Strategic thinking on the Court: The decision to retire


On May 1, 2009, Justice David Souter announced his intention to retire from the Supreme Court. The timing of Souter’s departure has raised anew the question of strategic departures from the Court. Because members of the nation’s highest court can retire whenever they choose, or not at all, their departures are not ordained by fixed terms, term limits, being voted out of office, or a host of other factors that play into other public officials’ thinking about whether the time has come to hang it up. A potential element in their thinking, then, is how their retirement is likely to affect the ideological balance on the Court. Thus, members of the Court’s conservative bloc may be determined to outlast a liberal president and a Democrat-dominated Senate in hopes that their successors will not be able to undo what they have done on the Court; and liberal justices who are considering retirement may decide that the time is ripe when the Democrats control the White House and the Senate.

Such, at any rate, is the conventional wisdom. But how accurate is it? At least one justice left no doubt that his thinking about whether to stay on the Court was influenced by precisely such considerations. Chief Justice William Howard Taft, pondering retirement, wrote:

I am older and slower and less acute and more confused. However, as long as things continue as they are, and I am able to answer in my place, I must stay on the Court in order to prevent the Bolsheviki from getting control.

That quote is contained in a study by Kjersten Nelson and Eve Ringsmuth that appears in the May 2009 issue of American Politics Research (gated; abstract here). In that study, Nelson and Ringsmuth present statistical tests of the notions that, all else equal, a justice’s likelihood of retiring in a given year is inversely related to his or her ideological distance from the president and the Senate. Analyzing annual data from 1947 through 2004 (a period in which there were 19 retirements), Nelson and Ringsmuth find that ideological distance from the justice to the median senator (and, for that matter, from the justice to the Senate’s filibuster pivot) did lessen the probability that the justice would retire. However, no parallel relationship emerged for the justice and the president. The authors don’t offer any compelling explanation of why distance from justice to president doesn’t seem to matter. They do note that presidents often anticipate Senate reactions to their nominations rather than acting independently, but there’s many an unexamined step between that tendency and justices’ retirement decisions. In any event, the link that Nelson and Ringsmuth do observe between the ideological composition of the Senate and justices’ ideological positions provides suggestive evidence that Chief Justice Taft was not alone in thinking strategically about the time had come to retire.

9 Responses to Strategic thinking on the Court: The decision to retire

  1. Emery June 5, 2009 at 5:48 am #

    It seems that the lack of a relationship to presidential ‘ideology’ actually cuts against the strategic retirement theory, no? I mean, the ‘distance’ to the ‘Senate median’ is the sort of thing that one would look at AFTER THE EXPECTED RELATIONSHIP TO PRESIDENTIAL IDEOLOGY DIDN’T PAN OUT. No?

  2. Lee Sigelman June 5, 2009 at 2:03 pm #


    Thanks for the link to that study. From a quick look, I see that it contains a comprehensive overview of research on this question and tests a fairly similar model to the one described in my post. Some key differences are: (1) time frame — this one goes back much farther, which may or may not be a strength; Nelson/Ringsmuth argue that the hypothesized effect should emerge in the modern (e.g., post-WWII) era; (2) conclusion about the presidential factor, which comes through pretty clearly in the study to which you linked; and (3) inclusion or exclusion of the Senate variables, which are absent from the study to which you linked.


    Yes, the lack of relationship with presidential ideology does cut against the strategic retirement hypothesis (I certainly wouldn’t call it a “theory”). But I don’t agree that the distance to the Senate median is just an ex post facto attempt to put things together after the main hypothesis has fallen flat. The Senate is a major player in the nomination/confirmation process.

  3. C. Zorn June 5, 2009 at 2:39 pm #

    Since we’re on the topic of when a political effect materializes, I’ll shamelessly (self-)promote this:

    (Sadly, it’s gated). It’s broadly consistent with N&R’s conclusions.

  4. Terry June 7, 2009 at 3:18 pm #

    Which president was Taft worried about as a Bolshevik? There were only Republicans in the White House while he was on the Supreme Court. Did he write this during an election year, worrying about Al Smith? As LBJ showed, you can’t expect to get to name a justice as a lame duck.

  5. Lee Sigelman June 7, 2009 at 4:13 pm #

    It’s not clear to me who Taft was talking about. It may not even have been a potential president. He could have been talking about other members of the Court, or about senators, or even about political trends in general. Maybe someone else will weigh in on the subject.

  6. Emery June 11, 2009 at 6:46 pm #

    As I understand the traditional “hypothesis,” ahem, judges and justices wait for a president of their own disposition (traditionally party-wise, but not necessarily) to retire. Maybe I am dense, but that seemed like the “hypothesis.” Or whatever (“theory” may be too strong, maybe not for a blog comment). Now we are told that judges (justices) don’t worry about the president’s ideology . . . but about the median senator’s. Hmm. That sounds post hoc to me.

    Again, there is a pretty small n here–I’m not sure what conclusions I would draw from 19 observations, given that there are major differences b/w a 1947 retirement and a 2004 retirement (that’s what the post suggests, but it must be a 1994 retirement, right? because post-1994, the next retirement was 2005, IIRC)–so, who knows.

  7. Lee Sigelman June 11, 2009 at 6:58 pm #


    It could be post hoc. I can’t speak for what went on in the course of the research project, as I don’t even know the researchers. Just as a side comment, though, I’d say that in my experience a great deal — I’ll even venture to say most — of the results that one finds written up in the journals are post hoc.

    As for the small-n problem and how it affects the ability to draw conclusions, I think this is a non-problem. The authors have the full population of cases they’re concerned about. They haven’t drawn a sample of just nine from some large population. Nine there are, and nine they’ve got. Whatever they observe pertains to the population in which they’re interested, with no sampling error whatever, no problem of nonrandomness, etc. (Note that they’re not even trying to generalize to earlier cases; they explicitly rule those out of consideration.)

  8. Emery June 11, 2009 at 11:49 pm #

    It may depend on what one understands as the small-n problem. I am not an expert here. Admittedly.

    One may start from the premise that, if in 57 years, there are 9 cases, then 9 cases is what there is. But if one assumes that one needs ten-twenty cases per variable, for reliable analyses, in a multivariate analysis, then with 9, one has enough observations for a single variable . . . and one is not doing multivariate analysis.

    It may be that the small-n problem doesn’t arise with a qualitative study. But with an analysis that looks at the median senator? Based on–I’m guessing here–NOMINATE scores . . . which are sensitive to historical factors . . . which cannot be accounted for with 9 cases. One certainly can’t dummy out years or presidents with 9 cases. (Or is it 19? Not clear.)

    The population-versus-sample problem frame assumes that one has enough cases to make reliable population estimates . . . I think. If the ‘population’ is very small, then non-generalizable findings may be apt but not publishable, in the widely accepted sense.

    Again, I will admit I’m no expert, certainly not with extremely small population sizes. But a published piece should include generalizations, no? Maybe one can extrapolate from nine or 19 observations to a wider world. One may rule out earlier cases. But even so, does nine or even 19 observations enable one to predict future cases? Based on–variable(s) that one has accounted for? That was, I think, the premise of the post?

    Can one even rule out randomness with nine cases? Cases that occur on average once every 5.2 years (most generous estimate). How would one know that one could?

    Again, I haven’t read the piece. Maybe I should. But I’m not seeing why I should so far.