Judicial Independence and Its Role in Judicial Review

Judicial independence, broadly defined, is a political construction that allows judges to make decisions free from outside influence. As a latent concept, scholars have taken a wide variety of approaches to measurement in empirical research. Many operationalizations focus either on expert descriptions of courts (Stephenson Reference Stephenson2003; Linzer and Staton Reference Linzer2015) or indicators that would allow judges to make decisions free from influence, such as the real salary of judges or the budgets provided to courts (Hayo and Voigt Reference Hayo2007; Ura and Wohlfarth Reference Ura2010). But others focus on judicial review, arguing that a constitutional court’s invalidation of a government action is indication of its independence from the other branches of government (Rios-Figueroa Reference Rios-Figueroa2007; Clark Reference Clark2009).

While many find judicial independence normatively appealing, independence is far from guaranteed, even within a democracy. Congress and the president, for example, can influence the Supreme Court by ignoring or circumventing previous decisions (Epstein and Knight Reference Epstein1998), starving the Court of resources like support staff and salary increases (Ura and Wohlfarth Reference Ura2010), limiting the amount of discretion the Court has in determining its docket (Harvey Reference Harvey2013), pursuing constitutional changes that would damage it, or even impeaching the justices (Whittington Reference Whittington2009). Given the many threats to judicial independence, scholars have identified a number of its determinants.

Perhaps the most important driver of judicial independence is the popularity of a court. Courts in modern democracies often have broad support among their publics (Gibson, Caldeira, and Baird Reference Gibson1998; Gibson, Caldeira, and Spence Reference Gibson2003). This support may cause voters to abandon officials who engage in court-curbing, which insulates courts from elected officials who are fearful of losing their jobs. In turn, constitutional courts are free to strike down laws (Carrubba Reference Carrubba2009). Comparative evidence demonstrates that constitutional courts are vulnerable to court-curbing when support is low (Helmke Reference Helmke2010). In the US context, supporters of the attitudinal model of judicial decision-making argue the Supreme Court’s enduring popularity allows it to ignore strategic considerations and make decisions based on its own preferences (Segal and Spaeth Reference Segal2002). More recent research shows, however, that variations in legitimacy weigh heavily even on the Supreme Court, influencing its decision to strike down statutes (Clark Reference Clark2009; Merrill, Conway, and Ura Reference Merrill2017).

Vanberg (Reference Vanberg2005) argues that transparency in the political environment moderates the relationship between popularity and judicial review. Public attention to a particular case makes it more difficult for elected officials to circumvent rulings. Similarly, a public that does not observe court-curbing efforts cannot publish them. Vanberg shows that the German constitutional court is more likely to strike laws in salient cases, a finding supported by additional analysis in Mexico (Staton Reference Staton2010). Cross-national analysis also finds that countries with higher degrees of press freedom also have more independent judiciaries (Hayo and Voigt Reference Hayo2007; Melton and Ginsburg Reference Melton2014).

Formal protections for constitutional courts, such as salary protections or life terms, also help secure judicial independence by removing tools that can be used to punish a court for unpopular decisions. This, in turn, insulates a court from political pressure and secures judicial independence. Comparative evidence supports that at least some formal protections for judges do bolster judicial independence (Hayo and Voigt Reference Hayo2007; Melton and Ginsburg Reference Melton2014). Supporters of the attitudinal model also argue that these formal protections enable the US Supreme Court to ignore strategic considerations (Segal Reference Segal1997). But analysis over the long term indicates that the Court’s institutional support was not always so high, and, as a result, it invalidated fewer laws (McGuire Reference McGuire2004).

Political fragmentation can also provide the independence courts need to exercise judicial review (Rios-Figueroa Reference Rios-Figueroa2007). Many modern democracies have a separation-of-powers system, in which the ability to govern is divided between multiple political entities. When these entities are not controlled by the same governing coalition, they may disagree on the desirability of court rulings and be unable to retaliate against a court that invalidates its policies. This, in turn, empowers courts to strike laws as they see fit. Similarly, insurance theory argues that judicial independence will be greater when political competition is high (Stephenson Reference Stephenson2003). Competition creates uncertainty for government officials evaluating whether they will be able to keep power. Fearing extreme policies by the opposition, governments support judicial review of their own actions as an insurance mechanism were they to lose power, which subsequently empowers courts to strike laws. While insurance theory has considerable empirical support in a variety of political contexts, evidence supporting the political fragmentation hypothesis is mixed (Rios-Figueroa Reference Rios-Figueroa2007; Helmke Reference Helmke2010; Vanberg Reference Vanberg2015).

In some ways, political fragmentation can be seen as a more basic formulation of Marks’s (Reference Marks2015) separation-of-powers model. Marks explained why Congress would tolerate a statutory Supreme Court decision inconsistent with its preferences; this logic was later extended to constitutional decisions (Spiller and Gely Reference Spiller1992; Bergara, Richman, and Spiller Reference Bergara2003). If the pivotal actors in the policy-making process—such as the median member of the House, the median member of the Senate, and the president—all support a law under review, the Court will not try to invalidate it for fear of backlash. But if a single pivotal member opposes the law, the court is free to strike it down, so long as doing so would not result in a policy environment more extreme than the ideal policy of the dissenting pivotal member(s). There is considerable debate as to whether the separation-of-powers model has empirical support; proponents of the attitudinal model in particular argue for a negligible relationship (Spiller and Gely Reference Spiller1992; Segal Reference Segal1997; Segal and Spaeth Reference Segal2002; Bergara et al. Reference Bergara2003; Segal et al. Reference Segal2011; Hall and Ura Reference Hall2015).

Independence as a Moderator of Preferences

As mentioned earlier, judicial review has been used as an indicator of judicial independence. To be sure, a truly independent court must not be afraid to strike down the decisions of other actors. But despite its correlation with independence, it is by no means a valid indicator of the concept. Scholars have long noted that governments may desire for courts to strike down laws under certain conditions (Rogers Reference Rogers2001; Whittington Reference Whittington2009). Striking down a law, therefore, is not a perfect indicator of an independent court, as it could just as easily be a court bowing to the pressure of another branch of government (Carrubba, Gabel, and Hankla Reference Carrubba2008). In a similar vein, failing to strike a law is not necessarily an indication of a weak judiciary; sometimes, judges uphold laws because they are consistent with their preferences. Indeed, these competing pressures obfuscate any potential for a direct relationship between judicial independence and judicial review and might actually lead to scholars observing a null or even negative relationship between the two at rates greater than one would expect by chance.

That judicial independence does not have an additive relationship with judicial review is an underappreciated prediction made by a number of formal models of judicial review. Consider Marks’s (Reference Marks2015) separation-of-powers model. Many scholars interpret this model to mean that when the elected branches are supportive of a particular law, the Court should be less likely to strike it down (Segal et al. Reference Segal2011). Conversely, the Court should be more likely to strike down a law when at least one pivotal elected official opposes the law. This characterization of the model omits an important relationship. When a pivotal actor opposes a law under review, the Court is free to strike down the law as unconstitutional and move the status quo toward its ideological preferences. But if the law reflects the preferences of the median member of the Court, then the Court will simply uphold the law to avoid costly decision-making. The Court is not forced to strike laws, but rather can “vote its own preferences” (Bergara et al. Reference Bergara2003).

A similar account can be given for Vanberg’s (Reference Vanberg2005) model of political transparency. A typical description of the model’s equilibria states that when a court is sufficiently popular and operates in an environment of political transparency, it is more likely to invalidate legislation (Staton Reference Staton2010). But this characterization misses an important part of the equilibria. Assuming that the government supports a statute under review, a court with divergent preferences from the legislature will strike down a statute when both popularity and transparency are high; otherwise, that court will uphold a statute out of fear of court-curbing. A court with convergent preferences to legislature, however, will always uphold a statute under review regardless of its popularity and transparency. Phrased differently, a court’s preferences are only relevant to its decision-making when the court is sufficiently popular and transparent.

I argue that none of the mechanisms of judicial independence discussed above implies simple direct effects on a constitutional court’s decision to strike down a statute, as previous studies have assumed in their empirical models. Rather, these mechanisms condition the effect that a court’s preferences have on striking a statute: the effect of preferences should be strongest when a court is protected from court-curbing, and that effect should decline as the court becomes more vulnerable. This leads to a general hypothesis:

This account is not to suggest that all empirical studies of judicial independence and judicial review miss this distinction, of course. Rios-Figueroa (Reference Rios-Figueroa2007) carefully details that over the time frame of his analysis, the judges of the Mexican Supreme Court would likely oppose all statutes being passed by the Institutional Revolutionary Party, thus obviating the need to model a conditional relationship. Yet such careful justification of research design is rare; most scholars model an additive relationship between the two variables, with little justification for doing so (Vanberg Reference Vanberg2005; Clark Reference Clark2009; Segal et al. Reference Segal2011). The implied theoretical approach of these research designs can be summarized as follows:

Additionally, the attitudinal model of judicial decision-making, as applied to judicial review, makes a distinct additive prediction. Proponents of the attitudinal model argue that the positive relationship between the US Supreme Court’s ideological predispositions and the ultimate decision in a case is constant. Because the Court is sufficiently protected, variance in the level of independence is inconsequential to judicial decision-making. This leads to an additional hypothesis:

Research Design

In order to test the above hypotheses, I need to analyze a set of constitutional court decisions with measures of both the court’s preferences toward invalidating a particular statute and different mechanisms that secure judicial independence. The US Supreme Court provides an excellent test case for two reasons. First, the Court provides a difficult test case for my theory. It is considered one of the most independent courts in the world, and so there is a strong expectation for finding evidence to support the attitudinal model of judicial review and a failure to find evidence for the influence of judicial independence (Segal and Spaeth Reference Segal2002; Gibson et al. Reference Gibson2003). Indeed, Harvey (Reference Harvey2013) argues that the strong evidence supporting the attitudinal model in the Supreme Court can be explained, at least in part, by confirmation bias in the coding of certain variables from the Supreme Court Database (Spaeth et al. Reference Spaeth2016) that are featured prominently in this analysis.Footnote ¹ Finding evidence that supports the Conditional Preference Hypothesis in the face of such a conservative test, then, would provide compelling support for my theory.

Second, the Court is the subject of a wealth of research on the ideological preferences of its members over a long period of time, with a particular emphasis on the measures of justices’ ideal points (Segal and Cover Reference Segal1989; Martin and Quinn Reference Martin2002; Bailey Reference Bailey2013). While comparative research efforts are growing in their ability to collect case data across countries, comparable ideal point estimates for courts or their judges are still unavailable. For these reasons, I analyze a subset of Supreme Court decisions.

Analysis

The analysis proceeds in two parts. The analysis is first conducted with a simple additive model in which mechanisms that secure judicial independence are included but not interacted with court ideology. The results of this analysis are contained in table 1.Footnote ¹³ For each model, a positive, statistically significant coefficient for the Court ideology variable indicates support for the Additive Preference Hypothesis. A positive, statistically significant coefficient for the independence variables indicates support for the Additive Independence Hypothesis.

Table 1.

Heckman Models of Decision to Invalidate an Important Federal Statute Passed between 1949 and 2011 That Test Additive Relationships

Note.—Grouped coefficients report Wald test of joint significance. Standard errors clustered on statute are in parentheses. One-tailed significance tests used for individual coefficients.

^a Independence variable subject to a linear transformation.

* p < .05.

** p < .01.

Court ideology is a robust and relatively stable predictor of whether the court will strike down a statute in the models. In most models, the average marginal effect for Court ideology in the sample is roughly 25%, meaning a one-unit increase in the Court’s ideological predisposition to striking down a statute results in a 25% increase in the probability of striking (Hanmer and Kalkan Reference Hanmer2013). To help illustrate, Justice O’Connor retired in 2005 as the median justice on the Court. Her replacement, Justice Alito, was decidedly more conservative: on the Bailey ideal point scale, his first ideal point measure in 2006 was roughly one unit larger than O’Connor’s in 2005. Thus, if Alito had become the median justice after replacing O’Connor, the Court would be 25% more likely to strike down liberal statutes and uphold conservative statutes.Footnote ¹⁴ This result supports the Additive Preference Hypothesis.

In contrast with court ideology, however, the various mechanisms of judicial independence are largely not statistically significant, even though most of them are signed in the correct direction. There are a couple of exceptions to these findings, however: the models of case salience and political fragmentation have positive, statistically significant coefficients in line with theoretical expectations. In the case salience model, the average marginal effect in the sample is roughly a 15% increase in the probability of striking a statute. In the political fragmentation model, the average discrete effect in the sample of moving from a unified government to a divided government is also roughly 15%.

The lack of statistical significance for the other seven mechanisms of judicial independence tested, however, casts doubt on the Additive Independence Hypothesis. While a lack of statistical significance does not necessarily indicate a negligible effect, at minimum it does indicate that the data does not support the Additive Independence Hypothesis (Rainey Reference Rainey2014). The results for court popularity and transparency, in particular, are either partial or total failures to replicate some of the findings in previous research. In addition, the analysis would also fail to find evidence for an implication of insurance theory. Were the analysis to end here, one would question whether the mechanisms of popularity and transparency, for whatever reason, do not hold up as well in the American context or whether there is a problem with previous or current analysis.

In addition to the analysis relevant to the articulated hypotheses, there are a number of other relevant pieces of information to be gleaned from the tables. The public’s ideology has a relatively stable, modest, and positive coefficient in the models. The p-value for this coefficient, while not significant at the 0.05 level in most models, as shown in the tables, is significant at the 0.10 level in most models. In contrast, the variable measuring the focus of the case before the Court is always positive and statistically significant. The average discrete effect in the sample of moving from a case focused on other issues to a case focused on the constitutionality of the federal statute under review is roughly a 20% increase in the probability of striking. Additionally, some measures of independence are statistically significant in the first stage of the equation, indicating that they have a net effect on the certiorari process. Specifically, there is a robust replication that the level of opposition to a statute among pivotal political actors is positively related to the probability that the Court will grant certiorari to a constitutional challenge to that statute (Hall and Ura Reference Hall2015).

While individual coefficients of the two-way fixed effects and the cubic polynomials are not reported, Wald tests of their joint significance by group are reported. Both sets of fixed effects are statistically significant in the first stage, but not the second. The cubic polynomials are also statistically significant. This is likely driven by the discrepancy in degrees of freedom in the first and second stages. Even so, the p-values for the policy area fixed effects and the cubic polynomials are statistically significant at the 0.001 level, indicating there is more than just sample size considerations driving these results. There are a few substantive observations we can glean from these control variables. First, the Court is much more likely to grant certiorari in some policy areas than others: the Court is more likely to hear cases on fiscal issues, crime and civil rights, good governance, and Social Security relative to the miscellaneous category. Second, the Court is much more likely to hear a challenge to a law right after it passes rather than waiting long periods of time, indicating that statutes with dubious constitutionality are considered by the Court swiftly.

Wald tests of selection effects are also reported in the analysis. There is no evidence of selection effects in any of these models. None of them was statistically significant at the 0.05 level, with the smallest p-value in any model being 0.57. While null results do not disprove the existence of selection effects in Supreme Court decision-making, they do cast some doubt on how they might affect inferences. Unfortunately, it is difficult to compare this result to the rest of the literature, as prior studies employing similar strategies do not report results concerning selection effects (Harvey and Friedman Reference Harvey2009; Hall and Ura Reference Hall2015). In future research, scholars should be more explicit about their findings.

The results in table 2 are similar to the models as those in table 1, but this time including an interaction term between court ideology and the independence variables. There is little change in the values of the control variables, including fixed effects and cubic polynomials. There is also no evidence of selection effects, with similarly large p-values for the Wald test on all models. A positive interaction term indicates support for the Conditional Preference Hypothesis, while a negligible interaction term—in addition to positive, statistically significant coefficients on the constitutive terms—indicates support for the other hypotheses.

Table 2.

Heckman Models of Decision to Invalidate an Important Federal Statute Passed between 1949 and 2011 That Test Multiplicative Relationships

Note.—Grouped coefficients report Wald test of joint significance.

Standard errors clustered on statute are in parentheses. One-tailed significance tests used for individual coefficients.

^a Independence variable subject to a linear transformation.

* p < .05.

** p < .01.

Like the previous models, the coefficient for the various mechanisms of judicial independence are rarely statistically significant. Unlike in the previous models, the coefficient for court ideology has wide variation: the coefficient almost doubles in the statute length model, is statistically insignificant and centered on zero in the pivotal support models, and is large and negative in the Court popularity model. These results are inconsistent with the Additive Preference and Additive Independence Hypotheses, which would predict positive, statistically significant coefficients for each set of constitutive variables and negligible interaction terms. In contrast, six of the nine interaction terms are positive, and four of those are statistically significant. This is again consistent with the Conditional Preference Hypothesis: as the level of independence increases, the Court’s ideological preferences play a stronger role in their decisions.

Yet scholars are not directly concerned with the coefficients of interaction terms, especially not those in probit models. Rather, scholars are concerned with marginal effect estimates at substantively interesting levels of conditioning variables (Brambor, Clark, and Golder Reference Brambor2006). In this study, we expect the Court’s ideological tilt to have a positive and statistically significant effect when independence is high and a smaller, negligible effect when independence is low. Figures 1 and 2 present the average marginal effect of court ideology at the empirical minimum and maximum levels of independence in the data.Footnote ¹⁵

Figure 1.

Average marginal effect of Court ideology of table 2, at the empirical minimum and maximum levels of independence (90% CI).

Figure 2.

Average marginal effect of court ideology of table 2, at the empirical minimum and maximum levels of independence (90% CI).

The results in figure 1 provide some support the Conditional Preference Hypothesis. When the Court is popular relative to Congress, a one-unit increase in the Court’s ideological predisposition to strike results in a 55% increase in the probability of striking in the sample. This relationship flips in direction and becomes statistically insignificant when the Court and Congress have similar levels of popularity. Similarly, when the Court considers a case that is salient to the public, a one-unit increase in court’s ideological predisposition to strike translates into roughly a 40% increase in the probability of striking; this effect becomes statistically indistinguishable from zero when a case is not salient.

The results in figure 2 more strongly support the Conditional Preference Hypothesis. When all pivotal policy-making members support a statute, for any of the pivotal policymaker models, Court ideology does not have a statistically significant influence on the Court’s decision to strike a statute. But when a single pivotal member is opposed to a statute, a one-unit increase in the court’s ideological predisposition to strike translates into roughly a 60% increase in the probability of striking in the sample. The partisan vote model results are even more extreme. For a bipartisan or nonpartisan statute, court ideology is statistically insignificant. For a strictly partisan statute, however, a one-unit increase in court ideology results in a 70% increase in the Court’s probability to strike. These results indicate that even small changes in the ideological disposition of the Court can have huge consequences on the Court’s propensity to strike a law when the Court is independent. These results disappear, however, when the Court is not independent.

The court-curbing, statute length, and political fragmentation models do not support the Conditional Preference Hypothesis: the models’ relevant coefficients are not statistically significant, and the direction of the relationship (as well as the interaction term) is in the opposite direction as would be expected by theory. While the results do not necessarily show that court-curbing or statutory complexity do not play roles in the Court’s decision to strike a statute, it does cast some doubt on those findings especially given the null results in the additive models. The political fragmentation model just barely misses out on statistical significance for its constitutive term and is near identical to its value in the additive model, which was statistically significant. This provides support to the Additive Independence Hypothesis for this model. But caution should be taken with this finding for a number of reasons. The Conditional Independence Hypothesis has more support overall. Political fragmentation also only matters if political parties have different policy preferences, a phenomenon that is better measured by other models in table 1.

Finally, a word about robustness checks. The research design discusses a robustness check that measures the ideological tilt of the Court in a case using Segal-Cover scores rather than Bailey ideal point estimates. The results are presented in the appendix. But to summarize, the substantive conclusions of the paper remain unchanged. These results demonstrate that my theoretical argument is robust to different estimation strategies.