Regression analysis, a common method of attempting to demonstrate racial and gender discrimination in hiring and other contexts, suffers from a number of shortcomings that invariably cast doubt on any incriminating results it produces. Most notably, regression models may fail to account for variables which correlate both with legitimate goals of a decisionmaker and with race/gender categories. In this article, Professors Ayres and Waldfogel offer an alternative method of measuring discrimination that overcomes the limitations of traditional regression tests. The authors present a market- based test of unjustified disparate impact using data from the bail bond market in New Haven, Connecticut, to demonstrate that New Haven courts systematically "overdeter" black and male Hispanic defendants from fleeing after release on bail by setting bail at seemingly unjustified high levels for these groups. Professors Ayres and Waldfogel discuss the validity of the assumptions underlying their model, and of possible nondiscriminatory explanations for their finding of disparate impact. They also offer suggestions for application of their methodology in contexts other than bail setting.

1. Market Forces that Might Cause Bond Rates not to Reflect Flight Probabilities
   1. Price discrimination
   2. Dealer search and monitoring costs
   3. Forfeiture rates
   4. Collateral
   5. Fixed costs
2. Alternative Judicial Goals
   1. Targeting higher probabilities of appearance for more serious offenses
   2. Color-blind bail setting
   3. Deterring pretrial misconduct
   4. Targeting desired bond fee instead of desired probability of appearance
3. Sample Selection
   1. Defendants who remain in jail
   2. Posting bail personally
   
   IV. HARNESSING THE MARKET IN THE BAIL SETTING PROCESS
   CONCLUSION
   APPENDIX A: EX ANTE MONITORING AND EX POST SEARCH
   1. Demonstrating How Monitoring and Search Can Deter Flight
   2. Investigating the Effects of Monitoring and Search Costs on the Bail Rate
   
   APPENDIX B: ESTIMATING RATE SCHEDULES

   
   

   INTRODUCTION
   [T]he professional bondsman system ... is odious at best. The effect of such a system is that the professional bondsmen hold the keys to the jail in their pockets.

   These words of Judge Skelly Wright capture the disdain often shown to bail bond dealers. Bail bond dealers are widely thought to be corrupt and to assert too much control over pretrial release decisions. Commentators offering bail reform proposals often advocate eliminating private bail bonding entirely.

   This article suggests, however, that the bail bond market may have an unforeseen usefulness. If competition induces bond dealers to charge fees equal to their average costs, the pricing behavior of bond dealers can be used to assess whether states discriminate on the basis of race or gender in setting bail. This article uses empirical evidence from the bail bond market in Connecticut to evaluate the determinants of bail setting.

   Connecticut law has traditionally mandated that bail be set at the smallest amount that will "reasonably assure the appearance of the arrested person in court." Implicit in this statutory command is the notion that higher bail tends to reduce the probability that a defendant will disappear. If the defendant posts bail herself, the potential forfeiture of the bail gives the defendant a direct financial incentive to appear at trial. If the defendant contracts with a bond dealer, however, the incentive to appear is indirect. A bond dealer charges the defendant a nonrefundable fee, and in return assumes the risk of paying the bail amount to the state should the defendant flee.

   Because the bond dealer's fee is nonrefundable and therefore is a "sunk cost," the size of the fee should not affect the likelihood of the defendant's appearance. Nevertheless, a higher bail amount can reduce the probability of flight: A higher bail amount credibly commits the bond dealer to expend more resources on (1) monitoring the defendant's whereabouts before a scheduled appearance; and (2) searching to reapprehend any defendant who flees.

   Thus, if a defendant fails to appear, a bond dealer with $20,000 at risk would likely search more extensively than a bond dealer with only $2000 at risk.

   Anticipating bond dealers' incentive to monitor and search more aggressively, defendants might perceive a lower likelihood of successful flight when the court sets a high bail amount. Even when defendants use bond dealers to secure release, high bail amounts can thus reduce the probability of successful flight and even deter flight attempts. Connecticut courts can "reasonably assure the appearance" of defendants by setting bail amounts high enough to reduce the defendants' flight probability to an acceptable level.

   Bail setting in Connecticut has come under fire recently amid allegations of racial bias. In 1991, the Hartford Courant reported that bail amounts for black defendants were, on average, more than 70 percent higher than for white defendants. Reports such as this are consistent with prior evidence of race discrimination in Connecticut and in other states.

   For example, Malcolm Feeley's classic analysis of the New Haven Court of Common Pleas found that black defendants were more often subject to monetary bail and were less likely to flee when released on bail than were white defendants.

   Nevertheless, providing unequivocal evidence of racial discrimination in bail setting has proven elusive to scholars and lawyers alike. The traditional way to statistically test for discrimination in bail setting would be to estimate in a regression how permissible factors (those related to a defendant's flight risk) affect the size of bail, and then to determine whether, after controlling for these permissible factors, race is still a significant determinant of the bail amount. Despite its popularity, this methodology is dogged by the problem of "omitted variable bias": Race may be correlated with unobserved variables that are not controlled for in the regression, but that legitimately increase bail size. If this is the case, the race effects estimated by regression analysis might not be caused by disparate racial treatment.

   For example, our analysis of 1118 New Haven arrests reveals that after controlling for eleven variables relating to the severity of the alleged offense, bail amounts set for black male defendants were 35 percent higher than those set for their white male counterparts. By itself, this result does not constitute very powerful evidence of race discrimination because the regression does not control for many other variables that might explain the racial disparity. For example, black male defendants might be less likely to be employed or to have other community ties, which might justify a higher bail amount. Judges might have set higher bail for black male defendants not because they were black, but because other characteristics we did not observe indicated that these defendants had a higher propensity to flee.

   The omitted variable problem has made it exceedingly difficult to use regression analysis to demonstrate racial discrimination. For example, in McCleskey v. Kemp the Supreme Court rejected a regression study that, after controlling for 230 variables, indicated that black defendants charged with killing whites were more likely to receive the death penalty than white defendants charged with killing blacks. While assuming that the regression study was "valid statistically," the Court nonetheless concluded that "[a]t most, the ... study indicates a discrepancy that appears to correlate with race."

   In contrast to the methodology used in the McCleskey v. Kemp study, this paper presents a test for discrimination in bail setting based on the rates bond dealers charge after bail is set. Underlying our approach is the familiar economic concept of competitive pricing. In a competitive market, the bail bond rate (i.e., the bond fee divided by the bail amount) should approximate the market's assessment of the defendant's probability of flight--because the bond dealer's expected cost of writing a bond is simply the amount of the bond multiplied by the probability that the defendant will flee. For instance, if there is a 10 percent chance that a defendant will fail to appear, bond dealers in a competitive market should charge a 10 percent bond rate.

   Our core finding is that bond dealers in New Haven charged significantly lower rates to minority defendants than to whites: For example, in our data, bond dealers charged black male defendants rates that were almost 19 percent lower than the rates charged to white male defendants. This race differential in the bond market, we argue, raises the specter of unjustified racial discrimination at the bail-setting stage. The lower minority rates indicate that bail reduced the probability of flight for minority males below the flight probability for white males. Our results showing significantly lower bond rates (alongside a traditional regression showing significantly higher bail amounts for black defendants in our sample) constitute powerful market evidence of unjustified racial discrimination in bail setting. The market evidence indicates that judges in setting bail demanded lower probabilities of flight from minority defendants. Judges could have reduced bail amounts for minority males without incurring flight risks higher than those deemed acceptable for white male defendants.

   Moreover, our analysis avoids the problem of omitted variable bias. Bond rates provide a market-disciplined assessment of a defendant's probability of flight, given her bail amount. As a result, bond rates obviate the need to observe and measure defendant characteristics which, in traditional discrimination studies, serve as indirect proxies for the defendant's flight probability.

   Knowledge about bond prices substitutes for the traditional requirement that the researcher control for everything that might have affected courts' decisions. Specifically, evidence that bond dealers charge blacks lower rates makes it implausible to contend that the black defendants in our sample had a higher propensity to flee and therefore needed higher bail amounts to induce a sufficiently low probability of flight. Put simply, evidence of lower market bond rates for blacks suggests that courts could set lower bail for black defendants without incurring abnormally high risks of flight.

   In this article, we emphasize that the racial disparity in bail amounts and bail bond rates might be explained by a number of alternative, nondiscriminatory rationales. But the presence of market information, at the very least, shifts the grounds of debate. In the face of this evidence, one cannot plausibly argue that our failure to control for certain variables induced the racial disparity, nor can one argue that this disparity serves the purpose of the law.

   Thus, while observers of the criminal justice system have traditionally regarded bond dealers with scorn, their pricing behavior, when disciplined by competition, is useful in evaluating government bail setting procedures.

   Moreover, our study suggests that bond dealers in competitive markets may provide an additional service of mitigating the effects of state discrimination. Although bail amounts for black male defendants average 35 percent higher than for white male defendants, the bond fees that black males pay are only 16.5 percent higher. For all its faults, this pariah industry may be responsible for mitigating more than half of the effects of racially disparate bail setting.

   This article has four parts. Part I describes the bail bonding system in Connecticut and provides a theory for the actions of courts, bond dealers, and defendants. Part II describes our empirical results. We find evidence that: (1) the bail bond market in New Haven, Connecticut, is reasonably competitive; (2) bail amounts for black and Hispanic male defendants are significantly higher than for white male defendants; and (3) bail bond rates for black and Hispanic male defendants are significantly lower than for their white counterparts. We argue that if three crucial assumptions are valid, the market test of discrimination provides evidence that bail setting has an unjustified disparate impact on minority males. Part III explores several ways that each of these three assumptions might fail and provides alternative, nondiscriminatory explanations for our results. Part IV argues that, while market competition may help to mitigate unjustified judicial bias, complete reliance on remedial market correction places too much confidence in the resilience of competitive forces, given the evidence currently available.

   In the concluding section, we suggest some of the broader implications of our findings and methodology, along with possible applications to other market settings.