Has Anthony Kennedy Moved to the Right? Maybe Not.

by John Sides on July 3, 2012 · 7 comments

in Judicial,Law,Methodology

This is a guest post by Michael Bailey.  See also his earlier post on the ACA decision.


Has Justice Kennedy moved dramatically to the right? Many believe so and Kennedy’s behavior on the health care case certainly reinforces this view. I want to express a bit of skepticism about such claims.

First, these claims rest on justice ideology scores developed by Andrew Martin and Kevin Quinn.  But these scores aren’t great at tracking preference change over time.  As the figure below shows, their method implies that the Court reached a conservative peak in the early 1970s, a time when the Court was creating constitutional rights to abortion and against the death penalty.  Quinn has (with Dan Ho) expressed concerns about using these scores to compare preferences over time and I agree.

Second, it’s not clear to me where the apparent shift to the right is coming from.  The Court is not ruling in a “conservative” direction more often, as the graph above demonstrates. (There are questions about what is conservative, of course, but I think it’s reasonable to take the coding in the Supreme Court database as a reasonable first cut for our purposes.).  Moreover, from 2008 to 2010, when Martin and Quinn estimate that Kennedy shifted to the right, some high-profile conservative cases were not major shifts Kennedy.  The conservative Citizens United decision continued Kennedy’s jurisprudence since Kennedy had voted against the original Austin decision that Citizens United overturned.  The Court’s decision in the McDonald gun case was supported by more than 300 members of Congress, who signed as amici advocating the eventual outcome (including many moderates such as Senators Baucus, Feingold, Snowe and Webb and Representatives DeFazio, Dingell, Giffords and Oberstar; only 56 members of Congress signed on as amici for the liberal side).

Of the remaining high-profile ideologically charged cases from 2008 to 2010 there is a rough parity between liberal and conservative votes for Kennedy.  By my count, Kennedy supported conservative outcomes on eight ideologically divisive cases (beyond what is discussed above and not counting this term, which does not factor in the Martin and Quinn scores).  These must be taken against nine ideologically divisive cases where Kennedy voted liberally.  That seems like the good old center-right Justice Kennedy we know rather than a new, hard-right one.  (I’ll share the specific cases I’m thinking of in comments.)

Third, the Court (which usually means Kennedy) is not that conservative compared to public opinion. Jessee and Malhotra (2012) polled Americans on specific Supreme Court cases, presenting them summaries of each position and asking which way the Court should have ruled.  Eight of the nine cases they polled had conservative outcomes (Citizens United, Heller, Salazar, Ricci, Crawford, Baze, Parents Involved, Gonzales v. Carhart) and respondents were, on average, in agreement with the Court’s decisions 71.6 percent of the time.  (The one liberal decision in the survey (Hamdan) had the lowest level of popular support, at only 29.9 percent agreement).  This is not proof that the Court hasn’t moved to the right, but I think it provides helpful context.  If the public agrees with the Court and the Court has indeed moved dramatically to the right, then the real story should probably be that the Court was dramatically more liberal than the public in 2007 before its shift to the right.  I don’t think that’s true; hence my skepticism about the claim that the Court has dramatically moved right.

Why, then, do the Martin and Quinn scores show such a move to the right by Kennedy?  I don’t know and have been thinking about various possibilities.  One idea is that there were a number of cases on Fourth and Sixth Amendment cases (e.g. Bullcoming, Melendez) in which Kennedy voted in a conservative direction and Scalia and Thomas voted “liberally” (in favor of defendants).  (If you want to appear conservative, just vote conservative on cases where Scalia and Thomas vote liberally!)  But are these votes really that conservative?  On every one of these cases, Kennedy voted with Breyer (and sometimes other liberals).  These votes seem to reflect less a shift to the right than the emergence of set of cases that aren’t on the standard left-right dimension. (The flaw in this idea is Martin and Quinn approach does not rely on liberal and conservative codings, but instead essentially infers the liberal/conservative nature of a decision from its voting coalition.  Hence the idea that the Fourth/Sixth Amendment cases are the cause of the shift in Kennedy’s estimates is only a guess; these cases may actually get coded as conservative by the Martin and Quinn approach simply by virtue of Thomas and Scalia’s votes.)

I’ve estimated Supreme Court preferences (which I’m calling “bridge estimates”) based on models that use additional information to pin down preference change over time.  While the Martin and Quinn model requires one to assume the spatial location of the Court’s agenda does not change over time, my approach depends on an assumption that my additional information is right (more information is here).  That’s a (looong) discussion for another time perhaps.  The upshot for now is that with this approach the median of the Court (who has been Kennedy since 2006) has been conservative, but nothing like the radically conservative Court depicted in the Martin and Quinn scores.  (Note that I normalized each set of scores by subtracting the mean to make them more directly comparable.)

It is possible that Kennedy could move to the right in response to the apparent hard feelings generated by the ACA case.  But through 2011 at least (which is the period covered by the Martin and Quinn scores) I’m not persuaded that he – and the Court—have moved dramatically to the right.

Finally, let me add a “meta” point. Martin and Quinn are incredibly smart scholars who have made huge contributions to the study of the Supreme Court.  After they put their results out, the rest of us mull them over and think about if and how we can improve them.  This post and the underlying paper are my two cents.  Rinse, lather and repeat and hopefully we can make progress on interesting questions.  This is how political science works.


Michael Bailey July 3, 2012 at 12:01 pm

Some additional details, for those interested:

Cases on which Kennedy supported conservative outcomes include District Attorney v. Osborne (finding no post-conviction right to potentially exculpatory DNA evidence), Connick v. Thompson (holding against holding prosecutor’s office liable for civil rights violations that arose because of poor training), Ashcroft v. Iqbal (holding that top government were not liable for alleged discriminatory activity of subordinates), Herring v. U.S. (allowing certain good faith exceptions to exclusionary rule), Salazar v. Buono (allowing a cross on public park), Ricci v. New Haven (disapproving of New Haven decision to nullify a firefighter’s exam on which whites did better than non-whites), 14 Penn Plaza v. Pyett (holding that contractual arbitration precluded judicial resolution of discrimination claims), and FCC v. Fox (upholding FCC ban on “fleeting expletives”).

Cases in which Kennedy wrote or joined liberal opinions include Graham v. Florida (holding juveniles cannot be sentenced to life in jail for non-homicide offenses), Cone v. Bell} and Harbison v. Bell (providing rights for death row inmates), Brown v. Plata (releasing California prisoners due to overcrowding), Wyeth v. Levine (holding drugs companies not shielded from state liability laws due to federal regulation), Altria v. Good (holding federal laws do not preempt ability of states to regulate tobacco), Caperton v. Massey (holding that a West Virginia judge had to recuse himself from case involving major campaign contributor), Safford Unified School District v. Redding (deeming a strip search of a junior high girl unconstitutional – unanimous on central point of case) and Brown v. Entertainment Merchants Association (striking California law against selling violent video games to minors).

For the purposes of evaluating the measures we should keep in mind that some of conservative votes by Kennedy that might be on our minds actually were decided in the 2006 or 2007 terms, terms in which Martin and Quinn scores indicate no change for Kennedy. These include Parents Involved v. Seattle School District, Hein v. Freedom from Religion Foundation, Morse v. Frederick, Ledbetter v. Goodyear, Bowles v. Russell, Medellin v. Texas and Gonzalez v. Carhart. Many of these are in the Jessee and Malhotra list of cases, too, so perhaps we can only say the Court was representative of public opinion before the 2008 term. That still leaves as a mystery as to what the set of cases are on which the Court did in fact deviate sharply to the right.

Andreas Moser July 3, 2012 at 12:26 pm

I don’t know where right and left are.

Mark July 3, 2012 at 2:26 pm

Part of the problem is that the right-left analysis does not fit Kennedy’s approach. Kennedy has increasingly stressed liberty interests in his jurisprudence. This has led to both his opinion in Lawrence v Texas (the sodomy case) which was a different approach than followed by the other “conservative” justices (I believe Thomas called the Texas statute “uncommonly silly” but voted to uphold) and the dissent in the Obamacare case where the vigorous language about liberty interests is almost certainly Kennedy’s and which contrasts with the approach of the liberal justices.

Kennedy’s focus on liberty does away with the post New Deal distinction the Court drew between Constitutional rights entitled to protection and to which “strict scrutiny” is applied and those which the Court basically does not care about any more. It is a distinction followed by the other justices regardless of where you place them on the left-right spectrum.

Kevin Quinn July 4, 2012 at 9:54 pm

It’s great to see such a rich discussion of substance and methodology in a public forum such as this. We’re also both big fans of Michael Bailey’s work on measuring ideal points in ways that are comparable across institutions. We do have some thoughts about the argument Mike is making—some of which are technical (read boring) and not of general interest here. There are three things, though, that we think are worth mentioning in this forum:

1] It seems odd to us to use “Percent Conservative” as a way to validate one measure over another. If that were the gold standard—and there are plenty of reasons why it isn’t—then we should just use that as our measure and not rely on measures that are statistically derived! Picking and choosing a handful of cases also doesn’t seem like a good way to validate a measure. While Roe is certainly an important case, all extant models treat it just like any other case.

2] It is true that the so-called Martin-Quinn model picks up a rightward movement of Justice Kennedy in recent terms. This is because a) he is agreeing with the liberals (Justices Breyer and Ginsburg) much less than in the past, b) he is agreeing with some conservatives (notably Justice Thomas) more than in the past, and c) he is agreeing with the other conservatives at relatively high, fairly stable, rates. The change in inter-agreements among the justices is what is driving Justice Kennedy’s trend. The fact that Mike’s model doesn’t pick this up is somewhat troubling. Why does Mike’s model have Kennedy moving slightly to the left when he is voting with Justices Breyer and Ginsburg less often? We’ve yet to fit the model using data from the 2011 term. It will be interesting to see what happens with Kennedy going forward.

3] The point about cardinal vs. ordinal measures in the Ho and Quinn article is that *any* measurement model that takes roll call votes (or the equivalent) as input will impose assumptions in order to produce cardinal measures. The cardinal measures are, thus, model dependent. On the other hand, the ordinal measures that emerge from most measurement models depend much less on modeling assumptions. All of this is true for NOMINATE, Clinton-Jackman-Rivers, Martin-Quinn, and Mike’s model as well. The use of bridging observations doesn’t ameliorate this limitation whatsoever. One could argue that the alternative modeling assumptions that Mike employs are more reasonable than other assumptions. Whether one believes such an argument or not doesn’t change the fact that these assumptions play an important role in producing the resulting cardinal measures.

In nearly any measurement exercise there will be a number of modeling assumptions that can, and should, be debated. We both firmly agree that this sort of debate is hugely important for improving our collective understanding of both politics and our models of politics. We look forward to a continued discussion about the best way to place Supreme Court justices on an ideological scale.

Kevin M. Quinn, UC Berkeley
Andrew D. Martin, WashU

Michael Bailey July 5, 2012 at 4:03 pm

Thanks, Kevin.

It’s an interesting question about how to validate statistical models. I don’t have a deep theory, but my general sense is that statistical models can get screwy for all kinds of reasons (from mispecification to coding anomalies to changing dimensionality — I’ve seen them all in my own work) and it’s a good practice to go back and look at the underlying phenomenon to make sure what we see there accords with results from a statistical model.

Percent conservative or cases-by-case analysis (or, for that matter, inter-justice agreement) are not the gold-standard for assessing statistical results. I do think they are helpful to calibrate our statistical results to what we see in the real world. I think a huge shift to the right w/o a shift in percent conservative is a signal to keep looking (if the issue space shifted to the right one could see exactly this phenomenon, but then the question is how the issue space shift is identified).

Inter-justice agreement belongs w/ percent conservative and case-by-case examination in this sense. I looked at the SCOTUS stat pack data (http://www.scotusblog.com/reference/stat-pack/) on this and don’t see the patterns you described:
Kennedy voted with Ginsburg 45% of the time in 2006 October term (mostly 2007), 45% in 2007, 59% in 2009 term, 50% in 2011;
Kennedy voted with Breyer 54, 61, 49 and 60 % in 2006, 2007, 2009, and 2011;
Kennedy voted with Scalia 57, 53, 50 and 53% in those terms.
Kennedy voted with Thomas 56, 44, 46 and 53 % in those terms.

Perhaps this is not fully representative (it is sloth, rather than some careful design that led me to look only at these 4 years). I looked at % full agreement on non-unanimous cases.

Kevin Quinn July 8, 2012 at 10:50 am

Thanks Mike. The agreement scores I based my comment on were also from the Scotusblog statpack. I was looking at agreement in full, in part, or judgment (1 minus the disagree in judgment rate) for the 2007, 2008, 2009, and 2010 terms. These are the most recent terms of the Court that are used to to compute the available Martin-Quinn scores.

The different definition of agreement and the difference in the terms used explains the discrepancy between what you report and what I described.

The R code below shows the agreement scores for select justice-pairs and plots these agreement rates on term.

I haven’t taken the time to evaluate this carefully, but I think part of the reason for Kennedy’s rightward movement is the fact that 2010 is the last term going into the model and this is the term where Kennedy looks most Thomas-like and least Breyer-like. The fact that there’s no data (yet) from 2011 or later going into the model to help smooth this out is what allows Kennedy to move fairly suddenly.

R code:
## Kennedy agreement scores (non-unan)
thomas <- c(56, 70, 52, 73)
scalia <- c(75, 72, 63, 67)
alito <- c(80, 81, 71, 76)
roberts <- c(83, 79, 76, 81)
breyer <- c(80, 66, 56, 50)
ginsburg <- c(62, 51, 56, 36)

## other relevant agreement scores (non-unan)
gins.brey <- c(75, 72, 76, 71)
scal.alit <- c(78, 81, 67, 74)
scal.thom <- c(84, 81, 85, 74)
scal.rob <- c(85, 81, 78, 81)

terms <- c(2007, 2008, 2009, 2010)

par(mfrow=c(2,1), mar=c(4,4,2,2)+.1)
plot(terms, scalia, xlab="Term", ylab="Pct. Agreement",
type="l", col="red", ylim=c(30,100), xaxt="n")
axis(1, at=terms)
lines(terms, thomas, col="darkgoldenrod1")
lines(terms, alito, col="brown4")
lines(terms, roberts, col="black")
lines(terms, breyer, col="blue")
lines(terms, ginsburg, col="dodgerblue")
legend(x=2008.75, y=100, legend=c("Kennedy-Thomas", "Kennedy-Scalia", "Kennedy-Alito", "Kennedy-Roberts", "Kennedy-Breyer", "Kennedy-Ginsburg"), lty=1,
col=c("darkgoldenrod1", "red", "brown4", "black", "blue", "dodgerblue"),

plot(terms, gins.brey, xlab="Term", ylab="Pct. Agreement",
type="l", col="black", ylim=c(30,100), xaxt="n")
axis(1, at=terms)
lines(terms, scal.alit, col="green")
lines(terms, scal.thom, col="darkorchid3")
lines(terms, scal.rob, col="deeppink1")

legend(x=2009, y=60, legend=c("Ginsburg-Breyer", "Scalia-Thomas", "Scalia-Alito", "Scalia-Roberts"), lty=1, col=c("black", "darkorchid3", "green", "deeppink1"), cex=.45)

Wilford April 22, 2013 at 9:37 pm

You should take part in a contest for one of the highest quality
blogs on the web. I will highly recommend this site!

Comments on this entry are closed.

Previous post:

Next post: