An Item Response Theory Approach to Estimating Survey Mode Effects
Analysis of Data from a Randomized Mode Experiment
Published in: Journal of Survey Statistics and Methodology, Volume 5, Issue 2 (June 2017), pages 233-253. doi: 10.1093/jssam/smw033
Posted on RAND.org on May 31, 2017
Read MoreAccess further information on this document at Journal of Survey Statistics and Methodology
This article was published outside of RAND. The full text of the article can be found at the link above.
When a survey is offered in more than one mode of administration, the potential for differences in the probability of selecting a response category attributable to the survey administration mode may threaten the cross-mode exchangeability of responses or comparability of results. We demonstrate the utility of item response theory (IRT) in quantifying the presence of mode effects, providing insight into the nature of the effects and adjusting cross-mode results. Such IRT applications are of interest when the survey instrument informs an underlying latent trait. We present a Bayesian hierarchical IRT model that can accommodate multiple modes of survey administration and provide cluster-level parameter estimates of the latent trait when observed groupings of respondents are of interest. We illustrate the model with data from a randomized survey mode experiment in which responding subjects within each of forty-five evaluated institutions were randomly assigned to one of four response modes: mail, telephone, interactive voice response (IVR), and a mixed mode of mail with telephone follow-up. Results indicate instrument-wide survey mode effects that differ across ordinal response categories and the underutilization of an interior response category in certain modes.