NEW YORK – New research suggests genetic factors can influence study participants' propensity for not answering questions in surveys used to assess social or behavioral traits — results that point to potential sources of bias in genome-wide association studies.
"We anticipate that these findings will provide insight into genetic variants associated with cognitive processes involved in item nonresponse and also provide a basis for evaluating the impact of nonresponse bias on GWAS of other traits and disorders," co-senior authors Andrea Ganna, a researcher affiliated with Massachusetts General Hospital, the Broad Institute, and the University of Helsinki, and Raymond Walters, an investigator at Mass General Hospital and the Broad, and their colleagues wrote in Nature Human Behaviour on Thursday.
For the study, the international research team used array-based genotyping profiles and survey data for more than 360,600 UK Biobank participants to search for genetic variants associated with nonresponse to 109 questions in touchscreen surveys used to study social or behavior-related traits. Overall, they noted that just over 67 percent of individuals answered "I don't know" to at least one of the questions, while nearly 9 percent had at least one "prefer not to answer" response.
"We are using well-powered genetic data to do more accurate and replicable social science and to explore what might be possible at the intersection of genetic and behavioral science," co-first author Robbee Wedow, a researcher affiliated with Purdue University and Indiana University, said in a statement. He added that "researchers in the field of sociogenomics want to study the genetics in order to do better social science."
In particular, the team saw genetic similarities between individuals who answered "prefer not to answer" and those who chose "I don't know" to survey questions, although each of the GWAS highlighted further genetic associations that were distinct to each of the two response types.
The questionnaire nonresponse patterns also coincided with socioeconomic factors such as education level or income, the team reported, along with a cluster of phenotypic features or health histories.
"It turns out that the genetics of people who either answer the survey question, or do not, overlaps with the genetics of other outcomes like education, income, or certain health behaviors," Wedow said.
The study investigators emphasized that their analyses not only relied on deidentified UK Biobank data, but also looked at response or nonresponse patterns across multiple questions to provide an additional level of privacy protection for participants.
Based on the UK Biobank results and a validation analysis that relied on data from the National Longitudinal Study of Adolescent to Adult Health, as well as polygenic scores developed for the nonresponse answers in the initial analysis, the team suggested that survey nonresponse clues could provide a window into participant behavior, health, genetics, and more.
In addition, the investigators noted the relationships uncovered might introduce previously unappreciated sources of bias to GWAS on certain traits or conditions.
Consequently, they suggested that findings from their genetics- and phenotypic-based analysis of nonresponse "should be considered when analyzing the UK Biobank, among other biobank-scale survey efforts, and when developing novel methods aimed at correcting and leveraging nonresponse in genetic analyses."