Does response propensity bias stress audits?

Abstract

Background Non-response bias in surveys occurs when non-respondents differ from respondents. Methods of dealing with this include measuring 'response propensity'. We propose that response propensity has a trait-like component, measurable within the survey. Covariance of this component with survey items could indicate non-response bias. Aims To measure and validate trait-like response propensity, to test whether it co-varied with survey items during a routine stress audit and to test whether a prior staff attitude survey may have been biased. Methods Stress survey sent to government employees during a routine audit. We designed two new items to measure trait-like response propensity and completion of a prior staff survey. We calculated an odds ratio, correlations and Mann-Whitney U-tests to assess relationships between trait response propensity, completion of the prior staff survey and current stress survey scores. Results There were 71 respondents; response rate 68%. The trait response propensity item predicted completion of a prior staff survey (odds ratio 8.75, 95% confidence interval 2.32-33.08). Trait response propensity significantly correlated with two of the 40 survey items and did not significantly correlate with any of the eight 'stressor' sub-scales. Non-respondents to the prior staff survey had a significantly lower risk of stress on two of the eight stressor scales. Conclusions Trait response propensity seems to be a valid construct. The weight of evidence does not provide an indication of non-response bias (yet neither does it assure freedom from non-response bias). The prior staff survey may have been affected by non-response bias.