COGNITIVE RESOURCES AND IDENTITY CHANGE

As explained in the preceding chapter, I conducted an experiment during May 2007 in which I manipulated disagreement with one's party as well as the cognitive resources available at the time of that disagreement. Partisan motivation was expected to drive identity justification, enabling partisans to reconcile the inconsistency between their party identity and their issue positions without changing their identity to reflect the disagreement. However, when cognitive resources were inhibited through the induction of cognitive load, participants were expected to have a more difficult time constructing a justification for continued identification with their party. The findings reported in Chapter 3 confirmed these expectations. This chapter investigates whether participants' inability to construct an identity justification resulted in party identification change.

In addition to the experimentally manipulated factors, I examined three additional variables likely to be associated with one's ability to justify maintaining one's party identity: age, political sophistication, and Democratic identity. Age is a well-known correlate of partisan stability, but several competing theories have been proposed to explain this empirical phenomenon (see Sears & Levy, 2003). The dual motivations theory suggests that older, more experienced partisans will exhibit greater identity stability because they are better equipped to justify their identity. Therefore, older partisans should be more resistant to identity change even in the face of disagreement with their party.

Likewise, individuals with higher levels of political sophistication are expected to have an easier time developing justifications for their identity. Therefore, they should also show greater identity resilience. Finally, as seen in the previous chapter, the political climate in May 2007 seems to have made it easier for Democratic identifiers than for Republican identifiers to justify their identity. Therefore, Democrats should also show more resistance to identity change even after disagreeing with their party.

Still, no one's cognitive resources are limitless, so the addition of cognitive load would be expected to challenge the resistance of even the most resourceful partisans. Figure 4.1 outlines the expected pattern of interactions.

This investigation assumed that the ability to justify one's identity exists on a continuum. Although cognitive load, age, political sophistication, and Democratic identity should all affect where an individual falls on this continuum, it is difficult to know at what point justification will give way to partisan change. For instance, partisans of all ages may be able to avoid updating their identity in response to disagreement as long they have sufficient cognitive resources (see Figure 4.1, Panel A). However, once cognitive load is induced, partisan updating may occur among younger partisans, whereas older partisans may show continued stability (see Figure 4.1, Panel B). On the other hand, disagreement may lead the least politically sophisticated partisans to change their identity even if their cognitive resources are unencumbered, while their more sophisticated counterparts may avoid identity change as long as they are not placed under cognitive load (see Figure 4.1, Panel B). After being placed under cognitive load, however, even the most sophisticated partisans may begin to update their identity in response to disagreement (see Figure 4.1, Panel C). In short, partisan stability should first give way among the most vulnerable. Those with greater defensive capacity should hold out longer until their defenses are eventually overtaken as well. Vulnerability may stem from youth, low political sophistication, an unaccommodating political environment, or a lack of cognitive resources. If all four variables converge on the expected pattern, the likelihood is reduced that any one alternative explanation will remain plausible.

Figure 4.1 Each panel represents a stage in the hypothesized progression of party identification change. Those with the fewest cognitive resources are the first to change their party identity in response to disagreement, but they are eventually matched by those with more cognitive resources. The panels are not meant to reflect actual patterns in the data but merely predictions about what these patterns will look like.

All analyses were conducted using ordered probit regression to account for potential variation in the size of intervals between levels of party identification strength.¹ To facilitate interpretation, results are reported as predicted probabilities. More specifically, I present the probability of starting out as a strong identifier at time t-1 and remaining a strong identifier at time t. Ordered probit regression tables with cut points are included in the Appendix for Chapter 4.

Measures

A variable measuring party identification strength prior to the stimulus was created by simply folding the standard seven-point party identification measure in half. The new scale runs from 0 to 3. Strong Democrats as well as strong Republicans were coded as 3; weak partisans were coded as 2; leaners were coded as 1; and pure Independents were excluded from analysis because they have no party with which to disagree. As described in Chapter 3, YouGov/ Polimetrix uses sample matching to obtain nationally representative samples and maintains a party identification on record for everyone in their respondent pool. Using these data, it is possible to avoid the inevitable biases that would result if party identification were obtained immediately before the treatment.

Current party identification strength served as the primary dependent variable of interest in this study and therefore was measured after exposure to the stimulus. It was coded identically to past party identification strength except that those partisans who crossed over from one party to the other between time t-1 and time t were coded as having zero strength.^{[1] [2]}

I measured political sophistication with the use of a seven-item battery. Therefore, the political sophistication variable runs from 0 (none correct) to

7 (all correct). This battery contains a variety of multiple-choice questions about political figures and institutions. More specifically, respondents were asked to identify the jobs filled by Nancy Pelosi, John Roberts, Tony Snow, and Tony Blair at the time of the study. Participants were also asked how many votes it takes to override a veto, which branch has the power to determine whether a law is constitutional, and which branch has the power of the purse. Surprisingly, the mean level of political sophistication was quite high (mean (M) = 5.43, standard deviation (SD) = 1.91), but there is substantial variation across the scale.^[3]

Results

Findings presented in Table 4.1 indicate that at least some partisans did change their party identity from pre-treatment to post-treatement. However, the question remains as to whether treatment exposure caused this variation (or at least some of it). I began by examining the effects of disagreement on

TABLE 4.1. CROSS-TABULATION OF PAST PARTY IDENTITY AND CURRENT PARTY IDENTITY

Note: Percentages are calculated so that rows sum to 100%.

party identification strength while controlling for past party identification strength (see Appendix Table 4.1). The results indicate that when cognitive resources were unencumbered, disagreement had no effect on party identification strength. Thus, the results appear at first to be consistent with the Michigan model, which portrays party identification as a stable socialized identity. However, when subjects were placed under cognitive load, the results appear more intriguing. In this group, a reasonably large coefficient (B = -.268; SE = .174) emerged for disagreement, but this effect narrowly missed the standard threshold for statistical significance (p = .12).^[4]

As discussed earlier, factors such as age, political sophistication, and Democratic identity were also expected to affect the ability to justify one's identity. When these sources of variation were included in the model, the effect of disagreement on party identification strength reached marginal significance among participants placed under cognitive load (p < .10). When cognitive resources were unencumbered, the coefficient on disagreement remained small and nonsignificant, as predicted. The predicted probability of maintaining a strong party identity is displayed in Figure 4.2. In the absence of disagreement, those placed under cognitive load had a 79% chance of maintaining a strong identity from pre-treatment to post-treatment. However,

Figure 4.2 The figure illustrates predicted probabilities obtained from the ordered probit regressions with controls reported in Appendix Table 4.1. For the purpose of prediction, pre-test party identification strength was set to 3, age over 35 years to 1, knowledge to the mean, and Democrat identification to 1. The results represent the probability of maintaining a strong party identity from pre-treatment to post-treatment.

when disagreement was experienced under cognitive load, this probability dropped to 69%.

Although the effect of disagreement under cognitive load is not enormous, it is noteworthy given the reputation of party identification as the “unmoved mover” in political behavior research. Consistent with the dual motivations theory, significant partisan weakening occurred only when cognitive resources were scarce (see Figure 4.1, Panel B). Still, the result is only suggestive, because the moderating influence of cognitive load was not statistically significant. In other words, party identification change was established in the cognitive load condition, but the possibility of an equally large change in the absence of cognitive load cannot be ruled out.

As explained previously, the theory implies that the effect of disagreement should be larger among groups who are less equipped to defend their identity, such as young people, those who lack political sophistication, and those whose justification abilities are impeded by the political environment (i.e., Republicans). We now turn to an examination of these potential moderating variables.

Age

Age is a well-known correlate of party identification stability, but this empirical regularity has been surrounded by theoretical controversy. Whereas some favor an “impressionable years” model, others favor a “lifelong openness” model (Sears & Levy, 2003). The former suggest that party identification is susceptible to change during an individual's impressionable years but crystallizes with age and exposure to successive campaigns (Alwin & Krosnick, 1991; Campbell et al., 1960; Sears & Valentino, 1997; Stoker & Jennings, 2009; Valentino & Sears, 1998). Once partisans reach their 30s, party identification becomes fully crystallized and is unlikely to change thereafter. However, revisionist scholars interpret the relationship between age and stability quite differently, arguing that as individuals gain experience, they become more confident in their assessments of where they stand relative to the parties. Thus, the need for partisan adjustment is reduced, and this is reflected in an increasingly stable party identity (Achen, 1992, 2002; Franklin, 1984). In other words, partisans may have a “lifelong openness” to identity change even though their party identity becomes more stable over time.

The dual motivations model suggests a third possibility. Age and experience may facilitate an individual's ability to justify his or her party identity. Therefore, stability increases with age, at least in part, because partisans get better at defending their identity against conflicting information as they gain experience.^[5] From this perspective, partisans are most susceptible to change during their impressionable years, yet they also possess a lifelong openness to change if their defenses give way.

To test this proposition, I analyzed the effects of age and disagreement in the presence and absence of cognitive load. For the purpose of this analysis, age was coded as a dummy variable in which those who were 35 years of age or younger were coded as a “0,” and those older than 35 were coded as "1." Age 35 serves as a rough cutoff for the end of the impressionable years life stage.

When the dummy variable "Over 35" was interacted with disagreement, no disagreement effects emerged among those whose cognitive abilities were unencumbered—even those younger than 35 years of age (see Appendix Table 4.2). However, among those placed under cognitive load, the story was much different. Partisans younger than 35 showed substantial partisan updating in response to disagreement, whereas their older counterparts remained significantly more steadfast in their identity. In other words, partisans younger than 35 are more vulnerable to disagreement but are able to resist updating as long as they have the cognitive resources to do so. It is also noteworthy that this effect seems to occur at the expense of past party identification, because the effect of past party identification on current party identification dropped off dramatically in the cognitive load condition, particularly among those who were age 35 or younger (p < .01) (see Appendix Table 4.3).

Looking back at the predictions made in Figure 4.1 (Panels A and B), the observed pattern appears to fit expectations extremely well. Disagreement had a much larger and more significant effect on younger partisans than on older partisans, but only in the presence of cognitive load (p < .05). In other words, cognitive load appears to be particularly disruptive to the justification process among those with the least political experience. This contrasts with existing models that portray age as a proxy for either life stage or information acquisition. These theories cannot account for different effects in the "cognitive load" versus "no cognitive load" conditions. The results presented suggest that partisans devote their cognitive resources to identity defense, and these resources are particularly critical for those still honing their defenses. The predicted probability of maintaining a strong party identity is categorized by age and displayed in Figure 4.3.

Once again, however, while the disagreement effects that emerged under cognitive load were significantly different from zero, the results were not significantly different from the null results obtained in the absence of cognitive load—although they were in the expected direction. This is indicated by the interaction between cognitive load and disagreement in the full triple interactive model (p = .23) (see Appendix Table 4.3). Again, we are left with results that are suggestive but not incontrovertible.

Figure 4.3 The figure illustrates predicted probabilities obtained from the ordered probit regressions reported in Appendix Tables 4.2 and 4.3. For the purpose of prediction, pre-test party identification strength was set to 3. Therefore, the results represent the probability of maintaining a strong party identity from pre-treatment to post-treatment.

• [1] As explained in more detail later, party identification strength serves as thedependent variable, rather than the standard seven-point measure of partyidentification. The reason for using this measure is that I was interested indetermining what weakens party identification, not in what makes a persona Republican or a Democrat. Therefore, a valence scale (strong identity to noidentity) is appropriate, rather than a bipolar scale (strong Republican to strongDemocrat). It is important to note that this is not the standard measure of partyidentification strength, because partisans who crossover from one party to theother are coded as having zero strength.
• [2] Tests were also run using a measure in which crossover partisans were codedas having negative identity strength; for example, a subject who identified asa Republican at t-1 and as a strong Democrat at time t was coded -3 to indicate strong identification with the other party. The results did not substantivelydiffer.
• [3] Prior and Lupia (2008) found that subjects scored substantially higher on political knowledge questions when given more time or monetary incentives forcorrect answers. This means that online studies such as those administered byYouGov/Polimetrix should yield substantially higher knowledge scores thantraditional telephone surveys or even face-to-face surveys in which interviewerswait for participants to respond. Prior and Lupia argue that such measures aresuperior because they capture the relevant skills necessary to acquire politicalinformation.
• [4] . The coefficient (B) and standard error (SE) reported in the text come from anordered probit model.
• [5] I acknowledge that this relationship is true only up to the point at which theexperience that comes with age is outweighed by the cognitive deterioration that comes with age. To avoid overcomplicating the model, age is simplydichotomized at 35 years in order to focus the comparison between those wholack political experience and those who do not, thereby avoiding the nonlinearity associated with cognitive deterioration in advanced age. When the modelwas run with age as a continuous variable and all respondents over the age of60 were excluded, the effects were almost identical.