Introduction

In a recent article (Reference Richards and KritzerRichards & Kritzer 2002), we proposed a new way of conceptualizing the role of law for use in modeling Supreme Court decisionmaking. We suggested that it is incorrect to think of law at the Supreme Court level as operating through the traditional mechanisms of plain meaning, precedent, or intent of the drafters. Given the Court's discretionary docket, the cases decided by the Court are precisely those that cannot be decided through the relatively mechanistic processes that Reference Segal and SpaethSegal and Spaeth (1993, Reference Segal and Spaeth2002) label the “legal model.” We posit that the influence of law is to be found in what we label “jurisprudential regimes,” which we define as “a key precedent, or a set of related precedents, that structures the way in which the Supreme Court justices evaluate key elements of cases in arriving at decisions in a parti-cular legal area” (Reference Richards and KritzerRichards & Kritzer 2002:308). The manifestation of jurisprudential regimes appears in the way that specific variables influence the justices' decisions. We propose that the way to test for the presence of regimes is to look for changes in how variables influence justices in a particular jurisprudential area.

In our earlier article, we tested this theory by examining Supreme Court decisions in the area of free expression. We hypothesized that the 1972 companion cases Chicago Police Department v. Mosley (408 U.S. 92) and Grayned v. Rockford (408 U.S. 104) demarcated a regime change that is reflected in a central distinction between regulation that is content-neutral and regulation that is content-based. Our statistical analysis provided strong support for our theory as applied in this area of Supreme Court jurisprudence. A central question we left for future research is whether the pattern we found for free expression cases can be found for other jurisprudential areas. In this research note, we extend our analysis to the Supreme Court's decisionmaking concerning the Establishment Clause.

Establishment Clause Jurisprudence

Modern Establishment Clause jurisprudence dates from Everson v. Board of Education (330 U.S. 1, 1947) when the Supreme Court, in a case involving reimbursing parents of schoolchildren for the costs of transportation to school even if the school involved was a parochial school, extended, by incorporation through the Fourteenth Amendment, the Establishment Clause strictures on Congress to the states. In Everson, Justice Black, even while upholding the aid involved in the case using a “child benefit” argument, enunciated what became known as the “no aid” test reflecting a “wall of separation between Church and State” (Reference LevyLevy 1994:152–54). Over the next fifteen years, the Court decided two Establishment Clause cases dealing with voluntary religious instruction during school hours, first striking down programs held in public school buildings (McCollum v. Board of Education, 333 U.S. 203, 1948) and then upholding off-premises programs using so-called released time arrangements (Zorach v. Clauson, 343 U.S. 306, 1952). In a set of three cases (McGowan v. Maryland, 366 U.S. 420, 1961; Two Guys v. McGinley, 366 U.S. 582, 1961; and Gallagher v. Crown Kosher Supermarket, 366 U.S. 617, 1961), the Court dealt with state laws forbidding various kinds of commercial activities on Sunday (so-called blue laws), with the Court rejecting the chall-enges to these laws in all the three cases.

The school prayer cases in 1962 (Engel v. Vitale, 370 U.S. 421) and 1963 (School District of Abington Township v. Schempp, 374 U.S. 203) raised issues not easily dealt with using the “no aid” test, and the Court turned to a two-pronged secular purpose and primary effect test. Eight years after laying out the two-pronged test, the Court added a third prong, first raising the entanglement issue in Walz v. Tax Commission (397 U.S. 664, 1970), a case challenging tax exemptions granted to church property, and then a year later, in Lemon v. Kurtzman (403 U.S. 602, 1971), spelling out what has become known as the “Lemon test.”

The Lemon test directs the decision maker to find that the statute or practice does not violate the Establishment Clause only if:

1. 1 The statute (or practice) has a secular purpose.

2. 2 Its principal or primary effect neither advances nor inhibits religion.

3. 3 It does not foster an excessive government entanglement with religion.

While the Lemon test has been much criticized, both by justices (see Burger and Rehnquist in Wallace v. Jaffree, 472 U.S. 38, 89, 110, 1985;Footnote ¹ O'Connor and Scalia in Kiryas Joel v. Grumet, 512 U.S. 687, 721, 750–01, 1985;Footnote ² Scalia in Lamb's Chapel v. Center Moriches School District, 508 U.S. 384, 398, 1993;Footnote ³ or Kennedy in Allegheny v. ACLU, 492 U.S. 573, 655–56, 1989Footnote ⁴) and by commentators (see, for example, Reference LevyLevy 1994:158; Reference McConnell, Stone, Epstein and SunsteinMcConnell 1992:119–20; Reference PaulsonPaulson 1993), it continues to be the core set of principles guiding decisionmaking in this area. The prominent First Amendment scholar Jesse Choper describes the Lemon test as “the governing approach to judging Establishment Clause issues” (Reference ChoperChoper 1995:165).

The Lemon test has continued to be influential even as a majority of the Court has shifted to an accommodationist stance in recent decisions.Footnote ⁵ In Zelman v. Simmons-Harris (122 S.Ct. 2460, 2002), Chief Justice Rehnquist, writing the opinion of the Court, utilized at least the first two prongs of Lemon:

In her concurring opinion in the Zelman case, Justice O'Connor explained that in Agostini v. Felton (521 U.S. 203, 204, 1997), the Court incorporated the entanglement prong into the “primary effect inquiry” of Lemon, and she declared, “Nor does today's decision signal a major departure from this Court's prior Establishment Clause jurisprudence. A central tool in our analysis of cases in this area has been the Lemon test” (Zelman v. Simmons-Harris, 122 S.Ct. 2460, 2476, 2002). Not surprisingly, the separationist dissenting opinions of Breyer and Souter in Zelman utilize Lemon as well.

Mitchell v. Helms (530 U.S. 793, 2000) is, among recent cases, perhaps the Court's most serious jurisprudential departure from Lemon in terms of its jurisprudential rationale, as Justice O'Connor, concurring, stated, “The plurality announces a rule of unprecedented breadth for the evaluation of Establishment Clause challenges to government school aid programs” (Mitchell v. Helms 793, 797, 2000). Justice Souter's dissent is replete with arguments and references about the inconsistency of Justice Thomas's opinion with Lemon. However, Thomas, writing the judgment of the Court for four justices (Mitchell v. Helms 793, 808, 2000), used Lemon as an analytic framework, explaining that Agostini revised Lemon's effect test and quoting a key portion of O'Connor's majority opinion in Agostini v. Felton (521 U.S. 203, 234, 1997) that relied heavily on Lemon:

The point we wish to make here is not that the justices always follow Lemon or that Lemon has dictated outcomes since it was established as precedent. Few, if any, legal or judicial scholars argue that law matters for the Supreme Court in such a deterministic manner. Instead, what we see is that the justices continue to reference Lemon's analytic categories such as entanglement or effect as relevant to their decisionmaking, even if those categories do not determine outcomes and if dissenting or concurring justices question whether the majority or plurality opinion has correctly interpreted Lemon and its analytic framework.

Does Lemon v. Kurtzman Define a Jurisprudential Regime?

The empirical question we test below is whether the Lemon test serves as the core of a jurisprudential regime underlying Supreme Court decisions concerning the Establishment Clause.Footnote ⁶ Specifically, we test the hypothesis that the factors predicting Supreme Court decisionmaking in Establishment Clause cases differed before and after Lemon v. Kurtzman (1971). Testing this hypothesis involves fitting a model to the justices' votes and comparing results before and after Lemon v. Kurtzman.

In doing this, we draw on the analysis (and data) of Joseph Ignagni, who published two articles reporting statistical analyses of Establishment Clause cases, one looking at Court decisions during the Burger Court (Reference IgnagniIgnagni 1994a), and one looking at the votes of the individual justices extending into the early years of the Rehnquist Court (Reference IgnagniIgnagni 1994b). One way to view Ignagni's analyses is that they focus specifically on what we hypothesize to be the Lemon regime. The variables included in Ignagni's analyses reflect the factors enunciated in Lemon along with other variables that he hypothesizes as relevant:

• • Whether the law or practice had a secular purpose.

• • Whether the law was neutral toward different religions.

• • Whether the law required substantial surveillance by government.

• • Whether the law dealt with a general government service (e.g., fire, police, etc.).

• • Whether the law reflected historical tradition.

• • Whether the law dealt with aid to institutions of higher education.

• • Whether the case involved one-time aid.

• • Whether the federal government took a separationist or accommodationist (or no) position on the case, either as a party or as amicus.

• • Whether the case also involved a free exercise issue.Footnote ⁷

In his analysis of court decisions, Ignagni finds that the Court is more likely to make accommodationist decisions if the case involves a general government service, does not require government surveillance, is neutral toward different religions, reflects historical practice, involves aid to an institution of higher education, and is one-time aid. Whether the law has a secular purpose, the position of the federal government, and the presence of an accompanying free exercise claim do not, according to Ignagni's analysis, significantly affect the Court's decision.

In his analysis of individual justices' votes, Ignagni drops the one-time aid variable but adds dummy variables for individual justices as a means of controlling for justices' ideology; he presents separate analyses for the Burger and Rehnquist Courts. For the Burger Court, his results are similar to that at the case level, with certain important exceptions. Specifically, aid to higher education is not significantly more likely to garner support, an accompanying free exercise claim does increase the likelihood of an accommodationist vote, and the position of the federal government is influential, with the justices more inclined to go along with the government. For the first five years of the Rehnquist Court, the results are somewhat different. Secular purpose does matter, while neutrality does not. The influence of the presence of a free exercise claim and the position of the federal government still matter, but in the opposite direction as in the Burger Court; that is, the Rehnquist justices are likely to disfavor the position of the federal government, a particularly perplexing finding given that the entire period involved there is under a Republican administration.

In the analysis reported below, we rely essentially on the same explanatory variables used by Ignagni. We drop some insignificant variables from our model and add to the model a standard measure of judicial attitudes (Reference Segal and CoverSegal & Cover 1989). Most important, we report analyses comparing decisions pre- and post-Lemon.Footnote ⁸

Data and Variables

Professor Ignagni graciously shared his data with us. We then expanded his data set, both backward to 1947 (beginning with Everson v. Board of Education), and forward through cases decided in the October 1999 Court term (i.e., through summer 2000, ending with Mitchell v. Helms, 530 U.S. 793). Our data set consists of 73 cases, involving 88 Establishment Clause issues and 760 votes by individual justices on those issues.Footnote ⁹

Our dependent variable is whether a justice cast an accommodationist vote (1) or a separationist vote (0). The independent variables included in our statistical analysis are the following:

• • Does the law have a secular purpose (1 secular purpose, 0 no secular purpose)?

• • Does the law require government monitoring or surveillance (1 surveillance required, 0 surveillance not required)?

• • Is the law neutral toward different religions (1 neutral, 0 not neutral)?

• • Does the law involve a general government service, including fire protection, police protection, school transportation, school lunches, loaning of textbooks, standardized testing, and ensuring school attendance (1 general government service involved, 0 no general government service involved)?

• • Was the aid or practice supported by historical practice (1 consistent with historical practice, 0 no historical practice)?

• • Was the issue involved aid to an institution of higher education (1 aid to higher education, 0 other issue)?

Justices' attitudes are measured by what are known as the “Segal-Cover” scores (Reference Segal and CoverSegal & Cover 1989; Reference Segal, Epstein, Cameron and SpaethSegal et al. 1995), which score individual justices on a liberal (+1.0) to conservative (−1.0) continuum based upon editorial commentary at the time justices were appointed. We expected liberal justices to be more separationist and conservatives to be more accommodationist; hence we hypothesized a negative sign for this relationship.Footnote ¹⁰

In some preliminary analyses, we also included variables for

• • Whether the aid was a one-time contribution or ongoing (1 one-time, 0 ongoing).

• • Whether the case also raised a free exercise question (1 accompanying free exercise question, 0 no free exercise question).

• • What stand, if any, the federal government took on the case, either as a party or as amicus (−1 separationist, 0 no position, 1 accommodationist).

We dropped these variables after preliminary analysis showed nothing approaching significant effects.Footnote ¹¹

Given the dichotomous nature of the justices' votes, our method of analysis was logistic regression. To test the core hypothesis of difference between the before and after Lemon periods, we employed a Chow test (Reference Hanushek and JacksonHanushek & Jackson 1977:128–29), which involves adding interaction terms to the basic model to allow coefficients to differ for the two periods and then testing to see if the set of interaction terms indicates a statistically significant improvement to the model's fit. And as a measure of control against the possibility that any before-after differences simply reflected personnel changes, we repeated the analysis limiting the data set to votes of the seven justices deciding cases in both the before and after periods.Footnote ¹²

Results

Table 1 shows the results of our analysis. The table shows the logistic regression coefficients for all cases, for cases decided before Lemon, and for cases decided after Lemon. The “all cases” panel is included primarily because it serves as the basis for the Chow test discussed above. In addition to the coefficients themselves, the table shows the multiplicative effects odds (“odds impact”) of an accommodationist versus a separationist vote (this value is simply the exponentiation of the logistic regression coefficient) and the impact on the probability (“prob. impact”) of an accommodationist vote assuming an otherwise 50-50 case (i.e., a case that otherwise has an equal probability of an accommodationist or separationist vote).

Table 1. Assessing the Lemon Regime: All Justices

Note: Dependent variable coded 1=Accommodationist; 0=Separationist; Lemon v. Kurtzman (1971) omitted from the analysis.

^* p<0.05

^** p<0.01

^*** p<0.001

The first thing to note in Table 1 is the result of the Chow test in the lower right-hand corner. The test produces a statistically significant chi-square of 35.287 (d.f.=6, p≤0.001). Above the Chow test results are indicators of which specific coefficients differ significantly before and after Lemon, which we will discuss below.

Let us look first at the results post-Lemon. Three of the variables in the model—the absence of a secular purpose, neutrality toward different religions, and the requirement for government monitoring—reflect the three prongs of the Lemon test. All three variables have significant effects, and these effects are in the direction suggested by the Lemon test. In terms of impact on the odds, the absence of a secular purpose or a requirement for government monitoring roughly halves the odds of an accommodationist vote, while neutrality toward different religions more than doubles it. The only other variable that influences the justices' votes after Lemon is the justices' attitude; the odds of an accommodationist vote by the most conservative justice are roughly nine times that of the most liberal justice.

The influences on the justices before Lemon are quite different. The only Lemon factor that has a significant effect is the government monitoring variable, and it has the opposite effect compared to after Lemon: required government monitoring increases the likelihood of an accommodationist vote. As indicated by the last column in the table, the difference in effect of government monitoring before and after Lemon is statistically significant. Neither secular purpose nor neutrality achieves statistical significance before Lemon; however, we cannot discern a statistically significant before/after difference in the coefficients for these two variables (i.e., the specific interaction terms for these variables do not achieve statistical significance). Judicial attitudes are also significant influences before Lemon, and this influence continues in the expected direction; the coefficient is smaller in magnitude than after Lemon (the odds of a accommodationist vote by the most conservative justices are only two-and-a-third times the odds of such a vote by the most liberal justices), although this specific before-after difference does not achieve significance.

As interesting as what is not significant before Lemon are the variables that are significant. Both the general government service variable and the historical practice variable are statistically significant, and the coefficients for both of these variables differ significantly from the corresponding coefficients from the after-Lemon period. Moreover, both variables have very large impacts on the odds.

The obvious question is whether the differences between before and after Lemon reflect anything more than personnel changes. Table 2 replicates Table 1, limiting the data to the seven justices who decided cases both before and after Lemon. While there are some differences between Table 1 and Table 2, the overall pattern is generally the same. Some variables do not achieve statistical significance in Table 2, but that at least in part reflects that the sample size is roughly a third of that used in Table 1. Despite these differences, Table 2 does eliminate that alternative explanation of simple personnel change. The overall test of difference (bottom right-hand corner of the table) before and after is significant (and achieves an only slightly diminished chi-square), and the coefficients that differed before and after differ in the same way (assuming one is willing to relax the criterion for judging statistical significance ever so slightly for the general government service variable).

Table 2. Assessing the Lemon Regime: Justices on Court When Lemon Decided

Note: Dependent variable coded 1=Accommodationist; 0=Separationist; includes only seven justices with significant numbers of cases both before and after Lemon; Lemon v. Kurtzman (1971) omitted from the analysis.

^(*) p<0.06

^* p<0.05

^** p<0.01

^*** p<0.001

Conclusion

Law does matter when the justices of the Supreme Court decide cases. Understanding how law influences the justices requires a theoretical framework that recognizes the institutional characteristics that set the Supreme Court apart from other courts, both in terms of the position occupied by the justices on the Court and the kinds of cases that the Court decides.

The influence of law is to be found in what we call jurisprudential regimes. Such regimes do not dictate outcomes in a mechanical way. Rather, jurisprudential regimes structure the decisionmaking process by establishing through law the parameters that justices, and other actors, should take into account in deciding cases. In this research note, we have replicated our previous analysis focusing on free expression by demonstrating that the Lemon test enunciated in the majority opinion in Lemon v. Kurtzman has served to provide a framework for the decisions in Establishment Clause cases decided over the last 30 years.