Minority Party Loyalty Under Challenge

Transcription

1 Minority Party Loyalty Under Challenge Michael R. Wolf Indiana University Purdue University Fort Wayne Numerous observers and scholars have taken notice of the regional separation of party dominance in the United States between red states (Republican-dominated) and blue states (Democrat-dominated) in recent presidential elections. It is not just a matter of states being dominated by one or another party, however. The percentage of counties in America where one party wins a landslide in a presidential election has increased by 47 percent between 1976 and 2000 (Noah, 2004), and has raised concerns. Today, most Americans live in communities that are becoming more political homogenous and, in effect, diminish dissenting views (Bishop, 2004, p.a1). Despite the alarm many have toward such results, however, they are not necessarily new. Indeed, one prominent political scientist, Warren Miller, in 1956 concluded that one-party dominance was the norm in American politics rather than exception (Miller, 1956). Left unresolved, however, is how minority partisans handle the often futile situation of living in an environment that is not conducive to their party ever taking power. Given the challenge of holding a partisan attitude that is out of step with one s surroundings, how well does a minority partisan s identification explain his or her voting behavior and how does it compare to party identifiers who live where their party dominates? Do partisans remain loyal to their party even when it has such a small chance of winning? Or, do minority partisans, socialized by the dominant partisan milieu, end up leaning more toward the dominant party in their voting behavior than do partisans of the same party in areas where the party is competitive or dominant? Even if minority partisans remain loyal to their party, do their issue-positions differ from their counterparts in more competitive areas? Political scientists have generally followed three approaches when discussing how a citizen s surroundings mediate or do not mediate his or her individual-level party identification. 1

2 In particular, these approaches tackle the question of why people may continue to hold tightly to their partisanship even in the face of strong partisan opposition. The first approach follows the social-psychological model from The American Voter (Campbell et al., 1960). It posits that perhaps the individual is so psychologically attached to their party that they do not mind that their party will always lose. A second approach, by other authors, counters that the dominance of the majority party weakens the loyalty of minority partisans, and these minority partisans end up leaning toward the majority party in voting and in their issue positions. A third approach argues that an individual may feel under siege from the dominant party and vigorously defend their partisanship against attack. Each of these approaches is tested using the Cross-National Election Studies data for the United States in 1992 and for Germany in Finally, because of the way these data were collected, another approach may provide a key component in explaining how minority partisans retain their party identification under stress. Political behavior is more than just a simple response to one s political attitudes. Rather, people actively expose themselves to information that conforms to their preexisting attitudes and they also actively filter out information that conflicts with their political beliefs. Political discussion networks provide an outlet for citizens to may reinforce their political beliefs in a protective environment, nested within what is a more hostile outer environment (Huckfeldt et al., 1998; Finifter, 1974). Therefore, after the discussion of the contending approaches concerning the mediating influence of the partisan context voters live within, the remainder of this study provides an illustration of how minority partisans reinforce their partisan attitudes in a network of like-minded partisans, despite the stress their political beliefs face in what may be a hostile partisan environment. Individual-Level Party Identification and the Partisan Environmental Context The approaches discussed above draw different conclusions about the role of predisposition vis-à-vis the contextual influences of one s social environment. The first 2

3 approach, which may be called the primacy of partisanship approach, closely follows the logic of the social-psychological approach from The American Voter, where party identification as a psychological attachment outweighs other factors in voters attitudes and behavior. Research within this approach argues that partisanship influences both voting decisions and views on political issues regardless of the dominance or competitiveness of political parties in a particular area. Miller s (1956) study of one-party dominance provides the blueprint for the logic of this approach. Miller breaks U.S. counties into four categories according to the Democratic percentage of the vote to judge whether they were Democratic dominated, slightly Democratic, slightly Republican, or Republican dominated as a measure of the partisan political environment. He then merges these data with individual-level National Election Study data. Because Miller finds only a few significant effects of party dominance on motivational forces, his conclusion highlights the resilience of minority partisanship in the face of majority dominance. The incidence of Party Identification does not vary with changes in political environment, at least insofar as minority voters are concerned. It appears to be the one factor of those examined which sustains the minor party vote, independent of the partisan complexion of the local political environment (Miller, 1956, p. 725). Others have challenged the view that the social-psychological model is so stable. A key critique of the social-psychological approach comes from Fiorina (1981). Instead of party identification being viewed as an unmoved mover, Fiorina (1981) pictures it as a running tally of past partisan identifications that are updated with retrospective evaluations of political performance. This suggests that partisanship may not be a sufficient independent influence in its own right, that other factors can mediate the influence of party identification if not promote change in party identification itself. A second approach, which could be referred to as the partisan contextual approach, relates to how one s party identification and political attitudes can be mediated by individual s surrounding environment. Others have directly noted how the partisan context of a specific 3

4 region may intervene in the relationship between party identification and political behavior. Butler and Stokes (1971) find that British working-class voters are abnormally supportive of the Conservatives in areas with a high concentration of middle-class citizens, while middle-class voters are abnormally supportive of Labour in areas with a high concentration of working-class citizens (Butler and Stokes, 1971, pp ). More specifically, the partisan contextual approach follows Putnam s (1966) findings. With U.S. counties again as the basis of aggregation, Putnam uses a similar research design as Miller s but finds that individual-level motivations and attitudes are malleable to the extent that an individual is exposed their community. Membership in primary and secondary social associations affects the political behavior of individuals; the more the individual is integrated into their community, the more their political attitudes reflect those of their community generally. So this counters Miller s findings and the basis of the primacy of partisanship approach, while still using a very similar method. The third approach, the threat approach, relates to the partisan contextual approach, but suggests that minority partisans identify more closely to their party when they face a majority party dominating their community. The threat these minority partisans feel has been captured in political behavior research in the past. Key (1949) and Wright (1977) both distinguished between different voting behaviors among American southern whites in the era of intense debate over segregation. Both found that whites who lived in counties with higher percentages of black populations tended to vote for candidates promoting racial segregation over other issues with a higher frequency than those did whites who lived in counties with lower percentages of blacks. These voters attitudes followed the contextual mood of their surroundings more than individuallevel features would suggest, and in turn they decided to ratchet-up the strength of their attitudes in opposition to what they viewed as a threat. Bowler, Lanoue, and Savoie (1994) make a similar argument but use the presence or dominance of an ideological party as their threat. They propose that the presence of a very 4

5 ideological opposition party may force minority party adherents to identify with their party more strongly. We argue, therefore, that the strength of partisan loyalty is influenced not just by personal experience and socialization (as is usually argued) but also by the nature of the party system. Similar voters with similar backgrounds may differ in their partisan attachments because of differences in the electoral choices with which they are presented. (Bowler et al., 1994, p. 993). Therefore, compared to their counterparts in areas where their party is dominant or where there is no dominant party, the threat of this ideological party should create stronger attachments to their party. A final approach may rectify all of these models. Instead of concentrating solely on the mediating contextual influences on individual-level behavior by voters surroundings, the partisan discussion approach has viewed how political discussion networks serve as a cushion against an inhospitable partisan environment. Like information in other realms of our lives, discussion networks provide important information on recipes, good movies, and workplace gossip about what to do and what not to do to succeed. Much of our political information also passes through our interpersonal communication networks. While disagreement on political matters cannot be purged from discussion networks (Huckfeldt et al., 2004), people tend to seek out discussion partners who share their political attitudes to reduce dissonance (see Festinger (1957) on dissonance and Huckfeldt et al., (2004) for discussion on contending arguments on disagreement within discussion dyads and networks). Thus, the influence of one s personal network may reinforce one s prior political attitudes if it takes place among people with similar attitudes. On the other hand, if a person lacks reinforcing information from political discussion, and receives discordant political messages from their environment, their attitudes could shift. Given the overwhelming presence of opposing arguments, three different reactions could be taken by minority partisans. First, minority partisans might suppress voicing their attitudes and contribute to a spiral of silence that overstates the level of the majority s dominance (Noelle- 5

6 Neumann, 1984). Second, individuals with minority opinions might seek each other out and segregate themselves from those with the majority opinion (Finifter, 1974). Finally, the individual might feel threatened by the dominance of the other party and become more attached to their partisan attitudes, as argued by the threat model above. These latter two reactions could occur in concert. Finifter (1974) studied the pattern of a Republican friendship network among Detroit autoworkers whose political attitudes were in the extreme minority given that they were labor union employees in a strongly Democratic city. The conclusion that minority partisans used their prior-held political attitudes to filter their political conversation network to protect against a disagreeable partisan environment fits well with the conditions minority partisans may face more generally. The more individuals are willing to censor their political communication, and share their opinions only those with whom they agree, the more they reduce the influence of what may be a hostile partisan environment (see Huckfeldt et al., 1998, p.998). Indeed, this political discussion network may protect their attitudes from the external context. There are many reasons why voters seek out fellow partisans to reinforce their attitudes in political discussion. Lodge and Taber (2000) explain how reasoning is motivated by previously held affective attitudes. Rather than integrating political information on a clean slate, one s previous affectively charged attitudes come to mind the moment the individual considers new information. The processing of the information is biased by the previously held emotional attitude. Consequently, it would not be surprising to find that party identifiers actively filter out opposing views from their discussion networks, which would in turn strengthen their attitudes and make it less likely to vote against their party. These four approaches leave many questions to be answered. The first is whether individual-level partisanship and attitudes sufficiently explain voting behavior as the partisan primacy model suggests, or whether individual s behavior is mediated by their surroundings. Secondly, if the individual s behavior is mediated by surroundings, how does the mediation work? Do an individual s attitudes become similar to the surrounding partisan environment as 6

7 posited by the partisan contextual model? Or, as the threat model argues, does the environment lead to a negative reaction to a perceived threat? Third, is the individual an active agent choosing what they are exposed to rather than a passive recipient of environmental influences? This is the key: how do minority partisans and other partisans for that matter discuss politics in a way that perhaps guards against other political views? To test each of these propositions, data from the 1983 German Panel Election Study, from the 1990 German, and from the 1992 American studies of the Cross National Election Studies are used that distinctively fit these models. The Individual and the Local Atmosphere of Competitive Party Politics To properly specify the mediation of one-party dominance on the voting behavior of minority partisans, individual-level and contextual variables are used together in a measure very similar to what Miller (1956) and Putnam (1966) used to capture the complexion of partisan dominance. The measure (context) reflects a community s partisan competitiveness. For the U.S. case, the aggregate vote outcome of the respondent s county reflects this partisan context. This is broken down further into a seven category variable depending on how well each major party did in the two-party vote. The categories range from the strongest Republican to the Strongest Democrat counties. 1 For Germany, the contextual proxy vote outcomes are presented in Regierungs Bezirke, which are merged with individual-level data. These are administrative districts roughly analogous to counties in the sense of being a relatively low level of aggregation. Instead of vote outcomes, the context for the Regierungs Bezirke are traced vote intention results throughout the election year from monthly German Politbarometer data. 2 The categories of these districts do not match perfectly with the U.S. case, but are similar enough, and range from: Very CDU, somewhat CDU, neither party, somewhat SPD, and Very SPD. 3 4 The individual level variables match Miller s notion of individual motivation: the respondents self-reported partisanship and issue partisanship. The former is self explanatory, and the latter is issue-voting consistent with one s party s positions. For Germany, partisanship is broken into four variables: CDUpart, FDPpart, GRNpart and Othpart, which reflect 7

8 identification with the CDU, FDP, Greens, and other smaller parties and independents. This is so the SPD serves as the unspecified comparison category for these dichotomous variables. Similarly, the U.S. case uses Repub and Indep as dichotomous identification variables that produce Democrats as the comparison category. This allows a direct comparison between the major parties in each country. A dichotomous variable for both Social Democrats in Germany and Democrats in the U.S. are produced but only as the basis for interactive variables created with the context variable. Issue partisanship is the respondent s evaluation of his/ her own position and the major parties positions on important political issues. It captures the difference between the individual s ideologically arrayed opinion toward an issue and that individual s placement of the CDU and SPD in Germany on the same issue, and between the individual and Bush, Clinton, and Perot in the U.S. on the issues. The types of issues are quite similar between cases as well. 5 To capture the possible mediation of party dominance on individual motivations, an interactive variable is created that captures the relative party dominance in each German administrative district or U.S. county (variable name: context) and individual-level party identification. The same interaction is produced between the context and individual-level issuepartisanship variable. Measuring the impact on voting attitudes provides a solid test of whether partisanship independently and sufficiently explains political behavior (as suggested by the primacy of partisanship approach), or whether it is mediated by context. According to the second argument the partisan context approach - a party s domination of a particular district or county should color individuals perceptions of issues and either attenuate or exaggerate the impact of that individuals party identification. Unfortunately, data shortcomings for Germany mean the respondent s stated second ballot vote intention serves as the dependent variable. 6 For the U.S. case, actual reported vote is used. Because the categories for both cases are nominal, multinomial logistic regression models are used. This allows comparison of vote intention propensity for each party as directly opposed 8

9 to the propensity to vote for each other party. 7 The first model tests how issue voting is effected by partisan context through the interaction between party context and issue partisanship and its influence on vote intention. (See Appendices 1 and 2 for more specific equations and operationalizations for the German and U.S. cases). The model includes party-issue distances (issue partisanship), the interactions between party dominance and party-issue distance, and individual party identification. For the partisan context to mediate a party identifier s issue positions and the party s issue positions, the interaction of context and issue positions would need to be statistically significant. Table 1 displays the results from the equation for the German case. The results of the model illustrate that district partisan environment does not affect issue partisanship and its influence on voting behavior. The interaction between context and the issue positions are never statistically significant except for the bizarre combination of partisan context and distance from the SPD s position influencing the vote choice between the CDU and Greens. Indeed issue voting does not emerge independently because the respondent s party identification typically dominates the vote intention. Similarly in Table 2, partisanship independently dominates the vote choice. There is some statistical significance of the interaction between the context and issue proximity in helping decide the vote for Clinton rather than Perot. In other words, the partisan strength of the county affects whether one s issue stances help them decide between Clinton and Perot. This is an interesting finding context can alter how issue positions lead to voting between partisans and an independent candidate, but this is not the case in the far more mainstream choice of Bush vs. Clinton. In both cases, partisanship is statistically significant and alone mainly explains the vote choice. Consequently, issue partisanship is not significantly altered by partisan context. <<TABLE 1>> <<TABLE 2>> Issue positions are not mediated by partisan context, but is the influence of party identification on voting affected by one s partisan surroundings? In a Republican-dominated 9

10 county or CDU-dominated district, the reader may expect that a Democrat or a Social Democrat will vote less often with their party. For each major party, an interactive variable between one s party identification and the partisan context of one s environment is created (see Appendix 3 for specific model information). If partisan context is influential, then the partisan contextual approach is supported. If the primacy of partisanship approach holds, one would expect party identification to influence voting or vote intention independently; the interaction with one s surroundings will have little or no influence. For the German case, the latter holds as seen in Table 3. None of the interactive variables are statistically significant, while the party identification variables predict the vote choice independently. The exact same findings occur for the U.S. case in Table 4. The nature of one s county does not alter the influence of party identification on one s voting behavior; the local partisan mood does not translate into changed individual motivations. Partisanship alone structures the respondents vote choices. The primacy of partisanship model trumps the partisan contextual model. <<TABLE 3>> <<TABLE 4>> Thus far, the findings suggest that partisans are as strongly partisan in areas where their party has little chance for success as they are in areas where the party is competitive or dominant. Perhaps these partisans who live where the opposition party dominates, as suggested by the threat model, feel threatened by the opposing party, especially if that party is ideologically extreme. As a result, these voters partisanship is actually strengthened. This cannot be measured in the U.S. case because no such extreme ideologically threatening party exists, but Bowler et al. (1994) argue that in the German case particularly Bavaria because the Bavarian Christian Socialist Union is more ideologically extreme than its sister party (the CDU) in the rest of Germany so Bavarian SPD identifiers may feel threatened. Bowler et al. (1994) use the 1983 German Election Study to measure whether Bavarian members of the SPD are more partisan due to this threat than other SPD partisans. 8 They use an ordered probit model because their dependent variable is ordinal with three categories: the 10

11 weakest SPD partisans, moderate SPD partisans, and the strongest SPD partisans. Their main variable of interest is a dichotomous variable that distinguishes between Bavarian and non- Bavarian SPD partisans. 9 They also include control variables for the typical partisan socialization theories: sex, age, education, previous vote, and whether a voter is Catholic. Their model is reproduced in Table 5. As Bowler et al. (1994) argue, Bavarian SPD members are more partisan than are non- Bavarian SPD members. Although the authors did not compute them, the predicted probabilities of partisan strength for both SPD partisans from Bavaria and for those not from Bavaria are included at the top of Table 5. This bolsters the authors argument because the probability that a Bavarian SPD partisan is strongly partisan is.94 compared to a.48 probability for non- Bavarian SPD partisans being strongly partisan. 10 Bowler et al s findings and the additional analyses provided appear to demonstrate the strength of the threat model. A major problem remains, however. Because of the way they had to operationalize the data to represent their theory, there were few respondents remaining in the data set (78 cases). This is not an overwhelming problem for Bowler et al. (1994) because Germany is one of three cases in their study and the similar findings in Australia and Canada reinforce the consistent yet limited data in the German case. As a single case, however, there is no way to truly conclude that this threat significantly influences party strength given problems in the data. <<TABLE 5>> There is an enormous distinction in strength of partisanship between Bavarian and non- Bavarian SPD members, but there are only two Bavarian respondents who claim that they very strongly lean toward the SPD. This was computed by cross-tabulating the original five category ordinal variable of strength of partisanship from the 1983 data set by the dichotomous variable Bavarian. There are only eleven cases of very strongly leaning SPD partisans of the 78 cases left in the data set and 9 are not in Bavaria. The lack of cases - two of eleven very strong partisans are in Bavaria takes from the findings. 11 Bowler et al. (1994) have provided a 11

12 promising argument concerning contextual influences on individual level partisanship due to a threat, but unfortunately, the limitations of the 1983 German Election Study do not allow firm conclusions about the contextual influences on partisanship. The threat model cannot be supported or refuted. Further, the lack of data in Bowler et al s Bavarian case means no firm conclusion can be drawn whether contextual influences exist, even though the evidence is suggestive. The data do not allow one to argue, to this point, against the conclusions of the primacy of partisanship model. Partisan Discussion Model There are other possible interpretations that might explain why the broader environment cannot be said to influence individual-level behavior. One test would be how a person s personal discussion network intersects with one s partisan contextual environment. An individual s political attitudes may lead them to construct political networks to reinforce their partisan attitudes, perhaps especially when one is in an inhospitable partisan environment. This argument provides the basis to the fourth approach: the partisan discussion approach. The model follows studies of social networks and their influence on individual motivations. As Finifter (1974) found in her study of Detroit-area Republican autoworkers during the 1960 election, Republican autoworkers picked other Republicans in reciprocal friendship dyads as frequently as had majority Democratic autoworkers picked Democratic reciprocal friendship dyads. The difference was, of course, that this occurred in an overwhelmingly Democratic context, which meant that this discussion among Democrats was similar to chance, whereas the likelihood of Republicans seeking out such Republican dyads was two-times greater than chance. Counter to her original hypothesis, workers with deviant partisan political attitudes made friends with each other and their propensity to form friendship groups was based on perceiving themselves as politically deviant. Thus, their prior attitudes were useful in choosing discussion and friendship partners, which stood in direct opposition to her hypothesis that political attitudes were influenced by surroundings. Instead, friendship patterns were 12

13 influenced by political attitudes; the hypothesis was turned around causally (Finifter, 1974, p. 623). Finifter hoped that others would provide tests of political attitudes in natural groups (Finifter, 1974, p. 614) and the data from the Cross-National Election Studies provides such an opportunity, though these data are imperfect because they are constructed ex post facto. Nevertheless, even with imperfect data, findings suggest that Finifter s argument may have validity given the curious finding above that minority partisans have as much partisan zeal in areas where their party never succeeds as well as where it typically succeeds. Further, Lodge and Taber (2000) might suggest that this is typical. Based on theories of motivated reasoning, people have biased ways of processing information. Consequently, people often follow directional rather than accuracy goals in information processing; rather than seeking the most accurate and objective information, some seek information that confirms their prior-held attitude whether consciously or not (Kunda, 1990). They discount or challenge opposing information by denial, or by selective exposure to information only that agrees with their beliefs. Given this, one would expect that minority partisans might seek each other out as will majority partisans. Further, they will avoid discordant political information. Following this logic, one would expect political discussion to be comprised mostly of fellow partisans. The German and American studies of the Cross-National Election Study allow such a measure of the frequency of political discussion and a number of characteristics of a survey respondent s discussion partners. Respondents could provide names of up to four people with whom they discussed important matters and if they discussed politics with each person. Further, they were asked if there were anyone else with whom they discussed political matters. This makes four important matters and one political discussant. 12 To measure the nature of political discussion in their network, partisans are broken down by their self-identified party, and they also report the party that they perceive that political discussant will have or did support in the election under study. The same partisan context measure used above is used again. 13 If 13

14 political conversation did not follow a biased process that is discussed by Lodge and Taber and illustrated by the Finifter study, then one would expect that those partisans who lived in a county or district dominated by their party would speak much more often with fellow partisans than would a similar partisan who lived in a county or district where they were a distinct minority. As Huckfeldt et al. (2004) note that the construction of a communication network occurs within pronounced constraints on supply (p. 21). The sheer opportunity of exchange from having so many fellow partisans at work and play should translate into a higher percentage of fellow partisans making up one s network where one s party dominates. If not, the minority partisan must undertake heroic attempts at seeking out and cobbling together discussion partners who are not as readily available. Does this hold, or do partisans pursue a biased political discussion network? To test this proposition, the individual-level reported dyadic network information is aggregated to measure how political discussion networks look in different contexts. The intersection of partisan context and political discussion networks should demonstrate the motivation of political discussants in very different contexts. Table 6 presents the nature of political discussion networks in different contexts in Germany. It breaks Social Democrats from Christian Democrats and averages the percentage of their political discussion network comprised of supporters for each of the major parties and presents them arrayed by the partisan context in which they live. At first glance, no matter what the partisan context is, partisans tend to overwhelmingly seek out fellow partisans with whom to discuss politics in all contexts. <<TABLE 6>> The slight differences between the discussion network composition in one s own partisan-controlled districts is by no means enormous in comparison to where one s own party falters, but are these patterns systematic? A chi-square test is run to test whether Social Democrats and/or Christian Democrats/Socialists have political discussion networks made up more of their own party where their party dominates compared to where it struggles. Given the 14

15 findings, a pattern appears to exist, though it only approaches statistical significance at the.05 level for Social Democrats and only slightly comes under the.10 level for Christian Democrats/Socialists. A t-test is run that tests the difference in means between each type of partisan context and each other type of partisan context to see where the differences may exist. However, none of these differences of means tests end up significant for either Social Democrats or Christian Democrats/Socialists. Further, the weak relationship between partisan context and make-up of political discussion networks is visible in the gamma measures of association. On a measure that varies between zero (weak relationship) and one (strong relationship), the effect of partisan context on make up of Social Democrats discussion network produced a value of only.1299 and only.1482 for Christian Democrats and Christian Socialists. 14 Between eyeballing the data, a chi-square test, the t-tests, and gamma results, no clear pattern exists: perhaps the partisan context may at best influence the choice of discussion partners just slightly, but the preponderance of the evidence suggests that political discussion occurs similarly across each and every partisan context. Partisans seek out those whom they perceive at least share their political leanings, no matter what the context. This same mixed finding holds for the American case in Table 7. By and large, the makeup of a partisan s political discussion network reflects that no matter where one lives, they discuss politics with fellow party supporters. Partisans tend to discuss politics with people who will support their candidate slightly more often where their party dominates. However, even where one s party has no chance at winning, a majority of voters networks are made up of supporters of their own parties candidates even though the 1992 election had a very strong third-party challenge. Indeed, the chi-square demonstrates that among Democrats, no statistically significant pattern arises, though context seems to have played a statistically significant role in discussion among Republicans discussion networks. Even here, however, only one statistically significant difference arises in the difference of means tests. Further, the gamma test demonstrates that no substantive relationship between the context and the make-up of discussion 15

16 networks; the gammas are.1621 for Democrats and.1668 for Republicans. These mixed results illustrate that even though partisan contextual realities make it difficult for minority partisans to find fellow party supporters, they tend to do nearly as well as fellow partisans who live where their party dominates. <<TABLE 7>> Voters construct political discussion networks to pursue information that strengthens their preexisting partisan attitudes. People also avoid information that could challenge their attitudes. The amount of political disagreement is also an important component of political discussion networks of German and American partisans in different contexts. As Huckfeldt, Johnson, and Sprague (2004) find, disagreement is ubiquitous in political discussion networks. Two things may drive disagreement: the number of people surrounding you with different views and/or the amount of people in your discussion network with whom one disagrees. To measure this, respondents are asked how frequently, if ever, they disagree with discussion partners about political matters, and here those respondents who claim they often disagree with someone on politics are concentrated upon. One might expect this figure to be highest in unfriendly partisan contexts. This is not the case for Germany, as three large trends stand out in Table 8. First, the N size suggests that people infrequently disagree with others about politics; people generally filter their sources of political discussion to avoid disagreement. Secondly, disagreement does not rise significantly with the partisan context with which one lives. Both the chi-square and the gamma tests demonstrate a weak statistical and substantive relationship. Thirdly, much of the most strident disagreement occurs with those who support another party as much as one s own party. This may not be a surprise, but when considering how infrequently people ever discuss politics with a partisan supporter of another party, it is noteworthy. That is, when looking at tables 6 and 8 together, people seldom discuss politics with partisans of other parties. However, when they do discuss politics with a supporter of another party, this tends to bring as much disagreement as 16

17 with the fellow partisans with whom they speak most frequently. The pattern appears as well for the United States as found in Table 7. Considering the small number of cases of strident disagreement and the fact that despite the overwhelming percentage of discussion partners being those with whom one shares partisanship, it is clear that people filter their political discussion. <<TABLE 8>> Discussion The partisan breakdown of voters communities in Germany and the United States do not appear to mediate their party identification or their position on political issues. Perhaps when an ideologically opposed majority party exists, minority partisans adhere more closely to their party, but the evidence is not entirely clear. What is clear from these findings is the key role that political conversation plays in reinforcing attitudes in all settings. At the very least, the findings of the partisan discussion model may illuminate the underlying process that explains or supports both the primacy of partisanship and the threat models. As Lodge and Taber (2000) suggest, people actively seek out biased information even if they are in an area dominated by their party or in an area where they are under challenge. Similar to Miller s (1956) argument, one could conclude that party identification was strong enough on its own to make a voter loyal, no matter what slim chances exist that a partisan s party actually has any shot at gaining office. This does not fully appreciate the effort that minority partisans put forward in protecting their party identification, however. It is quite remarkable that a party adherent, surrounded by voters who disagree with them about politics, still finds ways to bolster his or her partisan leanings. This highlights the underappreciated role political discussion plays in party politics. Some critics dismiss political discussion studies as something in the error term. Yet what this study has clearly shown is that there is a complex relationship between one s party identification, attitudes, and behaviors undertaken to guard these attitudes. Political discussion is not merely an issue of supply and demand partisans, and especially minority partisans, actively filter their political discussion networks to shield their attitudes from a hostile environment. While this can 17

18 easily appear as simply a result of party identification, the evidence here illustrates that party identification drives people to seek fellow partisans, and this conversation can serve many purposes, especially for minority partisans. The interaction of the individual with her surroundings, as well as her role in seeking discussion partners of the same party suggests a change in the way political science should understand partisanship s effects. The classic unmoved mover view of partisanship is that it is socialized and unlikely to be influenced by political events. Other critics suggest this is not the case. Party identification may be a running tally that can be altered by political events (Fiorina, 1981). Actively updating one s party identification through biased information processing (Lodge and Taber, 2000), particularly through filtering political discussion partners, reinforces party identification in this running tally. As a result, this may keep minority partisans engaged in political deliberation and keep their party alive, even if not viable electorally, in particular areas. 18

19 Works Cited Bishop, B. The schism in U.S. politics begins at home: Growing gaps found from county to county in presidential races. Austin American-Statesman. April 4, p. A1. Bowler, S. Lanoue, D. Savoie, P. Electoral Systems, Party Competition, and Strength of Partisan Attachment: Evidence from Three Countries. Journal of Politics. Vol. 56: P Butler, D. Stokes. D. Political Change in Britain. New York: St. Martin's Press Campbell, A. Converse, P. Miller, W. Stokes, D. The American Voter. New York: Wiley and Sons Finifter, A. The Friendship Group as a Protective Environment for Political Deviants. American Political Science Review. Vol. 68: Fiorina, M. Retrospective Voting in American National Elections. New Haven: Yale University Press Huckfeldt, R. Johnson, P. Sprague, J. Political Disagreement: The Survival of Opinions Within Communication Networks. Cambridge: Cambridge University Press Huckfeldt, R. Beck, P.A. Dalton, R. Levine, J. Morgan, W. Ambiguity, Distorted Messages, and Nested Environmental Effects on Political Communication. Journal of Politics. Vol pp Huckfeldt, R. Sprague, J. Citizens, Politics, and SocialCommunication: Information and Influence in an Election Campaign. Cambridge: Cambridge University Key, V.O. Southern Politics in State and Nation. Knoxville: University of Tennessee Press Lodge, M. Taber, C. Three Steps Toward a Theory of Motivated Political Reasoning in Lupia, A. McCubbins, M. Popkin, S. (eds) Elements of Reason: Cognition, Choice, and the Bounds of Rationality. Cambridge: Cambridge University Press Miller, W. One-Party Politics and the Voter. AmericanPolitical Science Review. Vol pp Noah, T. Mister Landslide s Neighborhood: Red Versus Blue States Isn t the Half of It. Slate. April 7, online: Putnam, R. Political Attitudes and the Local Community. American Political Science Review. Vol. 60: Wright, G. Contextual Models of Electoral Behavior: the Southern Wallace Vote. American Political Science Review. V.71: p

20 Table 1 ` Multinomial Logistic Regression Results of the Influence of a District s Partisan Context on Issue Voting in the 1990 German Election. N = 894 chi2(52) = Prob >.0000 Pseudo R2 = Log Likelihood Logit. Logit Vote Inten Coeff P> z Vote Inten Coeff P> z CDU - SPD CDU - FDP Cspdeast Cspdeast Ccdueast Ccdueast Cspdissu Cspdissu Ccduissu Ccduissu CDUpart *** CDUpart *** FDPpart *** FDPpart *** GRNpart ** GRNpart Othpart *** Othpart *** context context eastspd eastspd * eastcdu eastcdu issspd issspd isscdu isscdu intercept intrcept Logit Logit Vote Inten Coeff P> z Vote Inten Coeff P> z CDU - Greens CDU - Other Cspdeast ** Cspdeast Ccdueast Ccdueast Cspdissu Cspdissu Ccduissu Ccduissu CDUpart *** CDUpart * FDPpart FDPpart GRNpart *** GRNpart ** Othpart * Othpart *** context context eastspd eastspd eastcdu eastcdu issspd issspd isscdu isscdu intercept intercept Note: Each subtable offers the comparison between the bolded parties listed. Variables that are italicized are the interactive variables of relative party dominance in District and distance between the respondent and their view of where the party stands on that issue. * - significant passed the.10 level ** - significant passed the.05 level *** - significant passed the.01 level 20

21 Table 2 Multinomial Logistic Regression Results of the Influence of a District s Partisan Context on Issue Voting in the 1992 American Election. ` N = 989 chi2(52) = Prob >.0000 Pseudo R2 = Log Likelihood Logit. Logit Vote Inten Coeff P> z Vote Inten Coeff P> z Bush - Clinton Perot - Clinton Bisscntx Bisscntx Cisscntx Cisscntx ** Pisscntx Pisscntx * Repub *** Repub *** Indep *** Indep *** BushIssu ** BushIssu ClntIssu ClntIssu * PeroIssu * PeroIssu context ** context intercept *** intrcept *** Note: Each subtable offers the comparison between the bolded candidates listed. Variables that are italicized are the interactive variables of relative party dominance in county and distance between the respondent and their view of where the party stands on that issue. * - significant passed the.10 level ** - significant passed the.05 level *** - significant passed the.01 level 21

22 Table 3 Multinomial Logistic Regression Results of the Influence of District Partisan Context on Partisanship: Party Vote Intention in the 1990 German Election. N = 894 chi2(52) = Prob >.0000 Pseudo R2 = Log Likelihood Logit. Vote Inten Coeff P> z Logit Vote Inten Coeff P> z CDU - SPD CDU - FDP SPDcntxt SPDcntxt CDUcntxt CDUcntxt CDUpart *** CDUpart *** FDPpart *** FDPpart *** GRNpart GRNpart *** Othpart *** Othpart context context intercept intrcept Logit Vote Inten Coeff P> z CDU - Greens Logit Vote Inten Coeff P> z CDU - Other SPDcntxt SPDcntxt CDUcntxt CDUcntxt CDUpart CDUpart FDPpart FDPpart GRNpart *** GRNpart Othpart ** Othpart context context intercept intercept Note: Each subtable offers the comparison between the bolded parties listed. Variables that are italicized are the interactive variables of relative party dominance in District and respondent partisanship * - significant passed the.10 level ** - significant passed the.05 level *** - significant passed the.01 level 22

23 Table 4 Multinomial Logistic Regression Results of the Influence of County Partisan Context on Partisanship: Candidate Vote in the 1992 U.S. Presidential Election N = 991 chi2(52) = Prob > chi Pseudo R2 = Log Likelihood Logit. Logit Vote Inten Coeff P> z Vote Inten Coeff P> z Bush Clinton Perot Clinton Repcntx Repcntx Demcntx Demcntx Repub *** Repub *** Indep *** Indep ** context *** context intercept *** intrcept *** Note: Each subtable offers the comparison between the bolded parties listed. Variables that are italicized are the interactive variables of relative party dominance in County and respondent partisanship * - significant passed the.10 level ** - significant passed the.05 level *** - significant passed the.01 level 23

24 Table 5 Threat Model: Predicted Probability and Ordered Probit Results for Bavarian and Non-Bavarian Influence on Strength of SPD Partisanship. Predicted Probability of being a: StrongSPD ModerateSPD WeakSPD partisan partisan partisan Bavarian SPD Member Non-Bavarian SPD Member N = 78 chi2(6) = Prob >.03 Pseudo R2 = 0.11 Log Likelihood Probit Std strength Coeff Err P> z Bavaria Sex Age Educ Vote Catholic _cut Ancillary _cut parameters Strength Bavaria Sex Age Educ 1=weak SPD partisan 2=moderate SPD partisan 3=strongS SPD partisan 1=Bavarian 0=other Land resident 1=male 2=female listed year given 7 category Low to high level of school attended Vote80 1=CDU/CSU vote in =SPD vote in =FDP vote in =Greens vote in =Other(NPDorDKP) in 80 note: replication of 1983 German Election Study explained by Bowler et al. (1994) with predicted probabilities generated by this author in Stata. 24

25 Table 6 Make-Up of Discussion Networks in Germany Across Different Partisan Contexts Germany 1990 Social Democrat Respondents Christian Democrat Christian Socialist Respondents % of Network Made Up of Supporters of Each Party % of Network Made Up of Supporters of Each Party Partisan Context of Respondent s District SPD Green FDP CDU/CSU N SPD Green FDP CDU/CSU N Very CDU/CSU 78% 4% 2% 16% % 6% 3% 81% 159 Somewhat CDU/CSU 70% 14% 5% 11% % 2% 4% 73% 98 Neither SPD nor CDU 86% 0% 5% 10% 21 0% 8% 0% 92% 13 Somewhat SPD 72% 8% 3% 17% % 4% 2% 70% 186 Very SPD 84% 8% 3% 6% % 1% 7% 74% 74 Chi-Square Test: Gamma Test: Chi-Square probability = df = 12 Gamma =.1299 (SPD & CDU/CSU Discussion partners only) Chi-Square probability = df = 12 Gamma =.1482 (SPD & CDU/CSU Discussion partners only) Unpaired 2-tail test of difference of means: t value Probability that means significantly different t value Probability that means significantly different VeryCDU vs. VerySPD.287 P > [ t ] = P > [ t ] =.498 VeryCDU vs. SomeSPD P > [ t ] = P > [ t ] =.908 VeryCDU vs. Neither P > [ t ] = P > [ t ] =.243 VeryCDU vs. SomeCDU.015 P > [ t ] = P > [ t ] =.631 SomeCDU vs. VerySPD.302 P > [ t ] = P > [ t ] =.780 SomeCDU vs. SomeSPD P > [ t ] = P > [ t ] =.537 SomeCDU vs. Neither P > [ t ] = P > [ t ] =.248 SomeSPD vs. Neither P > [ t ] = P > [ t ] =.193 VerySPD vs. Neither.976 P > [ t ] = P > [ t ] =.278 VerySPD vs. SomeSPD P > [ t ] = P > [ t ] =