How much difference does the lawyer make? The effect of defense counsel on murder case outcomes.
ESSAY CONTENTS INTRODUCTION I. BACKGROUND ON INDIGENT DEFENSE IN PHILADELPHIA II. QUANTITATIVE ANALYSIS OF THE PERFORMANCE OF THE PUBLIC DEFENDER VERSUS APPOINTED COUNSEL A. Data and Sample Construction B. Methods: Counsel Assignment and the Preliminary Arraignment Process C. Results 1. Effects on Guilt 2. Effects on Sentencing III. EXPLANATIONS FOR THE DIFFERENCE IN OUTCOMES A. Conflicts of Interest B. Compensation for Lawyers, Investigators, and Experts C. Relative Isolation IV. PRELIMINARY IMPLICATIONS OF THE PERFORMANCE DISPARITY BETWEEN THE PUBLIC DEFENDER AND APPOINTED COUNSEL A. Constitutional Implications 1. Sixth Amendment 2. Eighth Amendment 3. Prospective Remedies B. Method of Providing Counsel to Indigent Defendants C. Improving the Process of Defense CONCLUSION APPENDIX
The mills of justice grind slowly, but they grind exceedingly fine. (1)
The idea that the inefficiency and slow speed of the justice system may somehow be justified by the system's ultimate precision is a reassuring one. It suggests that the justice system's vast creaky apparatus, for all its inefficiencies, will ultimately mete out the precise punishment that is necessary. It is also consistent with our goals of equal justice under the law and the idea that we are ruled by law rather than men. (3)
In this Essay, we examine one measure of the criminal justice system's "fineness"--its sensitivity to the defense counsel function. (4) Under nearly every normative theory of punishment or criminal responsibility, the characteristics of the offender's defense counsel should make no difference in the outcome of the process. Whether or not a defendant is found guilty and the extent to which the offender is sentenced to be punished should only depend upon facts about the offender and perhaps the possibility of and need to deter a particular crime. (5) The effect of the individual lawyer (and of the system for providing that lawyer) is pure "noise."
Usually the effect of the lawyer is hard to measure because lawyers and clients select one another. (6) It is difficult to determine whether the results obtained by a particular lawyer are attributable to the lawyer or simply to the characteristics of cases that the lawyer takes. Of course, most lawyers and clients act as though lawyers affect outcomes--lawyers brag about their abilities, (7) wealthy clients hire lawyers with the best reputations, and students compete to get into the best law school possible. (8) But because of this selection effect, it is usually impossible to isolate and measure the magnitude of the effect of the lawyer and the system for providing that lawyer. (9)
For the sake of the accuracy and fidelity of the criminal justice system--the fineness of the millstones of justice--one might hope that the differences in outcomes between lawyers are minimal. (10) This is particularly true in the most serious cases where the public interest in reliable adjudication is at its height. Perhaps the resources of the state are marshaled in such an effective way and the facts established so clearly by the government that what the defense lawyer does makes little difference. Perhaps, for example, those guilty of such a serious act as taking another's life are reliably and accurately punished irrespective of their lawyer. (11) It would be reassuring if the criminal justice system were this reliable in practice.
In this Essay, we take advantage of a natural experiment that allows us to measure the difference that defense counsel makes in the most serious cases. In Philadelphia, since April 1993, every fifth murder defendant is sequentially assigned at the preliminary arraignment to attorneys from the public defender's office. The other four defendants are assigned to appointed counsel. This sorting mechanism allows us to isolate the effect of the "treatment"--defendants represented by the public defenders--with the "control"--defendants represented by appointed counsel--by using an instrumental variables approach in cases from 1994 to 2005.
The differences in outcomes are striking. Compared to appointed counsel, public defenders in Philadelphia reduce their clients' murder conviction rate by 19%. They reduce the probability that their clients receive a life sentence by 62%. Public defenders reduce overall expected time served in prison by 24%.
These results suggest that defense counsel makes an enormous difference in the outcomes of cases, even in the most serious cases where one might hope that the particular type of defense lawyer would matter least.
Our findings, from the fifth-largest city in the United States, raise questions regarding the fundamental fairness of the criminal justice system and whether it provides equal justice under the law. The findings also raise questions as to whether current commonly used methods of providing indigent defense satisfy Sixth Amendment standards for effective assistance of counsel and Eighth Amendment prohibitions against arbitrariness in punishment. More generally, the strong impact of defense counsel suggests that the criminal justice system is, in practice, quite sensitive to the characteristics of the professionals involved. Policymakers may wish to consider efforts taken in other fields, like medicine, to increase reliability by reducing the system's dependence on the skill and performance of an individual professional.
We begin with an overview of indigent defense in Philadelphia. This is followed in Part II by a discussion of our methodology and our quantitative findings on the effect of counsel on the outcomes of murder prosecutions. In Part III, we discuss the qualitative interviews we conducted and previous research on indigent defense in Philadelphia. Finally, we discuss the constitutional and policy implications of these findings.
I. BACKGROUND ON INDIGENT DEFENSE IN PHILADELPHIA
In 2000, Philadelphia had a murder rate of 21 per 100,000 people, twelfth highest among large U.S. cities. (12) Most murder defendants, approximately 95%, cannot afford to hire private counsel and are therefore provided counsel by the county as required by the Sixth Amendment.
Pennsylvania is unique among the states in that the individual counties are solely responsible for the costs of indigent defense, in every other state, the state itself either funds a statewide public defender program or contributes to the costs of county public defender programs. (13) However, even with state help, counties bear a significant portion of the overall burden. In the one hundred largest counties in the United States, county and city funding made up 68.8% of total expenditures on indigent defense, with the states providing 25.3%. (14)
In Philadelphia, a nonprofit public defender organization, the Defender Association of Philadelphia, has long represented nearly all indigent defendants charged with any offense--except for murder. (15) Although its origin is somewhat murky, this exception apparently arose in the late 1960s or early 1970s as a way to maintain the private homicide defense bar and judges' power to appoint lawyers in murder cases. (16) In the mid-1980s, the Defender Association proposed representing some defendants accused of homicide, but the Philadelphia Bar Association opposed the measure and no change occurred. (17) After a change in bar and court leadership, the existing system began, and on April 1, 1993, the Defender Association began to represent one out of every five murder defendants. (18) The other four out of five defendants continued to be represented by counsel in private practice appointed by a judge ("appointed counsel") and paid by the county.
While some features of Philadelphia's indigent defense system are fairly unique, the basic approach of utilizing a mix of both public defenders and appointed counsel to represent indigent defendants is relatively common in the United States. In 2000, a survey of indigent defense systems conducted by the Bureau of Justice Statistics revealed that 80% of the one hundred largest U.S. counties employed both public defenders and appointed private attorneys as defense counsel in felony cases. (19)
The homicide unit of the Defender Association consists of a group of about ten experienced public defenders who have considerable experience practicing in the Philadelphia court system. (20) Every case is staffed with teams of two lawyers and one or more investigators and mitigation specialists (non-lawyer legal professionals, often social workers, trained to develop mitigation evidence usually introduced during the penalty phase of a capital trial) as needed. All members of the staff are salaried. The unit also has its own limited set of funds to hire expert witnesses directly without having to seek approval and funding from a judge, as appointed attorneys are required to do. (21) Unfortunately, we do not have the data necessary to calculate the cost per case of representation by the Defender Association.
Defendants who are not represented by the Defender Association are assigned counsel by one of the judges from the Philadelphia Court of Common Pleas, who each take turns assigning counsel in murder cases. (22) During the study period, the Court of Common Pleas of Philadelphia County required lawyers who wished to accept potential capital cases to have special qualifications based on the number of serious cases they had tried and the number of capital cases in which they had assisted. (23) In potential capital cases, two lawyers were often appointed, one to be responsible for the guilt phase of the case and the other to be responsible for the penalty phase of the case. (24)
During our study period, counsel appointed in murder cases--both capital and noncapital--in Philadelphia received flat fees for pretrial preparation: $1,333 if the case was resolved prior to trial and $2,000 if the case proceeded to trial. (25) The $2,000 trial fee also included the first half-day of trial. While on trial, lawyers received $200 for three hours of court time or less, and $400 per day for more than three hours in a given day. (26) Court appearances for continuances are not reimbursable.
Philadelphia's reimbursement rates for appointed attorneys during our study period were considered extremely low. Stephen Bright, president and senior counsel of the Southern Center for Human Rights, called Philadelphia's fee schedule "outrageous even by Southern standards." (27) Echoing the sentiment, a recent report issued by a homicide calendar judge in Philadelphia noted that "the compensation of court appointed capital defense lawyers in Philadelphia is grossly inadequate, both as to the dollar amount of the compensation and as to the compensation schedule provided by the present fee system...." (28)
Both capital and noncapital murder cases require numerous hours of preparation. (29) One examination of death-eligible murder cases in federal court in which the Department of Justice had not authorized prosecutors to seek the death penalty found that the median number of hours of preparation was 436, and the attorney cost per case from 1998 to 2004 was $42,148, which resulted in an hourly wage of approximately $97. (30) In capital cases during the same period, the median attorney hours were 2,014, and the cost was $273,901, which resulted in an hourly wage of approximately $136.(31)
Philadelphia's fee schedules have also been criticized for creating perverse incentives. (32) Counsel has no financial incentive to prepare for trial because there is a flat rate for preparation time. In addition, counsel may have an incentive to take a case to trial so that she can make as much in five days of trial as for the entire preparation period. As the homicide calendar judge noted in his report, "[this arrangement] increases the risk of ineffective assistance of counsel by maintaining a compensation system which punishes counsel for handling these cases correctly and rewards them only if they take every case to trial without adequate preparation or the exploration of appropriate non-trial options." (33) Numerous interviewees noted that because there is no cap on the number of cases that counsel can accept, the relatively small pool of attorneys who are willing to take appointed cases take on many more cases than they can adequately prepare. (34)
In short, the conditions in Philadelphia are conducive to an excellent test of how much the defense counsel function matters to outcomes. For the reasons discussed in this Part, the appointed counsel system seems very likely to result in comparatively poor defense counsel function. Conditions in Philadelphia allow us to empirically test our hope that, as one federal judge put it, "the facts--not the lawyers ... result in a substantially correct verdict." (35)
II. QUANTITATIVE ANALYSIS OF THE PERFORMANCE OF THE PUBLIC DEFENDER VERSUS APPOINTED COUNSEL
A. Data and Sample Construction
Murder defendants are initially charged in Municipal Court before being tried in the Court of Common Pleas. Our basic dataset includes a sample of 3,412 defendants charged with murder between 1994 and 2005 in Municipal Court. These data were provided to us by the Philadelphia Courts (First Judicial District of Pennsylvania). For each record, we observed the identity of the defendant, basic demographics (race, gender, and age), charges, attorney of record, and outcome. The Philadelphia Courts also provided us with a separate database with similar information tracking Court of Common Pleas cases that corresponded to these municipal cases and a database tracking changes in attorney assignments over time for a subset of defendants. (36) We supplemented these databases by collecting both the Municipal Court and Court of Common Pleas dockets for all of the cases in our sample from the Pennsylvania Judiciary's online docket database (37) and, as necessary, using data from the dockets to supplement information missing from the Philadelphia Court database. (38)
After eliminating forty-six defendants with missing data or ambiguous information on counsel assignment and 193 individuals (5.6%) in the sample who were ineligible for appointed counsel based on lack of indigency, we were left with 3,173 defendants. To identify individuals who were initially assigned to the public defender based on the one-in-five rule, we relied on logs provided to us by the public defender tracking the defendants in their murder cases, including both defendants initially assigned to the public defender and replacement defendants. (39) Of the 1,043 individuals listed in the public defender logs, we were able to find matches for 1,027 (98.5%) in the murder case records provided by the Philadelphia courts. (40) We also eliminated sixteen records involving cases that had not yet been resolved, that were missing Court of Common Pleas records, or that contained other data anomalies, leaving us with a total of 3,157 defendants.
One conceptual issue that arises in measuring the effects of representation is how to determine who represented a defendant who might have had multiple attorneys over the course of a case. One approach would be to count anyone who was represented by the public defender at any point in the process as having had public defender representation. A drawback of that rule is that it would include as public defender clients a large number of defendants initially assigned to the public defenders who were quickly reassigned due to a conflict of interest, and who therefore had essentially no interaction with the public defender.
The best approach would be to assign representation based upon the identity of counsel at the time the murder charge was resolved. Unfortunately, because our attorney history data are incomplete for most of our defendants, our ability to identify who was representing a defendant at case resolution is limited. (41) Moreover, if public defenders represent defendants at earlier stages of the case, such as at a preliminary hearing, they can arguably exert some influence over the outcome of the case even when defendants are ultimately represented by other counsel. As a compromise, we measure representation by the public defender based upon the identity of the attorney at the formal arraignment. (42) This approach has the advantage of measuring representation at the same point of case progression for all cases and at a point at which the attorney could have influenced case outcomes. An obvious drawback is that, to the extent that defendants change attorneys subsequent to the formal arraignment, our definition fails to account for such changes. This happens very infrequently, however, so representation at the formal arraignment makes an excellent proxy for representation at disposition. (43)
We also constructed synthetic criminal histories for each defendant by extracting information from the Pennsylvania Court of Common Pleas docket sheets for each prior case involving that defendant. (44) Although these histories provide useful information regarding the prior criminal involvement of the defendants in our sample, they do not fully capture prior criminal activity because they only include offenses that occurred in Pennsylvania, generated a court record, and occurred after electronic recordkeeping was instituted in each county in the state. (45) Although it seems likely that our measures understate the amount of prior criminal activity because some activity is not captured in available court dockets, we have no reason to suspect that the pattern of missing information would correlate with attorney assignment.
Our sentencing data report a maximum and minimum sentence for each defendant, and also identify life and death sentences. (46) Because life and death sentences are qualitatively different from other sentences, we consider these outcomes individually. Ideally, we would also like to calculate an overall effect on length of incarceration. This metric is complicated by the fact that those sentenced to life and death do not receive numeric sentences. Because of this issue, we consider two alternative measures of incarceration length as outcomes. First, we consider average sentence, which we define as the midpoint of the reported maximum and minimum sentences for those given a numeric sentence. For those sentenced to life or death, we set the average sentence equal to forty years, an admittedly arbitrary choice but one that seems sensible given that the bulk of those sentenced to life are in their early twenties (47) and life sentences in Pennsylvania carried no possibility of parole. (48)
Alternatively, to avoid the necessity of imputing an arbitrary sentence length for those sentenced to life or death, we also calculate the expected time served in prison and use this as an additional outcome measure for length of incarceration. To do this, we turn to data from the National Corrections Reporting Program (NGRP). (49) The NCRP contains individual-level information about state prison admissions and releases (including deaths) for participating states, and includes information about alleged offenses, sentencing, and time served. For the years between 1999 and 2003, the NCRP includes records for 15,721 defendants who were released from prison after serving a sentence for a murder conviction. (50) For each combination of age at prison admission and sentencing outcome, we compute the average time stowed across prisoners in our NCRP sample, which includes data from states other than Pennsylvania, and then apply that average to Philadelphia defendants who fall into that same age and sentence combination. (51) For example, among those in the NCRP with 30-year sentences imposed at age twenty-two to twenty-four who were released or died in prison between 1999 and 2003, the average actual time served was 16.1 years, (52) suggesting a newly convicted twenty-three-year-old murder defendant with a 30-year sentence might expect to spend around 16 years behind bars. For life sentences we only use NCRP data for prisoners who died in prison because life sentences in Pennsylvania do not carry the possibility of parole. (53) This approach offers a data-driven method for deciding how much incarceration to assign to those with life and death sentences. Conceptually, the expected time served in prison can be thought of as the response of a well-informed attorney if the defendant asked, immediately after receiving a particular sentence, how long he could actually expect to spend behind bars.
A drawback of using NCRP data to project actual time served is that because these projections require data on complete sentences, they require us to use individuals who were mostly sentenced during the 1980s and early 1990s. Because of the growth of truth-in-sentencing laws (54) and declines in mortality among prison inmates, the actual time served for individuals in our sample from Philadelphia will be greater than time served in the NCRP, meaning that our projections likely represent lower bounds on future time served. (55) However, there is no reason to suspect that the bias towards underprojection of time served inherent in our approach will differentially affect defendants represented by appointed counsel as compared to public defenders. As a result, we can use these projections to correctly measure the percentage difference in expected time served for defendants represented by the public defender.
B. Methods: Counsel Assignment and the Preliminary Arraignment Process
Appendix Figure 1 presents a flow chart illustrating the processing of murder cases in the Philadelphia courts. Shortly after arrest, defendants accused of murder receive a preliminary arraignment. This usually occurs by video conference before a court magistrate. The magistrate reviews the information about the defendant compiled by the court's pretrial unit to determine if the defendant can afford counsel. (56) If, in the magistrate's judgment, the defendant is unlikely to be able to afford counsel in a case with a murder charge, the magistrate appoints either the Defender Association of Philadelphia or to-be-determined appointed counsel to represent the defendant. In the vast majority of cases, it is clear that the defendant cannot afford private counsel. (57) The default is to assign counsel. Hearings in which counsel are assigned typically take approximately two to three minutes.
The Criminal Law Clerk maintains a log book of all cases of this type, and every fifth defendant with a murder charge is assigned to the public defender. (58) The other four defendants are not immediately assigned counsel, but their names are sent to court appointments for assignment to a court-appointed counsel. (59)
After assignment, there is some crossover between "treatment" (Defender Association defense counsel) and "control" (appointed counsel) groups. Some defendants hire private defense counsel who replace either appointed counsel or the public defender. In some cases assigned to the public defender, it is determined subsequent to the initial assignment that there is a conflict of interest and that the public defender cannot represent the defendant. When that occurs, the case is assigned to appointed counsel and the public defender receives another "replacement" case that had been assigned to appointed counsel at the preliminary arraignment. The goal is to ensure that the public defender ends up with 20% of cases, per its contract, in spite of the fact that some defendants change counsel subsequent to initial assignment. Although these replacement cases are nominally random, there is no mechanism comparable to the rotation at preliminary arraignment to ensure that they are, in fact, randomly selected. However, because these diversions occur after the initial one-in-five randomization, they are not problematic for our analysis, and we need not assume that replacement cases are randomly selected.
If compliance with random assignment were perfect, so that every defendant initially assigned appointed counsel were ultimately represented by appointed counsel, and the same were true for the public defender, the causal impact of public defender representation could be computed simply as the difference in mean outcomes between those represented by the public defender and those represented by appointed counsel. However, in actual practice, later representation varies from the assignment for numerous reasons. In some situations, such as cases involving multiple defendants, individuals initially assigned to the public defender must be appointed counsel to avoid conflicts of interest. When defendants are able to hire a private lawyer, they often progress partway through the adjudication process with appointed counsel before being able to assemble the financial means to pay for a private attorney.
It is possible that this crossover (or imperfect compliance, as it would be called in a clinical trial) is correlated with the identity of counsel and characteristics of the case. Suppose, for example, that defendants with very serious cases assigned to the public defenders are more likely to hire private counsel than defendants with equally serious cases assigned to appointed counsel. Simply comparing the mean outcomes between defender-assigned defendants and appointed-counsel-assigned defendants in that instance would then be misleading because the case mix would not be comparable--the public defenders would be left with a less serious set of cases.
Similarly, the operation of conflict-of-interest rules might also change the mix of cases. In general, a defense lawyer will not represent a defendant if an important witness in the case was previously represented by that lawyer because the duty to zealously represent the defendant's interest might conflict with an ongoing duty of loyalty to a former client, because of the need, for example, to attack the credibility of the former client. Because a conflict of interest is imputed to other attorneys in the organization, and because public defenders represent nearly all other criminal defendants, public defenders are much more likely to be conflicted out of a case than appointed counsel for any given set of witnesses in a case. Suppose that cases with numerous witnesses (in which the Defender Association attorney is more likely to be conflicted out) are more serious than cases with fewer witnesses. Once again, the case mixtures are no longer equivalent and the results of a simple comparison in outcomes are not valid.
To deal with this problem of crossover we employ an instrumental variables (IV) analysis. We use the initial random assignment as an instrumental variable for the later representation. The IV method permits us to exploit the randomness of initial assignment to estimate the causal impact of public defender representation. (60) This method essentially isolates the portion of variability in outcomes that is attributable to the initial random assignment. In estimation, this result is reached by regressing the case outcomes of interest on the predicted legal representation at arraignment, where the predicted value is determined by a first-stage regression of representation at arraignment on the legal representation at the point of random assignment (plus all other controls in the model). Because we use only the variation in legal representation status attributable to random assignment and not the actual representation, we can estimate the impact of public defender representation even when there is nonrandom sorting of defendants across different types of attorneys subsequent to the initial assignment. In light of the systematized assignment of counsel, even if a nonrepresentative subset of defendants switch counsel after the initial step in the process, we can still identify two groups of defendants--namely, those who were and those who were not initially assigned to the public defender--for whom the expected average sentence is the same except for the fact that they end up with different types of counsel. The IV approach compares the average outcomes across these groups (rather than groups based upon actual realized representation) and then scales this difference by the groups' difference in representation. This comparison allows us to control for the fact that there may be nonrandom sorting (e.g., by seriousness of case) between the time of initial assignment counsel at the preliminary arraignment and the time that the cases are ultimately resolved.
The key requirement for the IV analysis to deliver valid causal estimates is that the instrumental variable--in this case, initial counsel assignment--affects eventual representation but is otherwise uncorrelated with case outcomes. If the initial assignment of counsel is truly random, as we assume, this requirement will be satisfied. Fortunately, it is possible to examine the validity of this assumption directly using the available data. In particular, if counsel is assigned randomly, we would expect those assigned to appointed counsel and those assigned to the public defender to appear similar on observable characteristics determined prior to counsel assignment.
In Table 1, we summarize the characteristics of our sample, reporting average characteristics of defendants initially assigned to appointed counsel (Column I) and the public defender (Column II). We also report the t-statistic and associated p-value for a test of the null hypothesis of equal means across the two groups. The first row of the table indicates that of those who were initially assigned appointed counsel, 15.5% were ultimately represented by the public defender at their municipal court arraignment. Many of these cases represent individuals who normally would have been given court-appointed counsel based on the one-in-five assignment rule, but who were instead diverted to the public defender in order to provide replacement cases for clients initially assigned to the public defender who had subsequently found other representation. Only 59.2% of those initially assigned public defenders retained their public defenders through the municipal court arraignment. In other words, almost half of those assigned public defenders ultimately were represented by other attorneys, due to either conflicts or voluntary hiring of an outside attorney. Although substitutions away from the initial assignment were fairly commonplace, the t-test indicates that the initial assignment satisfies the first requirement of an instrument, namely, that it affects eventual representation.
The next rows of Table 1 report average demographics by initial assignment. Age, race, and gender are comparable across the two groups of defendants. Although available case records contain no additional direct demographic information, another way to assess the comparability of the background characteristics of defendants is to examine the population characteristics of the ZIP codes in which they reside.
The next rows of Table 1 compare economic and social characteristics of the residential ZIPs of indigent defendants using data drawn from the 2000 Census. (61) If the randomization is compromised so that certain types of defendants are more likely to receive Defender Association attorneys, we might expect to observe different neighborhood backgrounds for these defendants. A drawback of examining ZIP code characteristics is that ZIP information is missing for almost a third of the sample, although, as indicated in Table 1, rates of data availability are similar across the two groups.
Indigent homicide defendants are drawn disproportionately from disadvantaged areas. For example, 56% of households in the ZIP code of a typical defendant were female headed, versus 22% for the city as a whole and 12% nationally. (62) Unemployment rates in the defendants' ZIPs were more than 2.5 times the city average. (63) Although homicide defendants are clearly drawn from an unrepresentative sample of the city's neighborhoods, differences in the neighborhood characteristics of those assigned appointed versus public attorneys are negligible. (64)
Our criminal history data provide another way to assess the comparability of the two groups of defendants. As indicated in Table 1, average criminal involvement appears slightly higher among those assigned to the public defender, although none of the differences is statistically significant except for prior theft charges. Given that prior criminal history is one of the strongest predictors of case outcomes, (65) the fact that the two groups of defendants appear largely balanced in their prior criminal involvement is reassuring.
The next rows of Table 1 summarize the characteristics of these defendants' cases, including number and nature of charges and number of defendants involved in the case. Because attorney assignments are made prior to the formal arraignment, in theory the charge composition could adjust based on attorney characteristics. For example, if prosecutors believe that public defenders are likely to beat weapons or conspiracy charges, they may drop or decline to file such charges once they see that a particular defendant is represented by a public defender. As a practical matter we see little evidence of important differences in case characteristics by initial assignment, although there appears to be a slightly lower rate of weapons charges for defendants initially assigned public defenders.
There are statistically significant differences between the two populations across a handful of characteristics, such as prior theft, but even in the absence of true differences, we would expect to observe some statistically significant differences when examining this many characteristics due to sampling variation alone. One way to assess whether the overall pattern of group differences shown in Table 1 provides evidence of nonrandom assignment is to examine the distribution of p-values in the table. Under the null hypothesis of random assignment, we would expect these p-values to be uniformly distributed between 0 and 1. A Kolmogorov-Smirnov test (66) applied to the twenty-nine defendant characteristics listed in Table 1 that were determined prior to assignment of counsel yields a p-value of 0.17, indicating that we cannot reject the null hypothesis of random assignment. (67)
Of course, data-based tests of the independence of an instrument are limited to the available data. It is always possible that the proposed instrument is actually related to the outcome in other ways that are unobservable in our data. It is therefore important to examine the actual mechanism of the instrument.
Here, interviews with the Philadelphia court staff indicate that the assignment process is almost completely mechanical and ministerial-little human judgment (and possible conscious or unconscious bias) is involved. (68) A log book is kept by the clerk of the arraignment court and every fifth defendant with a murder charge who comes through is assigned to the public defender. This sorting mechanism is additional evidence of the independence of our instrument.
We find significant differences in the outcomes of the defendants represented by the Defender Association and appointed counsel. Table 2 reports defendant outcomes by initial attorney assignment.
Given that the two groups of defendants appear largely similar in terms of demographics, prior criminal involvement, and observable case characteristics, absent any effects of counsel, it seems reasonable to expect similar outcomes across the two groups. In the first row, for example, of the 2,677 defendants who were originally assigned appointed counsel, 80.1% were found guilty of any charge; the comparable number for defendants originally assigned to the public defender was 79.2%. The low t-statistic and p-value that is greater than 0.05 for this characteristic indicate that there is not a statistically significant difference in overall conviction rates across the two groups at a 95% confidence level.
However, we observe statistically significant and practically large disparities in some outcomes across the two groups. For all of the sentencing measures except for death verdicts--which, even among this population, are quite rare--those assigned to the public defender achieved better outcomes than those assigned to court-appointed defense counsel. The seven-percentage-point difference in the likelihood of receiving a life sentence and the difference in expected time served are particularly notable. The greater-than-one-year difference in expected time served is large relative to the overall expected time served of around eleven years.
One potential explanation for these differences in outcomes is that public defenders might use different strategies for determining whether to take cases to trial than appointed attorneys, particularly given that these two sets of attorneys have different financial incentives for trial. The bottom row of Table 2 indicates that defendants assigned to the public defender are appreciably more likely to plead guilty in their cases than those initially assigned appointed attorneys.
I. Effects on Guilt
The simple comparisons in Table 2 strongly suggest that public defender representation is associated with improved case outcomes. To estimate the causal impact of representation by the public defender, we turn to the instrumental variable (IV) analysis.
In Table 3 we report IV regression estimates of the impact of public defender representation on whether an individual was found guilty of various types of offenses (either at trial or because of a plea arrangement), with a murder defendant as the unit of observation. (69) The first column of the table reports the average outcome among those initially assigned private counsel, which offers a basis for judging the magnitude of the impact estimates. Each subsequent row entry reports an impact estimate obtained from estimating a particular regression model. The first row in the table, for example, indicates that defendants initially have a 0.801 average probability (or 80% chance) of pleading or being found guilty of some charge. Using the IV1 model, representation by the public defender is estimated to reduce the probability a defendant is found guilty of any charge by 0.02, but this difference is not statistically significant.
Column IV1 estimates a simple linear IV model with no controls; this setup is equivalent to dividing the mean difference in outcomes reported in Table 2 by the mean difference in representation (0.44).
Column IV2 adds to the IV regressions controls for defendant race, gender, age and age squared, and year of case; and indicators for the number of defendants, total number of charges, presence of a weapons or conspiracy charge, and total prior charges and prior arrest for assault, aggravated assault, weapons offenses, drug offenses, burglary, robbery, and theft. If randomization was successful, as is suggested by Table 1, the inclusion of additional controls in the regression model is not strictly required to obtain an unbiased estimate of the impact of public defender representation. However, controlling for additional covariates may yield more precise estimates of attorney effects, and the controls may also be helpful for addressing any unrecognized departures from randomization. Because the IV2 model includes a comprehensive set of controls and identifies the effect of public defender representation using the broadest set of cases, it is our preferred specification, although in general we obtain similar effects estimates whether or not we control for other factors.
Column IV3 limits the analysis to cases involving multiple defendants and adds a set of indicator variables, one for each case, as additional controls. (70) This IV analysis essentially identifies the impact of public defender representation by comparing the outcomes for codefendants who were involved in the same case, where one defendant was assigned to a public defender and other defendants were assigned appointed counsel. The main advantage of such a within-case analysis is that it holds constant factors that are determined at the case level-such as the quality of witnesses, investigative effort by the police, etc.--across those with different types of representation, even when such factors may be unobservable. The primary drawback of the models with case-level indicators is that these models appreciably reduce our sample size, excluding the 2,061 cases involving a single defendant from the analysis and focusing only on those cases with several defendants who differ in their initial assignment. Because of the smaller sample, these estimates are less precise than those using the full sample.
Although estimates of the impact of public defender representation on guilt for any charge are negative, these estimates are modest relative to the overall guilt rate of 80%, and none is statistically significant. More striking are disparities in murder conviction rates. The IV2 model, our preferred specification, demonstrates that those represented by public defenders are 0.11 points less likely to be convicted of murder, a 19% decline relative to the conviction probability among those with appointed counsel of 0.565. This difference is statistically significant.
2. Effects on Sentencing
We next turn to sentencing outcomes. The two most severe penalties for murder are life in prison, which in Pennsylvania carries no possibility of parole, and death. As shown in the next rows of Table 3, representation by the public defender reduces the probability of receiving a life sentence by 0.16 (Column IV2), or a remarkable 62%. This reduction in life sentences can be observed both in the full sample and when limiting the analysis to trials with multiple defendants. (71)
While no defendant represented by the public defender at trial has ever received a death sentence, our estimates of the effect of being represented by the public defender on receiving a death sentence are small. However, because fewer than 2% of defendants receive a death sentence, our estimates are highly imprecise. (72) The 95% confidence interval for these estimates encompasses values that would imply either a substantial reduction or a substantial increase in the probability of receiving a death sentence due to public defender representation. Thus, these data preclude drawing conclusions about the efficacy of public defenders in avoiding death sentences. (73)
We next turn to an analysis of sentence length and expected time served. We find substantial and highly statistically significant impacts of public defender representation on average sentence length. The causal impact of public defender representation on sentence length is a 6.4-year reduction (IV2), which represents a 31% decline relative to the mean sentence length for those assigned appointed counsel of 20.9 years. (74)
For those who are not sentenced to life imprisonment or death, we also examine minimum and maximum sentences. The W point estimates for these outcomes are negative and sizable, but only marginally statistically significant. The magnitudes of the estimated impacts, however, are large, implying a greater than one year reduction in minimum sentences and a more than three year reduction in maximum sentences. It appears that public defenders are successful at both reducing the likelihood of the most extreme sanctions and reducing the severity of less extreme sentences.
The final row of Table 3 uses expected time served as the outcome, where expected time served is calculated using the NCRP as described in Section II.A. Our analysis reveals statistically significant and practically large impacts of public defender representation on expected time served. The IV2 estimate of -2.6 implies that individuals represented by public defenders are expected to spend more than two-and-a-half fewer years in prison than otherwise similar defendants represented by appointed counsel. (75) This measure represents a 24% reduction in expected sentence. The magnitude of this effect in percentage terms is roughly comparable to our estimated impacts using average sentence length as an incarceration measure.
It is instructive to compare our findings with those of two other recent studies that use randomization to understand the influence of defense lawyers on case outcomes. (76) Rather than comparing appointed attorneys and public defenders as we do, Abrams and Yoon exploit the random assignment of defense attorneys within the public defender office in Clark County, Nevada to examine whether experience and other attorney characteristics affect outcomes. They find that an additional year of experience is associated with a 1.7% reduction in the sentence. (77) Thus, moving from a private appointed attorney to a public defender in Philadelphia, which we estimate reduces sentences by about 31%, is similar in effect to shifting from an attorney with no experience to one with over 18 years of experience in their study.
Iyengar compares public defenders and appointed counsel in the federal system and finds that public defenders reduce expected sentences by 16% relative to private assigned counsel. (78) Her estimates are sufficiently precise so as to statistically reject the 31% decrease in sentences we find in Philadelphia. The fact that public defenders in Philadelphia appear to have a larger impact on sentences than federal defenders may reflect the fact that attorneys play a larger role in murder cases than in other, less serious cases, or this difference may result from the compensation system and other factors more specific to Philadelphia. (79)
By way of contrast, the column of Table 3 labeled OLS presents estimates of the impact of public defender representation on outcomes that use ordinary least-squares (OLS) regression analysis that adjusts for observable differences in characteristics between those with private appointed counsel versus those with public defenders. The OLS method is the primary approach used in past studies of the impacts of public versus appointed counsel. (80) Its primary flaw is that it ignores the effect of postassignment, nonrandom sorting: the fact that defendants who start out with the public defender and move to appointed counsel (and vice-versa) may have particular nonrandom characteristics.
The OLS approach does provide some evidence that public defenders attain better outcomes than their appointed counterparts--for example, public defenders are estimated to reduce the number of guilty charges by an average of 0.2 charges across defendants and reduce the probability of receiving a life sentence by 0.05. However, differences between the OLS and IV estimates are noticeable for many outcomes. For example, properly accounting for nonrandom sorting to attorneys triples the estimated impacts of public defender representation on life sentences and increases the reduction in expected time served by two years. OLS estimates suggest public defenders do not affect murder convictions, whereas the more credible IV results show a strong effect.
To provide further insight into why OLS and IV estimates differ, in Appendix Table 1 we report coefficient estimates from a regression model where the dependent variable is an indicator for whether a defendant was represented by a public defender at the formal arraignment and the explanatory variables capture defendant demographics and prior criminal history. (81) These regressions provide insight into which types of defendants are ultimately most likely to keep their public defenders through the resolution of their cases. Appendix Table 1 demonstrates that those ultimately represented by public defenders are indeed a nonrandom subset of the total population--for example, older defendants are slightly more likely to remain represented by the public defenders. Defendants with current weapons charges are less likely to be ultimately represented by public defenders. Given this clear evidence of sorting based on observable characteristics, it seems reasonable to expect that sorting may also occur along dimensions that are unobservable to us but may affect how cases are ultimately decided. (82) These patterns demonstrate the difficulty of cleanly measuring attorney effects using traditional regression methods that cannot readily account for defendant sorting behavior.
One question raised by the large disparity in outcomes shown in Table 3 is the extent to which these differences reflect superior performance by public defenders in plea negotiations versus trials. Although the data clearly indicate public defenders must perform better at something, ideally one might wish to isolate whether the difference results from better handling of plea negotiations, better handling of trials, or both. Unfortunately, the usual notion of whether a public defender is "better"--(i.e., for a given defendant and fact pattern, does the public defender achieve a lower sentence in a trial or plea negotiation?)--cannot be measured using the available data. Whether a person pleads guilty or goes to trial is not randomly determined, but rather reflects a selection process involving the attorney, the client, and the prosecutor; we cannot simply reanalyze the subset of cases that involves guilty pleas or trials to measure the effect of attorney type on that particular class of case. (83) Put differently, even if, for any given defendant, public defenders are worse at both plea negotiations and trial representation, it would still be possible for public defenders to get shorter than average sentences for their clients if there is heterogeneity across defendants in the expected benefit of going to trial and public defenders are simply better at sorting appropriate defendants to pleas versus trials.
We can, however, examine whether there appear to be systematic differences in how different types of attorneys handle cases. Table 4 presents estimates of the impact of public defender representation on two measures of case handling: whether the defendant waives a jury trial--a strategy typically used to reduce the likelihood of a death sentence-and whether the defendant pleads guilty to at least some charges. (84) While use of waiver trials does not vary across the two types of attorneys, clients of public defenders are 21 percentage points (or 76%) more likely to plead guilty than clients of appointed private attorneys. These differences in willingness to plea bargain may at least partly explain the shorter sentences obtained by Defender Association attorneys for their clients.
Because public defender representation does not affect overall guilt rates, as shown in Table 3, but does appreciably increase the share of cases resolved by plea bargain, we know that the conviction rate must be lower among cases taken to trial by Defender Association attorneys than among cases taken to trial by private appointed counsel. This pattern might occur because public defenders are better at ensuring that cases with superior prospects for acquittal proceed to trial, or it may be that public defenders are better at arguing cases.
The following provides one interpretation of the combined evidence from Tables 3 and 4: public defenders are more successful at convincing their clients to plead guilty to lesser charges, and as a result, these clients are able to avoid being found guilty of murder and do not receive the most severe sanctions, such as life imprisonment. Clients of appointed counsel are more likely to go to trial, but overall guilt rates for this group end up being the same, meaning that these individuals lose the benefit of lessened sentences from plea bargaining without gaining much in terms of increased likelihood of acquittal.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Introduction through II. Quantitative Analysis of the Performance of the Public Defender Versus Appointed Counsel, p. 154-187|
|Author:||Anderson, James M.; Heaton, Paul|
|Publication:||Yale Law Journal|
|Date:||Oct 1, 2012|
|Previous Article:||Aggregation and law.|
|Next Article:||How much difference does the lawyer make? The effect of defense counsel on murder case outcomes.|