Response bias in opinion polls and American social welfare

Abstract

Response bias and notably response falsification undercut the usefulness of opinion polls in characterizing collective American attitudes toward social welfare. The attitudes polled in surveys such as the General Social Survey (GSS) and the National Election Studies (NSS) appear to be customarily sensitive, often evidencing very large amounts of response falsification that obscure the degree of consensus for America's bifurcated social welfare system. Response bias and falsification greatly affect self-reports of family income, drinking and drug abuse, sex behavior, voting, and a large number of other attitudes central to the provision of social welfare. The problems of response bias may be intractable and attention might profitably return to traditional scholarship that attends to actual collective choices, e.g., legislation, as still imperfect but more reliable estimates of the national will.

Notes

1 Moreover, it is worth recalling that the Poisson income distribution covers a disproportionately greater number of respondents with each increment from low to mean income.

2 These annual differences between actual and reported turnout may overestimate the actual disparities for the following reasons: Census estimates of the eligible voting population did not exclude noncitizens until 1998; they do not exclude the small number who had lost their franchise. At the same time, NES nonresponses to the turnout question ran between about ten and fifteen percent in each year; they were excluded from the denominator (respondents) of turnout. To the extent to which nonrespondents were nonvoters who simply wished to withhold that information, the turnout reported in the NES declines. On the other hand, NES turnout as calculated in Table 2 might actually underestimate error by accepting the accuracy of those who said they did not vote when in fact some voters may have misremembered or falsified, perhaps not wishing to record any sensitive information with an untrusted data base.

3 An enterprising graduate student might probe American attitudes to the immigration to the United States of Abfaluzians, an entirely imaginary group. What percent of respondents would feel their pain, what percent would adamantly insist that those anti-American terrorists stay at home?

4 Random response techniques, ingenious devices to overcome bias in response to sensitive questions, add an unconscionable cost to polls and with uncertain results. The present amount of response bias takes place in even the best designed surveys. See in particular Tracy and Fox (1981) and in response Miller (1981) and Umesh and Peterson (1991).