Total Survey Error in Practice

24The Effect of Nonresponse and Measurement Error on Wage Regression across Survey Modes: A Validation Study

Antje Kirchner¹ and Barbara Felderer^3,4

¹ Department of Sociology, University of Nebraska‐Lincoln, Lincoln, NE, USA and Survey Research Division, RTI International, Research Triangle Park, NC, USA

² Collaborative Research Center SBF 884, “Political Economy of Reforms,” University of Mannheim, Mannheim, Germany

24.1 Introduction

Nonresponse and measurement error can lead to serious bias in parameter estimates, especially if the mechanisms causing error are related to the survey variables of interest (Biemer, 2010; Kreuter et al., 2010b; Olson, 2006). Nonresponse bias is a function of the nonresponse rate and the systematic difference between respondents and nonrespondents with respect to the outcome variables of a survey (de Leeuw et al., 2008, p. 3; Groves, 2006; Groves and Couper, 1998). Measurement error can result in response bias if there is a systematic difference in the respondents’ survey reports and their (unobserved) true values (for a review, see, e.g., Biemer and Lyberg, 2003). For the remainder of this chapter, the terms measurement error and response error are used interchangeably.

Processes leading to nonresponse and response bias can be expected to differ across survey modes (for an extensive discussion of mixed‐mode designs, see Chapter 23). Being a cost‐ and time‐saving alternative to other interview modes, web surveys can be very attractive (Callegaro ...