Probing PR: Does proportional representation induce ethnic power sharing and reduce conflict?

Abstract

Recent work on power sharing and civil conflicts in ethnically divided societies emphasizes the crucial distinction between mandates and their implementation. Formal power-sharing rules reduce the risk of armed conflict, but power-sharing practices mediate this effect. Political scientists frequently categorize proportional electoral rules (PR) as part of the broader class of power-sharing institutions, which should induce power-sharing practices, and in turn reduce the likelihood of intrastate armed conflict. Empirical evidence for this claim is indirect at best. Using mediation analysis, we assess whether PR rules or executive power-sharing institutions (or their combination) engender elite power sharing in ethnically divided societies and reduce the risk of ethnic intrastate conflict. Using different datasets, we find no evidence for a positive effect of PR on power-sharing practices, and some weak evidence that PR reduces intrastate conflict through other mechanisms. Recommendations of PR to ethnically divided societies or post-conflict environments as means to foster inclusion and thus reduce conflict should be reconsidered.

Do power-sharing institutions induce peace in ethnically divided societies? The answer frequently depends on the definition of these institutions (see Bormann et al., 2019; Cederman et al., 2022: chapter 2). While some scholars draw on Arend Lijphart’s (1977) concept of consociationalism, the scope of power sharing has notably broadened since. ¹ Electoral rules and more precisely proportional representation (PR) take a central place in this broader understanding of power sharing, especially as an element of liberal power sharing, which encourages elite cooperation in government rather than guaranteeing it (e.g., McCulloch, 2014). Yet whether or not PR actually induces power-sharing behavior remains a theoretically and empirically disputed question (Horowitz, 2002; Lijphart, 2002; Reilly, 2005).

We provide a novel test of the link between PR and the likelihood of intrastate ethnic conflict. ² In contrast to existing work which investigates a direct link between PR rules and ethnic conflict onset, we test the mechanism by which PR affects intrastate ethnic conflict through power-sharing behavior in the central government. ³ We conceptualize power-sharing behavior or practices as the de facto inclusion of representatives of two or more ethnic groups in a state’s central government. Our minimalist understanding of ethnic power sharing differs from Lijphart’s more encompassing notion of the grand coalition.

We compare the effect of PR rules to institutions that mandate governmental representation of specific ethnic groups. The pre-determined inclusion of representatives from distinct ethnic communities in key governmental positions in Lebanon and Bosnia are the best-known examples of such formal institutions, but many peace agreements contain similar provisions (e.g., Hartzell and Hoddie, 2007). Specifically, we define executive power-sharing institutions as the de jure guarantee of ministerial positions to representatives of different ethnic groups or the right of group representatives to veto government decisions. ⁴

To study the mechanism that connects PR and executive power-sharing institutions to the onset of intrastate conflict via practices, we rely on mediation analysis. Our results support earlier findings that executive power-sharing guarantees strengthen power-sharing practices and thereby reduce the likelihood of conflict onset. In contrast, PR rules have a null or negative effect on power-sharing practices by elites, which in turn has a negative effect on the likelihood of conflict onset. We thus find a positive mediated effect of PR on conflict risk. However, we also discover that PR has a negative direct effect on conflict onset. Put differently, PR reduces conflict risk but not through increased inclusion in the central government. Our results thus question the argument by proponents of liberal power sharing that PR induces power-sharing behavior.

Executive power sharing, PR, and conflict onset

Investigating which institutions stabilize post-conflict states and ethnically divided societies, political scientists typically distinguish between political, economic, territorial, and military power sharing in peace agreements (e.g., Hartzell and Hoddie, 2007). Among these, political power sharing takes on central importance in overcoming the severe lack of trust among former combatants in post-conflict societies (Mattes and Savun, 2009). Unfortunately, little agreement exists over the exact institutions that induce political power sharing in post-conflict contexts. Whereas some scholars highlight guaranteed government inclusion for political antagonists (Jarstad and Nilsson, 2008; Johnson, 2021), others conceptualize power sharing as varying bundles of guaranteed government inclusion, proportionality in bureaucratic appointments, and PR rules (Hartzell and Hoddie, 2007; Mattes and Savun, 2009).

Beyond post-conflict settings, scholars distinguish “liberal” from “corporate” approaches to power sharing (Juon, 2020; McCulloch, 2014; McGarry & O’Leary, 2007). PR belongs to the family of liberal institutions that allow political elites to “self-determine” how to cooperate in the government. In contrast, corporate institutions “pre-determine” or guarantee political representation of political rivals (McGarry & O’Leary, 2007). Research in this tradition also bundles various institutions into broader categories. If individual components of the broad categories have countervailing effects on conflict risk, empirical investigations overlook pacifying effects of individual institutions (Binningsbø, 2013). Avoiding the aggregation of multiple institutions, we study the two most central power sharing solutions separately: (1) executive power-sharing institutions that guarantee ministerial positions or permit politicians to veto policies and (2) PR rules.

Existing scholarly work on the link between PR and the risk of conflict onset reaches inconclusive, if not contradictory, conclusions (Cammett and Malesky, 2012; Cohen, 1997; Reynal-Querol, 2002). Some scholars argue that this lack of agreement results from complex conditional effects of PR on conflict risk (Ishiyama and Shliek, 2020; Neudorfer et al., 2022; Schulte and Trinn, 2022; Selway and Templeman, 2012). We point to another issue: existing research on the effects of PR does not consider the underlying theoretical mechanism. Both executive power-sharing guarantees and PR rules should influence the risk of conflict onset through power-sharing practices, the de facto inclusion of multiple politically rival or ethnically distinct elites into the central government of a state.

Cohen (1997: 610) concisely describes the effect of PR: “Proportional regimes formally include political factions in decision-making institutions,” and by “dispersing authority and victory, proportional institutions endear dissatisfied ethnic groups groups to the regime” (ibid., 613). Lijphart (2002: 53) himself adds that “the degree of electoral proportionality is strongly related to the degree of multipartism, which in turn is closely related to [coalition cabinets].” Figure 1 captures this logic. PR and executive power sharing institutions should reduce the risk of armed conflict by increasing the practice of ethnic inclusion which decreases conflict risk.

Yet, we do not know any study that traces the entire mechanism from PR through power-sharing practices to conflict onset. Existing work often focuses on one link in the causal chain and finds only weak evidence that PR induces power sharing (Bormann, 2019; Pospieszna and Schneider, 2013). None of these studies investigate the link between practices and conflict risk. Our contribution is then two-fold: First, we compare two of the most prominent power-sharing institutions, executive guarantees and PR, rather than bundling multiple formal rules into an aggregate index. Second, we empirically trace the mechanism from these institutions via power-sharing practices to conflict risk.

Figure 1.

Theoretical Mechanism.

Data and methods

To study the mechanism that links PR and executive power-sharing guarantees to intrastate conflict onset via power-sharing practices, we employ mediation analysis at the country and group-level. We combine four different data sources in our analysis.

Our first explanatory variable, a binary indicator of proportional electoral rules derives from the “Democratic Electoral Systems” dataset. It covers all democratic, lower-house elections between 1945 and 2020, and defines proportional representation when a state employs “a quota or divisor system in multi-member districts” (Bormann and Golder, 2022). We fill in country-years between elections with the electoral rule used in the last election, drop all non-democratic country-years, and use majoritarian or mixed electoral systems as a baseline in our models. Our second explanatory variable, a dichotomous indicator of executive power-sharing institutions, captures the presence of one out of three mandates in constitutions and peace agreements: (1) a grand coalition, (2) reserved seats in the executive, or (3) mutual vetos (Strøm et al., 2017). The data encompass the period 1974 to 2009 at both the country and group-level (Cederman et al., 2022).

To operationalize our mediator, power-sharing practices, we draw on the Ethnic Power Relations (EPR) dataset (Vogt et al., 2015). It records the share of politically relevant ethnic groups that participate in the executive of a state as of January 1st of a given calendar year, provided that at least two groups share power. At the group-level, we use a simple indicator of group inclusion in the central government. We record power-sharing practices only if the government includes at least two groups, and drop all country-years with only one politically relevant group. Hence, our analysis focuses on cases where ethnic divisions are present.

Finally, we measure our outcome variable with a binary indicator of intrastate ethnic armed conflict. We draw on the UCDP Armed Conflict Database to identify the onset of armed conflicts with at least 25 battle-deaths in a calendar year (Gleditsch et al., 2002; Pettersson and Öberg, 2020). We retain only conflict onsets, when an armed organization claims to act on behalf of and recruits from a given ethnic group (Vogt et al., 2015). In our country-year analysis, we keep observations with ongoing conflict in our sample, as other ethnic groups might initiate additional armed conflicts. In our group-year analysis, we drop years with ongoing conflict, as the same group cannot fight the government again in that period. ⁵

To estimate the direct and indirect treatment effects of PR rules and executive power-sharing institutions via power-sharing practices on ethnic conflict onset, we rely on mediation analysis for multiple treatments (Hayes, 2017) and adapt it to a Bayesian framework. ⁶ We estimate three regression models: The first model, our mediator equation, links our treatments—PR and executive power-sharing institutions—to our mediator, power-sharing practices. The second and third models are different versions of the outcome equation. One links both the treatments and the mediator to our outcome, conflict onset. The other only links the treatments to the outcome, but drops the mediator. From these three models, we obtain estimates of two effects: (1) the indirect or mediated effects of PR and executive power-sharing institutions on the risk of ethnic conflict onset and (2) the total effect of those institutions on conflict onset.

We determine the treatment effects of PR and power-sharing guarantees from the posterior samples of the coefficients for our three regressions by applying a procedure that Muthén et al. (2017) call “counterfactually-defined causal effects.” To obtain the total treatment effect, we use the samples from the posterior distribution of our outcome model that omits the mediator, which allows us to calculate the average predicted probability for situations when either of the treatments is equal to 1 while the other is set to 0. Comparing these distributions to the one obtained for the control group (all treatment variables set to zero) gives us the relative “counterfactually-defined causal effects” for each treatment (Muthén et al., 2017).

We repeat this procedure for the posterior samples from the mediator model. Comparing the posterior distributions when a treatment variable is set to 1 compared to the distribution when all treatments are set to 0, we obtain treatment effects for our institutional measures on power-sharing practices. Finally, we use these predicted differences from the mediator equation to assess the change in conflict probability in the outcome model that contains the predicted mediator. These results provide the indirect or mediated treatment effects of PR and executive power-sharing institutions.

The procedure is straightforward for the country-level analysis because our mediator is continuous, and we obtain meaningful predictions for the share of included groups. At the group-level, our mediator of power-sharing practices is dichotomous, and we obtain probabilities rather than observable values of the mediator. We therefore use a binomial distribution to generate predicted probabilities for the mediator, which we then plug into our outcome equation to assess by how much the predicted probabilities of conflict onset change. Relying on the binomial distribution introduces additional uncertainty into our results. To obtain more reliable effects, we run the final step five times and average the results.

Analysis

To compare our analysis to previous research, we adapt the model from Bormann et al.’s (2019) mediation analysis including its control variables. We graphically depict the estimated mediated and total treatment effects.

Country level

In Figure 2 we depict the mediated and total effects of PR and power-sharing institutions on intrastate conflict onset, conditional on whether or not the respective other institution is present. ⁷ The average mediated effects of PR is virtually zero on average (hollow diamonds), regardless of whether executive power-sharing institutions are present or not. Moreover, we find no evidence for the long-held hypothesis that PR induces power-sharing behavior in the central government (coefficient mean: −0.00; CI: [−0.04,0.03]; only shown in Table A5). Even though our mediator, power-sharing practices, decreases conflict risk (−0.32 [−1.00; 0.33]), PR rules have no role in this conflict-dampening effect.

Figure 2.

Mediated and total effect of PR and power sharing institutions, democracies 1974–2009.

In contrast, the average mediated effects for executive power-sharing institutions in the bottom panel of Figure 2 are negative, irrespective of the presence of PR. Although the 95% credible intervals comprise some positive values, the results suggest a conflict-dampening effect of executive power-sharing guarantees via practices. ⁸

Additionally, Figure 2 displays the total effects of both PR and executive power-sharing institutions (solid circles). Whereas the total effects of executive power-sharing institutions are negative whether or not PR is present, the estimated total effect of PR on ethnic conflict onset is only negative in the absence of executive power-sharing institutions (top panel). Combining the insights from the mediated and the direct effects, we conclude that PR might decrease the risk of conflict onset, but not through power-sharing practices.

Group level

Shifting our analysis to the group-level has two consequences for the interpretation of our results. First, it decreases the uncertainty for the estimated effects of PR because we use the same country-year observation multiple times for different groups. Second, the group-level classification of executive power-sharing institutions is more conservative and biases our results against finding a credible effect. Cederman et al. (2022) only coded executive power-sharing institutions at the group-level if the formal rules made an explicit reference to any particular group but did not capture less formal applications of country-wide power-sharing rules that presumably benefit all groups. These cases end up in our control condition of no power-sharing institutions.

Figure 3 depicts total and mediated effects of PR and executive power-sharing guarantees. ⁹ Once again defying existing scholarly expectations, PR exerts a positive effect on conflict risk when it is mediated by power-sharing practices (hollow diamonds, top panel). This surprising result comes about because PR decreases the probability of ethnic inclusion in the central government by −0.12 [−0.10 -0.09] on average. In turn, power-sharing practices decrease conflict risk. Combining the two negative relationships leads to a positive mediated effect on conflict onset.

Figure 3.

Mediated and total effects of PR and power-sharing institutions, democracies 1974–2009.

In contrast, executive power-sharing institutions increase the probability of ethnic inclusion by 0.64 [0.34; 0.67] on average, and thus the mediated effect on ethnic armed conflict risk is negative on average, especially when a country does not run PR rules (hollow diamonds, bottom panel, Figure 3). Our group-level results provide the clearest evidence that executive power-sharing institutions on their own decrease the risk of ethnic conflict onset through power-sharing practices.

The total effects of executive power-sharing institutions and PR on conflict onset do not differ much from the country-level analysis (solid circles). Most of them decrease the risk of conflict. PR reduces the likelihood of ethnic conflict onset but not through power-sharing practices.

Robustness tests

Our results are robust to a wider time period (Appendix C.1), a broader sample of electoral rules that encompasses democracies, non-democracies, and hybrid regimes (C.2-3), alternative modeling approaches at the country-level (C.4), using ethnic parties as a mediator (C.7), and splitting the sample into pre and post-conflict periods (C.8). As our data is non-random, our estimated effects are only causal if no confounding is present, that is, when the “sequential ignorability assumption” holds (Imai et al., 2010: 310). Our sensitivity analyses suggest that the direction of the mediated effect of executive power-sharing institutions is robust, as long as these institutions are more likely to be introduced in cases where conflict is likely. The opposite holds for PR. In addition, we show that both the mediator and outcome model estimates are robust to unobserved confounders (C.5-6).

Conclusion

In this article, we analyzed the mechanism by which common power-sharing institutions influence the risk of armed conflict through de facto ethnic power sharing in the central government. In contrast to the widely held expectation that PR rules induce multi-ethnic power sharing and thus reduce conflict risk, we do not find any evidence for this mechanism. PR does not have the purported positive effect on power-sharing practices, and thus proves to be a counter-productive institutional intervention in ethnically divided societies. Formal guarantees of executive power sharing, however, are associated with a higher chance of power-sharing practices and a reduced risk of armed conflict.

We find some evidence that PR nevertheless reduces the risk of ethnic armed conflict onset but not through power-sharing practices. We can only speculate what underlies this negative relationship. Better governance outcomes in closed-list PR systems could offer one explanation (Cammett and Malesky, 2012). As this proposition has only been explored in one post-conflict study, evidence for it is scarce. Until proponents of PR present a convincing theoretical mechanism linking PR to peace and supportive empirical evidence, recommendations for the use of this electoral rule to prevent ethnic intrastate conflicts should be received cautiously.

Supplemental Material

Supplemental Material - Probing PR: Does proportional representation induce ethnic power sharing and reduce conflict?

Supplemental Material for Probing PR: Does proportional representation induce ethnic power sharing and reduce conflict? by Nils-Christian Bormann, Simon Hug in Research & Politics.