We compare nonresponse and measurement error bias across two modes of data collection--telephone and web. In an experimental setting we randomly assigned respondents to either telephone or web mode (n=3,482). Because the sampled persons were selected from German administrative records, record data are available for all sample units allowing us to disentangle nonresponse and measurement error.
This paper discusses how regression coefficients are affected by the biases we found in previous work examining univariate survey statistics. We demonstrate which error source exerts a larger effect and whether the errors reinforce each other or cancel each other out. Finally, the paper assesses whether the effects are the same for both modes, or whether there is a 'preferred' mode.
We find that while nonresponse and measurement error affect the effect size of the regression coefficients they do not substantively change the conclusions one would draw from the regression models. The findings also show that in the presence of measurement error nonresponse adjustment techniques should be used with caution.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.