SCOPE CONDITIONS: ELECTORAL AUTOCRACIES

The theory presented in this article pertains to electoral autocracies, the most common form of autocracy in the world today (Coppedge et al. Reference Coppedge, Edgell, Knutsen and Lindberg2022; Guriev and Treisman Reference Guriev and Treisman2022; Magaloni and Kricheli Reference Magaloni and Kricheli2010; Morse Reference Morse2012). Electoral autocracies are regimes that hold national-level, multiparty elections with an independent opposition, but where the likelihood of an opposition win is extremely unlikely because of structural impediments constructed by the regime (Schedler Reference Schedler2006). These impediments include gerrymandering, electoral fraud, the use of state resources for campaigning and propaganda, as well as harassment and intimidation of the opposition (Morgenbesser Reference Morgenbesser2020; Schedler Reference Schedler2002). Formally, the theory pertains to any country that has been consistently coded as an “electoral autocracy” by V-Dem’s “Regimes of the World” variable (Coppedge et al. Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Altman and Bernhard2023)Footnote ² for the past 10 years (2012–2022). Because this is a study of partisanship, I exclude regimes with no formal ruling party or where parties do not last more than one election cycle, as well as those that have experienced regime change within the past 10 years. A list of these cases is included in Supplementary Appendix A. Critical to the theory in this article, these regimes provide just enough freedom for ordinary citizens to associate with opposition parties—as well as enough freedom to make opposition partisanship a meaningful political division from ruling party partisanship—while simultaneously making it just difficult enough to do so that partisanship is heavily skewed toward the ruling party.

INDIVIDUAL-LEVEL APPROACHES TO UNDERSTANDING PARTISANSHIP UNDER AUTOCRACY

Following the proliferation of electoral autocracies since the 1990s (Levitsky and Way Reference Levitsky and Way2010), a growing literature has sought to explain the apparent puzzle of why ordinary citizens bother to get involved in autocratic politics. Given the apparent obviousness of the outcome of the election, why would anyone bother to support a party or vote in elections in the first place? Because, by definition, ruling parties dominate in electoral autocracies—and thus the large majority of partisans are ruling party partisans—this literature has heavily focused on why ordinary people choose to vote for ruling parties, especially when this means that they are supporting the very authoritarianism that presumably undermines their own rights and freedoms.Footnote ³

Overwhelmingly, the literature shows that people vote for the ruling party in expectation of a direct or community-based economic reward. For example, Blaydes (Reference Blaydes2011) and Magaloni (Reference Magaloni2006) demonstrate that electoral autocracies—specifically Mexico under the PRI and Hosni Mubarak’s Egypt—focus their social spending and infrastructure investments in ways that correlate with voting behavior. These regimes tended to invest more heavily in strongholds or swing districts, implying that ordinary citizens support the ruling party in expectations of these rewards. However, due to the limitations of the macro-level data, they cannot show whether or not vote choice at the individual level is actually affected by this logic of spending.

In contrast, studies by, for example, Lust-Okar (Reference Lust-Okar2006) and Miguel, Jamal, and Tessler (Reference Miguel, Jamal and Tessler2015) focus on the voter, arguing that ordinary citizens support candidates that they believe will offer them clientelist access to the regime, specifically through personal connections. Naturally, in an electoral autocracy, candidates with the best connections to patronage almost exclusively belong to the ruling party. Thus, most people vote for the ruling party in order to access personal benefits via clientelism. Taken together, these two hypotheses—that people support the ruling party in expectation of improved government spending in their district or personal clientelist exchange—are central to our understanding of political behavior in electoral autocracies. In fact, these explanations are so dominant in the literature that few other explanations have even been proposed, though a newer literature has begun to focus on some of the more ideological reasons people vote for ruling parties, specifically a sense of patriotism and civic duty (Letsa Reference Letsa2020; Reuter Reference Reuter2020).

A parallel literature has inversely proposed a different set of explanations as to why some people living in autocracies are more likely than others to prefer democracy (and therefore, by implication, oppose the autocratic regime). The core logic is that people prefer democracy for ideological reasons, stemming primarily from higher levels of education. This general idea was first articulated by Lipset (Reference Lipset1959a; Reference Lipset1959b), who famously argued that economic development was correlated with democracy because people with higher levels of education and socioeconomic status should be more likely to prefer democracy to authoritarianism. Of course, Lipset was not making an argument about partisanship under electoral autocracy, but his logic carries to the puzzle at hand: we might expect people with higher levels of wealth and education to support opposition parties because these parties campaign on platforms of democratization (Letsa Reference Letsa2019). Croke et al. (Reference Croke, Grossman, Larreguy and Marshall2016) make this argument most explicitly, finding that citizens with higher levels of education in Zimbabwe under the Mugabe regime were more likely to opt out of politics altogether (nonpartisanship) or support the opposition.Footnote ⁴ Recently, Rosenfeld (Reference Rosenfeld2020) has offered a compelling corrective to this theory, arguing that when the middle class is captured by the state, this relationship between socioeconomic status and opposition support disappears.Footnote ⁵

Taken together, these core hypotheses provide arguments for different sides of the same coin; the first about why some people might vote for ruling parties and the second about why some people would vote against them. However, neither provides a unified theory of vote choice in electoral autocracies. In other words, neither offers a compelling explanation for the inverse outcome, resulting in indeterminant predictions about whether or not socioeconomic status leads to support for the ruling party or support for the opposition. Further, despite the strong correlation between partisanship and both vote choice and support for democratization, very little of this literature directly considers partisanship itself. The lack of attention to partisanship in autocratic contexts is perhaps tied to an assumption that it is less meaningful than partisanship in democracies. Yet, if we are interested in understanding support for democracy itself, presumably such identities matter quite a bit (Jöst, Vergioglou, and Jacob Reference Jöst, Vergioglou and Jacob2022). Existing approaches tell us very little about the origins, meaning, or content of such identities.

A SOCIAL THEORY OF PARTISANSHIP

I argue that these individual-level explanations for participation in electoral autocracies cannot fully answer the question of partisanship because they miss the social context that has become key to our understanding of democratic politics (for an overview, see Campbell Reference Campbell2013). I propose that political socialization within social networks is central to understanding partisanship in electoral autocracies: what differentiates partisans as a group from nonpartisans is that they belong to social networks full of other people who support political parties and are therefore socialized to adopt the same partisan identities of their friends and families. The core argument of this article is that if “ruling party partisanship” and “opposition partisanship” are social identities built through a process of socialization, then a key observable implication will be that the best predictor of partisanship—both whether one is close to a party at all and whether one is close to the ruling party or an opposition party—is not education or access to patronage but the political nature of one’s social network.

Ordinary citizens tend to adopt the same identities of the other people in their social networks through the day-to-day interactions they have that influence their perceptions and beliefs (Tajfel Reference Tajfel1981). When people spend time with family and friends who bring value and meaning to their lives, they tend to take seriously their ideas, beliefs, and the meaning of their social identities. They are inclined to view the beliefs and identities of their loved ones through a positive lens and therefore approach these things with a more open mind than they would with a stranger or casual acquaintance. Although the qualitative data I present will lend some evidence to the argument that families play an outsized role in the adoption of partisan identities, parents are not the only source of political socialization in any given person’s life. I argue that political socialization—the “transmission” of a partisan identity—can occur between any close social contacts (Chazan Reference Chazan1978; Ventura Reference Ventura2001). However, the more homogenously partisan one’s social network, the more likely such transmission will occur.Footnote ⁶

This argument builds on several classic studies of partisanship and vote choice in democratic contexts, most specifically on Huckfeldt and Sprague’s (Reference Huckfeldt and Sprague1995) study of South Bend, Indiana, who find a high level of political congruence between discussion partners.Footnote ⁷ Using similar methods but with nationally representative samples, Sinclair (Reference Sinclair2012) shows a strong correlation between the partisanship of discussion partners and individual-level partisanship in the US. Further, using panel data, she finds considerable evidence that people even change parties when their social networks are discordant with their own partisanship, providing clear evidence that social networks influence the choice of partisan identity.Footnote ⁸

In addition to showing that political socialization within networks matters for understanding individual-level partisanship, I also argue that the partisan homogeneity of social networks itself is not driven by homophily (e.g., selecting into networks based on shared demographic characteristics, such as education or ethnicity, which are correlated with partisanship). In other words, it is plausible that because people seek relationships with people who are like themselves, the members of social networks may all possess similar levels of education (or ethnicity, socioeconomic status, etc.) and choose their political parties because of those characteristics and not because of political socialization or partisan contagion within networks. While most social networks feature some level of homophily based on education and other demographic factors, I show that these are not good predictors of individual-level partisanship.

Finally, building on the findings of Sinclair (Reference Sinclair2012) as well as Lazer et al. (Reference Lazer, Rubineau, Chetkovich, Katz and Neblo2010), I contend that not only is the most predictive feature of different networks partisanship itself (as opposed to, for example, education) but also that people are not exclusively self-selecting into partisan networks. In other words, network ecology—the origins of a social network—is not built upon shared partisanship. Instead, the mechanism that links partisan homogeneity of networks to individual partisanship is the process of political socialization; one’s partisanship is influenced and shaped by the people with whom one is already socializing. Evidence from democratic regimes suggests that political discussion with close contacts can and does change people’s political opinions (Beck Reference Beck2002; Beck et al. Reference Beck, Dalton, Greene and Huckfeldt2002; Klar Reference Klar2014; Mutz Reference Mutz2002), and I argue that this is the case in autocracies as well.

It is common sense that not only do we as humans choose to become close to people who are already like us but that as we become close to people, we also become more like them. Empirically, it is extremely difficult to pick apart these two separate processes (McPherson, Smith-Lovin, and Cook Reference McPherson, Smith-Lovin and Cook2001), and in fact Shalizi and Thomas (Reference Shalizi and Thomas2011) argue that it is functionally impossible to do so. But it is equally clear that the processes of network homophily and network contagion are not equivalent. For both individual and environmental reasons, some people are more likely than others to self-select into social networks that already look like them. Similarly, some people are more prone to social influence from their networks than are others. But for the purposes of this theory of partisanship, it is important to establish that partisan contagion can explain a high level of variation. If people are solely self-selecting into partisan networks, then clearly it is not the network that is causing partisanship, but vice versa.Footnote ⁹

CASE SELECTION: CAMEROON

I test this theory in Cameroon, a typical electoral autocracy for sub-Saharan Africa. Since independence from France and Britain in 1960–1 until its first multiparty elections in 1992, Cameroon was a single-party regime. By 1966, the first President, Ahmadou Ahidjo, had unified Cameroon’s multiparty system under the umbrella of his own political party, the Union nationale camerounaise (UNC) (Johnson Reference Johnson1970), and single-party elections were held every 5 years to re-elect Ahidjo and a national list of deputies to the National Assembly. Ahidjo stepped down from the Presidency in 1982, handing over power to his Prime Minister, Paul Biya. In 1985, as part of this succession process, Biya reorganized the UNC and renamed it the Rassemblement démocratique du peuple camerounais (RDPC), though it soon became clear that the RDPC was functionally equivalent to the UNC (Sindjoun Reference Sindjoun1999).

As pro-democracy protests swept sub-Saharan Africa in the early 1990s, an impressively large and unified protest movement hit Cameroon, mobilizing protests in its major cities and towns for months (Takougang and Krieger Reference Takougang and Krieger1998). Biya reluctantly conceded to legalizing opposition parties and set Cameroon’s first postcolonial multiparty elections for 1992. Paul Biya and the RDPC have won every election since 1992 with large majorities of the vote. As with any electoral autocracy, the regime has maintained its electoral advantage through various legal and extralegal channels. Access to the media (especially television broadcast) remains dominated by the state; the conflation between the ruling party and the state gives the RDPC unparalleled access to resources and campaign funding; gerrymandering continues to make winning structurally more difficult for the opposition (Albaugh Reference Albaugh2011; Takougang Reference Takougang2003). While vote choice in Cameroon has long been assumed to be tied to ethnic identities (Fonchingong Reference Fonchingong2005; Menthong Reference Menthong1998; Nyamnjoh Reference Nyamnjoh1999), partisanship has yet to be considered as a social identity itself in this context.

Thus, in the ways that are important to the argument in this article, Cameroon is a quite typical case of electoral autocracy. Most importantly, the regime has been around long enough—over 30 years—that the system has become predictable (Bombela and Daniel Reference Bombela and Daniel2023). For example, in a 2015 public opinion survey, 70% of Cameroonians reported that if an election were held tomorrow, the RDPC would “definitely win a clear majority of seats in the National Assembly” (Letsa Reference Letsa2020). At the same time, for the most part, ordinary citizens can support opposition parties without great risk to themselves, though highly visible support, such as protest, does increase this risk, especially since the outbreak of violence in the Anglophone regions in 2017 (Amin Reference Amin2021). It is possible that partisan socialization differs in cases of extreme repression, where any outward support for the opposition means risking life and limb.

QUALITATIVE EVIDENCE OF POLITICAL SOCIALIZATION

The theory presented in this article is primarily tested with original survey data collected in Cameroon, described in the following section. However, before formally testing the relationship between social networks and partisanship, I first present the stories of several Cameroonians who shared with me their reasons for supporting a political party. I collected their stories as a “plausibility check” on the concept of partisanship as a social identity and present them here in order to illustrate various processes of political socialization. These life histories do not focus on the full networks of the 12 subjects but instead on the mechanism linking individual identities to partisan homogeneity within networks: the process of political socialization.

Despite issues of generalizability, the qualitative data provide information on the process of political socialization that is extremely difficult to gather through large-n surveys. The nature of closed response options limits the ability of research subjects to share the life events and relationships that are at the heart of the construction of social identities. Even open-ended questions embedded in surveys are exceedingly limited in their ability to capture the relevant information necessary to understand how people come to support a political party. Most importantly of all, people themselves often are not cognizant of these processes. When asked outright why they support a party, people tend to draw upon superficial talking points—they report that they support a party “for change,” “for development,” or “for democracy.” Further, many people may find it socially undesirable to report that they have been influenced by other people, especially politically; even if unconsciously, we often like to think of ourselves as independent (Klar and Krupnikov Reference Klar and Krupnikov2016).

Thus, while the interviews provide rich detail of the proposed mechanism of political socialization, they are not representative of a larger sample of Cameroonians and are limited in their capacity to show a causal relationship between social networks and individual partisanship. However, if a significant proportion of the subjects had failed to identify key people in their lives who had influenced their political beliefs or had indicated that they had come to support a political party for clear ideological or economic reasons that were not tied to social considerations, it would cast considerable doubt upon the idea that partisanship is produced through socialization.

Thus, in the summer of 2022, I recruited 12 research subjects representing different types of partisanship—five ruling party partisans, five opposition partisans, and two nonpartisans—in three research sites across CameroonFootnote ¹⁰ to give an account of their early lives, how they came to learn about politics, and how they came to support one party or another (or no party at all, as the case may be). The life history portion of the interviews lasted between 1 and 2 hours. A short summary of the descriptive characteristics of each subject is presented in Table 1.Footnote ¹¹ Short descriptions of all 12 subjects can be found in Supplementary Appendix B.

Table 1. Description of the 12 Core Research Subjects

In sum, 9 of the 10 partisans clearly pinpointed people in their lives who “invited” them to join the party. For half of these partisans (Anita, Bertrand, Henri, Joseph, and Martin), this person was their father or a father-figure. For those who grew up without a father figure (Titus and Smart), it was friends who invited them later in life. Patience, Mireille, Jacques, and George all had nonpartisan parents; Jacques followed in the footsteps of his parents, while Patience and Mireille joined parties in their adulthood when they were invited by other important people in their lives. George was the only person I interviewed who joined a party—the RDPC—for purely materialist reasons.

Bertrand, a native of the RDPC stronghold Bafia, is the prototypical example of someone whose partisanship was strongly influenced by his family growing up (Jennings and Niemi Reference Jennings and Niemi1968). He was raised by his maternal grandfather, who had served in the French colonial army during WWII. At independence, his grandfather enthusiastically joined the UNC, serving as a Municipal Councilor for 25 years under first the UNC and then the RDPC after the 1985 transition. Bertrand reported that he idolized his grandfather; he was his grandfather’s namesake and “confidante.” There was no question in Bertrand’s mind that he would join—and eventually become active in—the RDPC, despite the fact that his own father, with whom he was not very close, was a Municipal Councilor for the largest opposition party in Cameroon, the Social Democratic Front (SDF). His father had even come to Bertrand to campaign for his vote, but Bertrand turned him down: “I was on my side and he was on his side, and I did not have any problem….[But] I did not vote for him because I had my aims in the RDPC and therefore I had nothing to do with the SDF” (June 21, 2022). Because of the disciplined stance of his beloved grandfather, Bertrand would support the RDPC no matter what—even if it meant voting against his own father.

Joseph, a lifelong supporter of the opposition SDF, described how two different uncles who had raised him influenced his choice. Ironically, both of these uncles were also members of the Cameroonian military. The first uncle (who Joseph lived with in Limbe for his primary education) was a marine, and the second (who he lived with in Yaoundé for his secondary education) was actually a gendarme.Footnote ¹² Because they were in the military, it was impossible for them to express their political beliefs outside of the house: “we discussed politics only at home, like if we were watching debates on the television. So that’s what I learned from him. That’s what even pushed me to support the party then” (June 25, 2022). It was not until he left his uncle’s house that he could openly support the SDF by going to meetings and rallies. For Joseph, it was a point of pride that he could openly support the SDF; almost as if through his open support he was not just representing his own political beliefs and identity but also those of the uncles who raised him, on their behalf.

For those who did not grow up in partisan households, friends and colleagues tended to play a larger role in the process of political socialization. For example, Smart, an Anglophone graduate student at the University of Yaoundé II, reported that his father passed away when he was just 2 years old and was raised by a single mother. It was not until he came to university in 2018 that he became interested in politics. When I asked Smart whether most of his friends at university support the regime or the opposition, he replied: “I think 99 percent of those I speak to are against the system” (June 27, 2022). Within this environment, Smart said that his roommate’s friend, Zachariah, used to come to his dorm every time there was a political event or debate on TV. Eventually, Zachariah, a supporter of the opposition Cameroon Renaissance Movement (MRC), invited Smart to a meeting of Stand Up for Cameroon (a consortium of opposition parties) where Smart discovered the Cameroon People’s Party (CPP) and became a member. When I asked Smart, “Do you think if not for Zachariah or for your other friends, you would not have gotten involved in politics? Or do you think you would have gone anyways?” He replied, “No, I was already into it, I think without them. I do not do this for people. I have been politically orientated from the onset. I just needed someone to show me the way—what Zachariah did in 2018, and that was enough” (June 27, 2022).

Of the 10 partisans that I interviewed, George was the only one who did not explain his partisanship in terms of the social relationships in his life. He was an RDPC partisan for pragmatic reasons, hoping that by joining the party he could advance his own career as well as the economic prospects of his hometown. In his 30s, he decided to stand as president of the youth wing of the RDPC subsection in his hometown, Bafut, with the thought of running for mayor. When I asked George about the fact that Bafut normally elects its representatives from the opposition SDF, he replied, “Yes, it is normally for SDF, so we wanted a change because we realized that when we are under SDF, so many things, they do not come to us. So we wanted to change that. We wanted the government to realize that now in Bafut they have the ruling party that is working there seriously so that they can bring so many other opportunities for us” (June 27, 2022). Just as the literature would suggest, it was clear that George—a young, highly educated, middle-class man from a family of bureaucrats—joined the ruling party in order to make a name for himself by using the party to provide public goods for his community. It is not entirely clear what makes George different from the other subjects I interviewed, though it is probably important that George did not grow up in a partisan household.

Apart from George, however, it is difficult to hear the stories of these 12 autocratic citizens—who collectively represent a wide range of socioeconomic, ethnic, and educational backgrounds—and conclude that political socialization was not a critical factor in explaining their support for a party. The subjects each pointed to people in their lives who clearly played an important role in their personal process of political socialization. The following section will show that as one’s social network becomes more homogenously partisan, the more likely this process of socialization is to occur.

TESTING THE RELATIONSHIP BETWEEN SOCIAL NETWORKS AND PARTISANSHIP

In order to test the relationship between social networks and partisanship, I fielded an original 1,200-respondent public opinion survey in Cameroon in January 2021. This large-n survey was implemented in four of Cameroon’s 10 regions: the Centre, Littoral, Ouest, and Sud.Footnote ¹³ Within these regions, enumeration areas (EAs) were selected based on two primary criteria. First, whether the EA was in an electoral district that voted consistently for the ruling party, voted consistently for the opposition, or was competitive between the ruling party and the opposition. Second, EAs were selected to produce variation within these three categories between rural and urban areas.Footnote ¹⁴ The survey was administered by enumerators who were native to the region, though oral comprehension of French is extremely high in Cameroon, so translation to local languages was not necessary. The full sampling schedule can be found in Supplementary Appendix C. In addition, 138 households were double sampled, meaning two people were selected to be interviewed in these households instead of just one. This is controlled for in the modeling by using two-way clustering of the standard errors at both the EA and household level. I also include standard demographic controls for the respondent’s age, gender, religion, ethnicity, interest in politics, and whether they were sampled in an urban, semi-urban, or rural area.Footnote ¹⁵ All models in the main text are ordinary least squares regressions.

Other Network Measures

In order to control for the effects of a network’s level of education, ethnicity, and proximity to patronage, several other questions were posed in regard to each of the discussion partners listed by the respondent. Similar to the measures of partisan homogeneity, an average was taken of each of these measures across each network. The average of these averages for the relevant measures, along with their standard deviations and minimum and maximum values, is listed in Table 2. The average number of discussion partners listed by a respondent was a little over two,Footnote ¹⁹ and the average level of education across discussion networks was the Brevet d’Études du Premier Cycle (BEPC)—a general education exam taken usually around 15 or 16 years of age. Second, a measure of “ethnic homogeneity” was constructed for each network, which measured the percentage of a respondent’s discussion network that matched the respondent’s own self-reported ethnicity. About 56% of these networks were entirely homogenous—the respondent reported that all of her discussion partners were the same ethnicity as her—20% were entirely heterogeneous, and 24% were mixed, with some co-ethnics and some non-co-ethnics.

Table 2. Descriptive Statistics of Network Characteristics

Finally, in order to control for the network’s proximity to patronage, the respondent was asked whether or not they thought each discussion partner had a larger political influence on the community, with response options ranging on a four-point scale from “no influence at all” to “very strong influence.” Presumably a discussion partner who has an outsized political influence on the general community is more likely to be connected to sources of patronage.

MAIN RESULTS

Generalized Partisanship

To estimate the relationship between an individual’s partisanship and their social network’s partisan homogeneity, the first set of models takes generalized partisanship of the respondent as their dependent variable; a value of zero indicates that the respondent does not feel close to a party, while a one indicates that the respondent feels close to any party (ruling party or opposition party). Table 3 presents the results. Model 1a includes just the network variables: average partisanship across all discussion partners (generalized partisan homogeneity), the size of the respondent’s network, the network’s average level of education, its ethnic homogeneity, and its average level of community influence. The coefficient on generalized partisan homogeneity is substantively very large: holding other network characteristics constant at their sample means, the average respondent with zero partisan discussion partners has only a 30% chance of feeling close to a party themselves. In contrast, the effect of having only partisan discussion partners more than doubles the odds of the respondent feeling close to a party—the average respondent with a homogenously partisan network has a 72% chance of being a partisan themselves, a difference of 45 percentage points. No other network-level variable even begins to approach the importance of generalized partisan homogeneity when it comes to explaining individual-level partisanship.

Table 3. Correlates of Generalized Partisanship

Note: Controls: political interest, age, gender, religion, ethnicity, urban/rural coefficients are not reported. Standard errors are given in parentheses. Two-way standard errors clustered at both the enumeration area and the household. Survey weights are included. * p < 0.10; ** p < 0.05; *** p < 0.01.

Model 1b includes the measure of generalized partisan homogeneity but compares it just to the individual-level variables from the existing literature: demographic factors as well as education, socioeconomic status, access to patronage, vote-buying, and experience with voter intimidation. The effect of generalized partisan homogeneity remains substantively large. However, many of the other factors also have significant relationships with generalized partisanship. There is a negative relationship between partisanship and socioeconomic status (although education seems less important). In addition, respondents who frequently contacted their mayor are slightly more likely to be partisans. Finally, someone who reported having been intimidated by a political party in the past is 71% likely to report feeling close to a party, compared to only 56% of respondents who have never been intimidated, a jump of 15%. However, this accounts for a very small subset of the sample—only 34 respondents in total (of 1,200) indicated that they have experienced voter intimidation, five of whom were nonpartisans.

Finally, Model 1c combines all of the variables into one regression. Figure 1 presents the marginal effects of partisan homogeneity from Model 1c. The bars along the x-axis display the distribution of the independent variable: each network’s generalized partisan homogeneity. Very little changes were seen between Models 1a, 1b, and 1c: generalized partisan homogeneity remains the strongest predictor of the respondent’s own partisan status by quite a bit, while the estimates for most of the control variables remain stable.

Figure 1. Marginal Effect of Network’s Partisanship on Individual Partisanship with Distribution of Independent Variable Along X-Axis

Note: Predicted values derived from Model 1c. Full results reported in Table D.1 in Supplementary Appendix D. Predicted values for “Network’s Partisan Homogeneity” (displayed) estimated from 0 to 1 at 0.1 intervals. All other variables are held at their sample means.

Opposition Partisanship

The relationship between a respondent’s opposition partisanship and their network’s opposition partisan homogeneity mirrors the relationship between individual-level generalized partisanship and average network-level partisanship. The models in Table 4 are identical to the models in the previous section, except that their dependent variable is opposition partisanship (where the zero category includes both ruling party partisans and nonpartisans) and their independent variable is each network’s average opposition partisan homogeneity. Again, far and away the biggest predictor of opposition partisanship is the percentage of one’s social network that supports an opposition party. Figure 2 presents the magnitude of this effect visually from Model 2c. Someone with zero opposition partisan discussion partners has only a 9% chance of supporting an opposition party themselves. In contrast, someone who is otherwise identical but who only discusses life and current events with opposition partisans has a 45% chance of also supporting an opposition party, an increase of 36 percentage points.

Table 4. Correlates of Opposition Partisanship

Note: Controls: political interest, age, gender, religion, ethnicity, urban/rural coefficients are not reported. Standard errors are given in parentheses. Two-way standard errors clustered at both the enumeration area and the household. Survey weights are included. * p < 0.10; ** p < 0.05; *** p < 0.01.

Figure 2. Marginal Effect of Network’s Opposition Partisanship on Individual Opposition Partisanship with Distribution of Independent Variable Along X-Axis

Note: Predicted values derived from Model 2c. Full results reported in Table D.2 in Supplementary Appendix D. Predicted values for “Network’s Opposition Partisan Homogeneity” (displayed) estimated from 0 to 1 at 0.1 intervals. All other variables are held at their sample means.

Controlling for the relevant confounders, the other factors that have a substantive effect on predicting opposition partisanship at the p < 0.01 level across all models are ethnic homophily, socioeconomic status, contact with a Municipal Councilor, belief in electoral patronage, and political intimidation. The networks of opposition partisans appear to be more ethnically homogenous than those of nonpartisans or ruling party partisans, and opposition partisans tend to have higher levels of socioeconomic status but are less likely to contact their municipal councilors, less likely to believe that voting for the ruling party will lead to local investments, and, unsurprisingly, more likely to have experienced political intimidation. These results lend evidence to the hypotheses proposed by Blaydes (Reference Blaydes2011) and Magaloni (Reference Magaloni2006) that ruling parties derive their support from promises of economic investment, though including these measures in the analysis does not affect the enormous magnitude of the effect of a network’s opposition homogeneity on individual partisanship.

NETWORK ECOLOGY

To what extent are these findings driven by ecological homophily (people create their social networks based on individual pre-existing partisan preferences) versus network contagion (peoples’ partisanship is influenced by their pre-existing social networks)? Although the large-n survey data cannot definitively adjudicate between these two processes, it does show that, at least to some extent, people’s partisanship is shaped by pre-existing social ties. The final statistical model mirrors Model 2c in Table 4 but analyzes only discussion partners who predate the respondent’s interest in the party they support.

When the respondent was originally asked whether or not they feel close to any political party, those who responded in the affirmative were further asked if they could recall approximately how old they were when they started feeling close to the party. Later in the survey, when asking about the respondent’s discussion partners, the survey also asked approximately how old the respondent was when they first met each discussion partner. Therefore, the final analysis only includes discussion partners whose relationship to the respondent predates their partisanship. In other words, it excludes people the respondent came to know after forming their partisan preferences, eliminating the possibility that the respondent felt close to a party and befriended a new discussion partner because they were copartisans.Footnote ²² Although many of these older relationships are family members, they are not exclusively so. In fact, 69% of the relationships that predate the respondent’s partisanship were reported as friends, neighbors, or work colleagues.

Controlling for all the measures from Model 2c, the effect of copartisanship remains high even when we look only at relationships that predate the respondent’s partisanship. Figure 3 shows the predictive margins of this relationship; the results of the full model are reported in Supplementary Appendix I. If the respondent’s discussion partner feels close to the ruling party, the respondent herself only has a 26% chance of feeling close to an opposition party. Everything else held equal; if the respondent’s partner instead feels close to an opposition party, the respondent has a 50% chance of also feeling close to an opposition party, a jump of 24 percentage points. Thus, at least in part, the social networks of these ordinary citizens are not being driven exclusively by partisan matching. Instead, as described in the qualitative data, there appears to be a process of political socialization occurring between friends and family. Partisan friends and family have the power to influence one another in politically salient ways.

Figure 3. Predictive Margins of Partisanship of Discussion Partner on Respondent’s Own Partisanship, Including Only Partners Who Predate Respondent’s Partisanship

Note: Predicted values derived from Model I.1, presented in Supplementary Appendix I. Predicted values for “Discussion Partner’s Partisanship” (displayed) estimated at 0 (ruling party partisan) and 1 (opposition partisan). All other variables are held at their sample means.