Consistent responses without a religious sponsor (left) and inconsistent responses with a religious sponsor (right)

A recent study by me and Tim Mayhall finds that mentioning to respondents that a survey is chaplaincy-sponsored, greatly affects the reliability of the survey. Thirty-six respondents given a chaplain-sponsored survey had less consistent responses about how depressed they were, compared to 133 respondents told that the survey was sponsored by a hospital. The shift holds steady using bootstrapped sampling to account for sample size differences:

We also emphasize that most of the differences would have been missed by looking only at mean response, and were instead most visible when looking instead at respondent variability. We show that it’s just as important to show differences in variation and model fit than mean responses, when assessing the effects of interviewers or sponsors on survey responses.