We know your survey is exciting and will divulge important findings; you know your survey is exciting and will divulge important findings. But darn those who don’t know this and so don’t participate. Or worse – those who do know this, but choose not to participate for some other reason. Their nonresponse can create uncertainty in how accurate our survey results actually are. If only we could survey robots or well-trained dogs who followed our instructions instead of people!
All kidding aside, nonresponse can and does make us wonder about the accuracy of survey results, going so far as even making all the effort we put into creating surveys for naught because we just can’t rely on the survey’s accuracy.
We don’t have to tell you how troublesome this can be for researchers and scientists: our studies rely heavily on survey results and we must be able to defend their accuracy.
And it’s not just that “not enough” people respond to our surveys: nonresponse can be problematic if those who do end up responding to our surveys turn out to be different from those who do not in some systematic way – especially if that systematic way is correlated with the topic of the study. When differences exist – we call this nonresponse bias. Studies have found that nonresponse bias is very difficult to pin down – it emerges differently with different topics and populations.
Download Evaluating Nonresponse Bias in a Longitudinal Study of Healthy Adults Receiving Genome Sequencing POSTER
We presented a poster in October at the ASHG Conference in Orlando, Florida on Evaluating Nonresponse Bias in a Longitudinal Study of Healthy Adults Receiving Genome Sequencing (You can download the poster by filling our the form on this page) and we discussed how the PeopleSeq Consortium study (designed to measure economic, medical and behavioral outcomes after elective genome sequencing of healthy individuals) may have been impacted by nonresponse.
Understanding nonresponse bias helps methodologists design questionnaires and studies to deal with the always-present chance such bias can become an issue in a survey research study.
In short, by conducting a nonresponse analysis within the PeopleSeq study, we found that white females were overrepresented in our survey respondents, meaning that nonresponders were more likely to be male and non-white – a difference that may matter for some analyses. However, we also found very little evidence for bias in age. (See the poster for more details).
The key takeaway? There are differences between those who respond to surveys and those who don’t.
Yet before we decide to throw up our hands and weep bitter tears at the chance our survey is biased in some ways, we need to delve deeper. So far, this study includes people who are a very select group to start with (those who have elected to participate in full genome sequencing). And, so far, we have not found any evidence that the bias is having an impact on any substantive outcomes being studied.
Armed with this information, we need to: (a) learn how to increase participation of those who turn out to be underrepresented, and (b) identify how much differential nonresponse may be actually influencing research findings.
Finally, the very existence of this evidence of nonresponse bias shows how important it is to have a survey tool and data collection design that is scientifically informed. And who has these nifty tools? SoundRocket of course! We help our clients create surveys that can tell nonresponse bias to stay home/you can’t sit with us, so that no matter who responds (and how many or few of them) results will be useful and actionable.
Contact us today to learn how we can help you save your surveys from the scourge or nonresponse bias!
(Yes, I’m now asking you yet again to download the poster for more information. Because I’m persistent like that.)