Abstract
In recent years, surveys in the United States have faced increasing refusal to answer questions about firearm ownership, even as other similar questions see no comparable up-tick in item nonresponse. Asymmetrical polarization, elite messaging, and changing media institutions all suggest that the surging nonresponse concerning gun-ownership questions may be increasingly concentrated among those with rightward political and partisan leanings, potentially skewing inferences about gun-related issues. Data from the General Social Survey confirms that the increase in probability of declining to answer firearm-ownership questions is particularly stark among those identifying as Republicans, particularly those with a conservative outlook skeptical of government.
Notes
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
1 This question produces different overall estimates of gun prevalence than do some others (CitationLegault, 2013), but for assessing changing response patterns internal consistency matters most.
2 “Refused to answer” here excludes those who respond “Don’t know,” as these respondents may be expressing genuine uncertainty about whether a co-householder has guns. The processes that generate “Don’t know” responses, in any event, typically differ from those leading to refusal to answer questions (CitationRiphahn & Serfling, 2005; CitationShoemaker, Eichholz, & Skewes, 2002).
3 Respondents’ being willing to fabricate an unreliable answer may have different antecedents than does merely refusing to answer a question (CitationBlasius & Thiessen, 2015).
4 includes “partisan leaners,” those who profess to be independents but lean towards a particular party, as members of that party rather than as independents: most research suggests that leaners in fact generally are more like partisans than like true independents (CitationTheodoridis, 2017; CitationValentine & Van Wingen 1980). Coding leaners as independents instead preserves the basic results of , although the larger number of observations categorized as independents mean that independents appear less volatile.
5 Reported conclusions persist using other functional forms for time (e.g., decade fixed effects).
6 The income question does not adjust for inflation. Instead using income percentile among the year’s survey respondents does not substantially change results for partisanship–year interaction.
7 Using a conventional, single-level logistic model with clustered standard errors produces similar results, albeit with generally smaller standard errors on the coefficients of interest.
8 When, in 2000, the GSS as a one-off asked how often respondents sought political information via television, the correlation with the number of hours of television watched was less than 0.03.
9 In regressions without the quadratic terms, Party identification × year consistently attains statistical significance, though the confidence-in-government variable shrinks the effect size.