Separated Institutions, Sharing Powers:
Congressional Influence on the Supreme Court’s Agenda
Anna Harvey
Department of Politics
New York University
Barry Friedman
School of Law
New York University
April 26, 2006
EARLY DRAFT
In recent years a great deal of attention has been devoted to the nature of the strategic
interaction between the Justices of the U.S. Supreme Court and members of Congress. In
particular, several studies have asked whether the Court moderates its rulings when faced with an
ideologically hostile Congress (Gely and Spiller 1990, Eskridge 1991, Spiller and Gely 1992,
Ferejohn and Weingast 1992, Segal 1997, Hansford and Damore 2000, Epstein, Knight and
Martin 2001, Spriggs and Hanford 2001, Segal and Spaeth 2002, Bergara et al 2003, Friedman
and Harvey 2003, Sala and Spriggs 2004, Martin 2005, Harvey and Friedman 2006).
Epstein et al (2002) extend this analysis of interbranch interaction to the agenda setting or
certiorari stage of the Court’s decisionmaking. They ask whether the Supreme Court is
constrained in its certiorari decisions by the ideological composition of the sitting Congress.
Specifically, they ask whether the Court is more likely to issue constitutional as opposed to
statutory rulings when it faces an ideologically distant Congress, or when it is unable to insulate
itself from congressional pressure via opinion supermajorities. They find evidence in support of
both these propositions, and conclude that the Court is indeed constrained by congressional
preferences at the certiorari stage.
We question the logic of the Epstein et al research design given recent evidence that the
Court is constrained by congressional preferences even in constitutional cases. We also
reexamine the Epstein et al findings using more sophisticated estimates of preferences scaled in
interinstitutional space, and find no evidence that the proportion of constitutional rulings issued
by the Court is responsive to congressional preferences. We then follow an alternative research
strategy of testing the influence of Congress on the Court’s agenda by estimating the probability
that the constitutionality of a congressional statute will be reviewed by the Supreme Court,
conditional on the expected utility gains to the median Justice from the final ruling on the merits.
1
Using this research design we find strong support for the hypothesis that the Court is constrained
by congressional preferences at the certiorari stage.
1
Congressional Influence on the Certiorari Decision
Previous empirical work on the decision to grant certiorari has been concerned largely
with two questions. The first is whether a Justice’s vote to grant certiorari is conditioned on the
likely outcome of the final merits decision. Boucher and Segal (1995), using the set of cases
granted certiorari during the 1946 through 1952 Terms, found that a Justice’s vote for certiorari
is dependent on the fraction of the Court voting with him on the merits when that Justice votes to
affirm the lower court’s judgment. Caldeira et al (1999), using the pool of cert petitions for the
1982 October Term, found that the probability of a Justice’s vote for certiorari increases when
the average ideology of the other eight Justices becomes more similar to his own.
The second question which has motivated empirical work on certiorari is whether cert is
used strategically by the Supreme Court to induce lower courts to adhere to the Court’s
preferences. For example, Boucher and Segal (1995) found that the Justices were more likely to
have voted for cert when they voted to reverse a lower court’s judgment. Caldeira and Wright
(1988) and Caldeira et al (1999) found in the 1982 sample of cert petitions that this relatively
conservative Supreme Court was more likely to grant cert when the lower court had issued a
liberal judgment. Cameron et al (2000), using a sample of 273 search and seizure cert petitions
from federal appeals courts between 1972 and 1986, found that cases are more likely to be
granted cert when the lower court is both ideologically distant from the Supreme Court and has
decided the case in a way that is inconsistent with Supreme Court preferences.
Other empirical work on certiorari has looked for case-specific factors which may
influence the probability of a grant of certiorari. For instance, Caldeira and Wright (1988) found
2
in the sample of 1982 cert petitions that amicus curiae briefs filed either in support of or in
opposition to cert increase the probability of cert being granted. They also found that actual or
alleged conflict in the lower courts, reversal in the lower courts, and the presence of the federal
government as a petitioner all increase the probability of cert being granted. McGuire and
Caldeira (1993) found in the 410 obscenity cert petitions filed between 1955 and 1987 that
amicus curiae briefs, the presence of the federal government as a petitioner, and the presence of
experienced litigators increased the probability of a cert grant.
With only one exception, however, scholars have not examined the possible influence of
congressional and presidential preferences on the certiorari decision. The exception is Epstein et
al (2002), who claim that the cert decision should be responsive to the ideological composition of
the political branches.1 Their argument follows from the median Justice’s incentives with respect
to the Court’s final decisions on the merits in cases involving statutory interpretation. If there is
an interpretation of a statute that the pivotal members of Congress prefer to the Court’s
interpretation, they can enact that interpretation following the Court's ruling. There thus may
well be situations in which the Court is better off issuing a statutory interpretation that is not the
preferred interpretation of the Court's median Justice, but is both unlikely to be overturned by a
sitting Congress, and is preferable to the interpretation which would be enacted by that Congress
were it to act (Gely and Spiller 1990, Eskridge 1991, Spiller and Gely 1992, Ferejohn and
Weingast 1992, Bergara et al 2003).
Extending this logic to the agenda setting stage, the Court may very well be less likely to
grant certiorari to cases involving statutory interpretation when it faces an ideologically hostile
1
Also see Provine 1980: 54-62, Epstein and Knight 1998: 83, Friedman article on Lopez: 797-
98, Cross and Nelson 2001, 1476.
3
Congress, given the deference to congressional preferences it may be forced to display. Instead,
Epstein et al argue, in this situation the Court may be more likely to grant cert to cases involving
constitutional interpretation, given the conventional wisdom that the Court is not constrained by
Congress in its constitutional rulings (Spiller and Gely 1992, Segal and Spaeth 2002: 96-97,
Bergara et al 2003). Epstein et al then observe that between the 1946 and 1992 terms the Court
decided a larger percentage of cases on constitutional grounds when it faced a more ideologically
distant Congress, unless the Court was so ideologically homogeneous that it could issue a
supermajority decision.2
However, two aspects of the Epstein et al study raise some concerns. The first is the
nature of the research design. Epstein et al assume that the Court is insulated from congressional
pressure in constitutional cases. It is true that under Article III of the Constitution, appointed
Supreme Court Justices are given life tenure, and are protected from congressional salary
reductions. In addition, constitutional decisions cannot be overturned by a simple congressional
statute; only a constitutional amendment can alter the Supreme Court's constitutional
determinations.
However, as Epstein et al acknowledge (page cite), there are other means by which the
Congress can influence the Court in its constitutional decisions. For instance, the Constitution
gives to Congress the power to regulate the Court’s appellate jurisdiction; stripping the Court of
2
Epstein et al acknowledge that their data do not permit them to distinguish between the
possibility that the Court is accepting relatively fewer petitions raising statutory claims when it
faces an ideologically distant Congress, and the possibility that the Court is more likely to choose
constitutional interpretation at the time of the merits decision in the same situation (page cite).
4
aspects of this jurisdiction is frequently threatened by members of Congress who disagree with
the Court’s decisions (Friedman 1990).3
Congress' control over the lower federal courts may also provide an incentive for the
Supreme Court to be attentive to congressional preferences. The number, composition, and
jurisdiction of the lower federal courts are arguably entirely at the discretion of Congress
(Friedman 1990; Segal and Spaeth 2002, 29-30).4 Congress’ ability to manipulate these aspects
of the federal courts may affect the degree to which the Court is effective in policing these courts
(Mcnollgast 1995, Segal and Spaeth 2002, 230).
The Congress also has considerable discretion concerning the Court's budget. Although
the Congress cannot diminish the Justices' salaries, it can fail to raise their salaries to keep pace
with inflation (Landes and Posner 1976, 885; Cross and Nelson 2001, 12). Similarly, Congress
could refuse to appropriate sufficient funds for the Court's supporting personnel, including the
3
For examples of jurisdiction stripping see Nagel 1964, 928; Rosenberg 1992, 377, 385, 387,
390; Mcnollgast 1995, 1664; Friedman 1998, 741, 745, 751-3; Cross and Nelson 2001, 11; Segal
and Spaeth 2002, 230; and Martin 2005, 10. More drastic congressional proposals to limit
judicial review have included allowing appeal from the Court to an elective body like the Senate
(Rosenberg 1992, 377, Friedman 1998, 740,) or the entire Congress (Friedman 1998, 749), and
requiring extraordinary majorities for declarations of unconstitutionality (Rosenberg 1992, 377,
387, Friedman 1998, 744, 748).
4
For examples of congressional manipulation of the number, composition, and jurisdiction of the
lower federal courts, see Nagel 1965, 926 fn. 3; Landes and Posner 1976, 885; Rosenberg 1992,
380-1; Mcnollgast 1995, 1648, 1663; Friedman 1998, 740; Cross and Nelson 2001, 11; Segal and
Spaeth 2002, 226 fn8, 227-228, 236, 236 fn 34.
5
clerks who are critical to the Justices' ability to manage their caseload (Cross and Nelson 2001,
12).
The Congress also has the authority to alter the number of Justices sitting on the Court, 5
and to remove Justices from the Court for "...Treason, Bribery, or other high Crimes and
Misdemeanors" (Article III, Section I). The power to impeach (and convict), given that it is only
vaguely defined, might be used to punish the federal judiciary for decisions with which Congress
disagrees (Stone et al 1991; Cross and Nelson 2001, 10; for examples see Friedman 1998, 740).
Finally, Congress also might simply refuse to implement or follow Court decisions, or
provide insufficient funds for effective implementation (Rosenberg 1991, Friedman 1998, 742,
Cross and Nelson 2001, 13). Many of the other weapons possessed by Congress are quite blunt,
and their use can attract public attention and debate. But Congress can wage a lower-level war of
attrition with regard to some constitutional decisions simply by failing to take heed of them.6
The foregoing opportunities for congressional action would seem to provide relatively
powerful incentives for the Justices to be mindful of congressional policy preferences, even in
constitutional cases. And while earlier analyses had largely failed to find evidence in support of
this claim (Spriggs and Hansford 2001, Sala and Spriggs 2004, Martin 2005), Harvey and
5
During the Civil War the size of the Supreme Court was for a time increased to 10, later
reduced to 8, and then restored to 9 after Grant's assumption of the Presidency (Rosenberg 1992,
381; McNollGast 1995, 1632, Friedman 1998, 743, 746). The New Deal Congress elected in
1936 seriously considered Roosevelt's proposal to "pack" the conservative Supreme Court with
new liberal members (Rosenberg 1992, 381, Friedman 1998 749-750).
6
See Fisher (1993) for a claim that this has happened with regard to the Supreme Court's
invalidation of the legislative veto.
6
Friedman (2006) report strong evidence that in fact the Court is quite sensitive to the preferences
of elected officials even in constitutional cases.
Even if the Court is responsive to congressional preferences at the certiorari stage, then,
we would not expect to find evidence of that responsiveness using the Epstein et al research
design. That is, if the Court is equally constrained by Congress in both statutory and
constitutional cases, then we would not expect to find variation in the proportions of these kinds
of cases as a function of congressional preferences. Yet, Epstein et al do find such a pattern. This
brings us to our second concern about the Epstein et al paper, namely the preference estimates
used therein.7
Despite the fact that the model being tested in the Epstein et al paper is one of interbranch
interaction, the ideal point estimates for the actors involved in these interactions are not scaled in
the same space. Instead, ADA support scores are used for members of Congress, and “predicted
annual liberalism support scores in non-unanimous civil liberties constitutional cases” are used
for the Justices (page cite). These scores are then simply assumed to be scaled in the same
interinstitutional space. However, as recent work by Michael Bailey has demonstrated, such ad
hoc scaling procedures can often lead one to erroneous conclusions (Bailey 2005). Bailey has
generated a large sample of interbranch “bridge” observations from which ideal point estimates
may be generated via Bayesian Markov chain simulation methods for members of the Congress,
7
There are other problems associated with the Epstein et al findings. There is no adjustment for
the selection bias inherent in just taking as data the sample of cases actually heard by the Court,
and there is no attempt to measure the utility gains for the Court from taking action on the
specific statute being litigated.
7
the Court, and the President between 1950 and 2002. To date, these are the best estimates we
have in which members of these institutions are scaled in the same ideological space.
To check whether the correlation between the mix of statutory and constitutional
decisions rendered by the Court and congressional preferences is a product of the scaling
technique (not) used by Epstein et al, we replicated their analysis using the Bailey interbranch
ideal point estimates. Table 1 reproduces the original Epstein et al estimates, our replication of
those estimates using a Prais-Winsten regression, and our replication of the estimates using the
Bailey ideal point measures.8 In these estimations the dependent variable is the percentage of all
constitutional and statutory decisions issued by the Court that are statutory.9 The Congressional
Constraint variable measures the degree of congressional constraint experienced by the Court
through the construction of winsets defined by the pivotal congressional actors. If the Court’s
median Justice lies within the congressional winset in a given year, then she is assumed to be
free to locate a statutory ruling at her own ideal point, and the congressional constraint is
8
The Epstein et al estimates are produced using an unspecified AR(1) maximum likelihood
function; we have chosen a Prais-Winsten regression as a reasonable approximation of the
Epstein et al estimator.
9
Percentages are derived from the United States Supreme Court Judicial Database (USSCJD),
available at http://www.as.uky.edu/polisci/ulmerproject/sctdata.htm. Constitutional decisions are
defined as those in which the primary authority for the decision, as coded by the USSCJD, is
judicial review at the national or state level. Statutory decisions are those in which the Court
interpreted a federal statute, treaty, court rule, executive order, administrative regulation, or
administrative rule. We thank Lee Epstein for making all data used in the 2002 article available
at http://epstein.wustl.edu/research/dynamic.html.
8
measured as 0. If the Court median’s ideal point lies outside of the relevant congressional winset,
then the degree of congressional constraint is measured by the distance between the median’s
ideal point and the nearest endpoint of the winset. The greater is the distance, the greater is the
degree of congressional constraint.
The winsets are constructed under two alternative theories of the legislative process.
Under the first, the Committee Gatekeeping Model (Committee-Power model in Epstein et al),
the Judiciary Committee medians decide whether to report legislation to the floor for
consideration under an open rule. In this model, the endpoints of the winset are the minimum and
maximum values of the following: the House and Senate veto pivots, the House and Senate
Judiciary Committee medians, the Judiciary Committee medians’ indifference points relative to
the midpoint between the House and Senate floor medians, and the House and Senate floor
medians.
Under the second theory of the legislative process used by Epstein et al, the Majority
Party Gatekeeping Model (Party-Caucus model in Epstein et al), majority party medians in each
house decide whether to allow legislation to reach the floor for consideration under an open rule.
In this model, the endpoints of the winset are the minimum and maximum values of the
following: the House and Senate veto pivots, the House and Senate majority party medians, the
majority party medians’ indifference points relative to the midpoint between the floor medians,
and the House and Senate floor medians.10
The Court Homogeneity variable is the standard deviation of the Justices’ ideal points in
any given term, multiplied by -1 (so that more homogeneous courts have larger values) and
10
For both models the indifference points are only relevant when the gatekeepers’ ideal points
lie outside the set created by the floor medians and the veto pivots.
9
added to a constant so that all values are positive.11 Finally, the Interaction variable is the
product of the Congressional Constraint and Court Homogeneity variables.
While our replication using the Epstein et al data is nearly identical to their original
estimates, the replication using the Bailey measures of ideology produces very different results.12
Here there appears to be no effect of either congressional preferences or ideological
heterogeneity among the Justices on the mix of statutory and constitutional decisions issued by
the Court [note that we have yet to replicate the Committee Gatekeeping Model].
Yet the original intuition in the Epstein et al paper appears to be a good one. The Court
chooses the cases it wishes to review through the decision to grant a writ of certiorari.
Presumably the median Justice prefers to review cases which promise her the greatest utility
gains. However, the structure of congressional preferences, in conjunction with congressional
control of policy instruments which may be used to punish the Court, may prevent the median
Justice from issuing her most preferred rulings in some cases (thus decreasing the utility she
11
In Epstein et al, ideal point estimates for the Court Homogeneity variable are taken from Segal
and Cover (1989) and Segal et al (1995). In our replication using the Bailey data we recompute
not only the Congressional Constraint variable but also the Court Homogeneity variable using
the Bailey ideal point estimates.
12
One measure of the difference between using the Epstein et al data on congressional constraint
and using the Bailey data is that, for the Majority Party Gatekeeping model, Epstein et al found
that the Supreme Court lay outside the congressional winsets in only 3 years between 1946 and
1992. Using the Bailey data, we find that the Court median lies outside the winsets in 11 years
between 1950 and 1992.
10
receives from those rulings). Consideration of congressional preferences may thus affect the
extent to which the median Justice supports granting cert in any given case.
However, estimating the effects of congressional preferences on the Court’s certiorari
decisions is not straightforward. Litigants may be strategic about their appeals to the Court
(McGuire et al 2005). If litigants can anticipate which cases the Court is less likely to take
because of congressional hostility, then they should be less likely to appeal those cases in the
first place. One would then be unlikely to observe the Court’s responsiveness to congressional
preferences in the sample of cases for which writs of certiorari are requested.
We take account of the possible selection bias in the pool of certiorari petitions by taking
as our sample the pool of congressional statutes enacted between 1987 and 2000 (Harvey and
Friedman 2006). We further restrict our attention to constitutional challenges to these statutes as
a means of testing the hypothesis of congressional constraint in what is arguably the least likely
place to find it. We estimate the probability that a statute is granted cert by the Court, as a
function of the Court’s predicted ruling on the merits. We find considerable evidence that the
certiorari decision is in fact affected by congressional preferences.
2
Model Setup
Our model of congressional constraint on the certiorari decision has several features that
distinguish it from other recent work on cert. First, we follow Epstein et al (2002) in modeling
the final merits decision as a choice in continuous rather than dichotomous space. All other
empirical work on certiorari has modeled the Justices’ final vote on the merits as a dichotomous
choice. For example, Boucher and Segal (1995) model the final merits decision as a choice
between reversing and affirming the lower court. Caldeira et al (1999) model the merits decision
as a choice between liberal and conservative judgments. Cameron et al (2000) model the final
11
merits decision as a choice between “correct” and “incorrect” judgments according to the
Justices’ own views on legal doctrine.13
These models of dichotomous choice assume that the most salient feature of a judicial
decision is the judgment in favor of one party or another (Kornhauser 1992 Int’l Review of Law
and Economics). But the judgment affects only the parties to the case, or, via stare decisis,
parties to future cases with identical fact situations. The legal rule announced in the case,
however, affects not only the instant case but also all future cases in the relevant area of law. We
therefore choose to model the creation of a legal rule as the most important feature of the Court’s
final decision on the merits. Since the choice of a legal rule is not a dichotomous choice, we
model the final merits decision as the selection of a point in one dimensional policy space
(Kornhauser 1992 page cite; Lax bargaining paper).
Second, we do not assume that the Court cares primarily about the direction of the
judgment issued by the lower court. Other empirical studies of certiorari have assumed that the
primary goal of the Court in granting cert is error correction, or policing the judgments of the
lower courts. The Justices are assumed to want to reverse erroneous lower court judgments more
than they want to affirm correct lower court judgments. Indeed, in some models affirms are
13
Lax (2004) follows Cameron et al (2000) in modeling the final choice on the merits as a
dichotomous choice, but introduces continuous preferences over this dichotomous choice. Each
Justice most prefers an “indifference” point in one dimensional issue space; this indifference
point sorts cases into dichotomous judgments (e.g. admit or exclude evidence). A case is located
in this issue space by virtue of its fact pattern. A Justice’s utility from a “correct” case outcome
is an increasing function of the distance between the Justice’s ideal point and the point
represented by the case’s fact pattern, and 0 otherwise.
12
simply mistakes; the Justices took the wrong cases (Cameron et al 2000, also see Baum 1997:
79).
This assumption about error correction makes sense if you think that the Justices care
only about judgments, not legal rules. A Court that cares only about judgments cares only about
the parties presently before the Court. Affirming a lower court judgment with which you agree
does not then add to your utility in any way.14
However, if you assume that Justices care about legal rules, then there can be a clear
value to affirming a lower court decision. A federal appellate court ruling creates precedent only
within that circuit (Perry 1991: 199). A legal rule announced by the Supreme Court, on the other
hand, creates binding precedent for all the circuits. Even if the judgment in a lower court ruling
comports with the way that a Justice would have decided the case, that Justice may still think that
the issue is important enough to want to take the case in order to create a nationally binding legal
rule.
There is anecdotal evidence from the Court to support this view of certiorari as not
primarily based in error correction. For example, Chief Justice Vinson declared in 1949 that
“The Supreme Court is not, and never has been, primarily concerned with the correction of errors
in lower court decisions…the Supreme Court must continue to decide only those cases which
present questions whose resolution will have immediate importance far beyond the particular
facts and parties involved.” (Perry 1991: 36). Justice Vinson’s view was repeated many times in
the interviews of Justices and their clerks conducted by H.W. Perry for his book on certiorari
(Ibid 37). Perry’s respondents reported that denial of cert should not be interpreted as agreement
with the lower court’s judgment (38); that cases in the same issue area are seen as
14
See Lax 2004 on strategic use of affirms.
13
interchangeable because the Justices care not about particular parties to a case but rather about
the legal rule being decided (220-221); that cases which merely present questions about the
applicability of a legal rule to a particular fact situation are much less likely to be granted cert
than cases which present direct challenges to legal rules (223-224); and that Justices look for
cases with very simple fact patterns so as to be able to “establish a principle of law pretty
cleanly” (235).15
The error correction approach assumes that Justices worry about accepting cert petitions
when they agree with the judgment in the lower court, for fear that the Court will end up
reversing that judgment. However, Justices may equally fear accepting cert petitions when they
disagree with the judgment announced in the lower court, for fear that the Court will affirm the
judgment and create a national precedent with which they disagree. One of Perry’s Justices made
this argument: “I might think the Nebraska Supreme Court made a horrible decision, but I
wouldn’t want to take the case, for if we take the case and affirm it, then it would become a
precedent” (Perry 1991: 200).
Moreover, the empirical results about error correction are mixed. Boucher and Segal
(1995) find that Justices are more likely to vote for certiorari when they wish to reverse.
Cameron et al (2000) find that a conservative Supreme Court is more likely to grant certiorari in
search and seizure cases when the lower court excluded the evidence. But Caldeira et al (1999)
show that, on average, an ideologically extreme Justice with a sympathetic Court is as likely to
grant certiorari to a case with whose judgment he agrees as he is to a case with whose judgment
he disagrees (302).
15
Justice Brennan’s clerks report that the Justice paid no attention to the disposition of
the case in the lower court in making his certiorari decisions (Perry 1991: 67-69).
14
Finally, we model the final decision on the merits as converging to the ideal point of the
median voter. We then assume that the median Justice is also the pivotal voter at the certiorari
stage, and will be more likely to take cases which promise larger utility gains for her in the
merits decision. This is not technically true, as the Supreme Court requires only 4 votes out of 9
in order to hear a case. However, Lax (2004) demonstrates that the effect of the Rule of Four in a
one dimensional spatial model is primarily to lower the threshold for the magnitude of utility
gains that must be realized by the median Justice before the Court agrees to hear a case. It will
still be the case under the Rule of Four that cases which promise larger utility gains for the
median Justice should be more likely to be granted cert than cases which promise smaller utility
gains for the median Justice.
4
The Merits Decision
We assume that at the time of the certiorari decision, the Justices vote based on their
expectations about the outcome of the merits decision. We thus begin by modeling the merits
decision, then address the certiorari decision.16
We assume that both Justices and members of Congress have symmetric single-peaked
preferences over a common left-right policy continuum. We further assume that each law
enacted by a given Congress reflects the midpoint between the ideal points of the pivotal
legislators in that Congress.17 Amendments to provisions of existing laws are treated as new laws
and are given the ideological midpoints of the enacting Congresses.
16
Our model of the merits decision draws heavily on Harvey and Friedman (2006).
17
The question of which members of Congress are pivotal depends upon how one models the
legislative process. We use a variety of models, as detailed below.
15
Enacted laws may then be challenged in the courts, including recourse to the Supreme
Court.18 Constitutional challenges may take a variety of forms. One might imagine that laws
enacted by Congresses on the left of this continuum would typically be challenged on the
grounds that they overstep constitutional limits on congressional powers to act. Laws enacted by
Congresses on the right of this continuum might typically be challenged on the grounds that they
insufficiently protect the constitutional rights of individuals.
Should a congressional law L be reviewed by the Court, the Court's median Justice
selects a point on the policy continuum (L’) as a standard of constitutionality against which the
law will be judged. We assume that the median Justice’s utility from L’ is a decreasing function
of the distance between it and her ideal point (C): U(L’) = -|C - L’|. In the absence of any
institutional constraint, the median Justice will select C as the constitutional standard to be
applied to the law.
If the Court faces institutional constraints, however, L’ may not equal C. In this case,
after the Court chooses L’, the Congress can choose to punish the Court with retributive
legislation. We assume that the Congress will do so only if all pivotal legislators prefer L to L’.
Should at least one pivotal legislator be closer to (or equidistant from) the Court's ruling than to
the original law, that member will choose either not to introduce, or to block, legislation
disciplining the Court. However, if the pivotal members are all closer to the law than to the
Court's chosen constitutional standard, those members will act to ensure passage of punitive
legislation.
18
This is true even for laws that have a finite time horizon, such as an annual appropriations bill.
Such a bill may still be litigated to the Supreme Court even after its formal authority has expired.
16
In this constrained case, the Court acts so as to avoid Court-punishing legislation. That is,
the median Justice will set L’ as close to her ideal point as possible, while yet forestalling
punitive congressional action. Under some configurations of ideal points and for some laws, the
median Justice of the Court will be constrained in her rulings of constitutionality by
congressional preferences. Under other configurations, she will not be constrained, and will
simply use her ideal point as a standard of constitutionality.
The specific conditions of a constrained Court will be given in the subsequent section,
with attention to alternative theories of legislative behavior. However, to illustrate the general
idea here, we present an example of a constrained Court in Figure 1. In this figure, the pivotal
members of Congress are assumed to be the median members of the House (H) and Senate (S).
The Court's median Justice (C) is located to the right of both houses' medians. The law at issue
(L) is located closer to both the median Representative and the median Senator than is the ideal
point of the median Justice, ensuring that the latter will be constrained in her constitutional ruling
on the law. Were she to choose her own ideal point as the standard of constitutionality in this
case, both the House and the Senate would support punitive legislation aimed at the Court. The
median Justice thus will moderate the constitutional standard (L’) to the point at which the
congressional pivot closest to her own ideal point is just indifferent between the standard
embodied by the law at issue and that articulated by the Court. In Figure 1, this implies that L’
will be set at the indifference point of the median Senator (I(S)). While the median
Representative will still support Court-attacking legislation, the median Senator will not support
that legislation, thus preventing its passage.
The second, unconstrained, case is represented by Figure 2. In this case, neither the Court
nor the law has changed ideological positions. However, the House and Senate medians have
17
shifted rightward, such that the latter's ideal point is now closer to the Court median than to the
law at issue. The median Justice may now set L’ at C, for the median Senator will prefer that
standard to the one embodied by the law at issue, and will again block any Court-attacking
legislation.
5
Theories of Congressional Behavior
The foregoing model of the Court’s merits decision requires some assumptions about the
legislative process. We here follow the literature on congressional behavior by deriving the
conditions under which the Court may be constrained from five prominent theories of
congressional pivotality (Shepsle and Weingast (1995), Krehbiel (1998)).
The Floor Median Model
In the Floor Median Model we assume that the pivotal members of Congress are the
House and Senate floor medians.19 The vision of congressional authority embodied in any given
law (L) thus may be represented by the midpoint between the floor medians of the two enacting
houses. Should a congressional law come before the Court, the constraint set for that Court in
that case will be defined by three points: the law, the sitting House median's indifference point,
and the sitting Senate median's indifference point. The leftmost boundary of the set is the
19
This model is motivated by three institutional details of the House and Senate: a majority may
order a committee to discharge a bill to the floor, thus preventing committee gatekeeping
(Krehbiel 1995, 1997, 1998, 233), any restrictive voting rules must be approved by majority
vote, thus ensuring open rules (Krehbiel 1997, 1998, 233), and committees do not possess ex
post vetoes in the conference stage (Krehbiel 1987, 1998, 233). According to proponents of this
hypothesis, these two rules imply that all legislative decisions will be made by the floor median
voters of both houses of Congress (Krehbiel 1987, 1995, 1997, 1998, Krehbiel and Rivers 1988).
18
minimum of these three points; the rightmost boundary of the set is the maximum of the three. If
the Court median lies within this set, then the Court is unconstrained and L’ will be set at the
ideal point of the median Justice. If the Court median lies to the left (right) of the leftmost
(rightmost) boundary of this set, then the Court is constrained and will set L’ at this leftmost
(rightmost) boundary.
The Committee Gatekeeping Model
The Committee Gatekeeping Model assumes that committees can prevent legislation
from reaching the floor, but that once legislation has reached the floor, an open rule obtains.20 As
with the previous model, then, the vision of congressional authority embodied in any given law
may be represented by the midpoint between the floor medians of the two enacting houses. The
Court's constraint set in any given case will be defined by five points: the law, the sitting House
median's indifference point, the sitting Senate median's indifference point, the sitting House
Judiciary Committee median's indifference point, and the sitting Senate Judiciary Committee
median's indifference point.21 An unconstrained Court median will set L’ at her own ideal point;
a constrained Court will set L’ at the leftmost or rightmost boundary of this set.
20
Gatekeeping models typically assume that congressional committees possess special
parliamentary powers that allow them to prevent legislation from reaching the floor and/or
protect such legislation from amendment by the floor (e.g., Shepsle and Weingast 1987, Ferejohn
and Shipan 1990, Dion and Huber 1996). The existence of such powers, however, is a matter of
some dispute (Krehbiel 1987, 1995, 1997, 1998, Krehbiel and Rivers 1988). Here we employ the
most minimal form of a gatekeeping model (Ferejohn and Shipan 1990).
21
We assume that legislation attacking the Court would most likely lie within the jurisdiction of
both houses' Judiciary Committees.
19
Majority Party Models
Some students of Congress assert that the legislative party organizations are able to
influence the votes of legislators such that outcomes are pulled away from the floor medians
toward the majority party medians (Rohde 1991, Aldrich 1995, Chapter 7, Dion and Huber 1996,
1997). There is a considerable amount of controversy about whether this is the case (Krehbiel
1993, 1995, 1998). We take two different approaches to modeling majority party influence. First
we assume that the majority party can gatekeep, but faces an open rule once legislation is
allowed onto the floor. Next we assume that the party can effect a closed rule on the floor
through influencing the votes of party members.
The Majority Party Gatekeeping Model
In the Majority Party Gatekeeping Model the relevant gatekeepers are not committee
medians but rather the majority party medians, acting as principals. Majority party committee
members hold back legislation when such action gets the majority party medians better outcomes
than would be attained on open floor votes (Krehbiel 1998, 234). But because legislation, once
released by a committee, is subject to an open rule, we assume -- again -- that the vision of
congressional authority embodied in any given law may be represented by the midpoint between
the floor medians of the two enacting houses. In any challenge to legislation, the Court's
constraint set will be defined by five points: the law, the sitting House median's indifference
point, the sitting Senate median's indifference point, the sitting House majority party median's
indifference point, and the sitting Senate majority party median's indifference point. An
unconstrained Court will set L’ at its own median; a constrained Court will set L’ at the leftmost
or rightmost boundary of this set.
The Majority Party Median Model
20
A second way of modeling majority party influence is to assume that legislative party
organizations can ensure that party members will only vote for party sponsored measures.
Majority party leaders thus can pull legislative outcomes to the majority party medians rather
than the floor medians (e.g. Dion and Huber 1996, 1997). In the Majority Party Median Model,
then, we assume that the vision of congressional authority embodied in any given law may be
represented by the midpoint between the majority party medians of the two enacting houses.
Should a congressional law come before the Court, the constraint set for that Court in that case
will be defined by three points: the law, the sitting House majority party median's indifference
point, and the sitting Senate majority party median's indifference point. An unconstrained court
will set L’ at its own median; a constrained Court will set L’ at the leftmost or rightmost
boundary of this set.
The Veto-Filibuster Model
Finally, we computed the constraint sets for a fifth theory of legislative behavior, namely
that of the Veto-Filibuster Model proposed by Krehbiel (1998). Krehbiel asserts the significance
of two institutional rules that may pull legislative outcomes away from floor medians: the
possibility of a presidential veto, necessitating a congressional override by 2/3 of both houses,
and the possibility of a senatorial filibuster, which can be broken only by mustering a 3/5
majority in the Senate. For a Democratic president, the relevant congressional pivots are thus the
146th Representative, the 34th Senator, the House and Senate medians, and the 60th Senator. For
a Republican president, the pivots are the 41st Senator, the House and Senate medians, the 290th
Representative, and the 67th Senator. Legislation is assumed to be represented by the midpoint
of the appropriate set. Should a congressional law be reviewed by the Court, the Court's
constraint set is composed of the point represented by the law and the indifference point of each
21
congressional pivot in the appropriate set. An unconstrained Court will set L’ at its own median,
while a constrained Court will set L’ at the leftmost or rightmost boundary of this set.
6
The Certiorari Decision
According to our model of the merits decision, when a congressional statute is reviewed
by the Court, the median Justice will choose an optimal value of L’. Under the hypothesis of an
unconstrained Court, L’ will always equal C, or the median’s ideal point. Under the hypothesis of
a constrained Court, L’ will be the equilibrium prediction resulting from a set of assumptions
about the legislative process.
At the time of the certiorari decision, then, the Justices can look forward to the merits
decision and predict the likely location of L’ for any congressional statute that they choose to
review. We model two different ways in which the Justices might prioritize granting review to
some cert petitions above others.
Our first approach is to assume that the median Justice prefers to hear cases which give
her the highest expected utility gains relative to her own ideal point. Under this approach, the
utility of granting certiorari for the median Justice is decreasing in the distance between her own
ideal point and the constitutional standard she will be able to set in the majority opinion:
U(Grant Certiorari) = -|C - L’|.22 If the Court is institutionally constrained, then the median
Justice prefers to hear cases wherein she will be able to set L’ as close to her own ideal point as
possible. If the Court is not institutionally constrained, then this expression collapses to zero and
provides no information about which cases the Court will take.
However, the median Justice may also care about the utility gains she is able to achieve
in a ruling, relative to the status quo represented by the statute under review. In this second
22
We could also assume a quadratic form here with little difference in the estimation results.
22
approach, the median Justice compares the utility gains she realizes from setting a new
constitutional standard to those she realizes from leaving the statute alone: U(Grant Certiorari) =
-|C - L’| - (-|C - L|) = |L’ - L|. If the Court is institutionally constrained, then the utility gains for
the median Justice from granting certiorari are increasing in the distance between the statute
being reviewed and the standard which the Court can set; the median Justice prefers to hear the
cases wherein she can move the status quo the furthest toward her own ideal point.
If the Court is not institutionally constrained, then the first part of this expression
collapses to zero and the utility to the Court from granting cert is equivalent to |C - L|, or the
ideological distance between the Court and the statute being reviewed. The unconstrained
median Justice prefers to review the statutes that are the most distant from her own ideal point;
she can move these status quos the furthest toward her own ideal point. We test both of these
approaches below.
7
Testable Implications
For each of the five theories of congressional behavior, we examine both approaches to
the Court’s certiorari decisions. Under the first approach, we estimate the probability that a
statute’s constitutionality is reviewed by the Court as a function of |C – L’|. If the Court is
unconstrained by Congress, then we should find no relationship between this variable and the
likelihood that a law is reviewed by the Court. If the Court is in fact constrained by Congress,
then we should find a negative relationship between this variable and the probability of review,
with the Court more likely to review laws when it can issue a ruling closer to the median’s ideal
point.
Under the second approach, we test the unconstrained model by estimating the
probability that a statute’s constitutionality is reviewed by the Court as a function of |C - L|. If
23
the Court is indeed unconstrained by Congress, then we should find a positive relationship
between this variable and the likelihood that a law is reviewed by the Court. We then test the
constrained model by estimating the probability that a statute’s constitutionality is reviewed by
the Court as a function of |L’ – L|. If the Court is in fact constrained by Congress, then we should
find a positive relationship between this variable and the probability of review.23
We recognize that these tests at best can provide only indirect evidence of a constrained
Court. That is, if we in fact find that the probability that a congressional law is reviewed is
responsive to the predicted location of L’, then we will have evidence that either the Court, or
actors who anticipated the Court's likely actions, was/were constrained by anticipated
congressional preferences in constitutional cases.
8
The Data
Our statute-centered analysis begins with the 100th Congress, elected in 1986, and
continues to track all public laws enacted through the 106th Congress, elected in 1998. The
public laws enacted by the 100th Congress were first available to be reviewed by the Supreme
Court sitting in October of 1987, the second Term of the Rehnquist Court.24 We follow the fate
23
Sala and Spriggs (2004) point out that the observational equivalence between the two models
for the observations for which the Court is predicted to be unconstrained will dampen the results
for the constrained model. Our estimation is thus a conservative test of the constrained model.
24
Although it usually takes several years for a challenge to a law to work its way to the Supreme
Court, it can happen quite quickly. Examples from our dataset include a November 18, 1988
amendment to the Communications Act of 1934, struck by the 1988 OT Court, the Flag
Protection Act of October 28, 1989, struck by the 1989 OT Court, and provisions of the
Communications Decency Act of February 8, 1996, struck by the 1996 OT Court.
24
of these laws through the Rehnquist Court's 2000 Term. Table 2 contains the numbers of public
laws enacted in each congressional year, and the number of years that the laws are followed for
each group.25 In total we follow the fate of 3725 laws over a range of 1 to 14 years. An
observation thus consists of law i observed in year t; we have 29,755 observations in all.
8.1
Dependent Variable
Because we are interested in whether (and when) these laws are reviewed by the Court,
our dependent variable is dichotomous: a law is either reviewed and coded 1 in a given year, or
is not reviewed and is coded 0 for that year. Struck laws are coded as 1 in the Term in which they
are struck down and are then removed from the dataset; upheld laws are coded as 1 in the Term
in which they are reviewed and then revert to 0 in subsequent Terms, giving the Court an
opportunity to change its mind.26
In order to determine which laws were reviewed by the Court during this period, we used
the United States Supreme Court Judicial Database (USSCJD) to identify all cases involving a
federal statute in which the authority of the decision was cited as judicial review. We then read
all cases to identify those in which a clear ruling of constitutionality (uphold/strike) was issued.27
25
The data in Table 2 were obtained from the Congressional Record's summaries of legislative
activity, also available from http://thomas.loc.gov/home/resume.html. The Congress also enacts
an increasingly small number of "private" bills every legislative session; these bills concern
topics of very narrow interest to individual members. Private bills are not included in our dataset.
26
Removing upheld laws from the dataset after they are reviewed by the Court makes no
appreciable differences to the results.
27
During this period there are 4 cases reported by the USSCJD as involving judicial review of a
federal law, in which the Court does not actually reach a decision on constitutional grounds
25
The cases involving struck laws were then checked against alternative sources, identifying some
errors in the USSCJD (Zeppos 1993, Library of Congress 2001, Epstein 2002).28
For each case we then identified the Congress which enacted the statute or part of a
statute whose constitutionality was at issue. These statutes have been frequently amended. In the
face of these frequent amendments, we adopted the following decision rule. First, we identified
the specific section or sections of the statute actually being reviewed by the Court. We then
looked at both the original enacting date and all reenactments of or amendments to this section or
sections. As long as the challenged language of the statute remained substantially intact through
all amendments and/or reenactments, we adopted the most recent reenacting or amending
Congress as the enacting Congress.29
(Austin v. U.S. (1993), TBS v. FCC et al (1994), Miller v. Albright (1998), and Friends of the
Earth v. Laidlaw Environmental Services (2000)). Including or excluding these cases does not
make any substantive difference to the results reported below; we report results excluding these
cases.
28
The USSCJD both includes cases that do not strike congressional laws, e.g., Saenz v. Roe, 143
L.Ed.2d 689 (1999) (involving the constitutionality of a California statute limiting the maximum
welfare benefits available to newly arrived residents), and excludes cases that do strike
congressional laws, e.g., Bartnicki et al v. Vopper et al, 532 U.S. 514 (2001) (involving a
provision of federal wiretapping law found to violate the First Amendment).
29
For example, in Greater New Orleans Broadcasting Assn. v. United States (1999), the
Supreme Court invalidated a 1934 congressional law prohibiting the advertisement of casinos. A
1988 amendment to the section of this law at issue before the Court had added the words "or
television," thus including television broadcasting as a medium through which casino advertising
26
In all, 35 of the 3725 laws were reviewed by the Court on constitutional grounds between
1987 and 2000. Tables 3 and 4 report the years in which, and the Congresses by which, the 35
reviewed laws were enacted, as determined by the foregoing decision rules. This table reveals
that 27 of these laws (77%) were enacted by the unified Democratic Congresses sitting between
the 1986 and 1994 congressional elections. Many of these 27 statutes do appear to be "liberal",
as that term is commonly used. They include protections from discrimination on the basis of
disabilities or gender, various forms of gun control legislation, and provisions for health care
benefits for retirees from the coal industry [add for upheld laws].
Moreover, among the 8 reviewed statutes which were passed by the unified Republican
Congresses after the 1994 elections, there are examples of "conservative" laws, as that term is
commonly used. For example, this set of struck laws includes an appropriations act prohibiting
the Legal Services Corporation from funding local organizations which represent clients who
challenge existing welfare laws [add from upholds]
The list also contains laws whose ideology is harder to ascertain at face value. For
instance, legislation restricting speech of various kinds (e.g. casino advertisements, flag
desecration, pornographic television broadcasts) is passed by the Congresses sitting both before
and after the 1994 elections. [add from upholds] By some accounts, these measures are
traditionally "liberal" in the sense that they extend the power of the federal government to
regulate private conduct. By other accounts these statutes are more properly understood as
"conservative", because they promote a socially conservative moral agenda.
was prohibited. The original prohibition on casino advertising remained intact through this
amendment, clearly signaling congressional support for the thrust of the original law. The later
Congress was thus designated as the enacting Congress.
27
Instead of trying to make these kinds of judgments about the ideological content of laws,
however, we turn to estimates of congressional preferences derived from roll call votes which
permit scaling the Court and the Congress in the same ideological "space".
8.2
Independent Variables
The testable implications of our empirical models require us to compute measures of
ideological distance between the Court and various pivotal members of Congress. Typically,
researchers testing separation of powers models use measures of ideology which are not scaled in
the same institutional "space" (Segal 1997, Segal and Spaeth 2002, Epstein et al 2002). Here we
use the measures developed by Michael Bailey and Kelly Chang, which use bridging
observations to scale the Court, the Congress, and the President in the same space (Bailey and
Chang 2001, Bailey 2005).
Using the Bailey estimates of judicial and congressional ideology, we can characterize
the constraint sets generated by the various models of congressional behavior. For the purposes
of illustration, Figure 3 [add] displays the congressional constraint set for 1987 statutes under the
Floor Median model. This figure illustrates a phenomenon common to all constraint sets for pre1995 statutes except for those generated by the Veto-Filibuster Model, namely that the Court lay
to the right of the rightmost boundaries of these constraint sets prior to 1994. After 1994, the
Court lay within these constraint sets. With the exception of the Majority Party Median Model,
the Court also lies within the predicted constraint sets for all laws enacted by the post-1994
Congresses. The Majority Party Median Model predicts that between 1995 and 1999 the Court
was actually mildly constrained by the leftmost boundary of the congressional constraint set for
laws enacted by the 104th-106th Congresses. Finally, the Majority Party Gatekeeping Model
generates a rightmost boundary for the congressional constraint set which is identical to that of
28
the Floor Median Model between 1987-1994. Its predictions are thus observationally equivalent
to those of the Floor Median Model.
These estimated constraint sets suggest that, if the Court is constrained by the elected
branches, prior to the 1994 elections the Rehnquist Court should have been hesitant to grant
certiorari to cases challenging the constitutionality of the liberal legislation enacted by the
Congresses sitting between 1987 and 1994. However, after those elections, the Court would have
been liberated to take those cases (and presumably then to strike those statutes essentially at
will).
This pattern is indeed reflected in Tables 3 and 4, which report the years and cases in
which congressional statutes were reviewed by the Court. This table reveals that only 7 of the 27
liberal laws reviewed between 1987 and 2000 (26%) were reviewed by the Court prior to the
1994 October Term.
The constraint set generated by the Veto-Filibuster Model is qualitatively different from
those generated by the other four models. This model predicts that, for all laws enacted prior to
the 1994 congressional elections, the Court was constrained by Congress and the President only
during the 103d Congress, or that sitting during 1993 and 1994. During this period the
Democratic Congress was joined by a Democratic president, and the Court lay to the right of the
rightmost boundary of the constraint set generated by the relevant pivotal actors. For all laws
enacted after the 1994 elections, the Veto-Filibuster Model predicts a Court unconstrained by the
Congress and President.
9
The Econometric Specifications and Results
Our data consist of individual laws observed over discrete units of time (years). In any
given year, a law may be reviewed by the Court (and generate a value of 1), or not (and generate
29
a value of 0). Laws that are reviewed and then struck are dropped from the dataset for subsequent
years. We observe the laws for only a limited period of time, ending our period of observation
with the close of 2000. Finally, there is some possibility that laws are less likely to be reviewed
the longer they survive.30 We thus require an empirical method which takes into account the facts
that we have a binary dependent variable, data which are "right censored," and potential temporal
dependence. The appropriate method for analyzing this kind of data is grouped event history
analysis (also known as duration, hazard or survival analysis) (Beck et al 1998). Grouped event
history models are derived from continuous time event history models, which estimate the
probability of an event occurring as a function of both the set of theoretically derived
independent variables, and a baseline "hazard" rate.
We here apply the grouped version of the most common continuous time event history
model, namely the Cox (1975) proportional hazards model. The Cox continuous time model,
reported in Equation 1, is widely used because it allows the estimation of a baseline hazard rate
which is unknown and possibly time varying.
ht|x i,t   h 0 te x i,t
(1)
In Equation 1, the instantaneous hazard or probability of an event occurring (h) as a
function of the time t and the vector of independent variables measured for unit i at time t (Xi,t)
depends both on the latter (through the e
30
x i,t 
term) and on the possibly time varying baseline
There is also the possibility of a nonlinear temporal relationship; Justices may prefer that
recently enacted statutes “percolate” in the lower courts until they are “ripe for review” (Perry
1991: 230-232).
30
hazard (h0(t)). Its grouped version is reported in Equation 2.
Py i,t  1|x i,t   ht|x i,t   1  expe x i,t t 
(2)
In Equation 2, yi,t is the binary indicator of whether an event occurred to unit i within year
t, Xi,t represents the observed values of the independent variables for the entire year t, and  t is a
dummy variable marking the length of time the unit has been "at risk."
The model reported in Equation 2 is identical to a binary dependent variable estimated
using either a complementary log-log (cloglog) or Poisson link function, with duration dummy
variables included (Beck et al 1998, Zorn 1998). We here estimate Equation 2 with a Poisson
link function. Following Beck (1998), we first included duration dummies in each estimation to
capture potential nonlinearities in the baseline hazard rate. We then tested these initial models
against ones which included simpler linear trend terms.31
For all 11 estimations reported in Tables 5 and 6, likelihood ratio tests failed to reject the
null hypothesis that the duration dummy variables were no better than linear trend terms to
capture the effects of time on the baseline hazard rate. Tables 5 and 6 thus report the results of
estimations including that linear term (Age of Law), whose values range from 1-14. The standard
errors are Huber (1967) robust standard errors, clustered by year of review. Finally, the Poisson
31
Beck (1998) and Beck et al (1998) also recommend a natural cubic spline as a way to capture
nonlinearities in the baseline hazard rate without using up as many degrees of freedom as
required for the duration dummy variables. However, our data show no evidence of nonlinear
baseline hazard rates.
31
goodness-of-fit statistics reported in Tables 5 and 6 confirm the appropriateness of a Poisson
model for all 11 estimations.32
The estimations reported in Table 5 test the hypothesis of a Court constrained in its cert
decisions using our first approach, namely that the Court selects cases which promise the greatest
utility gains for the median Justice relative to her own ideal point. Column 1 reports the results
from the Floor Median and Majority Party Gatekeeping models, whose predicted standards of
constitutionality are represented by L1’ and L2’, respectively; Column 2 reports the estimates
from the Committee Gatekeeping Model, whose predicted standard of constitutionality is
represented by L3’; Column 3 reports the results from the Majority Party Median Model, whose
predicted standard of constitutionality is represented by L4’, and Column 4 reports the estimates
from the Veto-Filibuster Model, whose predicted standard of constitutionality is represented by
L5’.
32
Beck et al (1998) report that Equation 2 may also be estimated using the more familiar probit
or logit link functions as long as the probability of an event occurring remains less than 50%.
Since in our dataset, the probability of a law being reviewed is considerably below this threshold,
we also used the probit link function to estimate Equation 2, as in Equation 3.
Py i,t  1|x i,t   ht|x i,t   x i,t    t 
(3)
These estimates were qualitatively identical to those reported in Tables 5 and 6. We also
reestimated all models in Tables 4 and 5 using the Rare Events Logistic Regression (RELOGIT)
estimator developed by Michael Tomz et al (Michael Tomz, Gary King, and Langche Zeng,
Version 1.1, Cambridge, MA: Harvard University, available at http://gking.harvard.edu/). We
found only negligible differences in the estimated coefficients, standard errors, predicted
probabilities and confidence intervals.
32
These tests provide strong support for the hypothesis of a constrained Court. In all four
estimations, the variable |C - L’| is in the predicted negative direction and is significant at
conventional levels; columns 1-2 and 4 report significantly better fits to the data than constantonly models. The Court appears to be less likely to review the constitutionality of congressional
laws when the standard it will be forced to set in its rulings is more distant from the median
Justice’s ideal point. If the Court were not constrained by congressional preferences, then we
would expect the median Justice to set the standard of constitutionality at her own ideal point in
every case. We then would not expect to gain any insight into the certiorari decision by looking
at L’, or the equilibrium standard of constitutionality which should result from the Court
considering the preferences of the sitting Congress. Finally, in all four estimations the trend
variable Age of Law is in the predicted negative direction but is not significant.
The estimations reported in Table 6 test the hypothesis of a Court constrained in its
certiorari decisions using our second approach, namely that the Court selects cases which
promise the greatest utility gains for the median Justice relative to the status quo of the statute
being reviewed. The first three columns of Table 6 directly test the hypothesis of an
unconstrained Court. In these estimations the probability that a law is reviewed by the Court is
modeled as a function of |C – L|, where C is the Court median and L represents the law being
reviewed by the Court. These estimations assume that congressional preferences play no role the
certiorari decision. If this assumption is correct, then we would expect to see a positive
relationship between this variable and the probability of review: the Court should prefer to
review statutes which are most ideologically distant from the median Justice’s ideal point.
The estimation in Column 1 corresponds to the theoretical predictions made by the Floor
Median Model, the Committee Gatekeeping Model, and the Majority Party Gatekeeping Model,
33
all of which assume that a law's ideological position may be represented by the midpoint
between the floor medians of the two enacting houses (L1). The estimation in Column 2
corresponds to the predictions made by the Majority Party Median Model, which assumes that a
law's ideological position may be represented by the midpoint between the majority party
medians of the two enacting houses (L2). The estimation in Column 3 corresponds to the
predictions made by the Veto-Filibuster Model, which assumes that a law represents the
midpoint between the senatorial filibuster pivot and the most distant veto pivot (L3).
All three estimations provide extremely poor fits to the data. The coefficients on the
variables measuring ideological distance are all in the opposite direction of that predicted,
although none reach conventional levels of significance. The trend variable Age of Law also fails
to generate a significant coefficient in any estimation. No model provides a better fit to the data
than a model including only the constant term.
Columns 4-7 of Table 6 instead model the probability that a law is reviewed by the Court
as a function of |L’ – L|, where L’ represents the predicted point at which the Court will set its
standard of constitutionality, given the constraint exercised by the sitting Congress, and L again
represents the law at issue. These estimations assume that the Court considers congressional
preferences in its certiorari decision. Column 4 reports the results from the Floor Median and
Majority Party Gatekeeping models, whose predicted standards of constitutionality are
represented by L1’ and L2’, respectively; Column 5 reports the estimates from the Committee
Gatekeeping Model, whose predicted standard of constitutionality is represented by L3’; Column
6 reports the results from the Majority Party Median Model, whose predicted standard of
constitutionality is represented by L4’, and Column 7 reports the estimates from the VetoFilibuster Model, whose predicted standard of constitutionality is represented by L5’.
34
In all four estimations, the coefficients on the |L’ – L| variable are now in the predicted
(positive) direction. In Columns 4 and 5, these positive coefficients reach conventional levels of
significance, and the models provide significantly better fits to the data than constant-only
models. For these models of the legislative process, the hypothesis of a Court constrained by
congressional preferences clearly outperforms the rival hypothesis.
For the Majority Party Median and Veto-Filibuster Models, however, the coefficients of
interest are indistinguishable from zero and the models perform no better than models containing
only the constants. Neither of these models of the legislative process appears to fit the data very
well under either the unconstrained or the constrained hypotheses, and we cannot distinguish
between the rival hypotheses using these estimations.
Finally, the trend variable measuring the age of a law is consistently in the predicted
negative direction in columns 4-7, although it fails to reach significance.33
These estimates thus give fairly robust support, across different models of the legislative
process, to the hypothesis of a constitutionally constrained Court. We can also convert the
33
One shortcoming of the grouped event history model used here is that it assumes that
all congressional laws will eventually be reviewed by the Supreme Court. As an alternative
strategy we estimated several split population event history models, which allow us to assume
that some laws will never be reviewed by the Court (Greene 2000, for an application to judicial
politics see Hettinger and Zorn 2005). We estimated these models using zero inflated Poisson
(ZIP) regressions with duration dummy variables included (Zorn 1998), with the same dependent
and independent variables as in Table 6. However, the reported Vuong statistics for each
estimation failed to provide any improvement in fit over the standard Poisson models used in
Table 5 (Vuong 1989, Greene 2000).
35
Poisson estimates into more meaningful quantities, namely the predicted probabilities that a
congressional law will be reviewed in any given October Term as a function of both the Court's
preferences over the law and the relevant congressional constraint. We hold the age of the law at
the sample mean. In Table 7 we report these probabilities for the relatively liberal laws enacted
in 1987, for both approaches reported in Tables 5 and 6. Figure 4 displays these predicted
probabilities graphically, along with their 95% confidence intervals, for the constrained Floor
Median/Majority Party Gatekeeping Model reported in Table 6.
As can be seen in both Table 7 and Figure 4, the changes in the predicted probability that
a liberal congressional law will be reviewed by the Rehnquist Court can be dramatic. In
particular, under our first approach, the predicted probability that a 1987 statute would be
reviewed increases between 63 and 204 percent immediately following the 1994 congressional
elections, depending on the assumed legislative model. Under our second approach, for the two
models which distinguished between the alternative hypotheses, the predicted probability of
review rises between 56 and 65 percent as a function of the 1994 congressional elections. Figure
3 illustrates the minimal overlap in the confidence intervals for the 1993 and 1994 point
predictions of the Floor Median/Majority Party Gatekeeping Model under this second approach;
the results for the other models are substantially similar.
10
Conclusion
Our results may shed some light on the failure of most empirical studies to find any
effects of congressional preferences on the Court’s final rulings on the merits (Segal 1997,
Hansford and Damore 2000, Spriggs and Hanford 2001, Segal and Spaeth 2002, Sala and
Spriggs 2004, Martin 2005) These studies have often been cited as evidence that the Court is
truly independent from the Congress (Segal 1997, Segal and Spaeth 2002, Hettinger and Zorn
36
2005). However, another possibility, and one which receives strong support in our data, is that
the criteria by which cases are selected for inclusion on the Court's docket include a
consideration of congressional and/or presidential policy preferences. The Court appears to be
less likely to grant certiorari when it anticipates that it will have to defer significantly to
congressional preferences. Cases that could systematically demonstrate the effects of
congressional constraints on the Court are likely then to be weeded out of the Court’s docket via
the certiorari decision. Studies which fail to take into account the fact that the Court selects
which cases it will hear in any given term may then fail to observe the fact that the Supreme
Court and the Congress are separate institutions which nonetheless share the judicial power
under the U.S. Constitution.
37
Table 1
Replications of Epstein et al (2002)
Prais-Winsten Regressions on Percent Statutory Decisions, 1946-1992
Epstein et al Original
Replication Using Epstein
Replication Using Bailey
Estimates
et al Data
Ideal Point Estimates
Committee
Majority
Committee
Majority
Committee
Majority
Gatekeeping Party
Gatekeeping Party
Gatekeeping Party
Model
Gatekeeping Model
Gatekeeping Model
Gatekeeping
Model
Model
Model
Constant
-5.41
(3.61)
78.14***
(14.10)
21.48
(15.94)
50.75
-7.09***
(2.68)
68.94***
(15.92)
30.23***
(11.89)
51.91
-7.07***
(2.66)
69.28***
(15.56)
30.08***
(11.77)
51.86
-24.86
(33.10)
-19.96
(20.57)
94.25
(95.18)
63.51
Rho
.33
.46
.44
.70
N
47
47
47
43
Congressional
Constraint
Court
Homogeneity
Interaction
Note: *α=.10; **α=.05; ***α=.01 (all two-tailed tests). Replication using Bailey data is
estimated for 1950-1992.
38
L’
H
L
S
I(S)
C
Figure 1
L’
L
H
Figure 2
39
S
C
I(S)
Table 2
Public Laws, 1987-2000
Congress
Years
Number of
public laws
enacted
Number of
years laws are
followed
Number of
observations
100
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
240
473
240
410
243
347
210
255
88
245
153
241
170
410
14
13
12
11
10
9
8
7
6
5
4
3
2
1
3360
6149
2880
4510
2430
3123
1680
1785
528
1225
612
723
340
410
1987-2000
3725
1-14
29755
101
102
103
104
105
106
Total
40
Table 3: Upheld Congressional Laws, 1987-2000
Statute
Ethics in Government Act
of 1978, 28 USC 49, PL
100-191
Internal Revenue Code, 26
USC 2057, PL 100-203
Sentencing Reform Act of
1984, 28 USC 994, PL
100-690
Federal Communications
Act, 18 USC 1304, 1307,
PL 100-625
Agricultural Adjustment
Act of 1933, 7 USC 608
c6I, PL 100-418
Immigration and
Naturalization Act, 8 USC
1409 (a), PL 100-525
Northwest Timber
Compromise of 1989, 103
S 745, b6A, PL 101-121
National Foundation on
the Arts and Humanities
Act, 20 USC 954(d)(1)
Organized Crime Control
Act of 1970, 18 USC 6002
Antiterrorism and
Effective Death Penalty
Act of 1996, 28 USC
2244b, 2254
Cable Television
Consumer Protection and
Competition Act of 1992,
47 USC 534
Year
Enacted/
Amended
Name of Case
Upholding Statute
Oct.
Term
Ideolo
gy of
Law
USSCJD
Direction
of
Decision
Court
Median
Predicted
Ruling
Predicted
RulingLaw
1987
Morrison v. Olson, 487
U.S. 654 (1988)
1987
-.07
Missing
.39
.03
.10
1993
-.07
Lib
.37
-.07
0
1988
-.07
Missing
.38
.03
.09
1992
-.07
Con
.48
-.06
0
1996
-.07
Con
.39
.39
.45
2000
-.07
Con
.47
.47
.53
1991
-.06
Con
.53
-.06
0
1997
-.06
Con
.41
.41
.48
1999
-.12
Lib
.44
.44
.56
1987
1988
1988
1988
1988
1989
1990
1994
United States v. Carlton,
512 U.S. 26 (1994)
Mistretta v. United
States, 488 U.S. 361
(1989)
United States v. Edge
Broadcasting Co., 509
U.S. 418 (1993)
Glickman v. Wilemen
Brothers (1997), 521
U.S. 457
Tuan Anh Nguyen et al.
v. Immigration and
Naturalization Service
(2001), 99-2071
Robertson v. Seattle
Audubon Society, 503
U.S. 429 (1992)
National Endowment for
the Arts et al v. Finley,
97-371 (1998)
United States v.
Hubbell, 99-166 (2000)
1996
Felker v. Turpin,
Warden, 518 U.S. 651
(1996)
1995
.35
Con
.38
.38
.03
1996
TBS, Inc. v. FCC et al,
520 U.S. 180 (1996)
1996
.35
Con
.39
.39
.05
1999
.52
Con
.44
.44
.08
1999
.47
Lib
.44
.44
.03
Prison Litigation Reform
Act of 1995, 18 USC
3626(e)(2)
1997
Drivers' Privacy Protection
Act of 1994, 18 USC 2721
1999
Miller, Superintendent,
Pendleton Correctional
Facility, et al v. French
et al, 99-224 (1999)
Janet Reno, Atty
General et al Petitioners
et al v. Charlie Condon,
Atty Gen of SC 98-1464
(2000)
41
Table 4: Struck Congressional Laws, 1987-2000
Statute (Name and
Pub. L. No.)
Indian Gaming
Regulatory Act
(Pub. L. 100-497, §
11(d)(7))
The
Communications
Act of 1934, as
amended (Pub. L.
100-625, § 3(a)(4))
The
Communications
Act of 1934, as
amended (Pub. L.
100-690)
Flag Protection Act
(Pub. L. 101-131)
Ethics Reform Act of
1989
(Pub. L. 101-194)
Americans with
Disabilities Act
(Pub. L. 101-336)
Mushroom
Promotion, Research
and Consumer
Information Act
(Pub. L. 101-624,
Title XIX, Subtitle B)
Gun-Free School
Zones Act of 1990
(Pub. L. 101-647)
Federal Deposit
Insurance
Corporation
Improvement Act of
1991
(Pub. L. 102-242, §
476)
Cable Television
Consumer
Protection and
Competition Act of
1992
(Pub. L. 102-385, §§
10(b) and 10(c))
Anti-Drug Abuse
Act, as amended
(Pub. L. 102-393, §
638(e))
Year Enacted/ Name of Case Striking
Amended
Law
Oct. Term
USSCJD
Ideology of Direction of Court
Decision
Law
Median
Predicted
Predicted
Ruling
Ruling- Law
1988
Seminole Tribe of
Florida v. Florida
1995
-.07
Con
.38
.38
.44
1988
Greater New Orleans
Broadcasting Assn. v.
United States
1998
-.07
Lib
.42
.42
.48
1988
Sable Communications
of California v. FCC
1988
-.07
Lib
.38
.03
.09
1989
-.06
Con
.37
-.02
.04
1994
-.06
Lib
.38
.38
.44
2000
-.06
Con
.47
.47
.54
1989
1989
1990
United States v.
Eichman
United States v.
National Treasury
Employees Union
Board of Trustees of
University of Alabama
v. Garrett
1990
United States vs. United
Foods, Inc.
2000
-.06
Lib
.47
.47
.54
1990
United States v. Lopez
1994
-.06
Con
.38
.38
.45
1991
Plaut v. Spendthrift
Farm, Inc.
1994
-.13
Missing
.38
.38
.51
1992
Denver Educational TV
Consortium v. FCC
1995
-.13
Lib
.38
.38
.52
1992
United States v.
Bajakajian
1997
-.13
Lib
.41
.41
.55
42
Statute (Name and
Pub. L. No.)
Year
Enacted/
Amended
Name of Case Striking
Statute
Coal Industry
Retiree Health
Benefit Act of 1992
(Pub. L. 102-486,
Title XIX, § 19143(a))
1992
Eastern Enterprises v.
Apfel
Trademark Remedy
Clarification Act
(Pub. L. 102-542)
Patent and Plant
Variety Remedy
Clarification Act
(Pub. L. 102-560)
Religious Freedom
Restoration Act
(Pub. L. 103-141)
Brady Handgun
Violence Prevention
Act
(Pub. L. 103-159)
Violence Against
Women Act
(Pub. L. 103-322, §
40302)
Communications
Assistance for Law
Enforcement Act
(Pub. L. 103-414)
Communications
Decency Act of 1996
(Pub. L. 104-104,
Title V, § 502)
Communications
Decency Act of 1996)
(Pub. L. 104-104, §
505)
Line Item Veto Act
(Pub. L. 104-130,
§2(a))
Omnibus
Consolidated
Rescissions and
Appropriations Act
(Pub. L. 104-134 §
504(a)(16))
1992
1992
College Savings Bank v.
Florida Prepaid
Education Expense
Board
Florida Prepaid
Education Expense
Board v. College
Savings Bank
Oct. Term
Ideology
of Law
USSCJD
Direction
of
Decision
Court
Median
Predicted
Ruling
Predicted
RulingLaw
1997
-.13
Con
.41
.41
.55
1998
-.13
Con
.42
.42
.56
1998
-.13
Con
.42
.42
.56
1993
Boerne v. Flores
1996
-.12
Con
.39
.39
.51
1993
Printz v. United States
1996
-.12
Con
.39
.39
.51
1994
United States v.
Morrison
1999
-.12
Con
.44
.44
.56
1994
Bartnicki v. Vopper
2000
-.12
Lib
.47
.47
.59
1996
Reno v. ACLU
1996
.35
Lib
.39
.39
.05
1996
United States v.
Playboy Entertainment
Group
1999
.35
Lib
.44
.44
.10
1996
Clinton v. New York
City
1997
.35
Missing
.41
.41
.07
1996
Legal Services Corp. v.
Velazquez
2000
.35
Lib
.47
.47
.13
43
Table 5
Grouped Event History Estimations
Poisson Link Function with Linear Time Term
Floor Median/
Party
Gatekeeping
Models
|C - L1’|
|C - L2’|
Committee
Gatekeeping
Model
Majority
Party
Model
VetoFilibuster
Model
-2.064***
(0.699)
-2.636***
(0.751)
|C - L3’|
-0.651*
(0.391)
|C - L4’|
-4.279***
(0.985)
|C - L5’|
Age of
Law
-0.079
(0.064)
-0.075
(0.062)
-0.075
(0.071)
-0.042
(0.055)
Constant
-6.107***
(0.365)
-6.126***
(0.351)
-6.159***
(0.427)
-6.405***
(0.324)
N
29668
29668
29668
29668
Wald
Chi2
9.09
14.43
2.80
22.23
0.011***
0.001***
0.246
0.00***
0.604
0.422
0.885
0.369
Prob >
Chi2
Poisson
GOF
Note: *α=.10; **α=.05; ***α=.01 (all two-tailed tests). Huber (1967) robust standard errors clustered by
year of review reported in parentheses.
44
Table 6
Grouped Event History Estimations
Poisson Link Function with Linear Time Term
Unconstrained
Floor Median/
Party
Gatekeeping/
Committee
Gatekeeping
Models
|C-L1|
Unconstrained
Majority
Party
Model
Unconstrained
VetoFilibuster
Model
Constrained
Floor
Median/
Party
Gatekeeping
Models
Constrained
Committee
Gatekeeping
Model
Constrained
Majority
Party
Model
Constrained
VetoFilibuster
Model
-1.432
(0.938)
-1.208
(0.794)
|C-L2|
-0.273
(1.315)
|C-L3|
1.160**
(0.575)
|L1’-L1|
|L2’-L1|
1.071*
(0.608)
|L3’-L1|
0.334
(0.424)
|L4’-L2|
0.920
(0.814)
|L5’-L3|
0.005
(0.051)
-0.029
(0.057)
-0.080
(0.059)
-0.068
(0.052)
-0.062
(0.059)
-0.032
(0.056)
Constant -6.060***
(0.492)
-5.356***
(0.901)
-6.481***
(0.452)
-6.698***
(0.355)
-6.749***
(0.393)
-6.687***
(0.441)
-6.780***
(0.370)
N
29668
29668
29668
29668
29668
29668
29668
Wald
Chi2
2.33
2.32
0.34
4.89
4.99
1.28
1.41
Prob >
Chi2
0.312
0.314
0.842
0.087*
0.082*
0.526
0.493
Poisson
GOF
0.483
0.678
0.538
0.756
0.615
0.632
0.709
Age of
Law
-0.002
(0.049)
Note: *α=.10; **α=.05; ***α=.01 (all two-tailed tests). Huber (1967) robust standard errors clustered by
year of review reported in parentheses.
Table 7
Predicted Probabilities of Reviewing 1987 Statutes
As a Function of Congressional Constraint Variable
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
Floor
Median/
Majority
Party
Gatekeeping
Model
Committee
Gatekeeping
Model
Majority
Party
Median
Model
VetoFilibuster
Model
Committee
Floor
Gatekeeping
Median/
Model
Majority
Party
Gatekeeping
Model
|C - L1’|
|C - L2’|
|C - L3’|
|C - L4’|
|C - L5’|
|L1’-L1|
|L2’-L1|
|L3’-L1|
.00075
.00078
.00072
.00062
.00048
.00052
.00064
.00155
.00155
.00155
.00155
.00155
.00155
.00155
.00116
.00090
.00087
.00077
.00048
.00039
.00051
.00155
.00155
.00155
.00155
.00155
.00155
.00155
.00092
.00092
.00093
.00092
.00086
.00087
.00092
.00150
.00150
.00150
.00150
.00150
.00150
.00150
.00137
.00137
.00137
.00137
.00137
.00040
.00064
.00137
.00137
.00137
.00137
.00137
.00137
.00137
.00097
.00098
.00093
.00088
.00088
.00088
.00088
.00145
.00145
.00146
.00150
.00152
.00155
.00161
.00125
.00112
.00109
.00106
.00101
.00089
.00089
.00139
.00139
.00140
.00143
.00145
.00148
.00153
Note: Predicted probabilities simulated by Clarify 2.1, available at http://gking.harvard.edu.
Statute age is held at the sample mean.
1
Bibliography
TO BE DONE LATER
Also see:
Glendon Schubert (1959) “Policy Without Law: An Extension of the Certiorari Game” 14
Stanford Law Review 284-327 scanty evidence of strategic voting on cert
Doris Marie Provine (1980) Case Selection in the United States Supreme Court Chicago:
University of Chicago Press. no strategic voting, cert based on jurisprudential criteria
Jan Palmer (1982) “An Econometric Analysis of the U.S. Supreme Court’s Certiorari Decisions”
39 Public Choice 387-398
Brenner, Saul and John F. Krol (1989) “Strategies in Certiorari Voting on the United States
Supreme Court, 51 JOP 828-840.
Krol and Brenner (1990) “Strategies in Certiorari Voting on the United States Supreme Court,”
43 Western Political Quarterly 335-342.
2