Political Behavior

, Volume 28, Issue 4, pp 349–366

Political Disagreement in Context: The Conditional Effect of Neighborhood Context, Disagreement and Political Talk on Electoral Participation

Authors

    • Department of Political ScienceSouthern Illinois University
Original Paper

DOI: 10.1007/s11109-006-9015-4

Cite this article as:
McClurg, S.D. Polit Behav (2006) 28: 349. doi:10.1007/s11109-006-9015-4

Abstract

Despite scholarly interest in determining how exposure to disagreeable political ideas influences political participation, existing research supports few firm conclusions. This paper argues that these varied findings stem from an implicit model of contextual influence that fails to account for the indirect effect of aggregate social contexts. A model of contextual influence is outlined which implies that the neighborhood partisan context moderates the effect of political disagreement in social networks on campaign participation. The evidence shows that network disagreement demobilizes people who are the political minority in their neighborhood, but has no influence on people in the majority. When viewed together, these findings indicate that a person’s relationship to the broader political environment sets distinctive network processes in motion.

Keywords

Political participationNeighborhood contextSocial networksPolitical disagreement

What is the impact of political disagreement on participation? Although it is frequently argued that exposure to different points of view in ones’ environment makes participation unlikely, progress on this issue is inhibited by how previous research conceptualizes and measures social context. While some studies treat exposure to disagreement as a trait of the aggregate context, others measure it as an attribute of interpersonal exchange. Such an approach implicitly assumes that either (1) one context matters but not the other or (2) these two contexts have equivalent effects, probably because they are theoretically related. However, as recognized by landmark contextual analyses, the social network is not simply a reflection of the local environment and the local environment is important beyond how it structures networks (Berelson, Lazarsfeld, & McPhee, 1954, Chapter 4; Huckfeldt & Sprague, 1995). This raises some additional questions that have gone unanswered in previous work. Does exposure to disagreement matter in one context but not the other? Has previous research mischaracterized the effect of being exposed to disagreement by overlooking the influence of multiple contexts?

To address these questions, this paper explores the joint effect of aggregate and interpersonal contexts on participation. There are two reasons why this investigation is worthwhile. First, previous research implies that there are normative tradeoffs associated with political disagreement. While it has positive effects on tolerance and the quality of opinions, it is often seen as having a negative impact on participation (e.g., Mutz, 2002a, 2002b; but see Leighley, 1990). If the effects of disagreement on participation are not always negative, it would imply that these tradeoffs are minimal. Second, much contemporary work on social context misses an important, but oft-overlooked, message in classic contextual theory, namely that the “social context” is a multilayered phenomenon. By exploring how multiple social environments combine to affect participation, we can better understand the subtleties in how context affects participation. The central role of exposure to different points of view in contextual theory makes this a good place to start sorting out these effects.

The theoretical framework developed in this paper posits that the effect of network disagreement differs according to the supportiveness of the aggregate environment. By virtue of their political status, political minorities are more sensitive to the level of support they garner in their social relations. In contrast, people who are in the political majority are less likely to need to support or to respond strongly to negative signals sent in their networks. I test this argument using survey data gathered specifically for studying cross-level social processes that provides information on both a respondent’s discussion partners and her neighborhood. I find that the effect of network disagreement is demobilizing for people who are minorities in their neighborhood, but that no equivalent effect exists for people in the majority. The results also demonstrate that another factor that connects political networks to involvement—the volume of political discussion—does not differ across contexts, a result that also holds implications for contextual models of behavior.

Political Disagreement, Social Context, and Political Participation

At the aggregate level, substantial research shows that economic, ethnic, and political diversity influences political participation (Campbell, 2004; Cho, Gimpel, & Dyck, 2006; Costa & Kahn, 2003; Gimpel, Lay, & Schuknecht, 2003; Oliver, 2001).1 In general terms, this body of work does not clearly explain how the potential for exposure to disagreeable political views influences behavior because it is unable to distinguish between different properties of the environment. Although some scholars argue that heterogeneity increases participation because it produces acute political competition among groups with different preferences (Campbell, 2004; Cox & Munger, 1989), others make arguments about the importance of participatory norms and exposure to diverse points of view (Cho et al., 2006; Oliver, 2001). Still other work implies that it is not simply aggregate heterogeneity that matters as much as whether individuals are members of the majority or not that is consequential for explaining political behavior (Huckfeldt & Sprague, 1987, 1988; Glynn, Hayes, & Shanahan, 1997; Noelle-Neumann, 1984). Since most studies fail to measure these various environmental properties simultaneously it difficult to discern which argument best applies to their analysis. As such, these studies often offer different interpretations of their evidence that tend to muddy, rather than clarify, the role of aggregate diversity on participation.

Networks studies are not susceptible to the same problem because measures of interpersonal disagreement can only reflect exposure to different points of view rather than some other property like competition.2 Moreover, such measures can only be understood in terms of how people receive and process information from their peers (Huckfeldt, 2001; Knoke, 1990; Lake & Huckfeldt, 1998; Leighley, 1990; McClurg, 2003, 2006; Mutz, 2002a, 2002b). Along these lines, most research on network disagreement finds that receipt of disagreeable political information demobilizes participation (McClurg, 2006; Mutz, 2002b; but see Leighley, 1990). Relying heavily on social–psychological concepts, the strongest explanation in this vein contends that disagreement stimulates attitudinal ambivalence and conflict avoidant predispositions (e.g., Mutz, 2002a), both of which make participation less likely.3

Nevertheless, network studies also provide an incomplete picture of how disagreement relates to participation. In particular, the findings cannot be reconciled with other research that finds a positive relationship between network politicization and participation (Lake & Huckfeldt, 1998; Leighley, 1990; McClurg, 2003). The problem is that, as a matter of simple probability, people with larger networks that are rich in political content are also more likely to expose people to political disagreement (Huckfeldt, Johnson, & Sprague, 2004). The same networks which seemingly discourage participation, then, overlap significantly with the ones that encourage it.

Finally, both types of studies typically focus on either aggregate or network context, but not both simultaneously (but see Huckfeldt & Sprague, 1992). Yet a review of classic studies in the contextual tradition illustrates their persistent concern with the effect of both aggregate and interpersonal contexts. For example, Berelson et al. (1954, Chapter 4) were struck by the fact that the political preferences of upper income Catholics in Elmira depended strongly on their integration into local churches. If they attended religious services regularly, they responded more strongly to the group norm; if they did not, they were “peeled off” in the direction of the community norm.4 This suggests that “contextual influences” on behavior are not monolithic, with different elements of social organization potentially pulling individuals in multiple directions.

Political Disagreement Across Multiple Contexts

What happens when people are exposed to different information in their multiple social contexts? Under what conditions should we expect them to participate? Are there circumstances in which disagreement from the network is unimportant? How about from the aggregate environment? This section addresses these questions. I start by explaining why networks and aggregates are overlapping, but not equivalent, contexts. This in turn supports the notion that research overlooking one or the other context may be misleading. I then develop a framework for understanding why political minorities should be more responsive to support from their discussion partners than political majorities.

Overlapping Social Contexts

Is it the case that aggregate and network contexts supply different types of information? Or, do they overlap significantly in meaningful ways? Looking beyond whether aggregate and interpersonal environments provide the same information purely by chance even when they are conceptually distinct, it is possible that these two environments are theoretically connected. Most simply, aggregate context may affect the nature of interpersonal interaction by shaping the supply of potential discussion partners. In doing so, there should be substantial overlap between the experience of disagreement in discussion and in the aggregate. If this is the case, then it may not be worthwhile trying to delineate conceptual and empirical distinctions between the two contexts. In support of this idea, some evidence shows that a person’s social network is influenced by the distribution of political preferences in the surrounding social context (Huckfeldt, 1983, 1984; Huckfeldt & Sprague, 1987).

Such evidence notwithstanding, there are at least two reasons to believe that aggregate and interpersonal contexts have differential effects. First, the effect of aggregate contexts on networks supposedly occurs by probabilistically affecting the supply of potential discussants. However, this process is constrained as most networks include a large proportion of spouses and family members (e.g., Marsden, 1987), limiting the amount of space in people’s core networks available to be filled. The dual processes of individual selection (Finifter, 1974; MacKuen, 1990) and exposure to multiple contexts (e.g., neighborhood, work church, etc; Huckfeldt, Plutzer, & Sprague, 1993) further restrict how likely it is that any one aggregate context strongly shapes who people include in their discussion networks. Second, even when an aggregate political bias seeps into the network, it in no way means that people then ignore information supplied by their aggregate contexts. For example, people who include a neighbor as a member of their discussion networks do not suddenly become ignorant of their observations about the political signs displayed in their neighborhood. Theoretically, these two processes create room for both contexts to simultaneously affect participation.5

A Conditional Model of Social Influence

Because there is separation between these two layers of social life, social networks and aggregate contexts can be treated as overlapping and coordinated, but also partly incongruent and separate, foundations for social influence. This re-shapes the question somewhat. Instead of focusing simply on how exposure to disagreement affects behavior, we must consider what happens when people receive different political cues from their multiple social contexts. In short, how do the multiple layers of social context combine to affect political participation? To answer this question, I focus on how experiencing disagreement in conversation differs according to aggregate context. In doing so, I relegate the context to a background role. Rather than having a direct impact on participation once its influence on the network is accounted for, I imply that it moderates the impact of disagreement in discussions by shaping how people perceive and react to them.

According to previous research, there are distinct differences in how political minorities and majorities perceive the world around them. Importantly, political minorities are acutely aware of their status vis-a-vis the people around them because the information they receive from the social context is, on balance, contradictory to their political views. As a consequence, the experience of disagreement in concrete social interactions takes on a different character for them. Foremost among the effects is that political minorities can more accurately gauge the political preferences of the people in their network, recognizing with acuity the people who do and do not share their views as well as those who do not. So in addition to being more likely to experience disagreement in social interaction, they also are more likely to perceive that disagreement accurately. Minorities also feel less freedom to express their political opinions because of the likelihood of receiving negative feedback (Noelle-Neumann, 1984). Such sanctions do not exist for those holding the majority view—they are less likely to be exposed to disagreement, recognize it in their social interactions, or feel constrained in expressing their views.

While this implies that disagreement in grounded interactions differs depending upon one’s political position in the context, it says nothing about the relative importance of agreement. Why would the above logic not just indicate that minorities are less likely to participate on average than majorities and that the specific nature of social interactions are inconsequential? Why must minorities also experience disagreement in conversation to be demobilized? Clues to an answer can be found in the literature on social conformity where experiments by Milgram (1974) provide telling insight. In these studies, they found that experimental subjects were unwilling to go against group norms when they had no support from others in the experiment. Conversely, if even one other person was willing to support their point of view, then the experimental subject was willing to report her true feelings. Much like the church-going Catholics in Elmira, people receiving support from their discussion partners are capable of resisting the pull of the larger context. Comparatively, the importance of support is not relevant for those in the majority because they do not need support for their perspective since it dominates the environment.

To summarize, support from one’s discussants is essential for understanding the political activities of people who are political minorities. When they receive it they are capable of resisting the negative pressure of the aggregate context; when they are exposed to disagreement instead, it should be demobilizing. Due to their political status, those in the minority are less cognizant of disagreement and in less need of support. Based on this discussion, we should find that political disagreement make participation less likely among those in the minority, but not among those in the majority.

Data

I investigate this hypothesis with data from a public opinion survey administered to residents of South Bend, Indiana during the 1984 election campaign. Known in the field as the South Bend Study (Huckfeldt & Sprague, 1985) these data are appropriate because they were gathered with an eye toward measurement of both local context and social networks. While other studies have adequate measures of either aggregate (e.g., the Social Capital Benchmark Study) or network (e.g., the 1992 Cross National Election Study) context, they do not allow for simultaneous measurement of both contexts. So while using this study limits the external validity of the results presented below, this is balanced by gains in internal validity.

The South Bend study employed a stratified sampling design. In the initial stage, sixteen distinct neighborhoods were identified.6 In the second stage, approximately one hundred respondents were randomly selected and interviewed within each neighborhood. This feature of the research design means that items measuring individual characteristics, such as the preference for a presidential candidate, can be aggregated to provide a measure of the neighborhood information environment. Although the neighborhood is only one type of aggregate context in which interaction takes place (Huckfeldt, Plutzer, & Sprague, 1993), it is influential for many people (Huckfeldt, 1986; Huckfeldt & Sprague, 1995).7

The survey instrument gathered information about each respondent’s social network. Specifically relevant to this project, the original researchers interviewed a random sample of the named discussants providing for valid measurement of respondent attitudes, discussant attitudes, and the level of social interaction between the two. 8 Therefore, each respondent is placed in two different social contexts—the discussion dyad and the neighborhood context—with the unit of analysis being the individual (Huckfeldt & Sprague, 1995). By drawing on actual discussant interviews, these data allow me to objectively measure the supply of political disagreement in both contexts.

Multiple indicators of electoral political participation were on the survey instrument. After the election, each respondent was asked if he or she participated in any of the following activities—working on a campaign, displaying a bumper sticker or sign, donating money, and attending political meetings. Valid data on voter turnout are also available in this study since official voter data (both primary and general) were merged with the survey file by the principal investigators (Huckfeldt & Sprague, 1985). In the analyses below, participation is examined as an index of the number of political acts in which the respondent engaged.9 This dependent variable ranges from zero to six, with a mean of 1.6 and a standard deviation of 1.30.10

To measure exposure to political dissonance in aggregate and network contexts, I examine the presidential vote preferences of respondents, discussants, and the respondent’s neighborhood. At the level of the discussion dyad, I contrast the self-reported vote preferences of the respondent and discussant. A score of one indicates disagreement, and a zero agreement. At the neighborhood level, I computed the proportion of people in each neighborhood preferring Reagan and Mondale. I then determined whether each respondent was in the neighborhood majority or not by contrasting their final vote preference with the neighborhood election outcome.

Aggregates, Networks, and Political Activity

Networks as Aggregate Contexts

A principal component of the argument outlined above is that disjoint exists between the information culled from political discussion and the aggregate environment. Although the hypothesis developed here in no way assumes that the two contexts are completely independent, it is still possible that we could find that dependency between them is strong enough to explain variations in participatory behavior. Therefore, to provide a baseline against which to judge the hypothesis that the impact of disagreement differs across aggregate environment it is first necessary to consider this possibility.

Toward this end, Table 1 tabulates disagreement in the discussion dyad with neighborhood political supportiveness. As this table shows, people who are in a neighborhood political minority are significantly more likely to experience disagreement in their dyad than are people who are members of the majority.11 While only 26% of all dyads are characterized by disagreement for respondents who are in the neighborhood majority, those in the minority experience disagreement in 42% of their interactions. This demonstrates that the neighborhood context has potentially a coercive influence on network interactions, seeping through whatever individual filters may be in place as people select discussion partners.
Table 1

The relationship between social context and disagreement in network dyads

Measure of dyad disagreement

Measure of social context supportiveness

Member of minority

Member of majority

Total

Same vote preference

58%

74%

67%

(210)

(360)

(570)

Different vote preference

42%

26%

33%

(151)

(125)

(276)

Total

100%

100%

100%*

43%

57%

100%

(361)

(485)

(846)

This table demonstrates that social context structures political disagreement in dyads

* χ2 = 24.2694, p < .01

Is it possible that this relationship accounts for contextual effects on participation? Some evidence of this is provided in Table 2, which compares the average level of political activity at each level of these independent variables separately. To test statistical significance, I conducted a means difference test with unequal variances for both variables. The results show that there are significant differences across type of discussion dyad, but not across types of neighborhood. There are two possible explanations for these results. The first is that aggregate context—in this case neighborhoods—does not matter for explaining participation. A second possibility consistent with the results in Table 1 is that neighborhoods only matter to the degree that they influence who people talk to about politics and whether they are likely to agree or not. Altogether this implies that networks and aggregates are, at least for these data, interrelated.
Table 2

Direct effect of context and network variables on political participation

Variable

Activity at independent variable minimum

Activity at independent variable maximum

Difference

Means difference test

μ

s

μ

S

μ

s

 

Dyad disagreementa

1.75

.06

1.43

.07

.32

.09

3.52**

Neighborhood preferencesb

1.59

.07

1.62

.06

.03

.09

.37

** < .01, two-tailed t-test

* < .05, two-tailed t-test

a0 = agreement, 1 = disagreement

b0 = member of minority, 1 = member of majority

Evidence for this argument notwithstanding, there is reason to be skeptical that it tells the whole story about the relationship between context, disagreement, and involvement. Most importantly, it does not rule out the possibility that even after influencing who we talk to about politics, neighborhoods contexts can still provide a backdrop against which those conversations are evaluated. Also it is often argued that neighborhoods are increasingly less central to American’s social life than they were in the past. While this may mean that social networks are less geographically-grounded than at some point in the past (Putnam, 2000), plenty of evidence suggests that aggregate context still affects behavior (Campbell, 2004; Cho & Rudolph, N.d.; Cho et al., 2006; Costa & Kahn, 2003; Gimpel et al., 2003; Oliver, 2001). This indirectly implies that the aggregate context still retains relevance for explaining behavior beyond its influence on who we talk to about political matters.

The Potential for Moderation

The next step, then, is to determine the potential for the aggregate context to moderate the influence of network context. To determine whether there is evidence that the influence of network agreement depends upon the context itself, the first panel in Table 3 reports the results of means difference tests of interpersonal disagreement while controlling for the respondent’s neighborhood political position. For those respondents who live in a neighborhood majority, moving from an agreeable dyad to a disagreeable dyad decreases the number of political acts by .20, a result that is statistically insignificant at the .05 level (two-tailed test). In contrast, the effect of disagreement is two-and-a-half times larger for respondents in the minority and is statistically significant. Panel B shows that, in contrast, networks do not mediate the influence of neighborhood context. For those respondents in agreeable dyads, there is a slight and statistically insignificant decrease (.09) in their participation when they go from being in the minority to the majority. Those respondents in disagreeable dyads exhibit a slight increase (0.17) that is also statistically insignificant.
Table 3

The potential for contextual mediation

Variable

Activity at independent variable minimum

Activity at independent variable maximum

Difference

Means difference test

μ

s

μ

s

μ

s

 

A: The effect of dyad disagreement when the respondent lives in the...

...majority

1.69

.07

1.49

.12

.20

.32

1.49

...minority

1.78

.09

1.32

.09

.47

.13

3.58**

B: The effect of neighborhood status when the respondent experiences...

... agreement

1.78

.09

1.69

.07

.09

.11

.81

... disagreement

1.32

.09

1.49

.12

−.17

.15

−1.16

** p < .01, two-tailed test

Network Effects on Political Activity

Though these findings support the conditional relationship between aggregate and interpersonal contexts outlined above, it is important to subject the argument to a more rigorous test. The first column of Table 4 reports estimates of the aforementioned contextual variables after adding controls for party contacting (Rosenstone & Hansen, 1993), political interest (Verba, Schlozman, & Brady, 1995), the strength of an individual’s partisan leanings, and education (Verba & Nie, 1972; Wolfinger & Rosenstone, 1980). Since the dependent variable is a count of the number of activities in which respondents participated, I estimate the coefficients using a Poisson regression model.12 To account for the dependence of respondents within the same neighborhood, standard errors were allowed to cluster with the neighborhood.13 Three sets of results are reported. The first results come from a pooled model where an individual’s relative political position in the neighborhood is included as a direct effect. The second and third models provide estimates for members of the minority and majority separately to investigate whether the influence of network variables is different across these two types of aggregate environments.14
Table 4

The effect of disagreement, political talk, and neighborhood context on political activity

Independent variables

Pooled model

Neighborhood minority

Neighborhood majority

β

Std. Error

β

Std. Error

β

Std. Error

Dyad disagreement

−.12

.06*

−.18

.11*

−.08

.08

Dyad political talk

.17

.04**

.15

.06**

.17

.04**

Neighborhood competitiveness

.26

.27

.57

.47

.14

.37

Party contact

.36

.09**

.25

.13*

.43

.10**

Political interest

.22

.07**

.31

.11**

.13

.09

Partisan strength

.10

.04**

.09

.05*

.10

.04**

Years of education

.06

.01**

.07

.01**

.05

.02**

Member of the majority

−.08

.07

    

Constant

−1.42

.02**

−1.70

.29**

−1.19

.30**

N

827

354

473

Wald χ2

268.09#

174.33#

80.33#

The Poisson regression models show the effect of social network and social context characteristics on levels of political activity. Social context only affects participation in that it affects the role of disagreement in social networks in explaining participation

** < .01, one-tailed t-test; * < .05, one-tailed t-test; #< .01, one-tailed chi-square test

The results for the pooled model largely reflect the conventional wisdom about the effect of social networks on participation. Despite the addition of control variables, disagreement decreases and discussion increases electoral political activity. There is no direct effect associated with the individual’s status as a member of the majority since the coefficient is statistically insignificant. Somewhat surprisingly, neighborhood competitiveness has no statistically significant effect on participation which may be a function of the aggregate unit of analysis here.

More important, the other models in Table 4 show that the effect of disagreement is contingent upon context. Dyad disagreement is a negative and statistically significant predictor of participation for respondents who are in the political minority, but not for respondents in the majority since dyad disagreement fails to reach conventional levels of statistical significance. This reinforces the findings from the previous section and is consistent with the first hypothesis. In short, the evidence implies that the effect of interpersonal disagreement is interpreted differently dependant upon whether the context is generally supportive or unsupportive of the individual.

Contrasting the second and third set of results also reveals another insight about contextual effects. Unlike disagreement, the impact of political discussion in a dyad does not depend upon context, implying that the stimulus it provides to people is uniformly available for minorities and majorities alike. This is an important finding for two reasons. First, it suggests that political discussion is associated with participation even when there is disagreement at both levels of social context. In other words, the effect of political support from social contexts on participation can be overstated even for minorities. Second, it further implies that there are distinct effects on involvement that emanate from different properties of social contexts, as well as from different social contexts themselves. In other words, we cannot simply assume that all network effects depend on aggregate context just as we cannot assume that they do not.

To investigate further the joint effect of political talk and dyad disagreement on involvement, I compute the expected number of activities for respondents who were average in all other respects. The top panel of Table 5 shows the effect of these variables for respondents who are in the neighborhood minority. For respondents who never discuss politics, disagreement reduces their level of expected involvement from 1.27 acts to 1.07 acts. Talking politics often raises these values to 2.02 and 1.69 expected political acts, illustrating the strong influence of political interaction on alleviating the negative consequences of network disagreement, even among minorities. Yet in both cases, dyad disagreement decreases the expected level of involvement by about 16-percent. This means that disagreement leads to about one less political activity for every five to three respondents who are political minorities, depending on how much political discussion takes place.
Table 5

Expected number of activities based on disagreement, political talk, and neighborhood context

Respondent and discussant political talk

Respondent and discussant preferences...

Agree

Disagree

Effect of disagreement

Respondent in neighborhood minority

Never

1.27

1.07

.20

Rarely

1.48

1.24

.24

Once in a while

1.73

1.45

.28

Often

2.02

1.69

.33

Effect of discussion (min to max)

.75

.62

.13

Respondent in neighborhood majority

Never

1.28

1.17

.11

Rarely

1.52

1.40

.12

Once in a while

1.81

1.67

.14

Often

2.15

1.98

.17

Effect of discussion (min to max)

.87

.81

.06

This table shows that the effect of disagreement is larger for respondents in neighborhood minorities. For those respondents who are in a neighborhood majority, political talk is the driving social network force of behavior

Being a member of the neighborhood majority accentuates the positive effects of political talk while minimizing the deleterious influence of network disagreement. Among such respondents, disagreement reduces the number of expected political acts from 1.28 to 1.17 if the respondent and discussant never talk politics. A similarly strong reduction occurs in dyads where politics is oft discussed. The expected number of political acts decreases by about 8%, from 2.15 to 1.98 in dyads where disagreement occurs. If the differences among these respondents were statistically significant they would be less consequential than for respondents in the minority since, at most, we expect one fewer political activity for every 10 to 5 majority-status respondents.

Conclusion

A connection exists between a person’s place in the social milieu and their political activity, though it is more subtle than previously recognized. The results demonstrate that the effect of exposure to divergent political views is conditioned by one’s political status in the neighborhood, hinting at a more complex contextual model than is typically used in examinations of political behavior. As conceived here, the influence of one context overlaps with the information provided by another. On the one hand, there are citizens who live in the world of political minorities. Although likely to participate in at least one activity, exposure to dissonant political opinions proves to be a demobilizing force for these people. No matter how much they discuss politics, they participate less than citizens who have agreeable discussion partners. On the other hand are the citizens who enjoy the support of the people in their neighborhood. Like those in the minority, these citizens are influenced by political talk but do not experience salient consequences from disagreeing with their political informants.

These findings add to the discussion on when and how contextually supplied disagreement affects participation. First, they imply that exposure to disagreeable political views is important for understanding political behavior but that it is not always an important factor. Disagreement is only influential when it is reinforced by the aggregate context. Indirectly, this suggests that inconsistent results in the literature may be a consequence of a failure to consider fully how social context affects behavior. More broadly, they suggest that processes of deliberation and discussion—which inevitably expose people to alternate points of view—do not necessarily erode the foundations of involvement. Considering the demonstrated effects of such discussions for political tolerance (Mutz, 2002b) and sophistication (Barabas, 2004; Mutz, 2002b), this is an encouraging result. Additionally, the framework provided here begins the discussion on how to generate new hypotheses for exploring these issues. Fully understanding the consequences of political disagreement requires additional empirical work, but this empirical work will provide more purchase if it moves beyond the question of whether it has negative or positive consequences to considering the conditions in which it is debilitating or not.

Second, the results have broad implications for sociological explanations of political behavior. A basic model of social influence argues that aggregate contexts are important because they shape communication in social networks. The theoretical story is that the social context affects the supply of political views and information in a geographically (neighborhood) or socially (workplace) defined unit (Books & Prysby, 1991; Huckfeldt, 1983, 1984; Huckfeldt & Sprague, 1987, 1988). As a result, the political views of our friends, neighbors, co-workers, and fellow travelers will reflect that supply, unless people purposefully choose discussants based on political criteria (MacKuen, 1990). This paper supports a return to a more nuanced model of social influence. The evidence re-affirms the notion that social networks affect behavior, but also shows that the influence of context does not stop at probabilistically structuring the preferences of political informants. Aggregate social environments also provide a backdrop against which information transferred in social networks is evaluated. Considering the overlapping effects of these environments is particularly important as we move to world where geographic context overlaps less significantly with interpersonal context.

Footnotes
1

Although diversity is not the same thing as disagreement, the two concepts are interrelated. As an aggregate environment becomes more diverse, the likelihood of encountering different minded people increases.

 
2

There are other mechanisms that potentially connect social networks to participation (e.g., Klofstad, N.d.; Lake & Huckfeldt, 1998; McClurg, 2006). The point here is that, in contrast to aggregate measures of heterogeneity, there is only one meaningful interpretation of measures of interpersonal disagreement.

 
3

The network studies on this subject are buttressed by findings from carefully designed experiments on conflict avoidance (Ulbig & Funk, 1999) and exposure to heterogeneous information (Barker & Hansen, 2005).

 
4

See Finifter’s (1974) discussion of Republican autoworkers for a similar example. Huckfeldt and Sprague’s (1995) simultaneous analysis of networks and contexts echoes these concerns, as does Cho and Rudolph’s (N.d.) findings that county-level turnout patterns are not solely explained by social diffusion.

 
5

A third reason bears mentioning, though it is less germane to the empirical analysis in this paper. Geographic contexts, such as neighborhoods, are increasing less important as spaces for social interaction. This probably means that networks are less influenced by them now than in the past. However, this does not imply that these types of contexts are unimportant to political decisions, as the earlier discussion of the literature shows. It does imply the influence of such contexts is increasingly unlikely to come from how they structure our discussion networks.

 
6

The principal investigators of the South Bend Study used a grounded approach for identifying and selecting neighborhoods. They first identified neighborhood boundaries using definitions provided by the South Bend, Mishawaka, and St. Joseph County planning departments. After eliminating small neighborhoods (those with fewer than 2000 residents), they drew on census data to maximize variation in the social status and ethnic (but not racial) composition. Finally, they drew on the expertise of Professor John Roos at the University of Notre Dame, an expert on South Bend, to validate their understandings of these neighborhoods. Although selection of neighborhoods is not random—an impossible task in the context of this study—it has the alternate virtue of maximizing variation on geographically-bound contexts. More information can be found in Huckfeldt and Sprague (1995, pp. 24–25, 37–38). For purposes of analysis, two adjacent neighborhoods with low numbers of interviews were combined in this paper. This follows the practice of the original investigators.

 
7

A reasonable criticism of this approach is that geographic location, particular in the guise of neighborhoods, may matter less now than in 1984. First, it is important to recognize that respondents had reasonably accurate impressions of this kind of context even though the U.S. was in the midst of suburbanization in the mid-1980s (Baybeck & McClurg, 2005). Second, there is plenty of evidence from recent years that political behavior varies according to different geographic contexts including some small enough to be considered rough approximations of neighborhoods (e.g., Baybeck, 2006; Branton & Jones, 2005; Cho et al., 2006; Oliver, 2001). Finally, the argument that neighborhoods are not important rests on the notion that people no longer interact with other people in their neighborhoods, not that they do not understand their neighborhoods. In other words, the argument is that there is more separation between network and aggregate context. This potentially makes the disjoint between contexts that is the focus of this study even more relevant. So while it is unclear whether geographic context matters more or less now, recent evidence implies that it still matters and that it’s disconnect from networks may create even more substantial room for differential effects of these contexts.

 
8

Discussion partners were solicited with a question asking respondents for the names of people with whom they discuss “political matters,” a variant of the “important matters” question used by the General Social Survey.

 
9

The reliability measure for this variable is Cronbach’s Alpha (α = .6021).

 
10

Variable descriptions and marginals are included in Appendix A.

 
11

Huckfeldt (1983, 1984) argues that this reflects the probabilistic impact of aggregate context on the availability of discussants.

 
12

The Poisson regression model assumes that there is constant variance in the dependent variable. Theoretically, the fact that some participation predicts further participation potentially violates this assumption. In analyses not reported here, it is shown that this assumption is not violated and a negative binomial model produces the same results.

 
13

I accomplished this using the cluster command implemented in Stata 8.0. This allows for with-in neighborhood dependence of the observations and estimates robust standard errors.

 
14

One interpretation of separating the models by the respondent’s political position in the neighborhood essentially is that it creates an interaction term between that variable and each independent variable in the pooled model, causing some loss in efficiency and making it difficult to test the statistical difference of coefficients across models. Despite these potential problems, I use this approach for two reasons. First, the hypothesis is about whether disagreement is statistically different than zero within each type of context. The coefficient for disagreement among respondents in the minority may be statistically different from zero, but not statistically different from the coefficient among respondents in the majority yet still be consistent with the argument offered here. Table 7 in Appendix B shows the estimates of a simple version of this interaction model estimated with Ordinary Least Squares. Using Friedrich’s (1982) method for estimating statistical significance (reported in Table 8), it is clear that the interaction term is not statistically significant but that disagreement is still statistically different from zero for respondents in the minority. Second, estimating the models separately has the advantage of not assuming that the respondent’s status as a majority or minority has no effect on the other independent variables. Thus, this approach makes fewer assumptions about the potential indirect effects of the aggregate context than an interaction model.

 

Acknowledgements

I gratefully acknowledge the comments of Mac Avery, David Campbell, Jim Gimpel, J. Tobin Grant, Jan Leighley, Diana Mutz, Thomas Rudolph, Ed Schatz, and John Sprague. The graduate students in my Seminar in Political Behavior deserve special credit for their careful read and demanding comments. All errors remain my responsibility.

Copyright information

© Springer Science+Business Media, LLC 2006