Canes-Wrone, B., Clark, T., and Kelly, J (2014). Judicial
Selection and Death Penalty Decisions. American
Political Science Review, 108(1):23-39.
Question 1: The authors test three hypotheses: first, that
partisan state supreme court judges (where “partisan” refers to selection by
partisan election) are least likely to uphold death sentences, compared to
nonpartisan (self-explanatory) and retention (subject to merit retention
election) judges; second, that there is a direct relationship between public
support for the death penalty and the likelihood that partisan judges will
affirm death sentences; and finally, that there is a direct relationship
between public support for the death penalty and the likelihood that a
reappointed judge (one who does not face direct election in any way, but is
reappointed by directly elected officials) will affirm death sentences. All of
the hypotheses are theoretically supported, not only in their assertions, but
in their limitations. The first hypothesis is supported by the notion that
partisan labels are powerful signals that can override interest group action.
Thus, judges who do not benefit from partisan labels may avoid action that
would trigger interest group action (such as attack ads for overturning a death
sentence). While the authors note that this could create collinearity problems
where judicial decisions are driven by support for the death penalty rather
than judicial selection systems, they point out that there is no variation in
the data as to “whether the death penalty is popular.” They differentiate this
from “level of support” to justify the phrasing of their first hypothesis
(which asserts a greater likelihood, rather than a direct relationship). As to
the second hypothesis, it is supported by literature that suggests that
partisan elected officials are likely to be responsive to lopsided public
opinion. Because the literature focuses only on partisan officials, the authors
limit this hypothesis to partisan selection systems, acknowledging that the
predictions could play out differently in nonpartisan contexts. The final
hypothesis cites to literature that “indirectly elected” officials behave
similarly to directed elected officials “when voters’ policy views are strong.”
This is the weakest linkage, as the authors a) assert that reappointment judges
are “indirectly elected” without assessing the institutional norms that
constrain judges, instead treating them like bureaucrats without civil service
protection; and b) they assert without evidence that the death penalty is
always salient, when they probably should have included salience as a control
variable for this hypothesis. While testing for salience is problematic, the
authors should have at least attempted it.
Question 2: The dependent variable is a judicial vote. It is
measured as a vote to uphold a death sentence is DV=1, a vote to grant relief
is DV=0. “Vote to grant relief” is broadly defined as a vote that supports a
“ruling that precludes inposition of a death sentence unless further action is
taken by some court.” This measure is internally valid as the hypotheses are
testing judicial behavior as measured through votes. It is reliable because
judicial votes can be discerned even for per
curiam opinions where judge’s positions are not listed, as per curiam opinions are presumably
unanimous.
Question 3: The independent variable is judicial selection
mechanism. The authors engage in a certain amount of lumping in coding this
variable, as reappointment mechanisms vary in their details as to the relative
involvements of the executive and legislative branches. All of those systems
are lumped together as “reappointment.” They also note, and code separately,
those states that changed their selection mechanisms during the period under
study, which provide for natural experiments. Finally, they exclude states
where supreme court judges are selected in districts, rather than by a
state-wide mechanism, and where the selection mechanism is a hybrid of multiple
coding options. They code via four dummy variables, where the state is coded as
IV=1 for its selection mechanism, and IV=0 for all others. This measure is
internally valid because it isolates the effect of the selection system on
death penalty votes, although they note that because of the lack of variance in
death penalty support, they cannot exclude the possibility that judges are more
responsive to a punitive public than otherwise. They also interact their
primary IV with the level of support for the death penalty to informally test
the second hypothesis for systems other than partisan selection. This measure
is reliable because judicial selection mechanisms are readily available for
anyone to find.
Question 4: The authors control for four judge-level
variables: partisan affiliation, proximity to reselection, proximity to
mandatory retirement, and lame-duck status. They control for nine case-specific
variables: murder of a police officer, murder in conjunction with rape or
robbery, multiple or female victims, the number of grounds presented in the
appeal, whether the state supreme court characterizes the appeal as rising
under a newly-decided United States Supreme Court decision, the state’s
homicide rate, and time (arguing that as states perfect their death penalty
processes to survive Eighth Amendment scrutiny, reversals drop). In an
appendix, they toy with a method for controlling for race of defendant and
victim, but conclude that the method is not sufficiently reliable to include in
the main paper, while noting the results. They also control for state-level
fixed effects, holding constant “the state’s general propensity to affirm death
sentences.” The state-level fixed effects model analyzes only judges in states
that changed their selection mechanisms to use the natural experiment method.
The authors also use a judge-level fixed effect model to examine the effect of
changes in selection mechanisms on individual judges. The fixed-effect models
are justified as natural experiments in states that changes their selection
mechanism.[1]
Two of the judge-specific control variables and eight of the nine case-specific
variables are derived from the literature that argues that they affect judicial
decision-making, and thus are necessary. The remaining three variables are not
supported by the literature. The case-specific variable – whether the United
States Supreme Court triggered the appeal – is theoretically unjustified, in
that the authors note that such decisions “tend to inspire a wave of successful
appeals, as well as some unsuccessful ones.” Because of this case-driven
variation in results, this variable, if included, should have only been
included as an interaction with other case-specific controls. However, it turns
out that this variable is tremendously significant, explaining over one-third
of the variation, and statistically significant at the p<.01 level. This
suggests that the variable may have initially been included as an interaction
variable, and was disaggregated after the authors saw its significance and a post hoc theoretical justification
offered, potentially on resubmission. With the judge-level variables, the
theory at least justifies their inclusion, as the authors argue that judges who
do not face reselection pressures do not suffer from the same public opinion
constraints. However, the numbers don’t add up: none of the models show that
these variables have any statistical significance, although lame-duck status
does explain a high level of variance, suggesting that judges who have been
rejected for reselection retaliate against the system that turfed them out.
Question 5: These authors, in describing judge-level control
variables, place themselves firmly in the attitudinal school of judicial
politics. They really, really want to treat judges as legislators (or at best, as
bureaucrats). But the attitudinal model has been, more or less, exploded for
twenty years. While strategic approaches to judicial behavior are predicated on
the court being treated as a closed system, which renders their theory moot,
the new institutionalism explores the role of institutional design and norms on
judicial behavior. I would like to have seen this article explore the
institutional norms (prior precedent, court composition, panel composition
rules, etc.) to determine if those constraints were driving judicial decisions.
For example, a Texas judge who believes that ineffective assistance of counsel
claims are valid where counsel slept through a trial would be constrained to
vote to uphold a death sentence by Texas law which forecloses such claims. This
would be true, even if public opinion shifted in favor of recognizing such
claims, unless a majority of the court shifted with it. In short, prior
precedent forecloses certain votes until public opinion sufficiently interacts
with judicial ideology to shift a majority of the court’s votes. But by
excluding such factors as the role of stare
decisis, the authors ignore the fact that courts are not legislatures.
Question 6: I would use content analysis to challenge this
article. I would approach my challenge by studying the opinions that decide death penalty cases, rather than simply votes.
This obviates the issue of treating judicial votes as equivalent to legislative
votes, and incorporates the power of concurrences – which these authors
conflate with majority votes.
[1]
This aspect of the response is thus not eligible for grading under the terms of
the exam, as the method for these models was experimental. These models will
not be discussed further.