Nonresponse Bias on Dimensions of Political Activity Amongst Political Elites

Virtually all political surveys of citizens’ political behavior suffer from unit nonresponse. Typically, researchers’ analyses implicitly assume that the nonrespondents do not differ systematically from the respondents on dimensions of political activity (King, Honaker, Joseph, & Scheve, 2001). For example, regressions fit with only respondents assume that the model for respondents applies equally to nonrespondents. When such similarity assumptions are not valid, conclusions from standard statistical analyses can lead to inaccurate inferences. The potential for nonresponse bias is especially worrisome in surveys of political elites, who sometimes have low response rates (e.g., McClosky, Hoffmann, & O’Hara, 1960; Hedges, 1984; Whitehead, Blankenship, & Wright, 1999; Goldstein, 2003).

Low response rates do not indicate bias in estimates per se (Groves, 2006). To assess the potential for bias, the survey researcher needs external information gathered independently of response decisions. For example, in some surveys it is possible to check the similarity assumptions partially by comparing known demographic data for the nonrespondents and respondents. Brehm (1993) compares the results of the General Social Survey (GSS) to the Current Population Studies (CPS) and finds that proportionately fewer youth, whites, and males respond to the GSS. Abraham, Maitland, and Bianchi (2006) perform a similar analysis in analyzing the American Time Use Survey.