsLicense This work is licensed under a Creative Commons Attribution 3.0 United States License. It was written by Steve Simon and was last modified on 04/01/2010.