Judicial Power and Legitimacy

We distinguish judicial power from the concept of institutional legitimacy—rightful authority to render rulings for the polity and a widespread belief that a court is “appropriate, proper, and just” (Tyler Reference Tyler2006a, 375; see also Bartels and Johnston Reference Bartels and Johnston2020; Gibson and Nelson Reference Gibson and Nelson2014). Legitimacy is a normative, more aggregate concept that develops when citizens and elites ascribe power to a court as an end in itself. Under this definition, when courts are legitimate, support for their power is not conditional on policy or partisan agreement with court outputs (e.g., Easton Reference Easton1965). It is important to distinguish judicial power (and related concepts such as trust) from legitimacy because conflating them can lead to analyses that suffer from an observational equivalence problem: People may have high support for judicial power because (1) they agree with a court’s rulings or expect partisan advantage or (2) they believe that a court has rightful authority regardless of partisan considerations. But those levels of support are derived from two drastically different processes. The first connotes instrumental support, while the latter connotes legitimacy (see also Bartels and Johnston Reference Bartels and Johnston2020).

By this logic, legitimacy is difficult to evaluate and is best inferred when tested in the face of political or partisan disagreement (e.g., Bartels and Johnston Reference Bartels and Johnston2020; Gibson Reference Gibson, Bornstein and Tomkins2015; Gibson, Caldeira, and Spence Reference Gibson, Caldeira and Spence2002). Support for judicial power, which can be directly conceptualized and measured, provides a basis for such a test. If people support judicial power only when they agree with a ruling or foresee partisan political advantage and they withhold power otherwise, that is a detriment to legitimacy. Legitimacy can therefore be evaluated on the basis of whether partisan or policy/ideological considerations influence support for judicial power (see Bartels and Johnston Reference Bartels and Johnston2020). More recent “revisionist” works in the US case have found such an effect on more diffuse forms of court support, a challenge to conventional wisdom (Bartels and Johnston Reference Bartels and Johnston2013, Reference Bartels and Johnston2020; Christenson and Glick Reference Christenson and Glick2015; Clark and Kastellec Reference Clark and Kastellec2015; Nicholson and Hansford Reference Nicholson and Hansford2014; Zilis Reference Zilis2018). But the literature (including some of those studies) has drifted on both conceptualization and measurement surrounding legitimacy, including the aforementioned issue of equating it with diffuse types of court support or conflating concepts like support for court-curbing, procedural fairness versus politicization, and trust into a single “legitimacy” or “diffuse support” measure (see, e.g., Gibson, Caldeira, and Spence Reference Gibson, Caldeira and Spence2003a; Bartels and Johnston Reference Bartels and Johnston2013; Gibson and Nelson Reference Gibson and Nelson2015).^{Footnote 4}

Comparison with Indicators of More Specific Support

“Support for judicial power”—our key outcome of focus—represents an evaluation over the preferred role of courts in a political system. In our analysis, we compare this more “diffuse” form of support with more evaluative perceptions—along the lines of “specific support,” or approval of judicial decisions. Evaluative concepts for courts include approval of rulings, confidence, and trust (see Gibson Reference Gibson2011; Gibson, Caldeira, and Spence Reference Gibson, Caldeira and Spence2003a; Staton Reference Staton2010). We focus on “trust” because it is both backward and forward looking. It can be formed on the basis of how good of a job people believe courts are doing in the present and will do in the future (Staton Reference Staton2010). It can also tap their experience with the courts. We distinguish this evaluative perception of courts from the assessment of “support” for judicial power, and moreover, it provides a key comparison with such support measures. We should expect “specific support” oriented measures, like trust, to be more rooted in partisan political foundations and thus track partisan preferences (e.g., Gibson, Caldeira, and Baird Reference Gibson, Caldeira and Baird1998; Gibson, Caldeira, and Spence Reference Gibson, Caldeira and Spence2003a). The extent to which support for judicial power is rooted in these same factors suggests a more politicized foundation to the broad concept of support for judicial power.

Measuring Attitudes about Courts

Afrobarometer conducts nationally representative surveys across Africa.^{Footnote 10} Seven have been conducted in Ghana since 1999 (Table 1).^{Footnote 11} We analyze data from the merged multi-country datasets from Afrobarometer rounds 1–6 (Afrobarometer Data 1999–2016) and from the round 7 survey in Ghana (Afrobarometer Data, 2018). Table 2 provides the question wording and coding for each outcome variable.

TABLE 2.

Summary of Variables and Measurement from Afrobarometer Surveys

The main outcome variables capture the distinction between support for horizontal and vertical power. The first, president should obey the courts, measures support for horizontal power. The second, people should abide by court decisions, measures support for vertical power. We create 5-point measures where higher values indicate greater support for judicial power (see Table 2). We rescale these variables to run on a 0–1 scale.^{Footnote 12}

We also measure trust in the courts of law. Trust taps an evaluative perception of the job courts are doing and expectations about how reliably they will carry out their functions. Trust may also capture individual experiences with the court system.^{Footnote 13} We create a 4-point scale where higher values indicate greater trust in the courts of law. Trust is also recoded to run on a 0–1 scale.

To enhance our reporting of substantive significance, we also create dichotomous outcome measures. For horizontal power and vertical power, we code those who agree or strongly agree with the statements as 1, and all others as 0, including those who are neutral. For trust, we code those who trust courts “a lot” or “somewhat” as 1, and those who trust courts “a little” or “not at all” as 0. In all analyses, we present results on the continuous and the binary measures.

Measuring Partisan Alignment with the President

The main explanatory variable captures respondents’ partisan alignment with the president at the time they are surveyed. Our central measure is dichotomous, close to president’s party, which takes a value of 1 of the respondent reports feeling close to the president’s party, and 0 otherwise. Independents are those who do not feel close to any political party. In Online Appendix D, we also analyze two other measures. President party voter indicates whether the respondent would vote for the president’s party if the election were held tomorrow. President coethnic indicates whether the respondent is a co-ethnic with the president.

Graphical Analysis

We first present the evidence graphically. Figure 1 presents the means of support for horizontal power among NDC and NPP supporters by year (survey round). The figure illustrates a partisan gap in support for horizontal judicial power.^{Footnote 14} When the NPP is in power, NPP supporters (dark circles) are less likely to believe the president should be bound by courts than are NDC supporters (light triangles). After an NDC president is elected in 2008, the partisan gap flips. For example, in 2005, 82% of NDC supporters support horizontal power, while 74% of NPP supporters do (Panel B). In 2012 following the election of an NDC president in 2008, support for horizontal power among NDC supporters drops to 71%, while it rises to 79% among NPP supporters. Support for horizontal power drops among both co-partisans and out-partisans in 2014, although the drop is more significant for co-partisans. This drop could be driven by popular reactions to the Supreme Court’s role in the 2012 election. Notably, the direction of the partisan gap reverses again after an NPP president is elected in 2016. This reverse trend is especially notable given that the survey was conducted in 2017 after the corruption scandal that shook public confidence in Ghana’s judicial system in 2015–16. This suggests that partisan considerations may be sufficiently strong to outweigh the effect of other important events that might otherwise reduce support court power. These results corroborate the “dynamic partisan alignment hypothesis.”

FIGURE 1.

Support for Horizontal Judicial Power Over Time in Ghana

Note: Data from Afrobarometer rounds 3–7. Figure presents means of support for horizontal power among NPP supporters (dark circles) and NDC supporters (light triangles). Vertical dotted lines indicate a change in the presidency.

Figure 2 presents evidence for support for vertical power.^{Footnote 15} Here, both partisan groups trend very similarly from 2002 to 2012, with attitudes seemingly unaffected by the transition from the NPP to the NDC. A partisan gap does emerge in 2014 (round 6), with the president’s (NDC) supporters expressing greater support for vertical power and out-partisans becoming significantly less supportive of vertical power. This drop among NPP supporters (out-partisans) likely results from their reaction to the Supreme Court’s role in the 2012 elections. NPP partisans do become somewhat more positive about vertical judicial power after an NPP president was elected in 2016. However, on balance the evidence suggests that, in Ghana, attitudes about courts’ vertical power are not conditioned by partisanship.

FIGURE 2.

Support for Vertical Judicial Power Over Time in Ghana

Note: Data from Afrobarometer rounds 2–7. Figure presents means of support for vertical power among NPP supporters (dark circles) and NDC supporters (light triangles). Vertical dotted lines indicate a change in the presidency.

Figure 3 presents evidence on trust in courts. The figure illustrates a substantial partisan gap in trust in the courts.^{Footnote 16} Notably, this partisan gap reverses in direction after all three transitions of presidential power, with the presidents’ co-partisans expressing greater trust in the courts. For example, NPP supporters’ trust in the courts increases substantially after the NDCs Jerry Rawlings—who had led the authoritarian government in the 1980s and had won the first two democratic elections in 1992 and 1996—was replaced by the NPPs John Kufuour: In 1999 (round 1), 50% of NPP supporters trust courts and in 2002 (round 2), 86% of them do (Panel B). By contrast, NDC supporters trust in courts drops from 67% in 1999 (round 1) to 41% in 2002 (round 2). After 2002, NPP supporters trend downward in trust, a trend that accelerates when NDC presidents come to power from 2008 to 2016. There was a sharp decline in NPP supporters’ (out-partisans) trust in courts in 2014 (round 6), which again was the likely result of the Supreme Court’s role in 2012 election. NPP supporters trust in the courts rebounds in 2017 (round 7) following the election of an NPP president in 2016. This sharp increase in support for the courts is particularly notable given that it occurred during the same period as the high profile corruption scandal in the judiciary that began in 2015. As noted above, this indicates that partisan considerations may be sufficiently strong that their impact on trust outweighs even the impact of an event that should reduce overall levels of trust in the courts. Overall, the evidence highlights that Ghanaians’ trust in the courts appears to be impacted by presidential turnovers.

FIGURE 3.

Trust in the Courts Over Time in Ghana

Note: Data from Afrobarometer rounds 1–7. Figure presents means of trust in courts among NPP supporters (dark circles) and NDC supporters (light triangles). Vertical dotted lines indicate a change in the presidency.

Difference-in-Differences Regression Analysis

To more formally analyze the impact of co-partisanship with the president, we estimate the following regression model using ordinary least squares.^{Footnote 17}

(1)$${Y_{ijt}} = \alpha + \beta {P_{ijt}} + \theta {X_{ijt}} + {\gamma _j} + {\lambda _t} + \delta {\gamma _j} \,\times\, {\lambda _t} + {\varepsilon _{ijt}}$$

Y _ijt is an outcome measure for individual i from partisan group j (NDC supporter, NPP supporter, or independent) at time t. P _ijt is a dummy variable indicating whether individual i of partisan group j is a supporter of the president’s party at time t. λ _t are fixed effects for each survey year, to control for time-specific factors that might impact support for courts. P _ijt is the “treatment” variable; in each survey round, some respondents are treated while others are not. Because the party of the president changes over time, the partisan group that is “treated” changes across survey rounds. Importantly, the model includes fixed effects for each partisan group, γ _j. These fixed effects control for time-invariant differences between members of these different partisan groups. Thus, if NDC voters or NPP voters are different in ways that might be relevant for attitudes about courts, the fixed effects control for these differences. Their inclusion, along with survey-year fixed effects, makes β analogous to the interaction term in a more typical (i.e., two-time period, one intervention) difference-in-differences model (for a similar specification, see Franck and Rainer Reference Franck and Rainer2012). β captures the average impact of changes in a partisan connection to the president (i.e., the average difference-in-differences). The model also includes an interaction between the partisan fixed effects and the year fixed effects (γ _j × λ _t), which controls for time-specific shocks that may impact different partisan groups in different ways. X _ijt is a vector of controls including age, gender, poverty, residence in a rural area, democratic values, and education.

To interpret β as a causal effect, we must invoke several assumptions. The first is an exogeneity assumption (Besley and Case Reference Besley and Case2000). This assumption is plausible, as it is unlikely that changing attitudes about courts among a party’s supporters would impact that party’s likelihood of winning a presidential election (reverse causation appears unlikely). Indeed, Ghanaian presidential elections are often decided by the choices of swing and independent voters (Fridy Reference Fridy2012). In addition, Ghanaian presidential elections have often been decided by extremely close vote margins. The 2008 presidential runoff election was decided by less than one percentage point: about 80,000 votes in an election in which nine million cast ballots.^{Footnote 18}

Second, we must invoke the “parallel trends assumption” (Angrist and Pischke Reference Angrist and Pischke2008): changes in the presidency are not confounded with partisan group–specific time trends. For example, if NPP and NDC supporters’ attitudes were trending in different directions before transitions of presidential power, this could produce a partisan gap in attitudes for reasons unrelated to the transition. For this assumption, we can examine Figures 1, 2, and 3. The figures make clear that time trends are not likely to be confounded with transitions of power. Indeed, we observe sometimes very sharp changes in trends following transitions of power. That such changes occur with every transition in Figures 1 and 3 further increases our confidence that the results are not being driven by time trends that are confounded with changes in the presidency. This highlights an advantage of studying Ghana; multiple alternations in power make it less likely that group-specific trends unrelated to which party controls the presidency are responsible for the partisan gaps we identify.

Third, while partisan group fixed effects controls for time-invariant variables, we must also consider time-varying factors that could be confounded with presidential transitions. To do so, we include individual-level controls for age, gender, residence in a rural area, democratic values, and education.^{Footnote 19} These latter two variables fall broadly within process-based approaches and prior shows that they can impact attitudes about courts. Online Appendix C describes the question wordings and construction of these variables. The inclusion of these controls helps us to rule out that NPP and NDC supporters are changing in ways that impact their views about courts. The possibility that NDC and NPP supporters are changing in ways that are confounded with presidential transitions highlights an additional advantage of studying Ghana. Because we are leveraging multiple transitions, our results are unlikely to be driven by unobserved changes among NDC and NPP supporters. For example, if one group were becoming more educated over time, this could help them secure the presidency and influence their attitudes about courts. However, a group’s increased education is very unlikely to explain why attitudes toward the courts reverse trend after their party loses the presidency, and then reverse trend again after their party comes to power again, as we observe in Figures 1 and 3.

Finally, we consider the possibility that judicial and other institutions are changing in ways that are confounded with presidential transitions and impact attitudes about courts. To control for time-specific changes that impact all Ghanaians equally, we include survey-year fixed effects (λ _t). Because presidential and opposition party supporters may have different reactions to changes in the judiciary—and time-specific effects generally—we also include interactions between the partisan group fixed effects and the year fixed effects. Thus, the specification controls for the possibility that time-specific shocks and factors, including those related to changing judicial institutions, may have a different impact on different partisan groups.

To further assess the possibility that judicial institutions are changing over time, Figure A.1 in Online Appendix A presents trends over time in Ghana on a series of judicial indicators as measured by VDEM (Coppedge et al. Reference Coppedge, Gerring, Lindberg, Skaaning, Teorell, Altman, Bernhard, Fish, Glynn, Hicken, Henrik Knutsen, Krusell, Lührmann, Marquardt, McMann, Mechkova, Olin, Paxton, Pemstein, Pernes, Petrarca, Römer, Saxer, Seim, Sigman, Staton, Stepanova and Wilson2017). Across the study time period, Ghana’s judicial institutions have been relatively stable: Judicial independence and executive compliance with the judiciary are relatively high, while executive attacks on the judiciary and politically motivated judicial purges or court packing have been rare, and most indicators remain constant across the study time period.

Regression Results

Table 3 presents the main results, with the coefficients in row 1 providing our estimates of β. Columns 1 and 2 show that co-partisanship with the president significantly reduces support for horizontal judicial power. The effect size in column 1 represents about a 15% decrease from the mean among opposition supporters. Column 2 shows that co-partisans of the president are about 8 percentage points less likely to believe that the president should obey the courts and the law (p = 0.058). The effect sizes in both columns are similar to the effect size of support for democracy, an indicator of democratic values, which is a factor that the existing literature highlights is an important driver of support for court power.

TABLE 3.

Partisan Alignment with the President and Public Support for Courts in Ghana, Difference-in-differences Analyses

Note: Robust standard errors in parentheses. ** p < 0.01, * p < 0.05. Omitted reference category for the education variables are those with No Formal Schooling. Omitted reference category for the partisanship variables are independents and those who support minor parties.

Columns 3 and 4 present the results on vertical power. In contrast to the results from all countries that we present below, co-partisanship with the president in Ghana has no impact on support for vertical judicial power. The partisan divisions on vertical power may be smaller because of survey question wording: The horizontal survey item directly mentions the president, which likely primes partisanship, while the vertical item does not directly reference politics. Columns 5 and 6 show that co-partisanship with the president significantly enhances trust in courts. The effect size in column 5 represents about a 16% increase in trust relative to opposition party supporters and independents. Column 6 shows that co-partisanship with the president increases trust by about 7 percentage points (p = 0.054), about a 14% increase over the mean among opposition supporters.

Table 3 also shows that support for democracy is a significant predictor of support for horizontal and vertical power, as well as trust in the courts. These patterns are consistent with process-based models that emphasize the importance of democratic values in shaping attitudes toward judicial institutions. While more educated Ghanaians are less trusting of courts—perhaps because of increased exposure or knowledge—education does not consistently predict attitudes about power.^{Footnote 20}

Table 4 presents the results from more stringent tests. We restrict each analysis to include only two survey rounds: the one that precedes and the one that follows each presidential transition (Table 1). In narrowing the time window, we can be more confident that contextual and judicial variables are being held constant while the party of the president changes. For example, as Online Appendix A shows, the characteristics of the judiciary in Ghana generally remain relatively constant before and after transitions. Moreover, we can be more confident that the composition of the judiciary remains more constant. For example, although the NPP president who was elected in 2016 has recently appointed new justices to the Supreme Court, the AB survey in 2017 was conducted well before these appointments.

TABLE 4.

Partisan Alignment with the President and Public Support for Courts in Ghana Narrowing Analysis to Time Window Around Presidential Transitions

Note: Robust standard errors in parentheses. ** p < 0.01, * p < 0.05. All models include partisian and survey-year fixed effects, and all controls from Table 3. Top panel analyzes change from NDC to NPP president in 2016 (surveys in 2014 and 2017). Middle panel analyzes change from NPP president to NDC president in 2008 (surveys in 2008, before the election, and in 2012). Bottom panel analyzes change from NDC president to NPP president in 2000 (surveys in 1999 and 2002). Only the trust measure was included in round 1.

The main results hold when we conduct these more narrowly focused analyses. Panels A and B, which examine the transitions that occurred in 2008 and 2016, show that presidential co-partisanship reduced support for horizontal power.^{Footnote 21} All three panels present the results on trust and show that presidential co-partisans are substantially more trusting of courts in the periods before and after all three presidential transitions.

Online Appendix J shows that these results are robust to the use of an alternative measure of presidential co-partisanship, president party voter; the inclusion of region fixed effects, which control for additional differences that may exist between the supporters of the major parties;^{Footnote 22} the inclusion of interactions between the round fixed effects and the individual-level controls, which control for changes over time in the impact of the individual-level variables; and a more conservative approach to clustering standard errors.

Generalizability: Evidence from 34 African Countries

To demonstrate generalizability, we analyze survey data gathered by the Afrobarometer in 34 African countries. We analyze data from rounds 1–6.^{Footnote 23} Table 5 displays the sample countries and their regime type at the time of the AB surveys, as measured by VDEM (Coppedge et al. Reference Coppedge, Gerring, Lindberg, Skaaning, Teorell, Altman, Bernhard, Fish, Glynn, Hicken, Henrik Knutsen, Krusell, Lührmann, Marquardt, McMann, Mechkova, Olin, Paxton, Pemstein, Pernes, Petrarca, Römer, Saxer, Seim, Sigman, Staton, Stepanova and Wilson2017). Online Appendix B presents further information about the sample.

TABLE 5.

Sample Countries by Regime Type (VDEM) and AB Rounds Surveyed

Note: Table indicates countries in the sample and their VDEM regime category for each AB survey round in which they enter the sample. Countries listed twice in the table changed regime categories across survey rounds.

Figure 4 presents the means of the outcome measures for three groups: the full sample (pooling all rounds and countries), president co-partisans, and opposition party supporters. The figure demonstrates that, on average, those in our sample have relatively high support for both horizontal and vertical judicial power, but relatively low trust in courts. About 69% agree or strongly agree that the president should always obey the courts (panel b), while about 25% believe that the president should be able to ignore the courts. Regarding vertical power, about 74% agree or strongly agree that courts have the right to make decisions that the people must abide by, while 18% do not. By contrast, 40% trust the courts just a little or not at all, while about 60% trust the courts somewhat or a lot (note that 29% trust the courts a lot).^{Footnote 24} Overall, this evidence suggests a gap in what the Africans in our sample desire from the courts and what they perceive they are actually getting; the latter is picked up by the trust measure.

FIGURE 4.

Means of Main Dependent Variables in the Full Sample and by Partisan Connection to the President

Note: Figure displays means of each outcome variable. Panel (a) presents the means for the continuous measures, rescaled to run from 0 to 1. Panel (b) presents the binary measures. Both panels present means in the full sample (circles), among president co-partisans (triangles), and among opposition and party supporters (squares).

Figure 4 also provides evidence of partisan gaps in public judgments about courts. Presidential co-partisans are about 4 percentage points less likely to support judicial power over the president, evidence that presidential supporters are more likely to reject horizontal power. Presidential co-partisans are, however, about 4 percentage points more supportive of power to make binding decisions for the people, an indicator of support for vertical power. The gaps in trust are substantially larger. Presidential co-partisans are substantially more trusting of courts than are independents and opposition supporters. They are about 14 percentage points more likely to express trust in the courts of law. Assessments of trust in courts appear to be politicized—via partisan alignment with the president—to a greater extent than support for judicial power.

Table 6 presents regression results. All models include AB survey round and country fixed effects. The survey-round fixed effects control for time-specific factors, while the country fixed effects control for country-specific differences, which include contextual differences, differences in political systems, differences in legal tradition, and colonial legacy. Together, the round and country controls also control for time and country-specific factors that might shape considerations driving attitudes about courts. For example, in countries with a recent history of violence or conflict, citizens might value courts that preserve peace and stability (Long et al. Reference Long, Kanyinga, Ferree and Gibson2013). We also include control variables that could be correlated with partisanship and can impact attitudes about courts. We control for democratic values, level of education, male/female, urban/rural, poverty, and age. We cluster standard errors by country–round–partisan affiliation with the president. We emphasize that a key difference between these models and those in the Ghana analysis is that we are unable to include partisan group fixed effects. Thus, these analyses are less causally identified than the Ghana analysis. However, they have the advantage of allowing us to test whether presidential co-partisanship is associated with attitudes about courts across a larger number of countries and time periods.

TABLE 6.

Partisan Alignment with the President and Public Support for Courts

Note: Robust standard errors in parentheses. ** p < 0.01, * p < 0.05. All models include survey-round and country fixed effects.

The first two columns show that co-partisanship with the president significantly reduces support for horizontal judicial power. The magnitude of the coefficient in Column 1 represents about a 5% reduction from the mean among opposition party supporters. Column 2 shows that presidential co-partisanship reduces the probability of supporting court power over the president by 5 percentage points, about an 8% reduction. To assess the political importance of these effect sizes, we compare these coefficients to the magnitude of coefficients on other variables that prior literature has deemed important in shaping attitudes toward courts. For example, the effect of co-partisanship with the president is equivalent to moving from having no formal education (the omitted reference category) to secondary or post-secondary education. The effect size is also about the same magnitude of the effect of support for democracy, a variable that has been shown to impact support for courts.

Columns 3 and 4 show that co-partisanship with the president is associated with a significant increase in support for vertical power. The effect size in Column 3 is modest, representing about a 3% increase over the mean among opposition supporters. The effect size is, however, about the same as the effect of democratic values and about the same as the effect of moving from no education to secondary. Consistent with the partisan alignment theory, co-partisans of the president are more likely to support vertical power.

Finally, Columns 5 and 6 show that those who support the president’s party have significantly greater trust in the courts of law. The coefficient in Column 5 represents an almost 20% increase in trust over the mean among opposition supporters. Column 6 shows that the president’s supporters are about 15 percentage points more likely to trust the courts. Trust in courts thus appears to be highly conditional on partisan connections to those in power—and to a greater extent than support for judicial power.^{Footnote 25}

Online Appendix F shows that the results are robust to controlling for perceptions of judicial corruption, a factor that may influence attitudes about courts. We do not include this variable in the main specification because perceptions of judicial corruption are likely endogenous to partisanship.

Online Appendix E replicates the main analyses on each AB Survey round individually. The results are the same in all survey rounds, which suggests that the findings are not particular to specific time periods. Online Appendix H presents country-by-country results. With respect to horizontal power, the co-partisanship coefficient is negative in 25 of the 34 countries, and statistically significant in 16. For vertical power, presidential co-partisanship is positive in 30 countries and statistically significant in 22. For trust, the coefficient on presidential co-partisanship is positive in all countries, and statistically significant in 32. These analyses highlight the generalizability of the findings.

Online Appendix F examines whether the results generalize across regime types and contexts with varying degrees of judicial independence.^{Footnote 26} The negative effect of presidential co-partisanship on support for horizontal power is biggest in countries that are more democratic and where judicial independence is strong (Figures I.1 and I.2). This suggests that the partisan political stakes regarding executive power are enhanced—and presidential co-partisanship is activated to a greater extent—in contexts where democracy is strong but courts have higher capacity to constrain elected leaders.

By contrast, the effect of co-partisanship on trust in courts is somewhat bigger in less democratic countries and those with less judicial independence (Figures I.5 and I.6)—although we emphasize that the trust results hold in almost all countries. Finally, democracy and judicial independence do not condition the effect of co-partisanship with the president on attitudes about vertical power (Figures I.3 and I.4).