Step 1

Step 2

Step 3




Reported issue attitudes

Received partisan cues (via video format)

Party Cues Control

Viewed advertisement Reported issue attitudes (no advertisement)

Received partisan cues (embedded in ad)

Reported issue attitudes

occurs when defenses break down. More specifically, this experiment induces conflict between individuals' party identities and the issue positions they espouse in order to understand how partisans go about reconciling this psychological inconsistency. The design is laid out in Table 2.1.


To induce identity conflict, subjects viewed an issue advocacy advertisement endorsing a fictitious bill. In one condition, subjects were provided with partisan cues telling them which party supported (or opposed) the bill. In another condition, subjects were not provided with party cues until after they had taken a position on the bill. Therefore, if they took the position advocated in the appeal, they discovered that this position conflicted with their party identity. In a control condition, subjects viewed no advertisement at all. Again, the idea here was to create pressure for partisan change in order to determine how subjects react to such pressure.


A total of 254 student participants were recruited during the fall of 2005 from the University of Michigan campus. Participants were offered 5 dollars to participate in a 30-minute public opinion study. No more than 12 subjects were allowed to participate at any one time, and usually only 1 to 4 were present in the laboratory. Results from seven subjects were excluded from analysis after it was discovered that they were not citizens of the United States. Given the political leanings of Michigan students, the stimuli were designed to target Democratic identifiers. Therefore, self-identified Republicans were excluded from analysis once it was determined that party identification had not been affected by treatment exposure.


The treatment was administered via a political advertisement created by the researcher. In an informal test, viewers were unable to distinguish the advertisement from an authentic political appeal. The advertisement appeared to be sponsored by the AFL-CIO and focused on a fictitious “Bankruptcy Abuse Bill." The Enron, Global Crossing, and WorldCom scandals had recently received considerable media attention, whereas legislation related to the issue had received much less attention. The appeal advocated passage of the Bankruptcy Abuse Bill and paired Enron, Global Crossing, and WorldCom job loss statistics with dramatic audio and visual effects to rouse the viewer. Clips borrowed from actual political advertisements showed an apparent corporate executive pleading his Fifth Amendment rights in front of the U.S. Senate. Another clip showed an apparently middle-class man rubbing his forehead as he looks over his bills at the kitchen table. A transcript of the advertisement with screen shots is included in the Appendix (see Appendix Figure 2.1).

Information about the parties' positions came in the form of three fictitious newspaper quotations. Each set of quotations commented on either Democratic opposition to the bill or Republican support for the bill. Because the stimuli were targeted toward Democrats, the quotes were attributed to sources Democrats would likely find credible.

By opposing the bankruptcy abuse bill...

“Democrats let corporate crooks off the hook"—The Washington Post “Democrats are simply wrong on bankruptcy abuse"—The Boston Globe

“Democrats are playing politics with people's lives"—The New York



By supporting the bankruptcy abuse bill.

“Republicans are keeping the heat on corporate crooks"—The Washington Post

“Republicans are right on bankruptcy abuse"—The Boston Globe “Republicans are putting people above politics"—The New York Times

Within each of the three experimental conditions, Democratic opposition and Republican support quotations were randomized to ensure that any effects would be attributable to inconsistency between issue attitudes and party identification and not to support versus opposition framing.


Initially, subjects were randomly assigned to one of four treatment groups or the control group. As mentioned earlier, the Republicans support and

Democrats oppose groups were consolidated. This design yielded three experimental conditions: a party cues condition, a disagreement condition, and a control condition.[1] As described previously, those assigned to the party cues condition viewed a version of the advertisement in which the party cues were embedded. Those assigned to the disagreement condition viewed the same advertisement but with the partisan cues removed. Subjects in this condition received these cues only after viewing the advertisement and reporting their support or opposition for the Bankruptcy Abuse Bill. Cues were presented in onscreen text in exactly the same format and for the same amount of time in both conditions. The control group viewed no advertisement at all.

Those assigned to the disagreement condition (in which participants received no party cues until after taking a position on the bill) were expected to experience the greatest partisan dissonance. It was expected that subjects assigned to the party cues condition (in which participants received party cues during the advertisement) might also experience dissonance but that most would avoid it by following the available cues. The control group served as a baseline—no threat—condition against which the other groups were compared.


Opinions on the Bankruptcy Abuse Bill were assessed on a seven-point scale ranging from strongly oppose (-3) to strongly support (3) with a neutral point at zero. Party identification was measured using the standard ANES (American National Election Studies) branching question yielding a seven-point scale ranging from strong Republican (-3) to strong Democrat (3). Partisan feeling thermometers allowed subjects to rate how warm or cold they felt toward the Republican Party and the Democratic Party. Smaller values correspond to colder (more negative) feelings, and larger values correspond to warmer (more positive) feelings. Feeling thermometers were rescaled to run from -50 to 50.

Four additional questions were posed to each subject: “Regardless of who you tend to vote for, how often do you find yourself supporting [opposing] what the Democratic [Republican] party stands for?” Subjects were asked to answer by placing themselves on a seven-point scale ranging from “never" to “always." In contrast to the bipolarity of feeling thermometers, these questions allowed subjects to express occasional support and occasional opposition for the same party.

An open-ended question was administered near the end of the study to allow subjects a chance to explain, in their own words, why they identified with a particular party. This item read, “You have already indicated that you usually think of yourself as a Republican, Democrat, Independent, or something else. In a few sentences, please explain why you consider yourself a Republican, Democrat, Independent, or something else. The study is nearly complete, so feel free to take your time." Responses to this item ranged from 1 to 327 words and were blind-coded so that there would be no way to determine which subjects had been assigned to which condition. This simple coding scheme was meant to capture lesser of two evils identity justification. Subjects who flatly stated that they identified with the lesser of two evils or that their identity was primarily based on negative attitudes toward the opposition party were coded as lesser of two evils identifiers. Also included in this group were subjects who made less overt statements yet explained their identity largely in terms of their negative attitudes toward the opposition (rather than positive attitudes toward their favored party). All other subjects were coded as “0" to create a dummy variable for lesser of two evils identity justification. Dummy variables called disagreement and party cues were created to correspond to treatment conditions. Each of these variables was coded “1" for that particular treatement condition and “0" for either of the other conditions.


Based on the standard seven-point measure of party identification, no partisan differences emerged between cells [F(4, 244) = .03].[2] Nor did any significant pairwise differences emerge. Group means ranged from .75 to .88 on a scale from -3 to 3. As previously mentioned, this study was designed for Democrats. Because party identification appears to be very stable across groups, Republicans could be excluded from further analyses without concern. More specifically, analyses were restricted to those who labeled themselves as a strong Democrat, weak Democrat, Independent-leaning Democrat, or Independent not leaning toward either political party. This


allowed me to focus on those subjects for whom the treatment was likely to be threatening.

Examination of the subjects' opinions on bankruptcy abuse indicated that the manipulation worked largely as expected. Of the 153 non-Republican subjects assigned to one of the two treatment conditions, 150 were able to correctly identify the Bankruptcy Abuse Bill as the issue on which the advertisement focused. Findings also indicate that subjects in the disagreement condition (mean (M) = 0.90, standard deviation (SD) = 1.53) and those in the control group (M = 0.89, SD = 1.29) expressed an almost identical amount of support for the bill. This suggests that, although subjects overall were supportive of the bill, the advertisement itself did not prove to be particularly persuasive. However, because subjects tended to support the bill anyway, this is of little consequence for the manipulation. Despite the weakness of the appeal, the vast majority of the subjects in the disagreement condition did express opinions inconsistent with those of their party. Moreover, those in the partisan cues condition seem to have followed those cues and avoided disagreement with their party—although they were apparently not able to avoid it completely because, on average, they still showed support for the bill. This means that we should expect to see some degree of party identity justification in this group, although probably not as much as in the disagreement condition. Finally, whereas those in the control group actually expressed support for the bill as well, they did not experience disagreement with their party because they were never provided with any partisan information (or other cueing information). In other words, the experimental manipulation appears to have operated as it was designed.

Given that disagreement was successfully evoked between subjects and the Democratic Party, yet no change in party identification appears to have occurred, attention can now be directed toward the original question of interest: Is the observed partisan stability a result of motivated identity justification, or was the stimulus simply not powerful enough to produce party identification change? To answer this question, I first examined responses obtained through an open-ended measure in which subjects were asked to explain why they identified with their party.

Open-ended responses were also coded for whether or not they contained evidence of a lesser of two evils justification. When participants' attitudes approach the indifference threshold (see Figure 2.1), they are expected to avoid crossing over by calling to mind negative attitudes and stereotypes of the opposition party to offset negative attitudes toward their own party. Because the justification process is not necessarily expected to occur on the conscious level, this is an extremely blunt measure. Therefore, observations of such overt lesser of two evils identity justification are expected to be low across conditions. Nonetheless, Figure 2.2 shows that subjects in both the disagreement condition and the party cues condition were significantly more likely to use lesser of two evils justifications to explain their identity than those assigned to the control condition. In fact, not a single subject in the control group was coded as having used a lesser of two evils identity justification.

I next examined attitude correlations within each of the three experimental conditions. More positive (less negative) correlations between party feeling

Cell entries represent percentage of responses coded "1" for lesser of two evils justification

Figure 2.2 Cell entries represent percentage of responses coded "1" for lesser of two evils justification. Significance levels are based on Fisher's Exact Tests—a non-parametric test used when dealing with dichotomous dependent variables and very low positive outcome frequencies. Comparisons are made with respect to the control group.

thermometer ratings indicate reliance on lesser of two evils identity justification. Again, what may appear in cross-sectional snapshots to be multidimensional party identification (Alvarez, 1990; Valentine & Van Wingen, 1980; Weisberg 1980) may actually be party identity justification. I expected that in the absence of disagreement, attitudes toward parties would line up on a single dimension but when inconsistencies arose between issue attitudes and party identification, individuals would venture off of this dimension and begin to line up on a second lesser of two evils dimension. Again, this should be reflected in increasingly positive (less negative) feeling thermometer correlations from the control group to the party cues condition to the disagreement condition.

The results in Table 2.2 fit this prediction extremely well. In the control group, the standard expectation of a large negative relationship appears to hold, but this is not the case in the other two conditions. As party identity threat intensifies from the control group to the party cues condition to the disagreement condition, the negative relationship between attitudes toward the Republican and Democratic parties becomes smaller in magnitude and actually passes zero to become (nonsignificantly) positive.

Are these more positive (less negative) correlations arising because individuals venture out onto a new (lesser of two evils) attitude dimension in order to avoid crossing over the indifference threshold? Factor analysis was used to answer this question, because it makes it possible to determine whether these correlation differences arise as a result of partisans' attitudes splitting into two orthogonal dimensions. If this is the case, a new justification dimension should begin to emerge as party identity threat increases from condition to condition. On this new dimension, liking one's own party less should correlate with liking the other party less as well. Factor analyses were conducted separately for each experimental condition to facilitate comparisons between conditions.


Correlation Between Attitudes Toward Republicans and Democrats

Control (No Disagreement)


(n = 35)

Party Cues


(n = 71)



(n = 77)

Note: Significance levels are calculated relative to the control group. The control group itself is significantly different from zero (p < .001).


Control (n = 34)

Party Cues (n = 68)

Disagreement (n = 75)









Support Democrats





Oppose Democrats


- .408

- .700

- .258

Support Republicans





Oppose Republicans




- .401

Democrats Thermometer





Republicans Thermometer


- .376

- .541


Party Identity Strength










Variance Explained





Variance Explained by Extracted Factors




Note: Eigenvalue cutoffs are set to 1.0 (Kaiser's rule). Cattel's scree test yields the same number of factors, as indicated by the Eigenvalues. Extractions are based on principal axis factoring.

Included in these factor analyses were feeling thermometer measures, party support and opposition measures, and strength of party identification. Results are displayed in Table 2.3.

As hypothesized, a single factor emerged in the control group. On this dimension, liking Democrats more means liking Republicans less—just as conventional wisdom suggests. This factor explained slightly more than 66% of the total variance within the condition. A single-factor solution also emerged in the party cues condition. However, this factor accounted for only about 48% of the variance in this condition—suggesting that subjects' attitudes no longer lined up as well on a single dimension. Finally, as predicted, two orthogonal factors emerged in the disagreement condition.[4] Before rotation, the first of these two factors was identical to the factor extracted in each of the other two conditions. As in the party cues condition, this factor explained about 48% of the total variance. Looking across conditions, it appeared that this factor explainedless variance as partisan dissonance increased.[5] On the second factor, liking one party less corresponds to liking the other party less, and liking one's party more corresponds to liking the other party more. In a cross-sectional survey context, it is understandable how this dimension might be interpreted as a unique independence or ambivalence dimension (Alvarez, 1990; Dennis, 1988; Greene, 2000; Kamieniecki, 1988; Valentine & Van Wingen, 1980; Weisberg, 1980). However, given that this second factor arose only in the disagreement condition and that strength of party identification still loaded onto the first factor, this experiment suggests a very different interpretation: identity justification. In sum, as party identification came under threat, partisans departed from zero-sum attitude reinforcement. As they approached indifference between the two parties, they called to mind negative evaluations and stereotypes of the opposition party in order to justify maintaining their party identity.

  • [1] Comparisons between the Republicans support and Democrats oppose groups—both of which conflict with Democratic identity—showed that effects of bothstimuli ran in the same direction, although Democrats oppose tends to producelarger effects, as one might expect.
  • [2] The reported F-statistic comes from a one way analysis of variance (ANOVA).
  • [3] p < .01, **p < .05, *p < .10
  • [4] . Eigenvalue cutoffs were set to 1.0 (Kaiser's rule), although the number ofunique factors was very clear in each condition. Cattel's scree test yielded thesame number of factors. In the control group, the first factor had an Eigenvalueof 4.624, compared with a value of .692 for the second factor. In the party cuescondition, a second factor with an Eigenvalue of .953 approached the thresholdof 1.0. However, this factor still explained relatively little variance compared to
  • [5] After rotating the matrix to maximize the variance of loadings for each factor, the two dimensions extracted in the disagreement conditions came to reflectattitudes toward the Democratic Party and attitudes toward the RepublicanParty, respectively. Strength of party identification was associated with both ofthese dimensions (negatively with the opposition party dimension), albeit morestrongly with the favored party dimension.
< Prev   CONTENTS   Source   Next >