Most online surveys start with screener questions, designed to target subsets of the population for specific research. Screeners can be used to qualify particular people, to enforce quotas, or to verify the accuracy of the list of survey participants. For instance, for one survey on cosmetics, we screened out men. For a campaign measurement study, we targeted respondents in four metropolitan areas and screened out everyone else; once we had more than 150 respondents for a metropolitan area, we screened out additional responses in that area as well (for being over the quota).
Members of a panel quickly learn that how they answer the first few questions of a survey determines whether or not they qualify for that study and its reward. As a result of the use of screeners, survey researchers often fear that people will lie to questions in the screener in order to qualify for the study.
To test this several years ago, we surveyed a panel and a river sample and asked respondents from each the following screener question:
(The choices were shown in random order to participants in both studies.)
Each of the choices listed were for drugs that were not yet on the market. These were names published as “under consideration” in June 2014 by the United States Adopted Names (USAN) Council of the American Medical Association. In other words, none of these drugs were in use at the time the survey was fielded, and the proper response for 100% of participants should be “None of the above.”
For the river sample, 94% of the 406 respondents answered “none of the above” and the 6% who didn’t each selected only one medicine. For the panel sample, 88% of the 408 respondents answered “none of the above”; however, 3% selected multiple medicines, and of those 14 respondents, 5 respondents selected all 5 medicines.
A fair hypothesis is that the 3% who selected multiple medicines were cheaters, fully aware they didn’t take any of these medicines but wanting to qualify for the study. Some of those who selected one medicine might be cheating; some might be confusing the choice they selected with the name of a medicine they actually take. In the river study, for example, Domeglicant and Tofeglicant were twice as likely to be selected as Aglatimagene Beradenovac and Susquetide, indicating such confusion might be at play (no one in the river sample selected Pegmodglutide).
Imagine that this screener was meant to find people taking a low incidence drug that 1% of the population might be administered. If a drug was selected 4% of the time by cheaters, fully 80% of the survey responses might be from people not taking the medicine (1% qualifying + 4% lying, divided by 5%).
The false positives to screeners can significantly skew results to surveys for which few participants qualify. Make sure to plan accordingly.
For more tips on addressing screener bias, see my post, 7 Best Practices for Writing Better Screeners.