LIFE TENURE RECONSIDERED: TERM LIMITS FOR … 2007 Revision Supremes.doc · Web viewRunning Head:...

Politicized Departure from the United States Supreme Court*

Running Head: Supreme Court Departures

May 14, 2007

Total Word Count 11,611

Ross M. Stolzenberg**Department of Sociology

University of [email protected]

Telephone: 847 408-0168

andJames LindgrenSchool of Law

Northwestern [email protected]

Telephone: 312-503-8374

*We thank Albert Yoon for making available to us some of the data analyzed herein, and Lee Epstein for advice.

**Corresponding author.

mailto:[email protected]

mailto:[email protected]

Politicized Departure from the United States Supreme Court

ABSTRACT

Nearly seven decades of quantitative analysis and narrative historical research investigate the

hypothesis that political factors influence the probability and timing of resignations and deaths-in-office

of U.S. Supreme Court justices (the Politicized Departure Hypothesis). Narrative historical studies mostly

report support for the hypothesis, quantitative analyses mostly report the opposite, and neither method

yields consistent findings. Reasons for these differences are uncertain because prior studies differ

dramatically in the data they examine and the methods they apply. Previous quantitative analysts criticize

each others’ methods and generally regret the lack of appropriate health (vitality) measures for most

justices. Here, we review prior research, reformulate hypotheses, and reconsider conceptual, measurement

and analytic issues. We include a widely-accepted vitality measure. We analyze data on every justice

from 1789 through 2006, using robust, cluster-corrected, discrete time, censored, event history methods.

We apply previously-used techniques to compare our results to previous studies. Findings are consistent

with the Politicized Departure Hypothesis and suggest that its effects are substantial. Most non-political

factors have nonlinear effects on resignations, some of which reverse direction as justices age.

2

Supreme Court Departures May 14 2007 Page 1

I. INTRODUCTION

Of U.S. Supreme Court Justices, two things are certain: They are appointed to the Court, and they

subsequently depart from it. By law, Court appointment is explicitly political, involving nomination and

appointment by the President “with the Advice and Consent of the Senate” (U.S. Const., Art. II, sec. II).

But law dictates no role for Presidents, politics, or public scrutiny in selecting the times at which justices

vacate the Court: Except in cases of treason, bribery, or other serious crimes, justices leave office only by

death, or when they themselves alone, individually, resign. Nonetheless, observers have long asserted that

the timing of justices’ resignations from the Court, and even the probability that they die in office, reflect

a highly politicized process that does, like their nominations to the Court, revolve around political

compatibility between the individual jurist and the incumbent President of the United States (Ward 2003;

Calabresi and Lindgren 2006; Zorn and Van Winkle 2000; Biskupic 2004). We call this assertion the

Politicized Departure Hypothesis, and we develop and test it below.

The Politicized Departure Hypothesis asserts that the timing of resignation from active service on

the Supreme Court tends to be influenced by political as well as personal circumstances of justices.

Personal circumstances likely would include the same factors that influence retirements and voluntary job

terminations of all workers -- vitality, age, personal finances, and job tenure (i.e. length of service on the

Court) (see, e.g., French 2005). Political circumstances are hypothesized to impinge on personal

considerations because the retirement of a justice – particularly if it occurs early in a President’s term of

office – allows the sitting President to nominate the replacement for that justice, and thereby prevents a

future President of the opposing party from doing so. The Politicized Departure Hypothesis reasons that,

regardless of differences in political philosophy, justices tend to be loyal to the political party of the

President who nominated them to the Court and that justices consequently tend to adjust the timing of

their resignations to give a President of that same party the opportunity to appoint their judicial successor

(see Epstein, Segal, Spaeth, and Walker 1996, 368).1 In particular, the Politicized Departure Hypothesis

asserts that:

1) Other things equal, if the incumbent President is of the same party as the President who

nominated the justice to the Court, and if the incumbent President is in the first two years of a

four-year Presidential term, then the justice is more likely to resign from the court than at times

when these two conditions are not met.

1


By removing live justices from office, accelerated departure reduces their opportunity to die in office.

Retarded departure does the opposite. Thus, the Politicized Departure Hypothesis implies that,

2) ) Other things equal, if the incumbent President is of the same party as the President who

nominated the justice to the Court, and if the incumbent President is in the first two years of a

four-year Presidential term, then the justice is less likely to die in office than at times when these

two conditions are not met.

This paper tests these two implications of the Politicized Departure Hypothesis. Section I considers

previous research and historical material that is related to the Politicized Departure Hypothesis. Sections

II and III consider methodological and conceptual issues, data, and measurement. Section IV reports

analytic results. And Section V discusses findings.

II. EXISTING EVIDENCE REGARDING POLITICIZED DEPARTURE

Table 1 summarizes key features of some previous considerations of the Politicized Departure

Hypothesis, starting with Fairman’s (1938, 406) Harvard Law Review article that considers both

quantitative data and narrative historical evidence. Subsequent narrative studies tend to be highly

selective, focusing on specific justices whose Court departures appear to have been timed politically. For

example, Goff (1960, 96), argues that a senile and incompetent Justice William Cushing remained in

office until death in 1810, solely in an attempt to stave off the appointment of a Democratic-Republican

replacement. Van Tassel (1993), Atkinson (1999), Garrow (2000) and Ward (2003) provide historical

evidence of politicized departure by particular justices. Farnsworth reports that Justices William

Rehnquist (2005, 448) and Sandra Day O’Connor (2005, 375) stated preferences that their replacements

be nominated by a President of the same party as the President who appointed them. Hutchinson (1998)

reports statements of similar preferences by Justice Byron White. Justices Black, Douglas, Marshall and

Brennan are all reported to have delayed departure in unsuccessful attempts to give a Democratic

President the opportunity to name their successors (Oliver 1986, 806-08; Woodward and Armstrong

1979, 161).

Although some narrative historical studies provide evidence that several justices acted according

to the predictions of the Politicized Departure Hypothesis, they leave three sources of uncertainty. First,

some narrative studies contradict the Politicized Departure Hypothesis. For example, Brenner (1999)

argues that politicized departure occurs rarely or never (but see Ward’s [2004] critique of Brenner).

2


Others cite the case of Justice Warren to assert that political philosophy, rather than party identification

per se, affects the timing of some Court resignations; Justice Warren was appointed by Republican

President Eisenhower, but is reported to have tried unsuccessfully to time his resignation so a Democrat

could nominate his successor (Oliver 1986, 805-06, citing White 1982, 306-08, Schwartz 1983, 680-83,

720-25). Second, most of these narrative accounts rely on justices’ recollections of their thoughts and

emotions, their subjective explanations of their own past behavior, and sometimes even their predictions

of their future behavior. Contemporary social science evidence suggests skepticism about the validity of

such reports (see Fazio 1986; Fishbein and Ajzen 1975; Cannell, Miller, and Oksenberg 1981;

Eisenhower, Mathiowetz, and Morganstein 1991; Kalton and Schuman 1982; Sudman and Bradburn

1982). Third, and perhaps most important, existing historical narrative studies disproportionately focus

on justices whose behavior is hypothesized or reputed to be dramatically consistent with the politicized

departure hypothesis. By selecting and focusing on judges whose departures were highly politicized, these

studies provide no evidence whatsoever concerning departures of the great majority of Supreme Court

justices. In short, historical studies provide nuanced understanding of specific cases, and they are a useful

basis for formulating the Politicized Departure Hypothesis, but they are too selective to test that

hypothesis, or to indicate if Politicized Departure is a general pattern, an occasional aberration, or a

hypothesized event that has never actually occurred.

In contrast to fine-grained, narrative historical research that seeks certainty about the personal

motives and retirement behavior of particular individuals, existing quantitative analyses consider

probabilistic assertions about patterns of behavior by Supreme Court justices in general, examine

observable, publicly available data; and consider hypotheses that observable political circumstances have

had, on average, some effect on the timing of all departures from the Court. Quantitative analyses of

Supreme Court departures vary widely in substantive conclusions and methodology. Substantively, Yoon

(2006), Brennner (1999) and Squire (1988) report finding no evidence of a general pattern of politicized

departure of justices. In contrast, Hagle (1993) and King (1987) report findings of political effects on

annual numbers of departures from the Supreme Court. Zorn and Van Winkle (2000), and Box-

Steffensmeier and Zorn (1998) present some statistical analyses that are consistent – and some that are

not consistent – with the hypothesis of politicized departure.2

While the contradictory findings of prior research permit no firm substantive conclusions,

3


methodological disagreements between previous analysts clearly delineate critical methodological issues.

These issues include the lumpability of death-in-office and retirement, aggregation of individual data into

frequency data for the Court as a whole, measurement and control of the vitality (health) of justices, and

the calculation of effect measures and statistical significance in analyses involving multiple indicators and

statistical interactions. We describe those issues briefly:

Lumpability. Some previous statistical studies (e.g. King 1987; Yoon 2006) lump resignations

and deaths-in-offices into undifferentiated departures from the Court. Hagle (1993), Spriggs and

Wahlbeck (1995, 575), and Nixon and Haskin (2000, 462) complain that lumping confuses interpretation.

These complaints are logical: Retirement is a voluntary action by justices who have not yet died, but

death-in-office is an involuntary biological event that befalls justices who have not yet chosen to retire.3

Thus, using Departure-by-any-means as a dependent variable appears to confound positive and negative

mortality risks, and to conflate processes that promote and discourage retirement. Logical problems

notwithstanding, the lumping problem is well-known in statistics to be sometimes disastrous, but

sometimes benign (Ledoux, Rubino, and Sericola 1994; Cotterman and Peracchi 1992). Accordingly, we

take an empirical approach to lumping: we estimate separate models of departures, resignations, and

deaths-in-office. If these models produce similar results for retirement and death-in-office, then they

demonstrate that lumping is permissible. If they produce different results for retirement and death-in-

office, then they show that separate analyses are required.

Aggregation. In some previous research, retirements and deaths-in-office of individual justices

are aggregated into annual frequencies of these events for the Court as a whole (e.g., Wallis 1936; Callen

and Leidecker 1971; Ulmer 1982; King 1987; Hagle, 1993). These annual frequencies are informative

descriptions of historical trends that permit appropriate tests of organization-level hypotheses about the

Court as a whole. However, departure from the Court is an individual-level phenomenon determined by

the decisions and deaths of individuals. Testing hypotheses about individual-level phenomena with

aggregated frequency data would commit the “ecological fallacy” under most circumstances, and so is

unacceptable (see King 1997). Therefore, we examine individual-level data. When it is useful to consider

frequency distributions for the Court as a whole, we can use probability laws to aggregate individual-level

results into frequency distributions for the court as a whole.

Vitality. Many previous labor force studies hypothesize and find that as workers’ subjective or

4


objective health declines, their probability of retirement increases (see French 2005; Bound 1991; Dwyer

and Mitchell 1999; Parsons 1982). Virtually all previous historical narrative studies of Supreme Court

departures consider the retirement effects of vitality, or its opposite, decrepitude. In quantitative analyses,

Squire (1988) includes a measure of justices’ poor health, but Hagle (1993, 35) criticizes Squire’s vitality

measure. Zorn and van Winkle (2000: 162) complain that Squire’s measure “is so narrow as to almost

reach the certainty of leaving office.” Previous analysts despair of the possibility of obtaining adequate

health data for quantitative analysis of Supreme Court departures: Citing Greenhouse (1984), Hagle

(1993:46) asserts that Supreme Court Justices are confirmed fibbers concerning their health. Zorn and van

Winkle (2000) use justices’ productivity (written opinion production per year) to measure their health, but

productivity is obviously different from health (see Green and Baker 1991), and seemingly subject to a

wide range of political, psychological and personal influences. To the best of our knowledge, no

subsequent quantitative analysis of Supreme Court departures (e.g., Yoon 2006) includes a direct measure

of vitality. Because previous research finds that health is an important determinant of retirement in the

general working population, the absence of an accepted health indicator in previous quantitative Supreme

Court studies remains a source of uncertainty.

Ironically, it appears that a standard, useful and available health measure exists for Supreme

Court justices, but has not been used previously. This measure is the number of remaining years of life for

each justice, measured in each year of each justice’s service on the Court. Remaining years of life has

been used as an objective health indicator in a long line of previous health research (e.g. Idler and Kasl 1991; Kaplan 1987; Mossey and Shapiro 1982) and studies of health effects on retirement of

the general population (e.g. Parsons 1982; Bound 1991; Baker, Stabile and Deri 2004; French 2005).

Further, remaining years of life has been used as an objective health indicator for establishing the validity

of self-reported, subjective health measures (Davies and Ware 1981; Mossey and Shapiro 1982; Ross and Wu 1995). In addition, Han, Phillips, Ferrucci, Bandeen-Roche, Jylha, Kasper,

and Guralnik (2005: 216) find significant association between increased mortality and changes in

subjective feelings of vitality. In short, future longevity is an empirically validated indicator of

“objective” and “subjective” vitality (health). As we write, all justices (except those now serving on the

Court and Justice Sandra Day O’Connor) are now dead, making it possible to calculate the years of

remaining life for all but one former justice in each and every year of their service on the Court.

5


Multiple Indicators. Previous studies of Supreme Court departures consider various hypotheses

involving the party of the sitting President, the party of the President who appointed the justice in

question, and additional variables combining various party affiliations. Even if there is no ambiguity in

the substantive hypotheses that motivate these variables, the variables themselves sometimes appear in

several interaction terms, each with its own coefficient, standard error, and error covariance with other

included terms. Calculation of effect measures and significance tests for each variable must therefore

accommodate all terms that involve the variable.

III. METHODS

Strategy. The Politicized Departure Hypothesis asserts that, other things equal, justices are more

likely to retire from the Court, and less likely to die in office, when the incumbent President is of the

same party as the President who nominated them to the Court, and when the President is in the first two

years of a four-year Presidential term. To test these assertions, we estimate discrete-time event-history

analyses of the probabilities of retirement and death-in-office by Supreme Court justices. Compared to

continuous time methods, discrete time analyses better accommodate the rounding of retirement dates and

co-occurrence of retirements -- the tendency of justices to withhold announcement of their retirement

until the end of a term, and the simultaneous retirement of more than one justice (see Yamaguchi 1991).

Dependent variables are the hazards of retirement and death-in-office for each justice in each year that

the justice served on the Court.4 Thus, the unit of analysis in this research is the justice-year. Because

each justice-year represents data pertaining to one justice sitting on the Court for one calendar year (or

part of it), time is intrinsic to all of our analyses. All variables are permitted to vary over time. Some

independent variables in these analyses represent the hypothesized causal variables: i.e., political parties

of the current President and President who appointed the justice, for the particular justice and specific

year that defines the justice-year. Remaining independent variables in these models are controls for

justices’ age, tenure on the Court, eligibility for Federal retirement pension benefits, secular trends in

probabilities of retirement and death-in-office and, in some models, the future longevity of the justice.

Nonlinear Probability Models. Although 44.5 percent of all justices have died in office and 47.3

percent have retired from office, death-in-office occurs in 2.6 percent of the justice-years, and retirement

occurs in 2.8 percent of the justice-years.5 The proportions of justice-years ending in death-in-office and

the proportion of justice-years ending in retirement are small enough to require nonlinear probability

6


models – linear models would predict nonsensical negative probabilities of retirement and death-in-office

for some justice-years. Therefore, most of our analyses use logistic regression (logit) analysis, which

constrains probability estimates to the interval (0,1). To permit comparisons to some previous analyses,

we also use multinomial probit analysis. Our analyses are estimated on a complete enumeration of each

and every year served on the Supreme Court by each and every justice ever appointed to the Court,

through 2006.

Right Censoring. At any moment while they serve on the Supreme Court, justices are “at risk” of

both retirement and death-in-office. That is, they have nonzero hazards of both retirement and death-in-

office. These hazards can be positively correlated, but the actual events of retirement and death-in-office

are mutually exclusive: Although death and retirement occur probabilistically – and the probabilities of

these events are the subject of our analyses – justices who actually retire no longer have offices in which

to die; and justices who have died in office no longer can retire. Thus, death-in-office right censors data

on retirement, and retirement right censors data on death-in-office (Allison 1995). Right censoring of

discrete time event history data is an easily accommodated commonplace of survival analysis and event-

history analysis: we include in our analyses the justice-years of each and every justice for each and every

year in which that justice serves any time on the Supreme Court. This method makes full use of all

available information on retirement and death-in-office, avoids bias, and produces consistent estimates of

model parameters (Allison 1995).

Competing Risks. We accommodate a previous argument that retirement and death-in-office of

Supreme Court justices should be conceptualized as “competing risks” (Zorn and Van Winkle 2000).

Box-Steffensmeier and Jones (1997: 1450) observe that a competing risks conceptualization implies the

use of a multinomial logit (MNL) analysis or, if computationally tractable, multinomial probit (MNP)

analysis. In retirement analyses of the U.S. labor force, the competing risks formulation with MNL is

commonplace (e.g. Blau and Gilleskie 2001; Cahill, Giandrea and Quinn 2006). In application of the

competing risks formulation to departures from the Supreme Court, there is no censoring. Rather,

continuation-in-office, death-in-office and retirement are conceived as alternative choices. Although we

apply the competing risks model for comparison and completeness, we note that continuation-in-office

and death-in-office are not different choices, but are both results of the choice to continue in office rather

than to retire. In the case of death-in-office, that choice is followed within a year by death, which is

7


experienced but not chosen.

Clustering. Persistent, unobserved and unmeasured characteristics of justices undoubtedly affect

their annual probabilities of retirement and death-in-office. For example, justices differ in their tastes for

work, family circumstances, and so on. Thus, the observations (justice-years) pertaining to a particular

justice constitute a cluster of observations with possibly correlated errors; the variance of errors also may

vary across justices. To accommodate these possible effects of clustering, we calculate robust standard

errors and significance tests, adjusted for sample clustering (Wooldridge 2002, 57).

Independent Variable Nonlinearities. Our analyses test hypotheses about the effects of political

variables on justices’ probabilities of retirement and death-in-office, after controlling for the confounding

effects of other variables. Previous studies suggest that these control variables may have nonlinear effects

on retirement, job termination and mortality. Lest we underestimate the impact of a control variable – and

thereby overestimate the effect of the political circumstances variables that are the subject of our

hypotheses – our models permit, but do not require, control variables to have nonlinear effects. We follow

an iterative algorithm to apply log-fractional polynomial transformations of these variables, and we retain

whatever transformations most improve the fit of control variables to the data (Royston and Altman 1994;

Gilmour and Trinca 2005).

Effects of Variables. Effects of variables follow from the logistic regression model. For brevity,

we state that model for retirement only. Let Yij equal 1 if the ith justice retires in the jth year, and let Yij = 0

if he or she does not retire in that year. Let Pij equal the probability that Yij equals 1. Let Oij = Pij/(1-Pij) be

the odds that Yij equals 1. Let Xij be a vector of independent variables, Xij = (xij1, xij2, …, xijK) , measuring

hypothesized causes and control variables for the ith justice in the jth year. ij is the error. ^ indicates an

estimate. The logistic regression model is:

(1) Ln(Pij/(1-Pij)) = Ln(Oij) = Xij +ij

Exponentiating equation (1) transforms the left side from the logit to the odds, , which has

greater intuitive appeal:

(2) /(1- ) = = e X ij

For more intuitive appeal, we solve (2) for the probability, as follows:

(3) = /(1+ ) =e X ij / (1+ e X ij)

The effect of Xk on , is the difference in the probability that Y=1 that is associated with some

8


convenient difference in Xk , on average and other things equal. If convenient differences in Xk are

minutely small, and the effect is expressed as a rate of change in probability per unit change Xk , then the

effect measure is the partial derivative of the probability with respect to Xk / Xk . These partial

derivatives (or marginal effects) obtain as follows: Differentiating (3) with respect to Xk indicates that the

marginal effect of any particular Xk on the odds is a constant proportion of the odds:

(4) / X..k = k

And, dividing both sides of (5) by yields:

(5)

Differentiating (4) yields the marginal effect of any particular x .k on the probability that Yij=1 :

(6) / Xk = k (1- )

If a “convenient difference” in Xk is not minutely small, then it is necessary to calculate the change in Pij

associated with that difference in Xk.

Methodological Strategy in Sum. Our statistical analyses estimate the effects, in each year, of

independent variables on the probabilities that each justice who sits on the Supreme Court retires that

year, dies in office that year or remains in office for all of that year. These probabilities are unobservable,

so we estimate them statistically, from observations of whether each justice retires, dies in office, or does

neither, in each year in which the justice serves at least some time on the Court. To test hypotheses, we

use statistical estimation to measure the effects of independent variables on these probabilities.

Independent variables in analyses described here are: 1) an indicator of whether or not the sitting U.S.

President is of the same political party as the President who appointed the justice (hereafter SameParty);

2) an indicator of whether or not the sitting U.S. President is in the first two years of his term (hereafter

Year12); 3) the combination of SameParty and Year12 (see discussion below); and 4) various control

variables, possibly correlated with the previous three, that are known or believed to affect probabilities of

mortality or retirement in the general working population.

To accommodate differences between the structures that produce probabilities of retirement and

death-in-office, we estimate separate models of retirement and death-in-office. To recognize that actual

retirement precludes future death-in-office, and to recognize that death-in-office precludes future

retirement, our analyses specify that actual retirement censors future observations of the probability of

9


death-in-office, and actual death-in-office censors future observations of the probability of retirement. To

avoid nonsensical probability estimates that are made likely by the low mean probabilities of death-in-

office and retirement (each about 2.5 percent per justice-year), we estimate model parameters with

logistic regression. To avoid confounding SameParty and Year12 with the other known or suspected

determinants of retirement and mortality, we include as independent variables in our analyses measures

justices’ age, tenure on the Court, pension-eligibility, calendar year, and, sometimes, future years left to

live, all in each justice-year. To further accommodate nonlinearities in effects of control variables, we use

log-fractional polynomial methods to find appropriate functions of control variables. To accommodate the

effects of unobserved differences between justices that could create heteroskedasticity and correlations of

estimation errors from the same judge, we compute robust, clustering-corrected standard errors and

significance tests. To overcome the intuitively unappealing metric in which logistic regression and probit

coefficients are measured, we find it useful to translate some of those effects into changes in the odds or

probabilities of retirement and death in office. Finally, to permit comparisons of our findings to those of

previous analysts, we perform additional analyses with multinomial probit analysis, and we perform

additional analyses in which deaths-in-office and retirements are lumped together as departures from the

Court, in spite of our expressed concerns for the appropriateness of doing so.

IV. DATA

We examine data on all justices of the United States Supreme Court from 1789 through 2006.6

Summary statistics for variables used in our analyses are described briefly in Table 2. These variables and

the reasons for including them are as described more fully as follows:

Retire. A dummy variable equal to 1 for a justice-year if the corresponding justice retired,

resigned or accepted “senior status” during the corresponding year, or retired, resigned or accepted senior

status before commencing service as a Supreme Court justice during the following year. Otherwise,

Retire equals 0. Justices who accept senior status no longer hear Supreme Court cases.

Death-in-office. A dummy variable equal to 1 for a justice-year if the corresponding justice began

the year sitting on the Supreme Court, and died without retiring before commencing service during the

subsequent year. Otherwise, Death-in-office equals zero.

Calendar year, Year1788, ln(Year1788). We include calendar year to hold constant trends in the

probabilities of retirement and death-in-office, as well as trends in the causes of retirement and death-in-

10


office. Subtracting 1788 from calendar year preserves all information and avoids rounding problems that

occurred in initial analyses with STATA version 8 that used calendar year. Year1788 is the calendar year

minus 1788. ln(Year1788) is the natural logarithm of Year1788; the logarithmic transformation improves

the fit of some models.

Age, Age-cubed, Age-squared. Age is the age of the justice in years. Probabilities of death and

retirement increase with age. In some analyses, we find it useful to add Age-squared and Age-cubed to the

analysis, to fit nonlinear age effects.

Tenure, Tenure-cubed, Tenure-cubed x ln(Tenure). Tenure is years of service on the Supreme

Court. The annual probability of job quitting in the working population is known to change nonlinearly

with tenure, first declining as tenure increases, and then increasing with additional tenure (Stolzenberg

1989). Tenure-cubed and Tenure-cubed x ln(Tenure) prove useful transformations of tenure.

Pension Eligible. A dummy variable equal to 1 if the justice is eligible for a Federal judicial

pension in the relevant year, and equal to 0 otherwise.

Future longevity, Future longevity-squared. Future longevity indicates the justice’s remaining

years of life. Future longevity for each justice-year is the difference between the calendar year of the

justice-year and the calendar year in which the justice ultimately dies. Future longevity-squared proves to

be a useful transformation of future longevity.

Political Circumstances Variables: These variables measure political circumstances hypothesized

to affect retirement and death in office probabilities.

SameParty. A dummy variable equal to 1 if the U.S. President who appointed the justice and the U.S.

President in the current year are affiliated with the same political party.7

Year12. A dummy variable equal to 1 if the U.S. President is in the first two years of a four-year

Presidential term. Because of a consistent pattern of midterm losses of Congressional seats to the

President’s party, Presidents tend to have more control over judicial appointments in the first two years of

their terms in office than in the third and fourth years (see e.g. Campbell 1991; Calabresi and Lindgren

2006b).

SameParty*Year12. The product of SameParty and Year12. SameParty*Year12 equals 0 unless

SameParty and Year12 both equal 1.

Together, the three variables SameParty, Year12 and SameParty*Year12 indicate all possible

11


combinations of values of SameParty and Year12.

In some models, the following alternative parameterization of political circumstances simplifies

interpretation. With no loss of information, we define NoSameParty as 1 minus SameParty, and Year34

as 1 minus Year12, and we parameterize political circumstances with the variables SameParty*Year12,

SameParty*Year34 and NoSameParty*Year12.8 Together, the three variables SameParty*Year12,

SameParty*Year34 and NoSameParty*Year12 indicate all possible combinations of values of SameParty

and Year12.

V. RESULTS

Historical Trends. Figure 1 presents historical trends in the mean age of sitting Supreme Court

justices, the mean age at which these sitting justices took the oath of office, the mean age at which they

eventually leave the Court and the mean age at which they ultimately die. Age-at-oath increased in the

19th Century and declined slightly (but insignificantly) in the 20th Century. Mean age of justices, mean age

at eventual departure from the Court, and mean age at ultimate death all have increased since 1789. At

any particular year, the gap between mean age at oath and mean age of justices indicates justices’ average

tenure on the Court. At each year, the gap between mean age and mean age at retirement indicates the

mean future years of service on the Court of justices serving in that year. Note that mean tenure and age

have increased in the 20th Century as consequences of the combination of declining age at oath and

increasing age at departure from the Court.

Analyses of Retirement: Table 3 presents five logit analyses of retirement from the Supreme

Court. The unit of analysis in these analyses is the justice-year; the dependent variable is the log of the

odds that the justice retires. Analyses 1, 2, 4 and 5 parameterize President’s political party and term year

with dummy variables SameParty*Year12, SameParty*Year34 and NoSameParty*Year12. Coefficients

of these variables indicate effects of combinations of values of SameParty and Year12, relative to the

condition that both SameParty and Year12 equal zero. Analysis 3 parameterizes Presidential party and

term year with dummy variables SameParty, Year12 and SameParty*Year12.

To dispel doubts that nonlinear transformations of control variables mask something unfavorable to

our argument, Analysis 1 includes only linear forms of independent variables. In column 1, Row 10 of

Table 3, the coefficient of SameParty*Year12 is 0.9850, with a robust, cluster-corrected z-statistic of

2.09 (≤0.02, 1-tailed test, 1879 d.f.). Thus, other things equal, the odds of retirement increase by 168

12


percent (=e0.9850 - 1) when the sitting President is from the same political party as the President who

nominated the justice and a Presidential administration is in its first or second year (compared to the odds

when the sitting President is from a different political party and a Presidential administration is in its third

or fourth year of a four-year cycle).

At the cost of some additional complexity, we translate this coefficient from an effect on odds to

an effect on the probability of retirement: Suppose a hypothetical justice had a retirement probability of 1

percent, the sitting President was of a party different from the party of the President who nominated the

justice; and the President was in the last two years of his four-year term (i.e., SameParty=0 and

Year12=0). Changing both SameParty and Year12 to 1 would increase the justice’s expected probability

of retirement by a multiple of slightly more than 2.6, from 1 percent to 2.6 percent, on average and other

things equal. If, in this same example, the justice’s initial probability were 10 percent, then the justice’s

expected retirement probability after the change of Presidents would increase by a multiple of about 2.3,

to 22.9 percent, on average and other things equal.

In Analysis 2 of Table 3, we use an iterative algorithm to fit log fractional polynomial

transformations of control variables. As control variables are better fitted to the data, they absorb more of

the variance in the dependent variable, leaving less variance to be explained by the political variables that

we hypothesize to affect the timing of retirement decisions and deaths in office. After estimating and

comparing 42 different sets of nonlinear transformations of calendar year, justice’s age, and justice’s

tenure on the Supreme Court, we replace Year1788 with ln(Year1788) and we replace Tenure with

Tenure-cubed and the product of Tenure-cubed and ln(Tenure). These transformations raise the pseudo-R2

from 0.1254 in Analysis 1 to 0.1581 in Analysis 2.

The coefficient for SameParty*Year12 in Analysis 2 is 0.9555 (significant, ≤0.02, 1-tailed test).

Thus, if both SameParty and Year12 changed from 0 to 1 (thereby also changing SameParty*Year12 from

0 to 1 and leaving all other political circumstances indicators unchanged), then the expected odds of

Retirement would increase by a multiple of 2.6. In short, effects of political circumstance variables found

in Analysis 2 are virtually the same as the effects obtained in Analysis 1 and discussed above.

In Analysis 3 of Table 3, we reparameterize the variables representing Presidential party and term

year. Rather than using three dummy variables to represent four combinations of these conditions, we use

one dummy for each condition (SameParty, Year12), and one dummy for the product of the two

13


(SameParty*Year12). The effect of having both conditions present is the sum of the coefficients for all

three dummies. The sum of those three coefficients is 0.955, which is identical to the coefficient for

SameParty*Year12 in analysis 2. Note that all other results for Analysis 3 are precisely identical to those

of Analysis 2.

Pension Benefits and Retirement. Coefficients of the pension benefits eligibility indicator range

from 2.0904 in Analysis 1 to 2.4514 in Analysis 5. The smallest of these estimates indicates that, other

things equal, pension benefit eligibility changes the odds of retirement in a justice-year by a multiple of

8.09. In the metric of probabilities, these effects are as follows: Other things equal, if a justice without

pension benefit eligibility had a retirement probability of 1 percent, then the addition of pension benefits,

would increase that probability to 7.55 percent, on average. If that justice had a 5 percent probability of

retirement before receiving benefits, then the addition of pension benefits eligibility would increase the

probability to 29.9 percent. Effects of this magnitude are remarkable by social science standards and need

no embellishment.

Tenure and Retirement. To interpret coefficients of Tenure-cubed and Tenure-cubed*ln(Tenure),

we use them to calculate the effect of an additional year of Supreme Court tenure on the odds of

retirement, other things equal. These effects are stated as proportional changes in odds, and are calculated

from Analysis 4 and graphed below in Figure 2(a). That Figure shows that these effects are not linear.

Figure 2(a) shows that for those with one year of service on the Supreme Court, the effect of an additional

year of tenure on the odds of retiring is negative but nearly zero. The effect of an additional year of tenure

becomes more strongly negative through the first 15 years on the Court. At 15 years of tenure, an

additional year of tenure decreases the odds of retirement by 12.5 percent. Thereafter, the retirement-

inhibiting effect of an added year of tenure decreases annually. At 25 years of service, tenure effects

become positive. At 28 years of tenure, an additional year would increase the odds of retirement by 11.2

percent. At 29 years, the increase would be 15.8 percent.

Age and Retirement. In Analysis 4, the coefficient of Age is .0548 (significant, ≤0.05, one-tailed

test). Thus, on average and other things equal, a difference of one year of age is associated with a 5.5

percent increase in the odds of retirement. In Analyses 2 and 3, which do not hold constant Future

longevity, the odds of retirement increase a rate of about 6.5 percent per additional year of age.

Future Longevity and Retirement. In Analysis 4 of Table 3, we add control variables based on the

14


years that a justice has to live: Future longevity-squared and Future longevity-squared x ln (future

longevity). Each future longevity term is statistically significant by itself (each ≤0.01, two-tailed test);

both terms are jointly significant ((2 d.f.)=7.88, ≤0.02, 2-tailed). Notice that controlling for Future

longevity does not substantially modify the estimated effect of SameParty*Year12 on the probability of

retirement. The coefficient of SameParty*Year12 in Analysis 4 is 0.9706, which trivially higher than the

estimate obtained in Analysis 2, trivially lower than the estimate obtained in Analysis 1, and not

significantly different from either at any meaningful level.

1 Political party identification is a meaningful but crude dimension of political orientation. Obviously,

justices and political parties may change their political orientations over time. Further, political views

vary among supporters of the same political party. Nonetheless, political party identification as described

here is a meaningful indicator of political orientation to researchers who have investigated the Politicized

Departure Hypothesis, beginning at least with Fairman (1938) and continuing as recently as Yoon (2006).

2 In related analyses, Nixon and Haskin (2000); Barrow, Gryski, and Zuk (1996); Spriggs and Wahlbeck

(1995); Barrow and Zuk (1990) and others examine departures from Federal district and appellate courts.

3 Suicide appears to be unprecedented among Supreme Court justices.

4 The hazard of an event is the probability that the event occurs to an individual in a period of time, given

that the individual has survived to the start of that period. The probabilities of retirement and death-in-

office examined here are hazards. In a continuous time framework, the hazard is the probability that the

event occurs per unit time, in the limit as time approaches zero, subject to the condition that the event has

not occurred before the start of the time period in question.

5 From the year 1789 through June, 2006, 110 Supreme Court justices had served all or part of 1,895

justice-years. 52 of these justices retired alive, 49 died in office, and 9 currently-serving justices had

neither. Two justices were appointed twice to the Court and retired from it twice.

6 We started with database kindly supplied by Professor Albert Yoon (see Yoon 2006), based on

information he obtained from the Administrative Office of the U.S. Courts (Federal Judicial Center 2006).

We checked some of those data against various sources, corrected errors and added more data from the

15


Because future longevity is known only for dead justices, Analysis 4 is not estimated for justice-

years of the nine living justices and one living former justice. Thus, Analysis 4 is computed over 161

fewer justice-years than Analyses 1, 2 and 3. To consider the hypothesis that results in Analysis 4 are

substantially affected by the loss of these 161 justice-years, we re-estimate as Analysis 5 the same model

that we estimated as Analysis 2, after excluding the 161 justice-years that are dropped to permit

computation of Analysis 4. If Analysis 5 produces estimates similar to those that we produced as Analysis

2, then there is evidence that removal of justice-years for surviving justices has not biased results reported

as Analysis 4. Looking at Analysis 5, notice coefficients are not meaningfully different from those in

Analysis 2. In particular, coefficients and z-statistics of Political Circumstances Indicators are virtually

identical in Analyses 2 and 5.

To interpret coefficients of Future longevity-squared and Future longevity-squared*ln(Future

longevity), we use them to calculate the proportional effect of a one-year increase in Future longevity on

the odds of retirement. We calculate these effects from Analysis 4 and graph them in Figure 2(b).

Looking at Figure 2(b) notice that if justices with less than 22 years of remaining life were granted

another year to live, then their odds of retirement would decrease, other things equal. That decrease is

Federal Judicial Center (2006) and the U.S. Supreme Court (2006) for the 1789-1868 and the 2003- 2006

periods.

7 As is conventional, Whigs, Federalists and Republicans are coded as if members of the same party.8 Other variables in the second parameterization are linear transformations of variables in the first

parameterization, as follows:

SameParty*Year34 = SameParty – SameParty*Year12

NoSameParty*Year12 = Year12 – SameParty*Year12

Thus, it makes no difference which parameterization is used. In any particular analysis, we report results

with the parameterization that simplifies the calculation and presentation of effects and significance tests.

For the unconvinced, Table 3 includes two renditions of the same analysis, each done with a different

parameterization; all results are absolutely identical. A footnote to that table shows how the results of

either analysis can be translated algebraically to the results of the other.

16


weakest for justices with 21 years remaining to live, for whom the odds of retirement are reduced by 0.5

percent, on average, other things equal; the decrease is strongest for those with 8 years of remaining life,

for whom an additional year of remaining life would decrease the odds of retirement by 7.5 percent, on

average, other things equal. These results are consistent with the interpretation of Future longevity as an

indicator of health – health declines more rapidly at the end of life than in middle age.

Secular Trend in Retirement Probability. Analysis 4 reports a coefficient of -.4951 for

ln(Calendar year – 1788). Thus, on average and other things equal, annual odds of retirement of Supreme

Court justices change at a constant elasticity of about one-half percentage point of decrease in the odds of

retirement for each percentage point of increase in the number of calendar years since 1788. Figure 2(c)

states these effects as the impact of 10-year increases in the calendar year. In 1789, an additional decade

would lower the annual odds of retirement by 69.5 percent. In 1879, an additional decade would bring a

5.0 percent reduction. In 2005, the effect of an additional decade was a 2.2 percent reduction in the odds

of retirement.

Analyses of Death-in-Office. Because death-in-office occurs when a justice both chooses to not

retire and dies, we expect negative effects on the probability of death-in-office from independent

variables that promote retirement, and positive effects from independent variables that mark elevated

mortality risk. If an independent variable increases the probabilities of both retirement and mortality, then

its negative effects (via retirement) and its positive effects (via mortality) would offset each other to some

degree, depending on the strength of each effect and the association between the probabilities of

retirement and mortality. Some of this offsetting seems likely for Age, Tenure and Pension eligibility.

However, we know of no suggestion from any source that Presidential party affiliation or the timing of

Presidential elections could alter a Supreme Court justice’s mortality risk – justices do not live longer just

because the President is a Republican (or a Democrat). So we expect negative effects on the annual

probability of death-in-office of the political circumstances indicators that show positive effects on the

annual probability of retirement.

Table 4 presents two logit analyses of the hazard of death-in-office. These analyses are analogous

to analyses of retirement probability shown in Table 3, except for two differences: First, we search for but

do not find nonlinear transformations of control variables that improve their fit to the hazard of death-in-

office. Second, we cannot use Future longevity as a control variable in analyses of death-in-office,

17


because future longevity is conceptually and empirically confounded with mortality. That is, death occurs

in the current year if and only if future longevity is zero. Consequently, models of death-in-office are

computationally intractable when future longevity is included among the independent variables.

Political Circumstances and Death in Office. In Table 4, Analysis 1 follows the parameterization

of political circumstances used in most of the Table 3 retirement analyses. The coefficient of

SameParty*Year34 is -1.1026 (significant, z = 2.30, <0.025, one-tailed). Thus, a change in value of

SameParty*Year34 from 0 to 1 is associated with a reduction by about two-thirds (2/3 ≈ 1 - e-1.1026) of the

odds of dying in office. (The reference category for this effect is NoSameParty*Year34=1.) Other results

are less clear in this parameterization: The coefficient of SameParty*Year12 is not significantly different

from zero (z=1.03), but it is also not significantly different from the coefficient of SameParty*Year34

(≤0.05, two-tailed).

Because we examine data on the entire universe of Supreme Court justices, it is at least arguable

that significance tests can be ignored in these analyses. If coefficients are interpreted without significance

tests, then Analysis 1 leads to the following conclusion: Other things equal, the coefficient of -.4229 for

SameParty*Year12 indicates that the odds of death-in-office for a justice in a year are reduced by about

one-third (.34=1- e-0.4229) if the sitting President is of the same party as the President who first nominated

the justice to the Court, and the sitting President is in the first two years of his term. The coefficient of -

1.1026 for SameParty*Year34 indicates that if the sitting President is instead in the third or fourth year of

his term, then this reduction in the odds of death-in-office doubles. But, by itself, the number of years left

in the sitting President’s term has virtually no effect on the probability of death-in-office.

Without abandoning significance testing, Analysis 2 clarifies political circumstances effects by

reparameterizing. In Analysis 2, the reference category is changed to SameParty*Year34 = 1. The

coefficient of NoSameParty*Year12 is 1.0824 (significant, z = 2.35, <0.01, one-tailed). The coefficient

for NoSameParty*Year34 is 1.1025, significant (z = 2.30, <0.015, one-tailed), almost identical to the

coefficient of NoSameParty*Year12 and not significantly different from it (at any meaningful

significance level, 2 < .01, 1 d.f.). We must reject the null hypothesis that both of these coefficients are

zero ( 2 =6.57, 2 d.f., ≤0.05, two-tailed). In contrast, the coefficient for SameParty*Year12 is not

distinguishable from zero at any meaningful significance level (i.e. insignificant, z=1.43, one- or two-

tailed test).

18


Interpretation of the political circumstances effects in Analysis 2 is straightforward: A change in

value of either NoSameParty*Year12 or NoSameParty*Year34 from 0 to 1 approximately triples the odds

of dying in office (3≈2.95 = e1.0824; 3≈3.01 = e1.1025) , other things equal. Thus, the odds of death in

office are about three times higher when the incumbent President is not of the same party as the President

who appointed the justice (compared to when the incumbent President is of the same party, other things

equal), but there is no statistically significant effect of Presidential term year on death in office.

These coefficients can be stated as probability effects: Other things equal, if a hypothetical justice

had a 1 percent probability of dying in office in a given year, then a change of NoSameParty*Year12

from 0 to 1 would increase that probability to 2.9 percent. If the hypothetical justice had an initial 10

percent probability of dying in office, then a change of NoSameParty*Year12 from 0 to 1 would increase

that probability to 24.7 percent. Effects would be nearly identical for changes in NoSameParty*Year34.

Effects of Control Variables on Death-in-Office. The odds of death-in-office decline at a rate of

1.17 percent per calendar year, on average, other things equal. Over a decade’s time, these annual

reductions cumulate to an 11 percent reduction in the odds of death-in-office. In contrast to our analysis

of retirement, nonlinear transformations of Year1788 do not improve its fit to death-in-office data. Thus,

the secular decline in these odds continues unabated. Other things equal, as time passes, death-in-office is

increasingly improbable. Each additional year of Age increases the odds of death-in-office by about 6.5

percent (the coefficient is 0.0646; significant, <.001, one-tailed), on average and other things equal.

However, the effects of Tenure and Pension Eligibility are not distinguishable from zero (i.e., not

significant, <.05, one-tailed). The absence of Tenure and Pension Eligibility effects stand in sharp

contrast to our findings for retirement hazard.

Analyses of Departure by Any Means. Table 5 presents two logit analyses of the odds of

Departure versus continuing in office. Again, the unit of analysis is the justice-year. Censoring cannot

occur in this analysis. Future Longevity is not included as an independent variable, because it is

confounded with death-in-office, a component of Departure. Other independent variables are those that

appeared in our analyses of retirement and death-in-office. We again use iteratively-fitted log-fractional-

polynomials to find transformations of control variables that maximize their fit to the data. Results for

control variables are discussed below, following discussion of political circumstance variables.

In Analysis 1, we measure political circumstances with dummy variables SameParty, Year34 and

19


their product, SameParty*Year34. Coefficients for these variables are all negative, and none is by itself

statistically significant. However, their joint effect is substantial and statistically significant: For a

particular justice-year, if the political party of the current U.S. President were the same as the political

party of the President who appointed the justice, and if the current President were in his third or fourth

term year of a Presidential term, then each of these dummy variables would equal one, and their

combined effect would be the sum of their coefficients. That sum is -0.7142 (statistically significant,

Z=2.44, <0.01, one-tailed test), and it corresponds to a 51 percent reduction in the annual odds of

vacating office, other things equal. Analysis 2 yields identical results, although it reparameterizes the

effects of Presidential party and term year with three interaction effects rather than two main effects and

one interaction: the coefficient for SameParty*Year34 in Analysis 2 is -0.7142 (statistically significant,

Z=2.44, <0.01, one-tailed test), and it corresponds to a 51 percent reduction in the annual odds of

vacating office, other things equal.

Three time-related control variables affect Departure in Table 5: First, the coefficient of

Year1788 is -.01020, indicating that the odds of vacating decline by about one percent with each

additional calendar year. Second, for justices with one year of Tenure on the Supreme Court, the effect of

an additional year of Tenure is a 10.0 percent reduction in the odds of Departure. Thereafter, the effect of

one additional year of Tenure increases by 0.6329 percent per year. After justices have accumulated 17

years of Tenure on the Court, the effect of an additional year of Tenure becomes positive. At 35 years of

Tenure, an additional year of Tenure increases the odds of Departure by 14.2 percent, on average and

other things equal. Third, effects of Justice’s Age are nonlinear and fitted with 1/Age-squared and Age-

cubed. As shown in Figure 2 (d) below, the effect of an additional year of Age on the odds of Departure is

highest at the very youngest ages (14.7 percent at age 35), declines to its minimum (6.8 percent at age

59), and rises thereafter, reaching 8.7 percent at age 80.

Competing Risks Analyses. Having stated above some limitations of the competing risks

conceptualization of departures from the Supreme Court, we apply it nonetheless, to permit comparison to

our own results in Tables 3 and 4. Table 6 presents results of a multinomial probit analysis of the

probability that in a given year a justice dies in office, remains in office, or retires from the Supreme

Court. Again, the unit of analysis is the justice-year. The multinomial probit yields two equations, one

predicting the probit of the hazard of Death-in-office (versus Continuation-in-office), the other predicting

20


the probit of the hazard of Retirement. The omitted reference category is Continuation-in-office. The

same independent variables must appear in all equations of the multinomial probit model, so we must

omit Future Longevity. We also add nonlinear transformations of control variables, but, because we have

no basis to evaluate transformations that improve the fit of one equation while weakening the fit of the

other, we cannot use automated algorithms to fit combinations and permutations of transformations.

Results of the Competing Risks model (Table 6) are similar to the single-equation results we

reported in Tables 3 and 4. The Table 6 retirement analysis finds a positive, statistically significant probit

coefficient of 0.5383 (significant, ≤0.03, one-tailed) for SameParty*Year12. To permit comparison of

this probit coefficient to the logit coefficient for the same variable in Table 3, we consider the same

hypothetical scenario that we applied to the logit coefficient: Suppose a hypothetical justice had a 5

percent probability of retirement in the current year, and suppose also that the President was in the last

two years of his term and of a different party than the justice (Year12 = 0, and SameParty = 0). Then, on

average, and other things equal, if Year12 and SameParty were both changed to 1 (it was the first two

years of a President of the same party as the President who first appointed the justice to the Court),

thereby making SameParty*Year12=1, the probit result indicates that the probability of retirement this

year by the hypothetical justice would rise from 5 to 13.4 percent, on average, other things equal. In the

logit analysis of Retirement in Column 2 of Table 3, the same hypothetical scenario indicates a

probability increase to 12.0 percent. Thus, the two retirement analyses estimate political climate effects

that are both statistically significant, of identical direction and similar magnitude.

Analyses of death-in-office yield much the same conclusion: In Table 6, the probit coefficient of

SameParty*Year34 is -.6053 (significant, ≤0.03, one-tailed). If a hypothetical justice has a five percent

probability of Death-in-office in the current year, and SameParty=0 and Year34 = 0, then changing

SameParty to 1 and Year34 to 1 would reduce the probability of Death-in-office from 5 percent to 1.22

percent, on average and other things equal. In Analysis 1 of Table 4, the hypothetical scenario just

described would reduce the probability of dying in office to 1.72 percent, on average and other things

equal. Again, in spite of our theoretical objections to the competing risks approach, it leads to the same

conclusions as our separate logit analyses of retirement and death in office.

VI. DISCUSSION

21


Can Retirement be lumped with Death-in-Office? Retirements and deaths in office both produce

opportunities for the President to nominate new justices to the Supreme Court. However, we argued

conceptually and logically that the hazards of retirement and death-in-office are different phenomena

produced by different processes, albeit with the same result. Our empirical findings are consistent with

those arguments. First, we find that political circumstances and control variables have different effects on

retirement hazard and the hazard of death-in-office. For examples:

SameParty reduces the hazard of death-in-office, but increases the hazard of retirement. These

SameParty effects interact with the effects of the President’s year in office, but, again, differently: If

Year12 =1, then the absolute size of the effect of SameParty on Retirement hazard increases; if

Year12 = 0 (that is, Year34=1), then the absolute size of the effect of SameParty on the death-in-

office hazard increases

Pension benefit eligibility increases the odds of retirement more than eight-fold, on average and other

things equal. But the effect of pension benefit eligibility on the odds of death-in-office is not

distinguishable from zero.

The odds of death-in-office declines since 1789. In contrast, the odds of retirement decline very

rapidly in the 18th Century, but taper off quickly, and now recede at a small annual rate.

Tenure shows no effect on the odds of death-in-office, on average and other things equal. But Tenure

shows negative effects on the odds of retirement for justices with less than 25 years of experience on

the Court, and increasing, positive effects thereafter.

In short, empirical evidence suggests different patterns of causation for the hazards of death-in-office and

retirement.

Second, our analyses also indicate that the hazard of retirement is empirically distinct from the hazard

of death-in-office. For each justice-year, we use Analyses 3 and 4 from Table 3 to calculate the estimated

the hazard of retirement.9 We use Analysis 1 from Table 4 to calculate the estimated the hazard of death-

in-office for each justice-year. Distributional characteristics of these hazards over all justice-years are

shown in Table 7. Figure 3(a) plots the logarithms of these hazards against each other and shows the

regression of ln(retirement hazard) on ln(death-in-office hazard): = -2.7441 + 0.3405 X (R2 = 0.14).

The standardized coefficient of ln(death-in-office hazard) is 0.37 (statistically significant, <.001, Robust

t=7.70, corrected for clustering, two-tailed test). For comparison, in status attainment models, the

22


standardized effect of father’s occupational status on son’s occupational status is about 0.3 and the

standardized effect of son’s schooling on his occupation is about 0.4 (Blau and Duncan 1967). Thus, the

hazards of retirement and death-in-office are substantially related, as expected, but they are no more

identical than father’s and son’s occupations, and no more identical than son’s education and son’s

occupational status.

Because the Supreme Court is a functioning organization as well as a collection of individuals, it is

useful to consider variation in the distributions of the hazards of retirement and death-in-office for sitting

justices on the Court.10 Figure 3(b) shows the time trend in the means of these hazards for sitting justices,

smoothed by Cleveland’s method of locally-weighted regression (LOWESS). Observe that mean hazards

of retirement and death-in-office tend to move in opposite directions over time, offsetting each other to

some extent. The standard deviations of these distributions also move in opposite directions over time. In

brief, trends in the mean and standard deviations of retirement hazards and death-in-office hazards for

sitting justices are well differentiated from each other, and lead to the same conclusion indicated by the

distributions of these hazards over justice-years and the effects of independent variables on these hazards:

9 Because retirement and death-in-office are disjoint events, the hazard that a justice retires or dies-in-

office is the sum of the hazards of death-in-office and retirement. To some extent, these differences in

predicted hazards are a consequence of the different coefficients just discussed. However, those

coefficients are not sufficiently identified to permit comparisons between different equations. However,

probabilities and odds that are estimated from different equations are identified and can be compared.

Finally, comparisons of individual coefficients leaves open the possibility that the effect of one

independent variable offsets the effect of another; unless otherwise indicated, the estimated probabilities

and odds reported here include effects of all independent variables at once.

10 Descriptive statistics for the annual estimated hazards of retirement, death-in-office, and either

outcome, 1789-2006 are:

Hazard ofRetirementHazard ofDeath-in-OfficeHazard of EitherMean0.03030.02590.0562Standard Deviation0.04680.03060.0615Minimum0.00070.00060.0028Maximum0.47430.25530.5873n (justice-years)189518951895

23


the hazards of death-in-office is empirically different from the hazard of retirement, and hypotheses about

the causes each should be tested with methods and data that permit that difference to manifest itself.

How Big are the Effects of Political Circumstances? We find the variable SameParty*Term12

multiplies the odds of retirement by a factor of more than 2.6. Although it seems unreasonable to call this

effect anything less than “big,” or even “very big,” if the hazards of retirement are very small, they will

still be small after being doubled or tripled. In fact, the average hazard of retirement is less than three

percent. Yet those hazards are sufficient to have dispensed with 103 justices of the Supreme Court in the

218 years from 1789 to 2006. It seems appropriate to consider the size of these annual probabilities in

relation to the likelihood of retirement and death-in-office over the four or eight year periods that

constitute the minimum and maximum lengths of a President’s tenancy in office. When integrated over

the interval of one or two Presidential terms, seemingly-small hazards of retirement and death-in-office

become substantial. For example, over an eight year period, a 1 percent annual hazard of retirement

cumulates to 7.7 percent – about one chance in 13 – a 3.53 percent annual hazard cumulates to 25

percent, and an 8.3 percent cumulates to 50 percent.11 Further yet, these cumulative hazards also cumulate

over the nine justices who have served on the Supreme Court at the same time,12 with each justice

separately subject to the hazards of retirement and death-in-office. When integrated over all nine justices

and the four or eight years of a presidency, the hazard of retirement is not small at all, and the

consequence of doubling or tripling that hazard seems to be appropriately described as a “big” effect.

Analyses of death-in-office lead to similar conclusions: When SameParty equals 1, the odds of death-in-

office are approximately tripled. Again, it seems appropriate to describe these effects as “big.”

Why is retirement behavior consistent with the Political Departure Hypothesis? The Politicized

Departure Hypothesis grows out of commonplace thinking that justices make rational decisions that

further their own wishes to serve the political party of the President who appointed them. However, the

research reported here includes no findings whatsoever on political wishes of justices, their perceptions of

political circumstances, or the rationality of their retirement decisions. Without such findings, we can

11 If the hazard of retirement in a year is p, then the probability of retirement in an n-year period is

1 – (1-p)n.

12 The number of Supreme Court seats has fluctuated, but was fixed at nine seats in 1869 (some of which

have been vacant from time to time).

24


report evidence in favor of the Politicized Departure Hypothesis, but we cannot dismiss alternative

explanations of Politicized Departure. For example, perhaps justices’ departures are politicized by the

application of external social pressure (rather than their own wishes) to repay their political debt to the

party of the President that nominated them to the Court. Alternatively, perhaps social pressure for

politicized retirement is not actually applied to justices. Perhaps instead justices anticipate that negative

social consequences would result from a retirement that denies political advantage to the party of the

President who appointed them. It seems that new data might be required to choose among these

alternative explanations. Because all but one former justice is now dead, the necessary data might well be

impossible to gather.

VII. CONCLUSION

Research on departures of justices from the Supreme Court spans seven decades and several

academic disciplines. Previous analysts are highly critical of each other’s conclusions and research

methods. One might despair of such dissensus, but we find it a useful guide to methodological and

substantive issues that must be accommodated to appropriately formulate and test the implications of the

politicized departure hypothesis. The analyses presented here attempt to accommodate the mutual

criticisms of earlier scholars of each others’ work on politicized departure from the Supreme Court. We

also attempt to evaluate those criticisms by comparing systematically a number of different methods that

have been used to study the Politicized Departure hypothesis. Very briefly, our analyses provide reasons

to agree that it is probably an error to lump together deaths in office and retirements, and it is probably an

error to use aggregate data on departure rates (i.e., the number of departures per year) to test hypotheses

about individual justices. However, we did not find evidence to support the claim that omission of a

vitality measure does fatal damage to estimates of the effects of political circumstances. Finally, we find

that length of tenure on the Court, future longevity, and calendar year all have markedly curvilinear

effects on retirement probability. Even if ignoring those nonlinearities would not alter our conclusions

about the effects of political circumstances, they are of substantive interest in their own right, and have

not been considered in previous research. In brief, if time matters for departures from the court, it seems

to matter in opposite ways, depending on how much time has passed already.

Finally, the main substantive news here is that our results are consistent with the Politicized

Departure hypothesis: In particular, we consider the annual probabilities of resignation from office and

25


death in office of each and every justice who served on the U.S. Supreme Court from 1789 through 2007.

Other things equal, and on average, we find that

If the incumbent President is of the same as the party of the President who nominated a Supreme

Court justice to the Court, and if the incumbent President is in the first two years of a four-year

Presidential term, then the annual resignation probability is about 2.6 times higher than at times

when these two conditions are not present. And,

If the incumbent President is not of the same as the party of the President who nominated a Supreme

Court justice to the Court, then the odds of death-in-office in any given year are approximately three

times higher than at times when the this condition is present.

In closing, we observe that although the Supreme Court has a uniquely important place in

American government, the behavior of Supreme Court justices may inform more general thinking about

the behavior of persons who serve on other governmental and nongovernmental boards and courts.

Further, research on other boards and courts may also inform thinking about the behavior of Supreme

Court justices. For example, Mace’s (1986) consideration of corporate boards of directors describes

behavior much like that predicted by the Politicized Departure Hypothesis. It remains to be seen if there is

much to be gained from connecting research on the Supreme Court to studies of other organizations.

However, connecting these different lines of research is certainly possible, and it seems a fruitful

direction for future studies. In the mean time, the analyses described here provide considerable empirical

support for the belief that the Politicized Departure Hypothesis does indeed describe the behavior of U.S.

Supreme Court justices, whether that hypothesis is uniquely applicable to the Supreme Court or more

generally descriptive of behavior of persons who are appointed to service on other organizations.

26


REFERENCES

Abraham, Henry J. 1999. Justices and Presidents and Senators: A History of the U.S. Supreme Court

Appointments from Washington to Clinton (revised ed.). New York: Oxford University Press.

Allison, Paul D. 1995. Survival Analysis Using the SAS System: A Practical Guide. Cary, North Carolina:

SAS Institute Inc.

Atkinson, David N. 1999. Leaving the Bench: Supreme Court Justices at the End. Lawrence: Kansas

University Press.

Baker, Michael; Mark Stabile and Catherine Deri. 2004 What Do Self-Reported, Objective, Measures of

Health Measure? Journal of Human Resources 39 (Fall, 4)) 1067-93.

Barrow, Deborah J., and Zuk, Gary. 1990. “An Institutional Analysis of Turnover in the Lower Federal

Courts, 1900-1987.” Journal of Politics 52: 457-476.

Barrow, Deborah J., Gryski, Gerard S., and Zuk, Gary. 1996. The Federal Judiciary and Institutional

Change. Ann Arbor, MI: University of Michigan Press.

Bazzoli, G.J. (1985): The Early Retirement Decision: New Empirical Evidence on the Influence of

Health, The Journal of Human Resources, 20 (2), pp. 214-234.

Belsley, David A. , Edwin Kuh and Roy E. Welsch. 1980. Regression Diagnostics: Identifying Influential

Data and Sources of Collinearity New York: Wiley.

Biskupic, Joan. 2004. “The Next President Could Tip High Court.” USA Today, September 29.

Blau, Peter and Otis Dudley Duncan. 1967. The American Occupational Structure. New York:

Wiley.

Blau, D, Gilleskie, DB 2001“Retiree health insurance and the labor force behavior of older men in the

1990's.” Review of Economics and Statistics, 2001, Vol. 83 pp. 64-80.

Bound, J. (1991): Self-Reported versus Objective Measures of Health in Retirement Models. The Journal

of Human Resources, 26, 1, 106-138.

Box-Steffensmeier, Janet M., and Zorn, Christopher J. W. 1998. “Duration Models and Proportional

Hazards in Political Science.” Midwest Political Science Assn. Annual Meeting, April 23-25.

Box-Steffensmeier, Janet and Bradford Jones 1997 “Time is of the essence: Event history models in

political science.” American Journal of Political Science, 41: 1414-61

Brenner, Saul. 1999. “The Myth that Justices Strategically Retire.” Social Science Journal 36: 431-439.

27


Cahill, Kevin, Michael Giandrea and Joseph Quinn 2006 “Retirement Patterns From Career

Employment” The Gerontologist Vol. 46, No. 4, 514–523.

Calabresi, Steven G., and Lindgren, James. 2006a. “Term Limits for the Supreme Court: Life Tenure

Reconsidered.” Harvard Journal of Law and Public Policy 29: 770-877.

Calabresi, Steven G., and Lindgren, James. 2006b. “The President: Lightning Rod or King?” Yale Law

Journal 115: 2611-2622.

Caldeira, Gregory A., Wright, John R., and Zorn, Christopher J. W. 1999. “Strategic Voting and

Gatekeeping in the Supreme Court.” Journal of Law, Economics and Organization 15: 549-572.

Callen, Earl and Henning Leidecker, Jr. 1971. “A Mean Life on the Supreme Court.” American Bar

Association Journal 57: 1188-1192.

Campbell, James E. 1991. “The Presidential Surge and its Midterm Decline in Congressional Elections,

1868-1988.” Journal of Politics 53:477-87.

Cannell, Charles, Peter Miller and Lois Oksenberg. 1981. “Research on Interviewing Techniques.” In

Samuel. Leinhardt (ed.), Sociological Methodology, 389–437. San Francisco: Jossey-Bass.

Cotterman, Robert and Franco Peracchi. 1992. “Classification and Aggregation: An Application to

Industrial Classification in CPS Data.” Journal of Applied Econometrics 7: 31-51.

Davies, Allison, and Ware, John. 1981. Measuring Health Perceptions in the Health Insurance

Experiment. Document R-2711-HHS. Santa Monica, CA: Rand Corporation.

Dwyer, D.S. and Mitchell, O.S. (1999): Health problems as determinants of retirement: Are self-rated

measures endogenous? Journal of Health Economics 18 (1999), p. 173-193.

Eisenhower, Donna, Nancy Mathiowetz and David Morganstein. 1991. “Recall Error: Sources and Bias

Reduction Techniques.” Pages , 127–144 in Paul P. Biemer, Robert M. Groves, Lars E. Lyberg,

Nancy A. Mathiowetz and Seymour Sudman (eds.), Measurement Errors in Surveys. New York:

Wiley.

Fairman, Charles. 1938. “The Retirement of Federal Judges.” Harvard Law Review 51: 397-443.

Farnsworth, Ward. 2005. “The Regulation Of Turnover On The Supreme Court.” University of Illinois

Law Review 2005: 407-453.

Fazio, Russell H. 1986. “How Do Attitudes Guide Human Behavior?” In Richard M. Sorrentino and E.

Tory Higgins (eds.), Handbook of Motivation and Cognition, vol. 19. New York: Guilford.

28


Federal Judicial Center. 2006. Biographical Directory of Federal Judges.

http://www.fjc.gov/public/home.nsf/hisj (last accessed April 5, 2006).

Fishbein, Martin, and Ajzen, Icek. 1975. Belief, Attitude, Intention and Behavior: An Introduction to

Theory and Research. Reading, MA: Addison-Wesley.

French, Eric. 2005 The Effects of Health, Wealth, and Wages on Labour Supply and Retirement

Behaviour. Review of Economic Studies 2005 72:2 39

Garrow, David J. 2000. “Mental Decrepitude on the U.S. Supreme Court: The Historical Case for a 28th

Amendment.” University of Chicago Law Review 67: 995.

Gilmour, Steven G., and Trinca, Luzia A. 2005. “Fractional Polynomial Response Surface Models.”

Journal of Agricultural, Biological and Environmental Statistics 10 (March): 50-60.

Goff, John S. 1960. “Old Age and the Supreme Court.” Journal of American History 4: 95-106.

Green, Gareth and Frank Baker. 1991. Work, Health and Productivity. New York: Oxford

Greenhouse, Linda. 1984. “Taking the Supreme Court’s Pulse.” New York Times. Section 1, p. 8, col. 3.

Hagle, Timothy M. 1993. “Strategic Retirements: A Political Model of Turnover on the United States

Supreme Court.” Political Behavior 15: 25-48.

Han, Beth, Phillips, Caroline, Ferrucci, Luigi, Bandeen-Roche, Karen, Jylha, Marja, Kasper, Judith, and

Guralnik, Jack M. 2005. “Change in Self-Rated Health and Mortality Among Community-Dwelling

Disabled Older Women.” Gerontologist 45: 216-221.

Hutchinson, Dennis. 1998. The Man Who Once Was Whizzer White. New York: Free Press.

Idler, Ellen, and Kasl, Stanislav. 1991. "Health Perceptions and Survival: Do Global Evaluations of

Health Status Really Predict Mortality?" Journal of Gerontology 46: 55-65.

Kalton, Graham,. and Schuman, Howard.. 1982. “The Effect of the Question on Survey Responses: A

Review.” Journal of the Royal Statistical Society 145: 42–57.

Kaplan, Sherrie. 1987. "Patient Reports of Health Status as Predictors of Physiologic Health Measures in

Chronic Disease." Journal of Chronic Disease 40 (suppl.): 27-35.

King, Gary. 1987. “Presidential Appointments to the Supreme Court, Adding Systematic Explanation to

Probabilistic Description.” American Politics Quarterly 15: 373-386.

King, Gary. 1997. A Solution to the Ecological Inference Problem: Reconstructing Individual Behavior

from Aggregate Data. Princeton, NJ: Princeton University Press.

29

http://www.ingentaconnect.com/content/asa/jabes;jsessionid=yv4mxst8wsi9.henrietta


Ledoux, James, Rubino, Gerardo, and Sericola, Bruno 1994. “Exact Aggregation of Absorbing Markov

Processes Using the Quasi-Stationary Distribution.” Journal of Applied Probability 31: 626-634.

Mace, Myles L. 1986. Directors: Myth and Reality. Revised Edition. Boston, MA: Harvard Business

School Press.

Mossey, Jana, and Shapiro, Evelyn. 1982. "Self-Rated Health: A Predictor of Mortality Among the

Elderly." American Journal of Public Health 72: 800-808.

Neumann, Richard K., Jr. 2003. “Conflicts of Interest in Bush v. Gore: Did Some Justices Vote

Illegally?” Georgetown Journal of Legal Ethics 16: 375-443.

Nixon, David C., and Haskin, J. David. 2000. “Judicial Retirement Strategies: The Judge’s Role in

Influencing Party Control of the Appellate Courts.” American Politics Quarterly 28: 458-489.

Oliver, Philip D. 1986. “Systematic Justice: A Proposed Constitutional Amendment to Establish Fixed,

Staggered Terms for Members of the United States Supreme Court.” Ohio State Law Journal 47: 799-

834.

Parsons, Donald O. (1982) The Male Labour Force Participation Decision: Health, Reported Health, and

Economic Incentives. Economica, New Series, Vol. 49, No. 193. (Feb.) 81-91.

Ross, Catherine, and Wu, Chia-ling. 1995. "The Links between Education and Health." American

Sociological Review 60: 719-45.

Royston, Patrick, and Altman, Douglas G. 1994. “Regression Using Fractional Polynomials of

Continuous Covariates.” Applied Statistics 3: 429–467.

Schmidhauser, John R. 1962. “When and Why Justices Leave the Supreme Court.” In Wilma Donahue

and Clark Tibbitts (eds.), Politics of Age, 117-34. Ann Arbor: University of Michigan Press.(1962)

Schwartz, Bernard. 1983. Super Chief: Earl Warren and his Supreme Court. New York: New York

University Press.

Schwarz, Norbert, and Knäuper, Barbel. 2000. “Cognition, Aging, and Self-Reports.” Denise Park and

Norbert Schwarz (eds.), Cognitive Aging. A Primer, 233-252. Philadelphia: Psychology Press.

Spriggs II, James F., and Wahlbeck, Paul J. 1995. “Calling It Quits: Strategic Retirement on the Federal

Courts of Appeals, 1893-1991.” Political Research Quarterly 48: 573-597.

Squire, Peverill. 1988. “Politics and Personal Factors in Retirement from the United States Supreme

Court.” Political Behavior 10: 180-190.

30


Stolzenberg, Ross M. 1989. "Job Quits in Empirical and Theoretical Perspective." In ed. Arne. L.

Kalleberg (ed.),. Research in Social Stratification and Mobility, 99-130. Greenwich, CT: JAI Press.

Sudman, Seymour and Bradburn, Norman M. 1982. Asking Questions: A Practical Guide to

Questionnaire Design. London: Jossey-Bass.

Taubman, Paul, and Rosen, Sherwin. 1982. "Economic Aspects of Health." In Victor R. Fuchs (ed.),

Economic Aspects of Health, 121-40. Chicago: University of Chicago Press.

Ulmer, S. Sidney. 1982. “Supreme Court Appointments as a Poisson Distribution.” American Journal of

Political Science. 26: 113-116.

United States Bureau of the Census. 2002. Current Population Survey. Technical Paper 63RV. Design

and Methodology. Washington, D.C.: U.S. Government Printing Office.

United States Constitution. Article II. 1789.

United States Supreme Court. 2006. Members of the Supreme Court (1789 to Present).

http://www.supremecourtus.gov/about/members.pdf (last accessed April 5, 2006).

Van Tassel, Field. 1993. “Resignations and Removals: A History of Federal Judicial Service—and

Disservice—1789-1992.” University of Pennsylvania. Law Review 142: 333.

Walker, Thomas G., Segal, Jeffrey A., Spaeth, Harold J., and Epstein, Lee (eds). 2006. Supreme Court

Compendium: Data, Decisions, And Developments (3d ed.).

Walker, Thomas G., Segal, Jeffrey A., Spaeth, Harold J., and Epstein, Lee (eds). 1996. Supreme Court

Compendium: Data, Decisions, And Developments (2d ed.).

Wallis, W. Allen. 1936. “The Poisson Distribution and the Supreme Court.” Journal of the American

Statistical Association 31: 376-80.

Ward, Artemus. 2003. Deciding to Leave: The Politics of Retirement from the United States Supreme

Court. Albany, NY: SUNY Press.

Ward, Artemus. 2004. “How One Mistake Leads to Another: On the Importance of

Verification/Replication,” Political Analysis 12: 199-200.

White, G. Edward. 1982. Earl Warren: A Public Life. New York: Oxford University Press.

Woodward, Bob and Scott Armstrong. 1979. The Brethren: Inside the Supreme Court. New York: Simon

and Schuster.

Wooldridge, Jeffrey M. 2002. Economic Analysis of Cross Section and Panel Data. Cambridge, MA:

31


MIT Press.

Yalof, David Alistair. 1999. Pursuit of Justices: Presidential Politics and the Selection of Supreme Court

Nominees. Chicago: University of Chicago Press.

Yamaguchi, Kazuo. 1991. Event History Analysis. Newbury Park: Sage.

Yoon, Albert. 2006. “Pensions, Politics, and Judicial Tenure, An Empirical Study of Federal Judges,

1869-2002.” American Law and Economics Review 8: 143-180.

Zorn, Christopher J. W. and Van Winkle, Steven R. 2000. “A Competing Risks Model of Supreme Court

Vacancies.” Political Behavior 22: 145-166.

32


40

50

60

70

80

90Y

ears

of A

ge

1800 1850 1900 1950 2000Calendar Year

Note: Lines fitted and smoothed by Cleveland's locally weighted regression (LOWESS).

from Court and Eventual Age at Death (in order listed), vs. Calendar YearFigure 1 - Means of Sitting Justices' Age at Oath, Age, Eventual Age at Departure

33


Figure 2 – Curvilinear Effects of Time Variables Figure 2a - Proportional Effect of an Additional Year of Tenure

on Annual Odds of Retirement, by Current Tenure

-15%

-10%

-5%

0%

5%

10%

15%

20%

25%

0 5 10 15 20 25 30

Tenure in Years

Prop

ortio

nal E

ffec

t on

Ann

ual O

dds

Figure 2b - Proportional Effect of an Additional Year of Future Longevity on Odds of Retirement in Current Year, by Future Longevity

-10%

-8%

-6%

-4%

-2%

0%

2%

4%

6%

0 5 10 15 20 25

Future Longevity (Years Left to Live beyond Current Year)

Prop

ortio

nal E

ffec

t on

Ann

ual O

dds

Figure 2c - Proportional Effect of an Additional Calendar-Year Decade on Annual Odds of Retirement, by Current Calendar Year

-70%

-60%

-50%

-40%

-30%

-20%

-10%

0%

1789 1809 1829 1849 1869 1889 1909 1929 1949 1969 1989

Calendar Year

Prop

ortio

nal E

ffec

t on

Ann

ual O

dds

Figure 2d - Proportional Effect of An Additional Year of Age on Annual Odds of Vacating Court, by Current Age

5%

10%

15%

35 40 45 50 55 60 65 70 75 80

Justice's Current Age in Years

Prop

ortio

nal E

ffect

on

Odd

s of

Vac

atin

g Co

urt

34


Figure 3 – Estimated Hazards of Court Departure by Death and Retirement1

.50

.30

.20

.10

.05

.025

.01

.005

.003

.001

Haz

ard

of R

etire

men

t (lo

g sc

ale)

1.50.30.20.10.05.025.01.005.003.001Hazard of Death-in-Office (log scale)

(n=1895 justice-years)

Figure 3a - Hazard of Retirement vs. Hazard of Death-in-Office, with Regression Line

0.0

2.0

4.0

6.0

8P

roba

bilit

y

1800 1850 1900 1950 2000Current Year

by Year, by Any Means -, by Retirement ... , or by Death - - -Figure 3b - Smoothed Mean Hazard of Leaving Supreme Court in Current Year,

35


36


37


38


39


40


41


42


43

LIFE TENURE RECONSIDERED: TERM LIMITS FOR … 2007 Revision Supremes.doc · Web viewRunning Head:...

Documents

Transcript of LIFE TENURE RECONSIDERED: TERM LIMITS FOR … 2007 Revision Supremes.doc · Web viewRunning Head:...