For a National Public Radio segment on term limits for Supreme Court justices this week, I was interviewed by Margot Adler of NPR’s Justice Talking. My segment is available online at the website for the show.
The interview focused mostly on mental and physical decrepitude. The staff of Justice Talking did a terrific job of editing the segment, though in response to a question asking about Justice John Paul Stevens they edited out my very positive comments on Justice Stevens, my favorite justice.
For background, one can read my article with Steve Calabresi published in the Harvard Journal of Law & Public Policy: “Term Limits for the Supreme Court: Life Tenure Reconsidered.” After documenting the increase in tenure over time, we argue for 18-year terms on the Court, staggered to allow a new appointment every two years.
The best piece on decrepitude on the Court is by Pultizer Prize winning historian, David J. Garrow in the Fall 2000 University of Chicago Law Review: “Mental Decrepitude on the U.S. Supreme Court: The Historical Case for a 28th Amendment.” Although Garrow’s abstract is here, the full text is available only to those with subscriptions to Westlaw, Lexis, or Heinonline (scroll down to p. 995).
This is just a reformulated attempt to change the makeup of the court in a different way than Roosevelt's attempt to pack the court by adding more compliant Justices. Just a different method. Does anyone doubt that the Party currently in power in Congress might find it just a little easier to boot out a Justice whose philosophy they don't agree with?
Justices can always be removed via impeachment if they're incompetent.
Just about every time the Constitution has been diddled with, the result has been for the worse. This would be no different.
Sorry about that; you are certainly correct about the Bill of Rights. I was assuming [always dangerous, I know] that the first 10 were pretty much concurrent with the Constitution. I admit that 'just about every time' was going too far out on a limb. But the 17th was such a disaster, which resulted in Senators answering to their respective national Parties - rather than to the voters who elected them - that it should have been abolished long ago.
Since it would take a Constitutional amendment to change the term from "good behavior," I suspect it would be, at the earliest, anyone nominated by the 45th President--and would not be retroactive (See Article III, Section 1).
Professors Saikrishna Prakash and Steven Smith of the University of San Diego Law School disagree--claiming "good behavior" is not as restrictive as life tenure. Of course, since judges will interpret the meaning of the term, don't count on the Prakash/Smith view gaining much traction.
Our proposal was to make term limits prospective, not applying to sitting justices or to the President in office when the amendment would be ratified.
Trevor:
1. In our article linked above, we demolished Stras &Scott’s earlier claim that the data looked random by pointing out that running almost any shaped line to the data on Supreme Court tenure yielded significant results.
Now in their latest piece Stras and Scott admit that the data are not random, but instead publish pictures of statistical models that they use to explore the question whether the recent trend since 1970 is unusual. They seem particularly enamored of the S-curve model.
As you can see by looking at every one of their data plots of curve-fitting models, the line for every model they offer (except the one they reject) does not even go through the last six data points. And the line for the model they like best, the S-curve, appears to be below ALL of the post-1970 data points. In other words, one can’t determine whether there is a major recent uptrend in a regression line that doesn’t fit ANY of the recent cases.
Among the models they plot, the only exception is the model that OBVIOUSLY fits the post-1970 cases much better than any of the others, the cubic model, but Stras and Scott spend the last third of the article attacking that model without ever realizing what any reader can easily see by glancing at the plots they publish: the cubic model is the only model they discuss that could be used to determine the recent trend because its line actually goes through the recent cases. A plot of the residuals would make the problems with their analyses even clearer.
Because no informed scholar could possibly think that one could assess the steepness of a post-1970 line that doesn’t fit any of the post-1970 cases, the last the third of their paper is one statistical howler after another. (BTW, in a draft manuscript with the University of Chicago methodologist Ross Stolzenberg, we use much more sophisticated techniques to fit time trends. You will have to wait until later to see those figures.)
2. In our article linked in the post above, we exploded the myth that our results were driven by the periods chosen by presenting our results alternatively without any periodization at all, reporting a moving average of the last nine justices to leave the Court (e.g., Chart 2, p. 781). Not surprisingly, that chart also shows unprecedented length of tenure in recent years.
I doubt that anyone familiar with our article who has even a modicum of understanding of statistics would take very seriously the latest paper by Stras and Scott. At some point, Steve Calabresi and I will probably get around to answering critics, but most of the statistical critiques of our paper so far have been just as full of embarrassing, cringe-inducing arguments as Stas and Scott’s.
Jim Lindgren
I want to first say that we have always felt that this debate should be held on a professional level. I feel that it is quite unfortunate that Jim, though I have tremendous respect for him and his empirical prowess, has repeatedly used loaded phrases to describe our arguments, such as an "odd paper," "no informed scholar could think," "anyone with a modicum of understanding of statistics could take seriously the latest paper by Stras and Scott." Let me say again that I have tremendous respect for Jim and I have never meant him any disrespect by engaging his arguments on the merits, nor has my co-author. I will attempt to respond to his arguments directly by not using any loaded phrases below.
Now to the problems with Jim's responses. The problem is that Jim and Steve focus much, if not all, of their attention on just the past thirty-seven years and seem to ignore in large part the possibility that the most recent period may just be anomalous. It is that focus on the last period that has led them to virtually ignore the preceding 180 years. Indeed, charting the data by date of appointment rather than date of departure completely destroys the trend in the data. If a trend is robust, then changing a simple assumption should not change the trend very much, if at all, and certainly should not essentially destroy the trend altogether.
The s-curve model fits the data better, has a smaller r-squared statistic, smaller prediction intervals, and is better as a predictive model. By any objective measure, there is absolutely no doubt that the s-curve model is superior. Indeed, of all the models they produced, only one --the cubic model--shows a rapid rise in the last period, and fully eight other models (other than the s-curve) that are statistically significant do not show a significant rise in that last period. That is, nine of the ten statistically significant models look very similar to the s-curve and do not show an unprecedented and dramatic increase in the final period.
We also fully engage the lagging averages argument by showing that only for certain groups of Justices (i.e. those selected by Calabresi and Lindgren) do the moving average charts show the "unprecedented" and "dramatic" increase in tenure that they claim. I cannot do justice to our arguments in the limited space here, so I would urge you to look at footnote 59 where we engage the claim fully and directly.
Finally, I find it notable that Jim keeps referring to our claim that the data were random. As we make clear in the draft, see pages 823-24, we did not say anything of the sort. Calabresi and Lindgren repeatedly ascribe to us the null hypothesis, which according to them, posits that the data are random and that there is NO trend in tenure at all post-1970. We could not make clearer in either paper that that is not, nor has it ever been, our position. We only assert that, while there is clearly a trend, the trend is not as steep as Calabresi and Lindgren claim. Again, I cannot do justice to our response here, but I am quite proud of our paper and think it draws substantial doubt on their claim. If Jim doesn't think so, he is certainly entitled to his opinion and we will just call it a professional difference on the matter. Again, I do not want to go back and forth on this issue, especially in the comments here, so I will let the paper speak for itself and not comment any further. I do appreciate Jim's and Steve's willingness to engage us on this issue.
]we argue for 18-year terms on the Court, staggered to allow a new appointment every two years.
How do you accomplish all of those things? Say you pass an amendment along those lines, and the next year 4 justices "retire." How do each of them get 18 year terms that are staggered by 2 years?
Thanks for the response.
After posting my comment, I went off to the store and planned to tone my comment down a little bit after I got back, but since you responded so quickly, I won't go back and do so.
As for other competent people missing the glaring problem in your paper that the S-curve you prefer and plot does not even go through any of the post 1970 data points, I believe you if you assert that they indeed missed this basic problem. That doesn't, however, make it go away.
But if (as I now believe) some otherwise statistically competent people really did read your article and thought its argument plausible, then I guess we will indeed have to write a response that includes your article among others so that other readers can make a more considered judgment.
Jim Lindgren
Under our proposal, there is an 18-year 2009 slot, an 18-year 2011 slot, an 18-year 2013 slot, and so on.
The phase-in works like this:
Anyone appointed in 2010 would get the 2011 slot plus an additional year. Anyone appointed gets the next open slot, plus any years until that slot starts. In most scenarios, the extra years are fairly trivial until the last one (or two) slots are filled for the first time.
Jim Lindgren
Again, I don't want this to go back and forth, and I only answer this because I think we are still talking on different frequencies. We do not, by any stretch of the imagination, adopt the s-curve model or claim that it accounts for length of tenure. I think we make that point clear in the text on pages 819-22.
None of the models, including the cubic model, are particularly good fits to the data. We use it only as a response to your claim that the cubic model was an elegant confirmation of rendering length of tenure data in thirty-year increments. A response that the s-curve is not a good fit would not really be responsive to our critique. We already figured that out; we only claim that it does better over all of the data than the cubic model. But a model such as the cubic model that predicts, and I am not kidding here, that by 2016, Justices will serve an average of 40 years and by 2036, will serve an average of 60 years, fares no better. That is why I am not particularly persuaded by the point that three of the twelve final data points come close to the cubic model plot. Our only point is that you rely on the cubic model to support your hypothesis, and we merely point out that there is another alternative model, the s-curve model, that fits ALL of the data--not just the last few data points--better than the cubic model. But although it is unlikely that I (or Ryan) will write another response, we look forward to your continuing contributions in this area.
And in general:
e_out_of_sample <= e_sample + sqrt((h(log(2N/h)+1) -log((1-p)/4))/N)
where p is the probability with which the relation holds (e.g. 95%), N is the number of sample points and h is the VC dimension. h >= order of the model.
oh, btw this calculation assumes that the structure of the model (e.g., linear, cubic, etc) has been selected a priori. Any iteration over many different models and selection thereof on the basis of in-sample error quite dramatically weakens the statement you'll be able to make about the out-of-sample error.
"Dramatically weakens" is an understatement here, isn't it?
What's the theoretical justification for fitting a straight line rather than a different curve? Why is the assumption of linearity justified? It's better not to make unnecessary assumptions about the shape of a curve. What you need is a hypothesis about the relevance of a variable, to which you then fit a line (curved or otherwise). Does a scholar need a theoretical justification for choosing a curved line such as a logit or probit to model binary data, or is it enough to choose a curve (logit or probit) that would fit the data better than a straight line?
That the substantial and statistically significant jump in the tenure of post-1970 retirees is likely to remain at very high levels for some time to come is suggested by the fact that only one justice appointed to the Supreme Court in the last 35 years has left it (Justice O'Connor).
Our paper suggests possible reasons for the shift in behavior around 1970: the addition of a third clerk for most justices and the aftermath of the political debates over the Warren Court in the late 1960s and early 1970s.
Jim Lindgren
This assumes you would still have 40-year olds nominated to the court after such a change. But the whole purpose of sending 40-year olds to the court (when they have relatively little experience compared to others who might get nominated) is so they can serve 40 years on the court. If they can serve only 18, why wouldn't you send 60- or 65-year olds to the court, on the ground that they have more experience? And those justices would be unlikely to be looking for work when their term expired.
Under the present system, though, the appointing presidents are pretty random. Some get zero, others get multiple appointments per term. This will distort the Court's composition. Plus, there is the problem of strategic retirement, in which justices try to ensure that a likeminded president selects their successor. There's no reason the ideological composition of the Court should be subject to the manipulation of a justice.
18 years means an appointment every two years or two per President, which seems about right. Something should be done to prevent Congress from preventing an appointment due the presidential term, though, and this could be a tricky problem.
So a president should have the power to choose an ideologically similar justice, but a justice shouldn't have the power to decide when he or she retires? What's to stop a justice from retiring early?
That seems a little silly.
Another interesting problem.
What happens in the situation where a (Retired) Supreme Court Justice comes before the Supreme Court as an advocate? It may or may not be substantive, but it certainly opens the court to allegations of bias.
and before you scoff, I would point out that it's happened before (albiet the former justice often lost). John A. Campbell, a former supreme court justice,who had resigned at the start of the civil war, represented the Butchers in the Slaughterhouse cases (among others)
The post-bench problem certainly exists, though, to be considered. There's an old antitrust decision where a former justice appeared for the defendant and persuaded the Court about the meaning of his earlier opinion. Lino Graglia has a new paper on this case.
Forgive me; I'm a mathematician, and I forget that laypeople don't think of lines as special curves. Linearity is certainly not an inherently more justified assumption than one leading to a curve.
The problem is that, as Laura S. mentioned, I can always fit a post hoc curve to a fixed amount of data — I mean, you might as well draw a crazy line through all the data points and be done with it. I can generate N random values at N distinct points, then draw a perfect (N-1)-degree polynomial through my data; the regression measures will all be perfect, but my curve has no predictive power at all — after all, it's random data! It simply is not meaningful to measure the predictive power of a model using the same data that generated it.
That might seem like a purely notional objection, given that you're not fitting 100-degree polynomials to the data, but consider: I could probably add a datapoint (2020,0) to your dataset, re-run a cubic regression, and find that the R-squared hasn't dropped by all that much, and yet the model's future predictions will have changed substantially (and for the absurd). The mathematics is not a black box here.
What you need is a hypothesis about the relevance of a variable, to which you then fit a line (curved or otherwise).
No. What you need is a hypothesis for how a variable is relevant. For this case, we might hypothesize that served time roughly follows the curve as - b, where s is the lifespan of the justice: b accounts for the time spent in education and practice, while a accounts for proportional factors like retiring when one feels too old to continue (presumably at a proportion of one's life span). We can then test that hypothesis over the known data using some accepted measure of regression. Now, average life spans have obviously been increasing over the duration of the Republic, but probably not at a constant rate — I'm sure there's a healthy debate over models for this in the appropriate literature, and we'd want to briefly summarize that debate and indicate why we've chosen some particular model. With all of that, we may be able to advance a valid claim predicting further increases in serving time; we'd then need to argue that this is Unfortunate and that The Fix Is Worth The Cure, and statistics doesn't have much to say about that.
Perhaps silly is the wrong word. More my thought is that this effect is either incredibly narrow, or in different cases non-existent.
Theoretically the constitution doesn't provide for retirements of supreme court justices at all. Surely we can't argue that justices don't have a right to "stay alive" through the term of a president they don't like. Or conversely should die, simply to allow a current president to appoint. Retirement seems to be little different. If we recognize the right of a judge to retire, I think we implicitly recognize their ability to choose the timing of that retirement.
If you you think about it, there's only two possible ways a judge may "influence" future candidates in this way.
1. the judge may resign early, to avoid a successor being chosen by a subsequent ideologically adverse president.
2. the judge may remain on the bench through the term of an ideologically adverse president in the hopes of getting a more sympathetic president in the future.
But both of these have significant problems.
In the first scenario, the situation can only problematically arise in circumstances similar to those that resulted in marbury v madison. You'd need a president and a congress sympathetic to the judge's viewpoint, that both have suffered significant electoral defeats. Otherwise, a hostile senate (or even one with a minority) could filibuster it into the next term.
If it's prior to elections, you'd need immense dedication on the part of a life term judge to resign on the off chance his "party" loses, and he wouldn't make it through another term.
Further, I think term limits for judges either don't change or possibly exacerbate this first scenario. Rather than a vague "retirement date" a judge may consider resigning and giving up an additional year on the court to make sure his appointment is by an ideologically friendly president.
Unless the plan is either
A. going to in some way prevent judges from willingly stepping down.
B. mandating that spots remain open for the remainder of a president's term (thereby handicapping the court into potential 4-4ness)
It doesn't solve the early retirement problem.
But going on to the second problem. A judge remaining on the bench through an adverse president's term.
given that the Judge is appointed to a life term and is still of sound mind and body. We have an unsolvable problem. Arguing that a judge still fit to occupy his post and entitled to it for life should exercise the voluntary decision to retire based on the politics of the current president puts the arguer or the judge in no better position than a judge who would voluntarily retire.
So now we come back to the usual scenario. A judge that is no longer "fit" to fill the post, but is legally entitled to it, who remains in the hope that his successor
Of course the problem here becomes determining what "Fit" is, as the situations where it becomes political are never going to be the easy ones. Certainly a judge that commits crimes or becomes fully incapacitated is an easy choice.
But where "fitness" is intrinsically related to reasoning in cases where political policies are at stake, many people might be prone to asserting that justices on the opposite side are no longer "fit" to occupy the post.
So, unless we have an unbiased method of determining "fitness," I think we have no choice but to admit that if Judges have the right to choose the time of their retirement , even if their considerations are political.
There's still the random death/retirement problem, where Presidents get different numbers of appointments. It's certainly possible that for a period in which the Presidency was equally divided D/R, one party's presidents might get significantly more selections.
And no, direct election does not make senators more responsive to their state's voters. It makes senators less responsive. The Framers knew what they were doing when they wrote the Constitution. Direct election has only made senators more responsive, once elected, to their national political Party in Washington, DC.
The problem now is that senators are beholden to their respective Parties over the interests of the citizens of their state. You see this every day, when the senators of one Party vote along party lines against the senators of the other Party when push comes to shove, instead of representing the best interests of their own state's citizens. Even if a senator wants to vote in his/her state's best interest, if the Party leadership in Washington instructs them to toe the Party line and vote contrary to their state's best interest, with rare exceptions they will vote in accordance with their national political Party's wishes.
If you would like to do a little reading about how this came about, you will see that there was no restriction on individual states directly electing their senators. In fact, Oregon provided for direct popular election of its senators. But certain factions pushed the demand that, whether other states wanted direct elections or not [and most definately did not - or they would have provided for them like Oregon did], they were going to get them anyway - thus giving out-of-state special interest groups in Washington control over each state's senators. That was the whole idea. But eventually, direct elections came about because of conniving and rabble-rousing by W.R. Hearst and others, thus forcing direct election of senators on all states - most of which had no desire whatever for direct elections of their senators.
The 17th allowed out-of-state interests to game the system. The result is that the electorate has lost a say in issues that affect them directly. [Example: L.O.S.T. probably would have had no chance of passing the senate if senators represented their state, instead of their political Party in Washington.]
Of course, Senate leadership isn't the only reason that a Senator moves in the direction of his party. The new Senator from Colorado has moved steadily to the left as he has spent time in Wash. D.C. He ran as a "moderate" claiming the ability to work both sides of the isle, but signed on to Harry Reid's letter to Clear Channel condemning Rush Limbaugh. Definitely, not something that many voting for him would agree with.
I have used two Democrats here, but am confident that there are just as many Republicans who move in the opposite direction once elected to the Senate.
the powers that be" in that state than direct election. Allowing the party in power to choose makes them a tool of that party. It certainly doesn't make them responsive to the people, that didn't elect them in the first place.
I am not sure how that would happen.
Presumably each of the slots would begin effective July 1, which means the preceding term would end June 30, of the odd number year.
Because presidential elections are in even numbered years, and except for the term that ends immediately after a president takes office, the justice would have to guess that a newly elected president would or would not be reelected so as to retire "early."
In either event, as I understand the Lindgren proposal, the terms are set based on time, not based on person. That is, for example, justice having the term ending June 30, 2019 retires January 20, 2019 1PM EST so the new president could nominate his replacement. The replacement would serve the remainder of the retiree's term, until June 30, 2019, whereupon to continue he would have had to be confirmed for the full July 1, 2019 term.
A question for Lindgren:
How does your proposal deal with justices dying or retiring in the middle of their terms? Do you have something similar to the 22d amend criteria for presidents?
2. People do not matter or are irrelevant if you have a good system in place. SCOTUS decisions will be written by the clerks, anyway.
If a justice leaves the bench before the 15 years expires, he or she may not be re-nominated. The next justice to fill that seat would also have a 15-year term; he or she would not be filling the remainder of the previous justice's term.
Fifteen years after they have left office is plenty of time for a president to be directly affecting national decisions.
While we are limiting the justices' terms, we can also institute a minimum age requirement. We already have one for representatives (25 years), senators (30 years), vice-presidents, and presidents (both 35 years). I think a minimum age of 50 years for a Supreme Court justice is just about right. It takes a long time to master the legal issues that have sprouted up over 200+ years of constitutional history in the US. This would be under the norm for the justices appointed since 1950, whose average age at the time their term began is 53.6 years.
As to the issue of the 17th amendment, by the time it was approved it was fast becoming a moot point. The majority of states had, by that time, instituted a non-binding referendum system in which the people chose the state's senators, and the state legislators, while not beholden by law to this vote, were often faced with the threat of loss of office if they didn't vote in accordance with the people's vote. It's obvious to me that almost all states would have eventually gone over to the referenda system.
And having the people vote for senator does make them more responsive to the people's voice. This of course does not necessarily equate to the people's interests. The American people have shown time and again that they are not always right about what is in their own best interests, which is how you get things like protective tariffs, fried Twinkies, and Dane Cook.