Presidential Selection of Supreme Court Nominees: The Characteristics Approach

Transcription

1 Presidential Selection of Supreme Court Nominees: The Characteristics Approach Charles M. Cameron Department of Politics & Woodrow Wilson School of Public and International Affairs Princeton University Jonathan P. Kastellec Department of Politics Princeton University Lauren A. Mattioli Department of Political Science Boston University July 20, 2018 Abstract Despite the importance of every nomination to the Supreme Court, a unified theory that illuminates presidential selection of nominees across the modern political era remains elusive. We propose a new theory the characteristics approach that envisions nominees as bundles of characteristics, such as ideology, policy reliability, and attributes of diversity. We formalize the theory, which emphasizes the political returns to presidents from a nominee s characteristics and the costs of finding and confirming such individuals, and derive explicit presidential demand functions for these characteristics. Using newly collected data on both nominees and short-list candidates, we estimate these demand functions. They reveal some striking and under-appreciated regularities in appointment politics. In particular, the substantial increase in presidential interest in the Supreme Court s policy output and the increased availability of candidates with desired characteristics has led to significant changes in appointment politics and the composition of the Court. We thank Joshua Fischman, George Kraus, Jeff Lax, and David Lewis for helpful comments, as well as seminar participants at Columbia, Princeton, Vanderbilt, and the 2018 Political Economy and Public Law Conference at the University of Connecticut Law School. We also thank Nina Sheridan for excellent research assistance. Finally, we thank Christine Nemacheck and Daphna Le Gall for generously sharing their data with us.

2 1 Introduction What explains how presidents select nominees for the U.S. Supreme Court? Given the importance and salience of every vacancy and nomination to the Court, one might expect this decision to be well understood by political scientists. Yet, while the literature on Supreme Court nominations is extensive, the literature on presidential selection of nominees is relatively and surprisingly small. To be sure, there are notable exceptions that span different approaches to the study of nomination and confirmation politics. On the historical side, many studies provide detailed accounts of specific presidential nominee choices (Abraham 2008, Danelski 1964, Mayer and Abramson 1995, Dean 2001, Greenburg 2007, inter alia). More systematically, Yalof (2001) scours the presidential libraries to describe the vetting operations and selection procedures of nine presidents (see also Maltese 1995). In an innovative quantitative analysis, Nemacheck (2008) uses the archives to reconstruct the nominee short lists for 48 nominations. She then examines variables that predict why one individual versus another was selected off the short list; she also shows that the overall process for selecting nominees has become more systematized and professional over time. Finally, Epstein and Segal (2005) place the goals of presidents when selecting Supreme Court justices in the larger context of the overall selection of judges to the federal judiciary. This valuable work notwithstanding, a unified theory that illuminates presidential choice of nominees across the modern political era remains elusive. Given the ideological stakes in recent Supreme Court nomination and confirmation battles, a theory of raw ideological calculation would seem likely to have predictive power. Yet, as Cameron and Kastellec (2016) show, the leading candidate in the ideological camp move-the-median (MTM) theory does very poorly in predicting the ideology of Supreme Court nominees. In particular, Cameron and Kastellec (2016) document two types of mistakes that MTM theory makes. First, presidents have nominated individuals who moved the median on the Court away from the president s ideal point that is, presidents have chosen nominees who have moved the 1

3 Court in the wrong direction. Second, presidents have nominated individuals who were more extreme than predicted by the theory, given the ideology of the Senate median. These results suggest that, even if ideology plays a significant role in the majority of Supreme Court nominations, a theory of presidential selection needs to incorporate additional factors into what presidents want from their nominees. In this paper, we develop the characteristics approach to presidential selection of Supreme Court nominees. The approach formalizes presidential demand for various attributes of nominees; these attributes include not only ideology but also factors like experience and background (which informs a nominee s policy reliability), as well as traits such as race and gender that increase the diversity of the Court. More specifically, using economic demand theory as a guidepost, we derive presidential demand functions for nominee characteristics. The characteristics approach emphasizes, on one hand, the political returns to presidents from a nominee s characteristics, and, on the other hand, the costs of finding and placing on the Court individuals with desirable characteristics. We then test the theory using data on the 54 Supreme Court nominees chosen between 1930 and 2018 a period that covers 15 presidents. We also extend the theory to 299 individuals on the president s short list across these nominations. The characteristics approach requires novel measures; accordingly much of the data is newly collected. We use the data to estimate empirical demand functions for three nominee characteristics: ideology, policy reliability, and diversity traits. In other words, we estimate empirical functions to predict the characteristics of short-list candidates and selected Supreme Court nominees. The data reveal some striking and under-appreciated regularities in appointment politics. First, we show that presidential interest in the Court has significantly increased over time, while the difficulty of finding more ideologically reliable candidates has decreased. Moreover, we show that the ratio of benefits to costs generally predicts the level of ideology that presidents choose among their candidates for the Court as well as the degree of policy reliability in their eventual nominees. We also find that presidents choose more ideological nominees 2

4 when the president is more distant from the ideological center of the current Court we also illustrate how success in moving the center of the Court undercuts this incentive. We also find that the increase in diversity among both short-list candidates and nominees over time has followed an increase in presidential demand for diversity on the Court as well as an increasingly diverse pool of potential nominees. These results illustrate the utility of applying a characteristics approach to the study of presidential selection, and help us understand the importance of changes in the selection process over the last nine decades. 2 The Characteristics Approach to Nominee Selection In 1971, President Richard Nixon was handed a rare opportunity to fill a double vacancy on the Court. Attempting to come up a strategy of selecting his two nominees, Nixon held the following conservation with his Attorney General, John Mitchell: 1 Mitchell: So, then, we may have a double play here, and of course what what you do in the northern slot might very well help what you do in the southern slot, and vice versa. Nixon: Well, even then, I don t want a liberal. Mitchell: Oh no, no. Nixon: I don t want a liberal. Mitchell: Absolutely not. Nixon: I just feel so strongly about that. I mean, when I think what the busing decisions have done to the South, and what it could do with de facto busing. Mitchell: I agree. Nixon: And forced integration of housing, I just feel that if the last thing we do, we have got to be exerted. Mitchell: You would be going back on the committments that you made during your presidential campaign. Nixon: Now can I ask you this? This is just to play an awful long shot. Is there a woman, yet? That would be a hell of a thing if we could do it. 1 This conversation, which is drawn verbatim from the infamous taping system in the Nixon oval office, is relayed in Dean (2001, 50). 3

5 Mitchell: I think there is, if there s a woman that s credible within an age limit. There s a couple of them, as you know, up in years. Nixon: They re old? Mitchell: It would be very hard for [senators] to vote against a woman. Nixon: Oh, hell yes. In this conversation, Nixon provides a vivid (if coarse) illustration of the multitude of distinguishable goals that presidents have when selecting nominees. It was primarily important to Nixon that his nominees be ideologically conservative clearly on Nixon s mind was the Supreme Court s decision earlier in 1971 allowing federal judges to impose busing policies to remedy past discrimination, 2 as well as his campaign pledges in 1968 to be a law and order president. At the same time, Nixon was enthusiastic about appointing a female nominee, both because he would be the first president to do so and it would (in his estimation) reduce the likelihood of defeat in the Senate. Yet, the nominee (male or female) could not be too up in years an older nominee would be less likely to have a longer term on the Court. In addition, the nominee had to be credible presumably meaning experienced or of sufficient legal quality. Finally, Nixon also expressed concerns about how his nominees would affect the geographic balance on the Court. We zoom out from this anecdote to make the following general observation. A Supreme Court nominee is a bundle of characteristics and it is these characteristics presidents value, rather than the nominee per se. One characteristic is the policy positions the nominee is likely to take on the Court. Another is how reliable the nominee is likely to be as a justice. And yet another is the salient characteristics of diversity she possesses, such as gender, race, or ethnicity. 3 Our theory formalizes these choices over characteristics. As a formal theory, the characteristics approach to Supreme Court nominations has roots in classic papers by Becker 2 Swann v. Charlotte-Mecklenburg Board of Education, 402 U.S. 1 3 Technically, diversity can only be a feature of a group and not an individual. For convenience, we at times will abuse this distinction and refer to a nominee as embodying diversity; this should be taken to mean that a female or minority candidate would increase the diversity of the Court, given that it has always comprised a majority of white males. 4

6 (1965), Lancaster (1966), and Gorman (1980), as well as the neoclassical theory of the firm. However, we know of no other application of these ideas in political science. In the classic formulation of Lancaster (1966, 133): We assume that consumption is an activity in which goods, singly or in combination, are inputs and in which the output is a collection of characteristics. Utility or preference orderings are assumed to rank collections of characteristics and only to rank collections of goods indirectly through the characteristics that they possess. A meal (treated as a single good) possesses nutritional characteristics but it also possesses aesthetic characteristics, and different meals will possess these characteristics in different relative proportions. Furthermore, a dinner party, a combination of two goods, a meal and a social setting, may possess nutritional, aesthetic, and perhaps intellectual characteristics different from the combination obtainable from a meal and a social gathering consumed separately. We begin with the characteristics view of nominees, but then go a considerable step further: we imagine a president assembling a nominee by purchasing nominee characteristics; and then selling or consuming those characteristics for political gain. This framework allows us to derive presidential demand functions for nominee characteristics. And, it allows an explicit statement of the conditions that lead to the careful (and costly) selection of a nominee and those that lead to a casual or even haphazard choice. The theory s predictions can readily be taken to data. The characteristics theory of nominee selection is novel and may at first seem somewhat outlandish. However, the following metaphor may be helpful. In essence, we imagine the president acting like a home builder, who builds a single house for sale. The house has attributes, such as its square footage, number of bed rooms, whether it has a swimming pool and so on. Thus, the house is a bundle of attributes, or an assemblage of characteristics. In turn, each characteristic has an implicit value on the market; these implicit values are often called a characteristic s shadow price. The market value of the house is just the summation of its attributes times their shadow prices. In turn, the shadow prices, along with the costs of materials for building the attributes (the factor input prices), determine the builder s demand for the materials used to build attributes in other words, how much 5

7 of those materials he buys. Increases in the market value of an attribute, and decreases in its construction cost, bring increased demand by the builder for input materials. We apply exactly this logic to presidential selection of nominees and their attributes. Each characteristic of a nominee has a value to the president (a shadow price), and each attribute has a political or effort-cost (a factor input price). The president s demand for the attributes indicates what and how much he buys given the factor input and shadow prices. Two assumptions are important for the characteristics theory of nominee selection. First, we implicitly assume that any given assemblage or package of nominee characteristics is technically feasible, albeit costly or difficult to procure. So, for example, we have never seen a highly experienced, conservative, African-American, female Supreme Court nominee but not because no such person exists. Rather, we see no such nominee because the input price of some of the characteristics is so high that a Republican president has usually found such a nominee uneconomical to assemble. (Of course, doesn t exist and prohibitively expensive can be seen as largely semantic differences.) Second, an empirical implementation of the characteristics theory depends on finding reasonable proxies for the political return on or value of nominee attributes (the shadow prices of attributes in the housing example) and proxies for the attributes costliness (the factor input prices in the housing example). Absent such proxies, the theory is an empty formalism. We believe natural and measurable proxies exist (though of variable quality), as we indicate below. Consequently, the theory affords a new, systematic, and rigorous way for thinking about presidential selection of nominees, a perspective that leads naturally to a dialogue with data. The characteristics approach departs in many key respects from move-the-median theory. Not only is MTM-theory one in which ideology is the sole consideration, but its role is quite stark: the president only cares about moving the median of the Court as far as possible, subject to the constraint that the Senate must confirm the nominee. No room is left for presidents (or senators) to care about any other attribute of a nominee, including quality. 6

8 Our theory incorporates these considerations. In addition, unlike MTM-models, we treat the president s choice of nominee as a decision-theoretic problem, and not game-theoretic. As noted earlier, Cameron and Kastellec (2016) show that the president has been much less constrained in his choice of nominee than MTM-theories (Moraski and Shipan 1999, Krehbiel 2007) would predict. Accordingly, we focus on how the president optimizes his choice of nominee across multiple characteristics. At the same time, operationalizing the costs of purchasing additional ideology may naturally lead one to consider factors such as the size of the opposition party in the Senate. The model here can be seen as presidential calculation of the first move in a two-period game, where action in the second period is captured in the cost of ideology. Thus, even as the theory moves away from a stark ideological focus, it is it not as distant from a game-theoretic separation-of-powers approach as it might seem at first impression. 3 The Model We begin with a general formulation of the theory, before turning to a more specific version that guides empirical testing of the theory. We model a president selecting a nominee n i at a given time i. A nominee comprises three characteristics: ideology, quality, which we often treat as policy reliability, and diversity traits. Formally, a nominee n i = ( x i,, q i, ŷ i ) is an ordered triple consisting of ideology, quality and diversity traits, respectively. (The subscript i denotes a concrete nominee but otherwise subscripts denote partial derivatives, while superscripts denote classes of variables). The quantities q i, ŷ i are non-negative scalars while x i R is a point on a liberal-conservative scale; we adopt the usual convention that 0 separates liberal (left) from conservative (right). A negative purchase of ideology (x i < 0) thus indicates a liberal buy, while a positive purchase indicates a conservative buy." 4 4 One additional characteristic that might seem natural to include is the age of the nominee, given the impression that recent presidents have tended to look for younger nominees who would be expected to have longer tenures on the Court (as it turns out, empirically the mean age of nominees has changed little over time, but the standard deviation has decreased). While we recognize the theoretical importance of age, formal analysis of presidential demand for nominee longevity is complex, both because the problem is 7

9 We divide each characteristic into two components: the first is a default component that the president can purchase for free. Denote x 0 i, q 0 i, and y 0 i as the default levels of ideology, quality and diversity (these levels will vary by nomination, and thus are not constants). Denote x i, q i, and y i as the additional amount of each characteristic that the president buys by exerting effort. A nominee s total level of attributes is simply the sum of the default and purchased levels define x i = x 0 i + x i, q i = qi 0 + q i, and ŷ i = yi 0 + y i. For each characteristic, the president takes a cost versus benefit approach to choosing a nominee. Let r(.) denote the political returns ( revenues ) accruing to the president from any assemblage of nominee characteristics, while c(.) is the effort or political cost associated with any assemblage of characteristics. Let π x i, π q i, πy i denote the shadow prices of the respective nominee characteristics that is, the political returns that accrue to the president from each unit of the associated nominee characteristic. Let w x i, w q i, wy i denote the factor input prices of the respective nominee characteristics that is, the effort cost or political cost required of the president to find ( assemble ) each unit of nominee characteristic. In our model, because the president cares about influencing the policy of the Court, he takes into account the characteristics of the 8-member court at the time of a nomination (we call this the extant Court). Let p i R denote the president s most-preferred ideological output of the Supreme Court, which he uses in evaluating the nominee s impact on the Court s policy. Let x i denote the overall ideological composition of the extant Court, while y i denotes the overall diversity of the extant Court. The extant Court s ideological composition influences the president s evaluation of a nominee s impact on the Court s policy outputs while the diversity of the Court may influence the president s evaluation of the perceived impact of adding an additional non-white-male to the Court. We assume that p i > x i that is, the president is more extreme than the center of the eight-member Court. We make this assumption for ease of exposition, but as an dynamic (it requires estimating the present value of nominees) and because nominee starting age interacts with nominee qualifications (younger nominees have greater expected longevity but lesser experience and reliability). We thus choose not to include age in the present model, and leave it as an interesting extension of the characteristics approach employed here. 8

10 empirical matter this seems to be true. Bailey (2007, Figure 7), for example, shows that Democratic (Republican) presidents are almost always more liberal (conservative) than the median Supreme Court justice. We assume the president has a utility function of the form: u(n i ; πi x, π q i, πy i, p i, x i, y i, wi x, w q i, wy i ) = r(n i, πi x, π y i, p i, x i, y i ) c(n }{{} i, wi x, w q i, wy i }{{} ) Returns Costs The president s objective is to select a nominee that maximizes this function. For the purposes of tractability, we focus on a somewhat less general formulation: u( ) = r x (f(ˆx i, x i ), p i, π x i ) + r q (ˆq i, π q i ) + ry (g(ŷ i, y i ), π y i ) wx i x i w q i q i w y i y i, (1) which can also be expressed as: u( ) = r x (f(ˆx i, x i ), p i, πi x ) wi x x i + r q (ˆq }{{} i, π q i ) wq i q i + r y (g(ŷ }{{} i, y i ), π y i ) wy i y i }{{} Ideology Quality Diversity (2) Here we break the general returns function r( ) into three additively separable sub-components. r x ( ) addresses policy returns, r q ( ) addresses quality, and r y ( ) addresses the returns from diversity. This formulation involves some loss of generality because additive separability precludes complex interactions between nominee attributes. 5 But, it brings major gains in tractability. We employ the simple cost function c( ) = w x i x i + w q q i + w y y i, so the cost of assembling a nominee is just the sum of the costs associated with each characteristic. We also impose more structure on the return functions, detailed explicitly in Appendix B. Several of these assumptions are worth noting here. First, we assume the three return functions are continuous and everywhere differentiable, while r x ( ) is concave in policy output, reaching a maximum when the Court s expected output corresponds to the president s ideal 5 The theory can accommodate complex interactions so that, for example, an increase in the cost of diversity might lower the amount of ideology demanded (as perhaps occurred with President Reagan s selection of Sandra Day O Connor). However, such a structure entails a much more complex analysis and often leads to ambiguous comparative statics, which could not be estimated with statistical precision absent a great deal more data than exists (one might need hundreds or thousands of Supreme Court nominations). We thus opt for the much more straightforward additively separable structure, and leave more general formulations for future work. 9

11 point p i. (An example, considered below, is a quadratic loss function.) Second, we employ the function x J = f(ˆx i, x i ) to capture the expected ideological tenor of the Court s output, conditional on the composition of the extant court and the the ideology of the nominee. The parameter x i captures of the ideological central tendency of the Court, based on the extant Court. The nature of the function f(ˆx i, x) remains an unsettled and vexing question in the study of courts, both theoretically and empirically. 6 While the precise form of f(ˆx i, x) is not crucial for the purposes of the theory, we do make the (weak) assumption that if ˆx i > x then f(ˆx i, x) > f(x, x) that is, the appointment of a nominee who is more conservative or liberal than the central tendency of the other members makes the Court s expected output more conservative or liberal, respectively. 7 Third, the function y J = g(ŷ i, y i ) indicates the perceived level of the Court s overall diversity given the nominee (recall y i is a measure of the diverse composition of the extant the Court). We assume this function is increasing in the diversity of the nominee and the other members of the Court, but at a decreasing rate. Thus, the addition of (say) the first woman to a Court possessing none may bring a large reward, while the addition of one more woman to a Court that already has eight may bring little. 8 We wish to examine the president s demand functions for nominee ideology, nominee quality, and nominee diversity, as expressed in the following implicit functions, respectively: 6 One possibility is the median function, such that x(ˆx i, x i ) = median(ˆx i, x i ). Empirical studies of the spatial location of the Court s final opinion tend to reject the median function, however (Carrubba et al. 2012, Clark and Lauderdale 2010). With respect to nomination politics, as noted above, Cameron and Kastellec (2016) show that move-the-median theory does not do a good job at capturing nomination and confirmation politics. 7 We assume the function is continuous and differentiable with fˆxi > 0, fˆxi ˆx i 0, f xi > 0, fˆxi ˆx i 0, and fˆxix i = 0. An example of such a function, employed later, is the weighted mean x = ax, +bˆx i, a, b > We assume the function is continuous, differentiable, with gŷi > 0, gŷiŷ i 0, g yi > 0, g yi y i 0, and gŷiy i = 0. 10

12 x i = x i (π x i, π q i, πy i, p i, x i, y i, w x i, w q i, wy i ) (3) q i = q i (π x i, π q i, πy i, p i, x i, y i, w x i, w q i, wy i ) y i = y i (π x i, π q i, πy i, p i, x i, y i, w x i, w q i, wy i ) The focus of the analysis is drawing out the empirical implications of the characteristics approach. Appendix B presents a full analysis. One additional matter deserves discussion. Given a low political return on an attribute, and considerable expense in procuring it, the equilibrium level of the demanded attribute may be zero or even negative. 9 In an economic setting, a producer would just not offer a good or attribute that cannot be sold profitably. But here, the president must produce a nominee. What then? In that case, the president chooses the zero-cost attribute, the default attribute, e.g., x i = x 0 i, q i = q 0 i, or ŷ i = y 0 i. In words, the president selects attributes that require no effort to find or verify. Conceptually one might imagine the president simply taking a random draw from a pool of potential nominees. 3.1 Estimating Equations Many of the most obvious equations one might be tempted to estimate are not compatible with maximization of Equation 1, a point well-understood in demand theory (Deaton and Muellbauer 1980). To assure that the functions we estimate are indeed compatible with our theory of nominee choice, we specify a tractable form of Equation 1 and explicitly derive the associated demands for ideology, quality, and diversity. We assume the nominee s ideology is a random variable χ i with mean x i and variance 1 q i. We assume nominee quality is not valued in and of itself, but only via its indirect effect on the expected ideological output of the nominee. Note also that from this point forward, for ease of presentation we assume that the president is a conservative (i.e. Republican) president (p i > 0), and that addi- 9 A positive (i.e. conservative) purchase of ideology is a sensible purchase for a conservative president; a negative (liberal) purchase is not. Similarly, a negative (liberal) purchase of ideology is a sensible purchase for a liberal president; a positive (conservative) purchase is not. 11

13 tional ideology equates to more conservatism. Below we note some differences for liberal presidents. The president s utility function is the quadratic-log utility function: u i ( ) = π x (p i (ax i + bχ i )) 2 + π y log (cy i + dŷ i ) w x i x i w q q i w y y i (4) In words, the president evaluates the Court s policy output using a quadratic loss function. Consequently, the Court s policy output is a weighted average of the ideologies of its members, as determined by the parameters a > 0 and b > 0, where b captures how much influence the nominee has on the Court s output relative to the existing justices. If, for example, the nine justices have equal influence, a equals 8 and b equals 1. Similarly, the diversity of the 9 9 Court is determined by a weighted average of the justices, as given by the parameters c > 0 and d > 0. Again, the most natural interpretation is that c equals 8 9 and d equals 1 9. Taking expectations, the above utility function becomes: Eu i ( ) = π x (p i (ax i + b x i )) 2 b2 π x q i + π y log (cy i + dŷ i ) w x i x i w q q i w y y i (5) It is useful to divide the cases where the president does or does not purchase any attributes. If he does not, Eq. 5 reduces to: π x i ( pi (ax i + bx 0 i ) ) 2 b 2 π x i q 0 i + π y log (cy i + dŷ i ) If the president does purchase attributes, Eq. 5 reduces to: π x i ( pi (ax i + b(x 0 i + x i ) ) 2 b 2 π x i qi 0 + q + π y i log ( cy i + d(yi 0 + y i ) ) wi x x i w q i q i w y i y i i This utility function has the same form as Equation 1 and obeys all the assumptions for general utility indicated in Appendix B. Note that π q i = πx i ; as noted above, nominee quality is valued but only for its contribution to ideological reliability. 3.2 Demand for Nominee Ideology We may now straightforwardly derive presidential demand for ideology (see Appendix B.3): 12

14 where x i if x i > 0 x i = 0 if x i 0 x i = 1 b p i 1 2b 2 w x i π x i a b x i x 0 i (6) Thus, the president will only purchase additional ideology (x i ) if his equilibrium level is above zero; it not, he will opt for the default level of ideology. Thus, we can think of x i as a left-censored latent variable, where we only observe the values of x i if it exceeds a threshold. 10 The comparative statics of demand for ideology follow from simple inspection of Equation 6. First, assume x i is positive. If so, Equation 6 leads to these predictions: Presidential ideology: Demand is increasing in p i as the president becomes more conservative, he chooses a more conservative nominee. Cost of ideology: Demand for nominee ideology is decreasing in w x i, the cost of ideology. Presidential interest in the court: Demand for nominee ideology increases as the saliency to the president of the Court s ideology (π x i ) increases. The existing Court s ideology: Demand for nominee ideology decreases as the ideology of the 8-member court becomes more conservative (i.e. as x i increases); recall our assumption that p i > x i, which means that this prediction is equivalent to saying the president demands less ideology as the extant Court moves toward him. Default level of ideology: Demand decreases as the default level of ideology increases (i.e. x 0 i is more conservative). All of these hypotheses (except the first) are new to the literature on appointment politics, but they are quite intuitive within the characteristics framework. Now, suppose x i 0, x i = 0 and hence ˆx i = x 0 i ; that is, the demand for ideology is so low that a conservative president does not purchase any. In this scenario, Equation 6 provides information about when one should see the president opt for the default level of ideology. Here, the comparative statics are the opposite of those indicated in the previous 10 This structure suggests a Tobit model we return to this point below in the empirical section. 13

15 paragraph. For example, a less conservative president is more likely to choose the default level of ideology, since a lower value of p i leads to less demand for ideology. Similarly, a more conservative existing Court increases the likelihood of a conservative president opting for the default level of ideology. Several subtleties deserve mention. First, as the price of ideology goes to zero, the value of x i goes to 1p b i ax b i x 0 i. This is exactly the value of nominee ideology needed to move the Court s expected policy output to the ideal point of the president. Second, with plausible values for the weights a and b and the other parameters, x i may well exceed the value of p i, in other words, a conservative president may nominate someone whose expected ideology is even more conservative than his own, or a liberal president may nominate someone whose expected ideology is more liberal than his own. The reason is, such an extremist nominee drags the Court s expected output closer to the president s ideal point. 11 However, if the president did not care about the nominee s impact on judicial policy but only about the nominee s proximity to himself, the president would never nominate someone more extreme than himself since the same policy utility could be achieved at lower cost with a more moderate nominee. A numerical example of ideological selection The following numerical example provides intuition for how the president chooses ideology. Assume that p =.55, x 0 =1/4, x=1/2, a=8/9, and b=1/9. What would the president do? As seen in Figure 1A, the expected ideological location of the overall Court with zero buy (i.e. ax i + bx 0 i ) is at.47. If the price of 11 This feature of the model is similar to Bailey and Spitzer (2017), though the logic is slightly different. Bailey and Spitzer present a move-the-median (MTM) model, so the Court s output is given by the median justice. The largest possible movement of the median justice is constrained in a MTM framework (the median can move only as far left as the 4th most liberal justice on the old nine-member court and as right as the 4th most conservative justice on the old Court). They also assume nominees are random variables. Hence, if a conservative president overshoots with a nominee who turns out to be super-conservative, the effect is negligible since the Court s median will just shift modestly in a conservative direction. But if a conservative president undershoots with a nominee who turns out to be somewhat liberal, the effect is to move the Court median in the liberal direction. Hence, presidents have an incentive to nominate extremists, since the downside is negligible while the upside is substantial. This reasoning depends heavily on the properties of moving the median. The model we consider more closely resembles a move-the-mean (or move-the-center) game. An extremist may yield a favorable move of the Court s central tendency, but there are limits too much movement can be bad. Hence, the tendency to nominate extremists remains, but somewhat circumscribed relative to a MTM game. 14

16 A) Example of ideology purchase ax i + bx i 0 p b a b x i x i 0 0 x 0 x p 0.7 Additional ideology purchased B) Ideology against costs Cost of purchasing additional ideology Figure 1: Example of ideological selection. A) Shown is the ideal point of the president (p), the ideological location of a zero-effort nominee (x 0 ), the central tendency of the 8-member Court (x), and the expected policy output of that Court given the zero-effort nominee ax i +bx 0 i. Also shown is the location of a zero-cost nominee the nominee that would bring the Court s output to the president s ideal point. Given costly actions, the president would never nominate such an extremist, but he might nominate someone more conservative than himself. The example assumes x 0 =1/4, x=1/2, a=8/9, b=1/9, p =.55; the expected Court with zero buy is at.47, and the cost-free nominee is at.7. B) The president s purchase of additional ideology (beyond the default level) decreases as the costs of doing so increase. If the cost is too high, the president purchases no additional ideology and opts for the zero-effort ideology x 0 i. ideology were free, the president would choose a cost-free nominee at.7 (i.e. p ax b b i x 0 i ). What happens if the purchase of additional ideology is not costless? Figure 1B depicts how the president s choice of ideology changes as the cost increases, using the same simulated values as in Figure 1A. If ideology is free, the president would choose x i =.7. As the cost increases, the president s choice decreases linearly. At a certain threshold (here around.15), 15

17 the costs become too great for the president to buy ideology, and hence he does not, and instead opts for the default ideology. Liberal presidents We have focused on conservative presidents. Let us briefly consider liberal presidents i.e. pi < x 0 i and p i < x i. If so, the president purchases negative ideology, and a nominee is a random variable with mean ˆx i = x 0 i x i. Solving for the president s demand for ideology results results in: x i = 1 b p i + 1 2b 2 w x i π x i a b x i x 0 i Because in this configuration the president s purchase of ideology is negative, a comparative static of the form (say) x i p i > 0 indicates that purchased ideology becomes less negative (i.e. less liberal) as the president becomes less liberal in other words, the absolute amount of purchased ideology decreases. Conversely, a comparative static of the form (say) x 0 i x i < 0 indicates that purchased ideology becomes more negative as the baseline level of ideology available for free moves in a conservative direction. In other words, the absolute volume of purchased ideology increases. 3.3 Demand for Quality/Policy Reliability Next, we turn to the president s demand for nominee quality/policy reliability, which also is straightforward (see Appendix B.3): qi if qi > 0 q i = 0 if qi 0 where q i = b πi x w q i q 0 i (7) As with ideology, q i is a left-censored variable, where we only observe the values of q i if it exceeds a threshold. First, assume that q i is positive. If so, Equation 7 leads to these predictions: 16

18 Presidential interest in the Court s ideology: Demand for reliability increases as the Court s ideology becomes more salient to the president (i.e. as πi x increases). Cost of reliability: Demand is decreasing in w x i. Default level of reliability: Demand for reliability decreases as the level of default reliability (q 0 i ) increases. Again, all of these hypotheses are novel but intuitive within the characteristics framework. As with ideology, if the demand for reliability is too low, the president opts for its default level. The comparative statics of the threshold are the opposite of those listed above. That is, the president is more likely to opt for the default as his interest in the Court s ideology decreases, the price of quality increases, and the level of default quality increases. 3.4 Demand for Diversity Finally, we turn to the president s demand for diversity, which is given by: where yi if yi > 0 y i = 0 if yi 0 y i = πy i w y i c d y i y 0 i (8) Again, y i functions as a left-censored latent variable, which we only observe if y i meets some threshold. First, assume that y i is positive. If so, Equation 8 leads to these predictions: Presidential interest in diversity: Demand for diversity is increasing in the saliency of or presidential interest in diversity (i.e. as π y i increases). Cost of diversity: Demand is decreasing in the cost of finding a nominee who is a not a white male (w y i ). The Court s overall diversity: Demand is decreasing in the number of non-whitemale justices already on the Court (y i ). Default level of diversity: Demand for diversity decreases as the diversity of the default nominee increases (i.e. yi 0 increases). 17

19 If the demand for diversity traits is too low, the president opts for the default level of diversity available without effort (in practice, this will be a white male). The president is more likely to opt for the default when the saliency of diversity falls, the cost of diversity traits increases, the pre-existing level of diversity on the Court is higher, and the default level of diversity is higher. 4 Data and Measures We apply the theory to the 54 unique nominations (which comprise 50 unique individuals) made between 1930 and 2018, including the nomination of Brett Kavanaugh to replace Anthony Kennedy. 12 Our analyses proceed on two tracks, depending on the quantity of interest. We are, of course, interested in who the president ultimately selects to join the Court. In addition, we follow the lead of Nemacheck (2008) and examine the short-list of candidates for every vacancy. 13 In this time period, there have been 299 total short-list candidates, comprising 208 unique individuals. Analyzing the short list provides multiple benefits. First, the theory arguably applies to the composition of the short list, if the incentives that produces the short-list also apply to the final selection of the nominee. 14 Second, and related, it is an interesting question in and of itself whether the selected candidates differ on relevant dimensions from the candidates who make the short list but are not ultimately chosen (as we show below, they do). Third, for many of our analyses adding the short list provides much needed statistical power; for example, while there have only only been seven nominees who were not white males, there have been 62 women and/or minority candidates on the short lists We treat the nominations of John Roberts to become an associate justice (to replace Justice O Connor) and his subsequent nomination to be Chief Justice as separate nominations. See Appendix A.1 for complete details on the data collection. 13 Nemacheck coded the short-list for most nominations between 1930 and We have updated her data to include all nominations through 2018 see Appendix A.1 for further details. 14 One way the theory would not apply would be if the composition of the short list was strategic in part that is, if presidents placed certain names on the list not because the president was seriously considering them for the Court, but rather he was interested in placating a certain constituency. If the analyses of the short-list candidates and the nominees diverge, one might consider strategic posturing in the short lists as a possible cause. But if the two track together, the factors shaping the short lists and choice of nominee would appear similar. 15 To clarify our terminology, candidates refers to anyone on the short-list, while nominees are the people 18

20 The characteristics approach to presidential selection of Supreme Court nominees utilizes some measures that are quite familiar (at least conceptually), such as the ideology of the president. But the theory emphasizes many other variables that are novel and that have never been calculated, much less employed in statistical analyses of nominations. Novel measures include presidential interest in Supreme Court policy, the political and search costs of purchasing ideology, the policy reliability of potential justices, the cost of finding reliable nominees, the political return to the president from a diverse nominee, and the difficulty of finding a diverse nominee. Because of the sheer number of measures we present and tests we conduct, rather than presenting descriptive statistics first and then moving to regressions, we organize our analyses as follows. We begin by describing how we measure the president s interest in the Court. We then turn to the three main outcomes of interest in the theory (ideology, reliability, and diversity), summaring the relevant variables first and then presenting regression analyses. After summarizing the regressions, we use the overall results from each model to interpret changes in nomination politics substantively. 4.1 The president s policy interest in the court The model requires a measure of the president s policy interest in the court. We combine two variables to measure this interest. First, in line with recent work from the UCLA School of political parties, we treat the president as an extension of his political party and tie his interests to the Court to that of the larger party (Cohen et al. 2009, Bawn et al. 2012, Noel 2013). As part of a larger project on the judicial agenda of the parties, we collected the texts of the Republican Party and Democratic Party platforms, as well as the acceptance speeches of the presidential candidates from each party. We coded statements about the Supreme Court with respect to both particular cases and more general statements and/or promises about favored types of actually put forward by the presidents. Nominees are thus a subset of candidates, while the sets of unselected candidates" and nominees are mutually exclusive in a given nomination (some short-list candidates in earlier nomination are ultimately selected in a subsequent nomination). 19

21 appointments, including litmus tests for nominees. 16 From these statements, we counted the number of statements dedicated to the Court, and developed a straightforward aggregate index of presidential/party interest in the court. The indices for each party are displayed in Appendix Figure A-1. Early on, there was little party interest in the Court; most platforms before 1968 featured zero or one references to the Court. Since the late 1960s, there has been a significant increase in the focus of the parties on the Court. However, consistent with Grossmann and Hopkins (2016), there exists a striking asymmetry between the parties: starting in the early 1990s the Republican Party became much more focused on the Court than the Democratic Party. In 2016, Republican statements on the Court were nearly three times that of the Democrats. Presidents, of course, may have interests in the Court that differ from the larger party. And, whereas the platform and acceptance speeches are measured only at four-year intervals, there exists year-to-year variation in presidential interest. We thus employ a more granular measure of presidential interest: public rhetoric (Eshbaugh-Soha and Collins Jr 2015). We assume the president allocates more rhetoric to issues that he considers important. In particular, we assume the President displays greater interest in Supreme Court policy-making when he voluntarily allocates more of his total rhetorical agenda to Supreme Court policy, in contrast to other topics. The voluntary element is important, since often the president s rhetoric is reactive, for example, in response to questions put to him in press conferences. We collected new data measuring the share of the president s rhetoric in which he voluntarily speaks on Supreme Court policy. We explain this measure in detail in Appendix Section A.1.5; Appendix Figure A-2 depicts the measure over time. The theory posits a single measure of presidential interest in Supreme Court policy (π x i ). To accomplish this, we conducted a principal components analysis of the platform-based measure and the presidential rhetoric measure, and employ the first principal component For example, since 1980 the Republican platform has often featured a statement along the lines of We support the appointment of judges who respect traditional family values and the sanctity of innocent human life," which is a litmus test to appoint pro-life nominees to the Court. 17 Specifically, we extrapolated the platform data to cover the four year-interval (inclusive) after the re- 20

22 Presidential interest in Supreme Court Figure 2: Combined factor score of presidential Interest in Supreme Court Policy, Shaded regions indicate Democratic presidents, while the dashed (blue) line is a loess line. The solid (red) points indicate years with a nomination. The resulting variable, which we call presidential interest, is displayed in Figure 2. Shaded regions indicate Democratic presidents, while the dashed (blue) line is a loess line. The solid (red) points indicate years with a nomination. The figure shows a clear upward trend in presidential interest, but interest tends to be higher in Republican administrations in recent years. 4.2 The president s demand for ideology Recall from the president s demand for ideology (Eq. 6) that we require a measure x i, which is the additional amount of ideology purchased by the president above the free" default level, x 0 i. This, in turn, requires measures of x 0 i and ˆx i, the total level of ideology. To measure nominee ideology, we require a measure that meets two criteria. First, it must spective year (e.g. the 1932 Democratic platform is duplicated for ). This data exists for both parties. However, we only have rhetoric data from the incumbent president. We thus conducted separate analyses for Democrats and Republicans, which are based on the years in which each party is in office that is, the Democratic principal component analysis is based only on years in which there was a Democratic president, and vice versa. We then merged the variables into a single measure, based on the party of the president at the time. 21

23 exist for all short-list candidates, not just nominees. This rules out vote-based measures of justice ideology (see e.g. Martin and Quinn 2002, Bailey 2007), which have two shortcomings: they are inadequate for testing theories of presidential selection and only exist for nominees who are confirmed. Second, we must be able to place the measure on the same scale as our measure of the default level of ideology. 18 Fortunately, Nemacheck (2008) has constructed an inferential measure of ideology based on the average DW-Nominate of a candidate s copartisan members of her home state congressional delegation. (See Appendix Section A.1.3 for details). 19 We use this measure to capture ˆx. As it exists in NOMINATE space, lower (higher) values of the measure indicate that a candidate is more liberal (conservative). 20 Next, we require a measure of x 0 i, the default level of ideology. To construct this measure, we turn to the Courts of Appeals. This institution has been the stepping stone for a significant number of Supreme Court justices, and thus can serve as an effective proxy for the larger pool of potential justices. For each active judge who served on the Courts of Appeals from 1930 to 2018, we calculated the DW-NOMINATE equivalent of the Giles-Hettinger-Pepper scores, which are based on the ideology of a judge s appointing president and/or home-state senators (Giles, Hettinger and Peppers 2001) see Appendix Section A.1.4 for further details. 21 Then, for each year, we calculated the mean DW-NOMINATE score of all Courts of Appeals judges; we use this measure to capture x 0 i. The idea is that if the president were to randomly select a nominee, this would be his or her expected ideology; equivalently, the president can acquire this amount of ideology for free This constraint rules out using Segal-Cover scores, which are not vote-based but only exist for nominees. 19 While using an inferential measure is not ideal, Nemacheck (2008) shows that distance between the president and a candidate is predictive of whom the president selects from the short list, which helps demonstrate the utility of the measure. 20 For nominees, this measure of ideology correlates with the Segal-Cover scores at The GHP scores are usually measured based on the Common-Space NOMINATE scores; we map the scores into DW-NOMINATE to make them comparable with our other DW-based measures 22 A second possibility is to employ the mean score of the president s co-partisan judges" the idea being the president would randomly select among judges appointed by presidents of his own party. We do not employ this measure for two reasons, one conceptual, one practical. Conceptually, there have been notable examples in U.S. history of presidents making cross-party" appointments to the Court when they did not have much interest in the ideological output of the Court for instance, Eisenhower s appointment of William Brennan. Practically, we attempted to use the co-partisan measure of x 0 i ; when we did so, the procedure by which we calculate x i (described shortly) results in a large majority of observations being censored, which 22