« Will Germany always really best the US (and the world) in doctrinal legal scholarship? | Main | Show me plead on, plead off »

Wednesday, February 19, 2014

The myth of the trial penalty?

Every now and then, I like to spotlight some articles that unsettle the conventional wisdom, particularly in criminal law. Add this one to the file. Almost every teacher of criminal procedure is aware of the idea of the "trial penalty," which conveys the sense that defendants who exercise their right to a trial will invariably get a worse result if convicted than if they plea bargain. The leverage prosecutors have in exploiting the trial penalty dynamic was described by my friend Rich Oppel in a front page NYT story he wrote a few years back.

Comes now (or relatively recently at least) David Abrams from Penn with an article that slays the sacred cow of the trial penalty by providing, you know, data. And the data is the best kind of data because inasmuch as it's true, it is SURPRISING data. Specifically, Abrams argues that based on the study he performed (which originally appeared in JELS and now appears in a more accessible form in Duquesne Law Review), the data supports the view that in fact there's a trial discount not a trial penalty. Fascinating stuff. Abrams offers some suggestions for what might explain this surprise: possibly a salience/availability bias on the part of the lawyers who remember the long penalties imposed after dramatic trials. Regardless of what explains the conventional wisdom, the competing claims should be ventilated in virtually every crim pro adjudication course.

Since this empirical stuff is far outside my bailiwick, I wonder if those who are in the know have a view about how Abrams' research intersects with the Anderson and Heaton study in the YLJ, which argued that public defenders get better results in murder cases than court appointed defense counsel, or Bellin's critique of that YLJ study here.  Anderson and Heaton basically argue that public defenders get better results because they get their clients to plea bargain more frequently than court appointed counsel and that explains the outcome. As I recall dimly, that conclusion may have been true for the murder cases but the study didn't purport to make the claim that PDs were better across the board and maybe that's consistent with Abrams' views too. It would be odd (wouldn't it?) if comparatively fewer murder cases involve a trial penalty while the many other cases do not and in fact show a trial discount. Granted, these studies took place in different cities, etc., so I am also wondering if the various studies can be reconciled. Thoughts?

Posted by Administrators on February 19, 2014 at 11:30 AM in Blogging, Criminal Law, Dan Markel, Legal Theory | Permalink


TrackBack URL for this entry:

Listed below are links to weblogs that reference The myth of the trial penalty?:


Oh, and for those who are wondering, the "JELS for Dummies" appellation (used in my post above) was first articulated by Dan on a Criminal Procedure professor listserve, in which he also introduced the same topic. I had thought he used it here too, but alas, I was wrong. I certainly mean no offense against David or the law review that published his more accessible piece (which is very readable, for those who are interested). Were the JELS folks to publish a "For Dummies" version of all their articles on an annual basis, I would be the first person to purchase such a copy.

Posted by: Miriam Baer | Feb 20, 2014 2:21:26 PM

Re value of dismissal data: it strikes me that pleading guilty is not a one-off decision, but rather a choice that periodically recurs during the life of the case. (I recall making contact with attorneys several times throughout a case before their clients finally decided they were ready to plead guilty - granted, many of these cases were white collar cases where the defendants were out on bail and had some incentive to take their time).

So, before a motion is filed, the dismissal data might be relevant - but even then, it seems to me that the data on how often cases are dismissed will offer fairly little insight that MY case will be dismissed (or that my suppression motion will win etc). This is, I think, a variant of Alschuler's argument above, when he discusses the District of Dershowitz.

But in any event, however helpful the dismissal data might be to me before I have filed a motion (and even then, i don't think it tells me very much), it certainly is not helpful AFTER I have lost all of my motions and trial is impending. I think this is important because David's paper (and I admit - I only read the JELS-for-dummies version) speculates that perhaps defendants overemphasize the salience of trial outcomes, or that their attorneys may be bad agents who prefer quick pleas to time-intensive trials. All of this possible, but if you include this dismissal data, then you will likely include outcomes that are no longer a possibility for the defendant whose trial is about to begin. (True, a dismissal could still occur during the trial itself, but this could only be a subset of the category denominated "dismissal.") If we are going to question either the defendant's decision-making or the extent to which his attorney is acting like a faithful agent, it seems odd that we would include outcomes that, by the time the defendant pleads guilty, have become impossible.

Posted by: Miriam Baer | Feb 20, 2014 1:05:55 PM

Including dismissals makes sense if what you are measuring is not the “trial penalty” (as conceptualized in Jonathan, Ellen and my comments, as the difference between the sentence you get for a post-guilty-plea conviction versus the sentence you get for a post-trial conviction), but rather the wisdom of pleading guilty – which I think is the point of David Abrams' study. But you would need to differentiate between types of dismissals if, as it appears, these are driving the result. Al Alschuler’s response above, says that the study counted as an acquittal (i.e., a sentence of “0” for purposes of assessing the “trial penalty”) cases where the defendant fled or was extradited (perhaps including cases “resolved” through transfer to INS). That would be problematic as those are just postponements not avoidance of sentence. Other dismissals that probably shouldn’t count as evidence of the wisdom of rejecting a plea offer, would be cases where the defendant dies before trial; picks up another charge and pleads guilty to that charge, with the prosecution dismissing the original charge; cases dismissed because the defendant is deemed to lack competence to stand trial; cases “resolved” through parole revocation; or cases dismissed as a result of various diversion programs like drug court, which are analogous to accepting a (favorable) plea. Dismissals that probably should count as acquittals (i.e., post-plea negotiation, defense victories) for this purpose would primarily include cases dismissed for want of prosecution because witnesses fail to show up at trial, etc., assuming that the Cook County prosecutors did not refile those cases at a later date and dismissals that result from the prosecutor’s assessment of the evidence as insufficient.

Posted by: Jeff Bellin | Feb 20, 2014 11:06:28 AM

Timing plays a role here that seems not to have been addressed. If it is two weeks before trial and my attorney has received a final plea offer from the prosecutor, then we likely are no longer talking about the possibility of dismissal - presumably that opportunity has already passed (ie, my attorney would have filed whatever motion, etc., and it likely failed). Nor, at this point, would we want to include statistics for transferred defendants, or absconded defendants. Two weeks before trial, if I am a defendant, I know there are two and only two options: plead or take the case to trial. At that point, I would want to know, were I rational a defendant: "What are my chances of winning (in which case the sentence is "0"), and how much more badly will I do at sentencing if I lose?"

For the first question (chances of winning/losing guilt phase), I would expect my attorney to answer, not simply by reference to general statistics ("well, this jurisdiction has a 70% conviction rate...."), but by examining the particular strengths of MY case. So it may well be that 30 out of 100 murder cases in my jurisdiction were acquitted last year, but if my case involves particularly damning proof, its not clear to me why I should place any stock in that 30% acquittal rate. The acquittals may have involved a factor that is not present in my case. Accordingly, if my attorney affirms from her experience that she thinks the case against me is quite strong, I might conclude that the chances of a finding of guilt are close to, but not exactly 100% (because nothing is certain).

If that's the case, then notwithstanding other concerns (remorse, acceptance of responsibility, concern for family members, costs of trial), my choice whether to go to trial comes down to whether the judge will give me a worse sentence if I lose. (If he gives me the same sentence, then I ought to go for it and take the case to trial). And if I am assuming already that I will likely lose at the guilt phase, then it makes total sense to me to look at studies that compare plea-bargained cases only to those cases in which sentencing occurred after a jury's finding of guilt. In other words, if I ask my attorney about the discount, I very likely have already performed a separate calculation - based on my own circumstances - on whether I would win at trial. If I have decided that the likelihood of losing is certain or near-certain, then what I really want to know is how likely I will receive a worse sentence if I go to trial and lose.

By the way, I can imagine scenarios where the defendant loses and DOES benefit from going to trial (albeit in a jurisdiction that uses indeterminate sentencing and does not have many mandatory minimums). Or, to be more specific: I can imagine scenarios in which a trial judge sentences Defendant A, who went to trial and lost, to a more lenient sentence than Defendant B, who entered a guilty plea. (Yes, this assumes that the prosecutor's charges remain constant, but work with me). If that is the case, then we should worry that judges spend more time considering the equities of a given sentence post-trial, but perhaps gloss over those cases that are delivered to them on a platter through a quick disposition. None of this is to say that defendants should take their cases to trial, but it would suggest that judges who preside over guilty pleas may not be doing their jobs - if the data were to actually demonstrate such a differential.

The only way I see in proving the above scenario would be to do some sort of matched study wherein one group of cases with certain strengths and weaknesses pleaded guilty, and another set with roughly the same set of strengths and weaknesses took the case to trial and lost, but received equivalent or lesser sentences. The problem with doing this kind of study would be to find "true" matches because the variables change in so many ways (different attorneys, judges, juries, etc), although one could at least look for similar fact patterns in the evidence (e.g., forensic evidence, number and type of witnesses, nature of victim, location of crime, etc).

Thanks for throwing this one out there, Dan. Interesting topic!

Posted by: Miriam Baer | Feb 20, 2014 11:02:46 AM

Human Rights Watch has an interesting report titled, An Offer You Can't Refuse - How US Federal Prosecutors Force Drug Defendants to Plead Guilty - http://www.hrw.org/reports/2013/12/05/offer-you-can-t-refuse . Page 2 of this report notes statistics from the US Sentencing Commission on the average sentence for federal drug defendants by plea/trial show:
plea - 5 years, 4 months (# 24,018); trial - 16 years (# 747).

Posted by: Ellen S. Podgor | Feb 20, 2014 10:06:58 AM

Many of the criticisms above make sense to me, but I'd love to know more about why it is wrong to include dismissals as a form of acquittal.

I take it that the underlying intuition would be that in these cases deciding to go to trial is basically calling a prosecution bluff. The prosecution isn't willing to spend effort to pursue the case, but you don't know that until you forego the deal (and thus force the state to spend effort to prepare for and go to trial). Now I know more about the federal fast-track regime, and maybe that is not what most of the dismissals in Cook County look like, I don't know; does anyone?

Posted by: Will Baude | Feb 20, 2014 1:13:57 AM

The Abrams empirical finding also conflicts with the qualitative research. When one study from one jurisdiction (based on some pretty funky acquittal rates) conflicts with so many quantitative and qualitative studies, it needs to be near-flawless and soon replicated by others to be persuasive. The inclusion of dismissed cases in the same category as acquittals is hard to justify in a study that tries to sort out pre-trial from post-trial outcomes. At most, the study leaves me in a wait-and-see posture.

Posted by: Ron Wright | Feb 19, 2014 10:10:44 PM

In response to Dan's query about other empirical research, Anderson and Heaton found that murder defendants represented by Public Defenders in Philly received lower sentences than those represented by private-court-appointed counsel. They conclude that these results are due to "less preparation by appointed counsel" who "investigate and prepare cases less thoroughly" than PDs. My forthcoming essay (which Dan graciously links above) argues – among other fascinating (!) things – that the findings might illustrate the "trial penalty" in action (as Jonathan characterizes the concept in his comment), since clients of PDs were 76% more likely to plead guilty, and there was no distinction in overall conviction rates. In other words, PD clients were convicted at the same rates as clients of private court-appointed counsel, but sentenced less severely because (I suggest) many more of their convictions resulted from guilty pleas. So Anderson and Heaton's study does, I think, provide an empirical counterpoint in favor of the existence of the "trial penalty," although they might disagree.

Posted by: Jeff Bellin | Feb 19, 2014 6:05:48 PM

Another critique (maybe Prof. Alschuler said this as well in his Duqesne piece, which I could not find online anywhere):

When we talk about the "trial penalty," we ordinarily do not mean it in the way Abrams uses it. Thus when he starts his article by stating that the concept of a “trial penalty” is widely accepted by all actors in the criminal justice system—and that those actors are all wrong—he proceeds to “disprove” the concept of “trial penalty” by using a different concept than the one widely accepted by participants in the criminal justice system.

As Abrams explains, there are two different concepts here: (1) the unconditional expected sentence (i.e., the expected sentence taking into account the chance that the defendant will not be convicted); or (2) the conditional expected sentence (i.e., the expected sentence assuming a conviction). Participants in the criminal justice system use the phrase “trial penalty” to refer to the latter, not the former. Abrams attempts to show there is no “trial penalty” under the former concept, not the latter.

So for example: from a defendant’s perspective, there is a “trial penalty” even if the plea offer is precisely the unconditional expected sentence (or even if it is below). Assume a defendant has a 50% chance of being convicted at trial, will receive a 4-year sentence if convicted after trial, and is being offered a 2-year sentence. That is the unconditional expected sentence, neither too low nor too high. If you were to tell this defendant that there is no “trial penalty,” he would scratch his head: of course there is a penalty, he would say—I’ll do two extra years if I’m convicted at trial rather than taking the plea! From the defendant’s view, the “trial penalty” is based on the conditional expected sentence, not the unconditional.

Relatedly: a major concern over a "trial penalty" is that an individual defendant who is innocent will be induced to plead guilty. (A related concern is that the guilty defendant gets an unjustified break from what he deserves by pleading guilty.) Insofar as this is the concern, our interest again lies with the conditional expected penalty, not the unconditional. That is because a properly-calculated unconditional expected penalty may indeed place serious pressure on a defendant, including an innocent defendant, to plead guilty. The defendant above, with the 50% chance of conviction, should take any plea offer below 2 years. That’s true whether he is guilty or innocent.

This is because our concern in evaluating the criminal justice system(s) is not primarily whether defendants as a group are receiving better (or worse) outcomes through plea bargaining than they would receive through a trial-only process. Our primary concern is whether individual defendants are convicted when they are guilty, not convicted when they are not guilty, and (when guilty) receive appropriate sentences.

The “trial penalty” (conditional) becomes particularly troublesome when the possible sentence is very high, and so there is a lot of pressure on a defendant to plead. Take a 25-year old defendant with two prior felony drug offenses, charged with selling 300g of crack cocaine. He’s facing a mandatory minimum life sentence if convicted at trial. The AUSA offers him 30 years. Is there a “trial penalty”? Well, in the way we should be concerned about, yes—there is enormous pressure on that individual defendant to plead guilty, even if he is innocent, and forego the chance to establish his innocence to bypass the risk of life imprisonment.

I am trying to think of an example of when the concept of “trial penalty” is used by the actors in the criminal justice system to mean “unconditional expected sentence.” I can’t think of one—can anybody else?

Posted by: Jonathan Witmer-Rich | Feb 19, 2014 4:17:20 PM

Thanks to Dan for posting my comments on the Abrams study. To answer Owen's question, the footnotes describe the response that Abrams made at the symposium at which both of us presented our papers, and I believe it's his only response to date: Abrams said that his findings were robust, & he was confident that if he had actually studied defendants acquitted at trial (rather than all unconvicted defendants), those findings would have stood up.

I don't consider this response implausible. Remember: Abrams agrees that sentences imposed following conviction at trial are far more severe than those imposed following guilty pleas. But if you assume a 50% acquittal rate (as Abrams mistakenly did), pleas would not look like bargains to him unless they led to sentences less than half as severe as those imposed following convictions at trial. I discuss all of this in the section of the article following the one that Dan posted. See 51 Duq. L. Rev. at 691-96. Forgive me for saying so myself, but this additional section is worth reading if you're seriously interested in this topic. Among other things, I say: "Frequent departures from the economic model occur because the personal interests of defense attorneys encourage them to recommend guilty pleas to their clients. A prosecutor need not offer a sentence that will overbalance the defendant's chances of acquittal when the defense attorney does not insist on such an offer and back his demand with a credible threat of trial. When defense attorneys sell out their clients too cheaply, findings like those reported by David Abrams should come as no surprise. . . At the same time, a lawyer who demands an offer minimizing his client's expected punishment may not get it, for prosecutors may be the ones who forsake the economic model. . . .[T]hey may recoil from placing the greatest pressure to plead guilty on defendants who may be innocent and/or from offering sentences far lighter than they believe honest-to-God offenders deserve." Folks interested in this topic might also be interested in the article's extended discussion of whether the sentence differential "rewards" defendants who plead guilty or instead "penalizes" defendants convicted at trial. I think that this question can actually be answered.

Posted by: Al Alschuler | Feb 19, 2014 3:22:43 PM

Seconding Sara - if somebody spends a year in jail before the case is dismissed, that's a one-year sentence for all practical purposes.

If they spend a year in jail, are convicted and get the exact same sentence for which they could have plea-bargained, that's a one-year 'trial penalty'.

Posted by: Barry | Feb 19, 2014 1:47:52 PM

Whether there is a trial penalty or a trial discount, there is also the issue that many defendants won't find it in their interest to wait and see, because they can't afford bail and therefore plead guilty to get out of jail (particularly in jurisdictions that rely on bail bondsmen where your 10% to the bondsman is nonrefundable). I haven't had a chance yet to dig into the Abrams data or Alschuler's critique, but bail seems to me an issue that should be factored in to the general discussion on this issue, particularly for lower-level crimes in jurisdictions with a high backlog/long wait to trial, where the time served awaiting trial might wind up approximating your sentencing exposure anyway.

Posted by: Sara Mayeux | Feb 19, 2014 1:08:06 PM

Has David responded to Al's critique? Such a response would be interesting to read, too.

Posted by: Orin Kerr | Feb 19, 2014 12:12:12 PM

Well, here's one response that I received via a friend, and it's a tip to Al Alschuler's piece in DLR that explains why he thinks Abrams' thesis is unconvincing. Here it is as copied and pasted from WL:
B. An Unconvincing Study

On one point, David Abrams' empirical findings match almost everyone else's. He reports that the sentences imposed following convictions at trial are substantially more severe than those imposed following pleas of guilty. [FN58] Abrams maintains, however, that this frequently repeated finding answers the wrong question.
In Abrams' view, earlier studies erred by considering only defendants who were convicted by guilty plea or at trial. [FN59] His study includes acquitted defendants as well. It treats the sentences of these defendants as zero. With these “sentences” of zero included in the mix of post-trial sentences, the post-trial sentences no longer appear to be more severe than those imposed following guilty pleas. To the contrary, they appear to be much less severe. [FN60] As a group, defendants who have taken their cases to trial appear to have achieved better results than those who have pleaded guilty. [FN61]
Abrams does not spell out what lesson he would draw from this finding, but he seems to suggest that plea bargains are usually bad deals for defendants. A risk-neutral defendant who discounts his probable post-conviction sentence by the likelihood of acquittal and compares it to the prosecutor's offer should usually reject the offer and go to trial. This defendant then will either win big or *688 lose big, but insisting on trial will on balance reduce his expected punishment. [FN62]
For reasons I will explain shortly, I do not credit Abrams' principal finding. [FN63] Even if I accepted this finding, however, I would not draw the lesson Abrams invites readers to draw. The discovery that defendants who went to trial achieved better results as a group than defendants who pleaded guilty does not show that any defendant in either group miscalculated. This finding may indicate only that plea bargains were good deals for defendants who took them and bad deals for those who didn't.
Every defendant in Abrams' sample who was acquitted might have discounted his expected post-trial sentence by the likelihood of acquittal and might have matched this discounted sentence against the prosecutor's offer, and every one of them might have made the correct choice. Every defendant who pleaded guilty also might have discounted the expected post-trial sentence by the likelihood of acquittal, and every one of them might have made the correct choice as well. [FN64]
Imagine, for example, a jurisdiction with a marvelous public defender system, the County of Dershowitz. Capably advised defendants in Dershowitz accept only offers that are truly in their interests. The District Attorney of Dershowtiz, however, is perverse. One day, he charges ten thousand men of Harvard with mayhem simply because a near-sighted witness, Magoo, reported seeing these men covered in crimson. The ten thousand men of Harvard demand trials, and all of them are acquitted. Abrams records their sentences as zero, and ten thousand sentences of zero bring the mean post-trial sentence to a tiny fraction of the mean *689 post-guilty-plea sentence. The existence of ten thousand wrongly accused Harvard men, however, reveals nothing about the wisdom of the choices made by the ably advised defendants who pleaded guilty before the district attorney went berserk. It also reveals nothing about the wisdom of the many defendants who are scheduled to plead guilty in Dershowitz tomorrow. The agreements these defendants entered with the district attorney may indeed have been bargains.
In any event, Abrams does not convince me that, with acquittals included, post-trial sentences are less severe than post-plea sentences. Abrams studied official court data from Cook County, Illinois—“a data set containing 42,552 cases initiated between 1997 and 2001 that were completed by the end of 2004.” [FN65] Because more than 40,000 felony cases were resolved annually in the Circuit Court of Cook County during this period, [FN66] the data set does not appear to include all of the cases that began and ended within the study period. Indeed, Abrams acknowledged in his presentation at this symposium that his data set did not include all cases. [FN67] Abrams also did not reveal whether the data included misdemeanor as well as felony cases, but they seem to consist almost entirely of felony cases.
Most of Abrams' study is presented in the obscure language of empirical economists, but I believe I understand Table 2. [FN68] This table divides defendants into those who pleaded guilty and those who stood trial. One column then reveals what proportion of defendants in various offense categories were convicted.
It comes as no surprise that 100% of the defendants who pleaded guilty were convicted, but the percentage of convicted defendants in the group that stood trial could raise eyebrows. Abrams reports conviction rates of 50% for defendants tried on drug possession charges, 48% for those tried on drug distribution charges, 36% for those tried on weapons charges, 47% for those tried on *690 theft charges, 53% for those tried on burglary charges, 43% for those tried on robbery charges, 47% for those tried on car theft charges, and 55% for those tried on charges of assault and battery. In most offense categories, more defendants appear to have been acquitted than convicted. In no offense category did the conviction rate exceed 55%, and in one category it was only 36%. [FN69]
Abrams evidently included a large number of sentences of zero in his mix of post-trial sentences. Although he did not supply the number, it seems a reasonable guess that roughly half of the sentences in his post-trial mix were sentences of zero. It was as though the 10,000 wrongly accused men of Harvard had marched from the County of Dershowitz to the County of Cook.
I have lived in Cook County; I have practiced and taught law in Cook County; I have studied Cook County; and Professor Abrams' Cook County is not the one I know. Statistics included in the Annual Report of the Illinois Courts permit one to calculate the conviction rates in felony cases tried in Cook County in 2001, a year that was both typical and included in Abrams' study period. In that year, 75% of the 301 felony defendants tried by juries and 81% of the 10,996 felony defendants tried by judges were convicted. [FN70] These conviction rates are similar to those reported by other American jurisdictions. [FN71]
The disparity between the normal Cook County conviction rates reported by the Illinois courts and the surprisingly low conviction rates reported by Abrams is apparently explained by the fact that Abrams treated all defendants whose cases ended without conviction as though they had been acquitted. Abrams in fact acknowledged during his presentation at this symposium that many of the defendants whom his article described as “acquitted at trial” were not acquitted at trial at all.
The Annual Report of the Illinois Courts includes a catchall category of terminated cases called “remaining balance.” This category “includes such dispositions as transfers to inactive/fugitive warrant calendar, extradition proceedings, and dismissed on motion of state.” The number of cases in this category in Cook County*691 in 2001 dwarfed the number of acquittals at trial, 7,848 to 2,134. When the “remaining balance” cases are treated as acquittals or “non-convictions,” Cook County's overall conviction rate in cases not resolved by guilty pleas drops to a rate resembling those that Abrams reported, 48%. [FN72]
Abrams' study thus reveals that defendants who abscond receive lower sentences than defendants who plead guilty—that is, until they are caught. It also teaches us that extradited defendants receive lower sentences than defendants who plead guilty—that is, until they are tried in the jurisdictions to which they are sent. And it reminds lawyers never to urge clients to enter plea agreements when prosecutors are willing to dismiss their cases outright. [FN73]
The low conviction rates Abrams reported enabled him to place a large number of sentences of zero in the cauldron of sentences imposed following trials. This number was not typical of the American criminal justice system or even of Cook County. Abrams' study does not provide convincing evidence that defendants who stand trial generally achieve better results than those who plead guilty, and even if it did, it would not show that any of the defendants who entered plea agreements had been snookered. [FN74]

Posted by: Dan Markel | Feb 19, 2014 11:58:53 AM

The comments to this entry are closed.