MULTIDIMENSIONAL IDENTITY, MEASUREMENT ERROR, OR JUSTIFICATION?
Various authors have argued that party identification cannot be adequately measured on a single dimension (Alvarez, 1990; Greene, 2005; Kamieniecki, 1988; Katz, 1979; Valentine & Van Wingen, 1980; Weisberg, 1980). Proponents
of such models often cite the small negative and sometimes positive correlations between “feeling thermometer” ratings of the Republican and Democratic parties (Alvarez, 1990; Weisberg, 1980). However, I argue that what appears to some as evidence of multidimensional party identification may actually constitute evidence of party identity justification.
First, as explained in the previous chapter, attitudes are conceptually distinct from identities (Green et al., 2002; Groenendyk, 2012; Rosema, 2006). Whereas attitudes are evaluative in nature (Eagly & Chaiken, 1993), identities are rooted in self-conceptualization (Monroe, Hankin, & Van Vechten, 2000). Therefore, departure from the assumed one-dimensional negative relationship between attitudes toward the two parties does not necessarily constitute evidence of multidimensional party identification. Instead, it may result from party identity defense. As attitudes approach indifference between the two parties, a second attitude dimension should emerge as a result of individuals' attempts to justify continued identification with their party. Lesser of two evils identity justification entails a positive relationship between attitudes toward the two parties. Individuals who like their own party less will come to like the opposition party less as well.
In arguing against the claim that party identification is multidimensional, Green (1988) pointed out that deviation from the expected strong negative correlation between party feeling thermometer ratings may result simply from measurement error. Random measurement error drives correlation coefficients toward zero, whereas systematic error or “charitability bias” might actually lead to a positive correlation (Green, 1988; Green & Citrin, 1994). In other words, some individuals are likely to have a more charitable nature and therefore to rate both parties higher than average, whereas less charitable individuals will rate both parties lower than average. Because this charitability trait dimension is an omitted variable, it may bias correlation coefficients between
party feeling thermometer ratings in a positive direction. After accounting for measurement error, Green showed that Republican and Democratic Party feeling thermometer ratings were more negatively correlated.
Although Green (1988) offered a strong rebuttal to the multidimensional partisanship literature, there remains a third potential explanation for this pattern. The type of correlational dynamism hypothesized in the discussion of lesser of two evils identity justification is indistinguishable from measurement error in a survey context. Just as variation in respondent charitability may drive the correlation between party feeling thermometer ratings in a more positive direction, so might the motivation to justify one's party identity. Therefore, what first appeared to be evidence of multidimensional party identification, and then appeared to be measurement error, may actually be evidence of party identity justification.
I take a multimethod approach to understanding whether the relationship (or lack thereof) between party feeling thermometer ratings is best explained by a multidimensional model of party identification, measurement error, or identity justification. Experiments provide researchers with leverage over both random and systematic error through random assignment. This ensures that differences between groups can only be attributed to the treatment and not to measurement error. By pairing these tests with aggregate level analyses, it is possible to assess how these microlevel processes manifest themselves at the macro level. When making aggregate-level comparisons within a single population over time, the effects of measurement error should be greatly reduced. In large samples, random errors should cancel out. Moreover, if each cross-section is truly representative of the population of interest, then aggregate charitability biases cannot explain differences between samples.
When partisans disagree with their party, they will attempt to justify their existing party identity.