The booker decision and discrimination in federal criminal sentences.

Nutting, Andrew W.

I. INTRODUCTION

The Sentencing Reform Act, passed by Congress in 1984, established the federal sentencing guidelines. The guidelines were designed to improve standardization in punishments handed out to convicted criminals by limiting what had been federal judges' nearly unfettered discretion in sentencing (Anderson, Kling, and Stith 1999; Stith and Cabranes 1998). For federal crimes committed after November 1, 1987, the Act established that sentences of incarceration were to be determined by a table mapping the severity of the offense (a level between 1 and 43) and the defendant's criminal history (a level between I and VI) to a potential range of months imprisonment. Table 1 shows the United States Sentencing Commission's (USSC) Sentencing Table. A defendant, for example, with a "final offense level" of 20 and a "criminal history category" of IH should, according to the guidelines, be sentenced to somewhere between 41 and 51 months in federal prison. (The actual sentence would be an integer within that range). (1) The Act also abolished parole for federal crimes, though up to 15% of an inmate's sentence could be reduced for good behavior. (2)

The Sentencing Reform Act declared that Table 1's guideline ranges were "mandatory"--federal judges had to incur a nontrivial cost to sentence a defendant outside them. But on January 12, 2005, the United States Supreme Court ruled in the two-part decision U.S. vs. Booker that (a)the sentencing guidelines violated a defendant's constitutional fight to a jury trial; (3) and (b) in order to maintain constitutionality, the guidelines were to be "effectively advisory" instead of mandatory. (4) Legal observers had expected the overturning of the federal sentencing guidelines system since June 2004, when the Supreme Court decision Blakely v. Washington invalidated Washington state's sentencing guidelines system because it violated a defendant's right to a jury trial (Denniston 2004).

By making the sentencing guidelines advisory rather than mandatory, Booker reduced the cost of sentencing outside the guidelines and increased the ambiguity of federal judges' sentencing decisions. According to Bertrand, Chugh, and Mullainathan (2005), "ambiguity" is one of three situations in which implicit discrimination is likely to occur. (The other two situations are "inattentiveness to task" and "time pressure.") Implicit discrimination, unlike taste-based discrimination (Becker 1957) or statistical discrimination (Aigner and Cain 1977), reflects implicit attitudes that Bertrand, Chugh, and Mullainathan (2005) state are "unintentional and outside of the discriminator's awareness." Research indicating implicit discrimination include findings that National Basketball Association referees call fewer fouls against players of their own race (Price and Wolfers 2010), that Major League Baseball umpires are more likely to call strikes when the pitcher is of their own race (Parsons et al. 2007), and that screeners of resumes are more likely to ignore job applications with "black" names than ones with "white" names (Bertrand and Mullainathan 2004). Each of these scenarios describes a situation where ambiguity, inattentiveness, or time pressure is a factor, and where a statistically significant amount of discriminatory treatment is detected. (5)

This paper tests whether the Booker decision, by increasing the ambiguity of federal judges' sentencing decisions, increased manifestations of discrimination in accordance with the hypothesis of Bertrand, Chugh, and Mullainathan (2005). Difference-in-difference results show that conditional on a defendant's Table 1 guideline cell, Booker reduced sentences significantly for the default group of defendants--white, male, U.S. citizens who never earned a high school diploma. Female defendants and defendants with terminal high school degrees received significantly larger Booker-related sentence reductions and college graduates received significantly smaller ones. Evidence that blacks received smaller reduced guideline-conditional sentence reductions than whites is weaker. All of these results are consistent with implicit discrimination. There is also some weak evidence that Booker resulted in significant guideline-conditional changes of blacks' substantial assistance sentences, even though Booker did not change the law concerning substantial assistance sentences. This may indicate that Booker spilled over into substantial assistance sentencings.

When estimations account for judges' control over offense levels, evidence that post-Booker sentences fell for the default group and for women weakens substantially. This suggests that, for women and the default group, final offense levels changed post-Booker in the direction opposite their guideline-conditional sentence changes. A possible explanation is that judges, who pre-Booker had been known to manipulate the sentencing guidelines to achieve desired (usually shorter) sentences, implicitly adopted a new sentencing system after Booker loosened the guidelines. Booker lowered the cost of sentencing outside the guidelines, so judges could let offense levels rise and then deliver lower sentences conditional on the guidelines. Therefore, the composition of sentences significantly changed, but overall length of sentences for women and the default group did not change much. Results showing that relative sentences fell for high school graduates and rose for college graduates post-Booker remain significant when accounting for judges' control over offense levels.

This paper is organized as follows. Section II discusses the federal sentencing guidelines and reviews research regarding discrimination under the guidelines regime. Section III discusses data. Section IV describes empirical methods. Section V describes results. Section VI concludes.

II. THE FEDERAL GUIDELINES (6) AND DISCRIMINATION IN SENTENCING

In the federal system, a defendant who has been convicted of a specific federal offense has his sentence determined by a judge at a sentencing hearing, which takes place after the criminal conviction. The criminal conviction yields statutory minimum and maximum sentences that are exogenous to the judge's actions at the sentencing hearing.

At the sentencing hearing, the judge determines the sentence using three factors: the final offense level, which captures the severity of the defendant's crime; the criminal history category, which captures the defendant's criminal history; and what I call the guideline-conditional sentence, which is the sentence a judge gives a defendant after the final offense level and criminal history category have been determined.

The severity of a crime is captured by the final offense level, a score between 1 and 43. (7) More severe offenses have higher final offense levels. The final offense level has three components: the base offense level, specific offense characteristics (SOCs), and adjustments. The base offense level is an offense level score that corresponds with the specific crime of which a defendant has been convicted beyond a reasonable doubt, either through trial or guilty plea. For some crimes, the base offense level is explicitly determined by the conviction and is therefore exogenous to judicial control at the sentencing hearing. For example, insider trading has a base offense level of 8, racketeering has a base offense level of 19, and treason has a base offense level of 43. For other crimes, base offense level is endogenous to judicial control at the sentencing hearing. For defendants convicted of drug trafficking, for example, the sentencing judge determines how much drugs (in weight) the defendant trafficked and this weight is mapped to a base offense level using the guidelines' "Drug Quantity Table."

After determining the base offense level, a judge adds or subtracts SOCs and adjustments. These are added and subtracted based on the facts of the case by the legal standard of "preponderance of the evidence," and are endogenous to judicial control at the sentencing hearing. (8) SOCs are factors that can affect the offense level of certain, but not all, crimes. A defendant convicted of fraud, for example, has more points added to his offense level if he defrauded victims of more money, but a defendant convicted of manslaughter--a crime unrelated to money--could not face that money-related SOC. Adjustments are similar to SOCs, but are potentially applicable to all federal crimes. Examples include whether the defendant obstructed justice, the size of his role in a conspiracy, and whether he accepted responsibility for his actions. (9)

Although the sentencing judge is charged with determining final offense levels, he acts upon the advice of a federal probation officer, whose role entails determining the facts of a case and applying them to the guidelines (Stith and Cabranes 1998). In some cases, judges apply final offense levels that are jointly agreed upon by the prosecution and the defense in a practice referred to as "fact bargaining" (Stith and Cabranes 1998, 132). The dataset used in this paper does not discern whether an offense level change resulted from a probation officer's recommendation, a judge's imposition, or bargaining between the prosecution and the defense, (10) If Booker changed offense levels, it could be because of changes in the behavior of any of these actors, not only the judge. For the sake of simplicity, for the rest of this paper I refer to offense levels as being exclusively under the judge's control.

The second part of the Sentencing Table is the defendant's criminal history category, which is exogenous to judicial control at the sentencing hearing. (11) It is formed by assigning point values to previous criminal convictions and periods of incarceration. (12) The defendant's points are summed and assigned to the categories outlined in the columns of Table 1: 0-1 points is Category I, 2-3 points is Category II, etc.

Once the final offense level and criminal history category have been determined, they are mapped to a guideline range of months imprisonment using the Sentencing Table in Table 1. In this paper, I call the sentence chosen after final offense level and criminal history category have been determined the guideline-conditional sentence. Pre-Booker, the sentencing guideline ranges were deigned "mandatory," but federal judges were permitted to depart from them in certain situations as long as they stayed within the statutory minimum and maximum sentence. Judges who felt the guidelines mandated a range of imprisonment that inadequately reflected a defendant's criminal conduct could impose a downward departure (sentence a defendant to less than the lowest guideline sentence) or upward departure (more than the highest guideline sentence). Such departures imposed nontrivial costs upon judges. They had to be explained in writing and could not be justified by the defendant's race, creed, sex, education level, socioeconomic status, or by factors already covered by the base offense level, SOCs, or adjustments. After Booker made the guidelines "advisory," departures from the guidelines became less costly for judges.

Booker had no legal effect on substantial assistance departures from guideline ranges. Substantial assistance departures occur when government prosecutors inform a judge that the defendant cooperated with the government and provided evidence against at least one other individual. In such a circumstance, a federal judge is permitted to sentence a defendant beneath the guidelines, perhaps even giving him a sentence of no incarceration (Weinstein 1999). Substantial assistance departures became especially important with the Anti-Drug Abuse Act of 1986, which created mandatory minimum terms of imprisonment--usually 5 or 10 years, but also possibly 20 years or life without parole--for defendants convicted of violating specific drug trafficking laws. The Anti-Drug Abuse Act made it illegal for a judge to sentence a defendant below a mandatory minimum term of imprisonment in the absence of a prosecutor's substantial assistance motion. For example, suppose a defendant faced a guideline sentence of 70-87 months and stood convicted of a statute that mandated 60 months imprisonment. The sentencing judge could grant the defendant a downward departure, but could not sentence the defendant to less than the mandatory minimum sentence of 60 months unless prosecutors presented a substantial assistance motion. A "safety valve" provision included in the Violent Crime Control and Law Enforcement Act of 1994 made defendants with no more than one criminal history point and no violent history, who had not played a high-level role in a criminal conspiracy, and who had cooperated with the government--even if that cooperation did not provide new information that aided in the prosecution of another--eligible for sentences below the mandatory minimum term of imprisonment.

The sentencing guidelines and the rich dataset that the USSC has developed in accordance with them have been used by researchers to explore demographic discrimination in sentencing. Mustard (2001) controls for each of the 258 final guideline cells in the Sentencing Table and finds that among federal defendants sentenced when the judge exerts more guideline-conditional control-either because of a downward departure or a substantial assistance departure (the groups are combined into a single estimation)--blacks and Hispanics receive significantly longer sentences than whites by 4.3 and 2.4 months, respectively. Among defendants sentenced within the guidelines, the respective differences are 2.2 and -0.9 months (both significant). Women receive 6.9-month shorter sentences than men when judges depart from the guidelines, but only 1.8month shorter sentences when judges adhere to them. High school dropouts receive significant 1.2-month longer sentences than better-educated defendants when judges depart from the guidelines, but insignificant 0.2-month shorter sentences when judges adhere to them. That blacks, Hispanics, males, and high school dropouts experience relatively harsher sentences when judges have more sentencing discretion suggests that judges disfavor these groups, and that the guidelines rein in judges' discrimination. Studies of the federal sentencing guidelines in other disciplines (Albonetti 1997; Steffensmeier and Demuth 2000) show similar results. Schanzenbach (2005), incorporating district-level information on judicial demographics, finds that male judges sentence female defendants more leniently than female judges do, but that discrimination against particular races does not vary with the race of judges. Schanzenbach and Tiller (2007) show that federal judges appointed by Democratic presidents sentence defendants convicted of violent crimes and drug crimes to shorter sentences than those appointed by Republican presidents. Using a different dataset, Abrams, Bertrand, and Mullainathan (2008) find that Cook County, Illinois judges exhibit significant heterogeneity in the extent to which race influences sentencing.

III. DATA

The data in this paper come from Monitoring Federal Criminal Sentences (MFCS), compiled annually by the USSC. I use data coveting federal criminal sentences from July 1, 2004 to July 31, 2005. All of these sentencings took place after Blakely v. Washington nullified the Washington state sentencing guidelines. If Blakely, the precursor to Booker, caused federal judges to alter sentences in anticipation of the federal guidelines being declared unconstitutional, limiting observations to post-Blakely decisions would better identify the effect of Booker on sentences. I remove from the analysis defendants who were convicted at trial and those sentenced to more than 50 years imprisonment or statutory life. Data include prison sentence in months, final offense level, base offense level, criminal history category, offense type, (13) statutory minimum and maximum sentence, whether the defendant received a safety-valve sentence reduction, whether the defendant received a substantial assistance departure, federal district (but not exact judge) of sentencing, (14) month and year (but not exact day) of sentencing, the role a defendant played in a criminal conspiracy, (15) whether the defendant used a weapon in his crime, (16) the type of drug a defendant trafficked or possessed, and the defendant's age, race, sex, citizenship status, education level, and number of dependents. (17) It also includes the relevant amendment year of the sentencing guidelines, i.e., the set of base offense levels/SOCs/adjustments under which the defendant was sentenced.

Table 2 shows summary statistics, separating defendants who received a substantial assistance departure from those who did not. Post-Booker sentences account for approximately 55% of both substantial assistance and non-substantial-assistance sentencings in the dataset. (18) Among non-substantial-assistance cases, the average sentence handed out was a month short of 4 years. Only 13% of defendants were women, but 45% were Hispanic, 23% were black, and 31% were illegal aliens. Fewer than half had at least a high school diploma.

Sentences in substantial assistance cases were slightly longer than those in non-substantial-assistance cases, and their defendants faced higher offense levels and statutory minimum and maximum sentences. (19) Women, blacks, and better-educated defendants were more common in substantial assistance cases than other cases, while Hispanic and illegal alien defendants were less common.

IV. ESTIMATION STRATEGY

Where i represents individual defendant, t represents time, and d represents federal district, I estimate the equation

(1)

ln(1 + [M.sub.itd]) = [lambda] [X.sub.itd] + [[theta].sub.d] + [[eta].sub.t] + [psi] [TIME.sub.t] x [X.sub.itd] + [gamma] [Z.sub.itd] + [rho] [BOOKER.sub.itd] + [omega] [BOOKER.sub.itd] x [X.sub.itd] + [phi] (C.sub.itd]) + [[epsilon].sub.itd].

[M.sub.itd] is months imprisonment. Adding one to [M.sub.itd] identifies the log of sentences of 0 months imprisonment. [X.sub.itd] represents demographic controls capturing a defendant's age, sex, race, citizenship status, education level, and number of dependents. [[theta].sub.d] represents fixed effects for federal district, allowing severity of sentence to vary with each district's judges. A defendant's assigned sentencing judge, which is unobserved, is assumed to be random conditional on district, [[eta].sub.t] represents fixed effects for the month and year of sentencing. Its inclusion captures potential changes in sentences caused by factors other than Booker, such as more conservative judges being appointed by the George W. Bush administration (Munson 2007; Schanzenbach and Tiller 2007). [TIME.sub.t] x [X.sub.itd] is an interaction of each demographic group with a linear monthly time trend. This permits each group's relative sentence to change over time for reasons other than Booker. (Age and number of dependents controls are not included in this set of interactions.)

[Z.sub.itd] represents controls correlated with a defendant's crime that are largely exogenous to judicial control at the sentencing hearing. It includes controls for the defendant's statutory minimum and maximum sentences, (20) type of drug trafficked or possessed (if any), and whether a defendant qualified for a safety-valve reduction. To control for offense-specific USSC amendments to the guidelines, and to control for changes in judicial attitudes that may have accompanied such changes, [Z.sub.itd] includes interactions between offense type dummies and amendment year dummies. (21) In some estimations (which will be discussed shortly) it includes dummies for role in a conspiracy and use of a weapon, which may be endogenous to judicial control at the sentencing hearing. (22) [BOOKER.sub.itd] is a dummy equal to 1 if the defendant was sentenced after Booker and 0 otherwise. (Throughout this paper, BOOKER refers to the dummy variable in Equation (1) and Booker refers to the Supreme Court decision U.S.v. Booker.) The coefficient on [BOOKER.sub.itd] captures how Booker affected sentences of the default group--white male U.S. citizens with less than a high school education. The coefficients on the interaction term [BOOKER.sub.itd] x [X.sub.itd] show whether Booker altered the sentences of different race, gender, citizenship, and education level groups differently. (I do not include age or number of dependents controls in this interaction term either.) If the increased ambiguity of federal sentencing brought about by Booker affected sentences differently for different demographic groups, it would indicate that judges possessed implicit discrimination as described by Bertrand, Chugh, and Mullainathan (2005).

[PHI] ([C.sub.itd]) represents controls for the federal sentencing guidelines. Section II outlined that some sentencing guidelines factors--mainly SOCs and adjustments, but also base offense levels in certain circumstances--are endogenous to the sentencing hearing. Thus these factors may be a function of the sentencing judge's discrimination, and they too might have changed with Booker. Including them as right-hand-side controls could potentially yield biased coefficients on BOOKER and its interaction terms. (23) Therefore, I estimate Equation (1) using three different versions of [PHI]([C.sub.itd]).

In the first version, [PHI]([C.sub.itd]) follows Mustard (2001) and consists of fixed effects for final guideline cell, i.e., a vector of 258 dummy variables representing each of the final guideline cells shown in Table 1. When the guidelines are accounted for in this fashion, the coefficients on [BOOKER.sub.itd] and its interactions capture the guideline-conditional effect of Booker on sentences, i.e., how Booker affected judges' sentences assuming Booker did not affect final offense levels.

Final offense levels, though, are endogenous to judicial control at the sentencing heating. Thus, they may be a function of judicial discrimination and controlling for them may yield biased coefficients on BOOKER and its interactions. Therefore, like Schazenbach and Tiller (2007), I use two other versions of [PHI]([C.sub.itd]). One includes controls for what I call baseline cell instead of final guideline cell. This technique, which will be discussed more thoroughly in the next section, captures the effect of Booker assuming that base offense level is beyond the control of the sentencing judge and exogenous to judicial discrimination.

In the last version, base offense level itself--which in certain circumstances is determined by the judge at the sentencing hearing--is allowed to be endogenous to judicial control and subject to judicial discrimination. In these estimations, [PHI]([C.sub.itd]) omits offense level and only includes criminal history category dummies. In these estimations, all right-hand-side controls are exogenous to the judge at the sentencing hearing, and the coefficients on BOOKER and its interactions capture the total effect Booker had on sentences.

When [PHI]([C.sub.itd]) controls for baseline cell or criminal history category only, role in the conspiracy dummies and weapons dummy are removed from [Z.sub.itd] because they are endogenous to the sentencing hearing. All estimations of Equation (1) include standard errors robust to heteroskedasticity. Since I am examining whether judges' sentences changed with Booker, ideally standard errors would be clustered by individual judge. Since judge is unobserved, I cluster standard errors by district.

V. RESULTS

A. Controlling for Final Guideline Cell

Table 3 shows coefficients from guideline-conditional Equation (1) estimations on nonsubstantial-assistance cases. Estimations include 258 dummy variables controlling for Table 1 final guideline cell. Column 1 contains all observations from July 1, 2004 through July 31, 2005. BOOKER is significantly negative, indicating that the mean guideline-conditional sentence for white, male, U.S. citizen, high school dropout federal defendants who pleaded guilty without a substantial assistance departure fell 7.2% (= exp(-0.075)- 1) after Booker. The significantly negative coefficient is unsurprising. Prior to Booker judges granted many more downward departures than upward departures (Mustard 2001) and many judges publicly complained about the severity of guideline sentences (Bradley 2004; Caher 2006; Gustafson 2002). Booker, by reducing the cost of sentencing outside the guidelines, allowed judges to increase their issuance of downward departures. Coefficients on interaction terms show that Booker reduced the guideline-conditional sentences of females by 8.6% more than those of males and sentences of high school graduates by 7.7% more than high school dropouts. Recall that Mustard (2001) showed that, prior to Booker, females and better-educated defendants were sentenced relatively more leniently when judges possessed more sentencing discretion. The BOOKER x FEMALE and BOOKER x HS_GRAD coefficients are consistent with Mustard (2001) and support the idea that sentences became more reflective of judges' implicit discrimination when Booker increased the ambiguity of guideline-conditional sentences.

My next estimations narrow the number of observations around Booker to strengthen the results of the difference-in-difference estimation (Bertrand, Duflo, and Mullainathan 2004; Black 2003). Column 2 drops July 2004 and July 2005 sentencings from the observations, so that the sample now consists of an 11-month window around the January Booker decision. BOOKER x BLACK becomes significantly positive at the 10% level, indicating that judges may possess implicit discrimination against blacks. This too is consistent with Mustard (2001). BOOKER x FEMALE and BOOKER x HS_GRAD maintain their significance and continue to indicate implicit discrimination. Columns 3-7 continue shaving one pre-Booker and one post-Booker month from the sample, so that Column 3 includes sentences from September 1, 2004 through May 31, 2005, Column 4 includes sentences from October 1, 2004 through April 30, 2005, etc. The coefficient on BOOKER is significantly negative in six of the seven columns--including Column 7, when only January 2005 observations are included--strongly suggesting that Booker significantly reduced guideline-conditional sentences for the default group. BOOKER x FEMALE is significantly negative in six columns, providing strong evidence that judges implicitly favor female defendants. BOOKER x BLACK is always positive, but only significant in Columns 2 and 6, so the evidence that judges disfavor blacks is not particularly strong. BOOKER x H S_G R A D is significantly negative in Columns 1-5, but BOOKER x SOME COLLEGE and BOOKER x COLLEGE_GRAD are never significantly negative. This suggests that judges implicitly favor defendants who graduate from high school, but not those who attend or graduate from college. In Columns 6-7, BOOKER x COLLEGE_GRAD is significantly positive, indicating that judges may actually disfavor college graduates who go on to commit federal crimes. The sum of BOOKER and BOOKER x COLLEGE_GRAD is not significant in either column, though, so college graduates did not experience significant guideline-conditional sentence increases after Booker. There is also some evidence, in Columns 3-5, that judges disfavor legal aliens.

Table 4 repeats Table 3 but performs the estimation on defendants receiving substantial assistance departures. Recall that Booker did not alter judicial control over these sentences. Most of the coefficients that were significant in Table 3 are not significant in Table 4. BOOKER and BOOKER x FEMALE are negative and often large, but insignificant in all columns. BOOKER x BLACK is significantly positive in two of the seven columns. The general insignificance of the results in Table 4 is expected, given that Booker did not change laws concerning substantial assistance sentences. But that some of the BOOKER x BLACK coefficients are significant, despite the estimations' small sample sizes and the utter lack of any legal change to substantial assistance sentencings, is somewhat surprising. It could indicate spillovers into substantial assistance cases. Substantial assistance sentences account for a minority of sentencings (16.2% of this dataset, which omits convictions at trial) and are determined by the very same judges who determine non-substantial-assistance sentences, so spillovers are a possibility. (24)

Even though most of the results in Table 4 are insignificant, the large coefficients make it difficult to present a "difference-in-difference-in-difference" test of Booker's impact on discrimination in sentencing. Such a test would have used non-substantial-assistance cases as a treatment group and substantial assistance cases as a control group. If substantial assistance cases changed during the treatment period, then they do not form a valid control group, rendering the test invalid. (25)

B. Controlling for Baseline Cell

The guideline-conditional estimations in Tables 3 and 4 assume Booker had no effect on sentencing judges' determination of defendants' final offense levels. But sentencing judges exert control over SOCs and adjustments (Schanzenbach and Tiller 2007), so final offense level is endogenous to the sentencing hearing and may be a function of the sentencing judge's discrimination. Booker may have changed judges' application of SOCs and adjustments differently for different demographic groups. If it did so, the Table 3 guideline-conditional estimates of the coefficients on BOOKER and its interactions would be biased.

To correct for this possible bias, I estimate in this subsection whether Booker altered what I call baseline-conditional sentences. Tables 3-4 employed fixed effects for final guideline cell, i.e., the interaction of final offense level and criminal history category. Estimations in this subsection instead use fixed effects for the interaction of base offense level and criminal history category. (26) These estimations assume that base offense level--which is a function of a defendant's criminal conviction--is beyond a sentencing judge's control and therefore exogenous to judicial bias. Judges' applications of SOCs and adjustments, as well as guideline-conditional sentences, are allowed to be functions of their bias. (27)

Table 5 shows results of baseline-conditional estimations when the dependent variable is log months imprisonment. Table 5 coefficients on BOOKER and its interactions capture the combined effect Booker had on SOCs/adjustments and guideline-conditional sentences. (28) BOOKER is small and insignificant in all seven columns, suggesting that Booker did not decrease baseline-conditional sentences for the default group, even though it decreased guideline-conditional sentences. (29) BOOKER x BLACK is never significantly positive, and BOOKER x FEMALE is only significantly negative in Column 6. So while Table 3 showed evidence that judges' implicit attitudes toward women, blacks, and the default group significantly affected their guideline-conditional sentences, there is little evidence in Table 5 that these groups' baseline-conditional sentences changed. Presumably, changes in final offense levels counteracted changes in guideline-conditional sentences. The significantly negative coefficients on BOOKER x HS GRAD in Columns 1-5 again show that sentences for defendants with terminal high school degrees fell post-Booker, and evidence that sentences increased for defendants with college degrees is actually stronger than it was in Table 3. (30)

In Table 6, I re-estimate the equations from Table 5 but change the dependent variable to final offense level. These estimations capture whether Booker differently impacted the final offense levels of different demographic groups conditional on their base offense level, i.e., they show whether Booker significantly changed judges' applications of SOCs and adjustments differently for different groups. Table 6 shows that Booker increased, but not significantly, the default group's final offense level. BOOKER x FEMALE is insignificantly positive in Columns 1-5 and insignificantly negative in Columns 6-7. Although these results are insignificant, any changes in SOCs and adjustments apparently were substantial enough to change the significant Table 3 guideline-conditional results into the largely insignificant baseline-conditional results in Table 5. The only significant coefficients in Table 6 show final offense level increases for other race (Columns 1-4) and college graduates (Columns 6-7). The significant increases in college graduates' final offense levels contributed to the significant increases in their Table 5 baseline-conditional sentences.

C. Omitting Offense Level Controls

Schanzenbach and Tiller (2007) note that defendants convicted of certain crimes (most importantly drug trafficking) have their base offense levels subject to judicial control at the sentencing hearing. Booker may have changed judges' application of base offense levels as well as their application of SOCs, adjustments, and guideline-conditional sentences. To fully test for the effect Booker had on sentences, this section performs offense-level-unconditional estimations. These estimates omit all controls for final offense level and base offense level. [PHI] ([C.sub.itd]), the Equation (1) control for the federal guidelines, consists only of criminal history category dummies. Every right-hand-side variable is now exogenous to the judge's actions at the sentencing hearing.

Table 7 shows results when the dependent variable is log months imprisonment. The findings largely reaffirm the baseline-conditional estimations of Table 5. BOOKER and BOOKER x FEMALE are insignificant in all columns. The only groups with significant coefficients are high school graduates (significantly negative in Columns 1-5) and college graduates (significantly positive in Columns 6-7). No coefficients are significant when the dependent variable is changed to final offense level (Table 8).

To summarize Tables 3-8: Booker resulted in guideline-conditional sentence reductions that were significant for the default group, significantly larger for women and high school graduates, and significantly smaller for college graduates and (perhaps) blacks. This suggests implicit discrimination among judges. When accounting for the fact that judges exert control over final offense level, there remains evidence that Booker reduced sentences for high school graduates and increased them for college graduates, but there is little evidence that Booker resulted in significant overall sentence reductions for the default group or for women. This latter result suggests that changes in these groups' final offense levels counteracted falls in their guideline-conditional sentences, even though increases in their final offense levels were not statistically significant.

D. Explaining Results

Tables 3, 5, and 7 provide fairly persistent evidence that sentencing under Booker was associated with relatively shorter prison sentences for high school graduates and relatively longer sentences for college graduates. Results for the default group and for female defendants, who both experienced decreases in guideline-conditional sentences but largely insignificant changes in baseline-conditional and offense-level-unconditional sentences, are more complex. Leipold (2005) and Weinstein (1992) may help. These papers claim that, prior to Booker, some judges determined offense levels for personal reasons and not the legal standard of "preponderance of evidence." Oftentimes, they intentionally kept an offense level low in order to shorten what they perceived to be a defendant's unjustly long prison sentence. Leipold (2005) writes that judges had a "tendency to be influenced by the likely consequences when making decisions formally unrelated to the punishment," and quotes Weinstein (1992), who writes that some judges, "with respect to facts that, if proved, would significantly enhance a defendant's sentence ... would require clear and convincing evidence or, in some cases, proof beyond a reasonable doubt" instead of just preponderance of the evidence.

Booker may have reduced judges' need to manipulate offense levels in this fashion. When Booker made guidelines advisory, judges did not have to remain as concerned about keeping the final offense level low, because the "advisory" guidelines were less determinant of prison sentence. Prior to Booker a sentencing judge may have spent much time keeping an offense level low for a white male defendant, so that he did not have to give him a long sentence or have to justify giving him a downward departure in writing. After Booker, the judge could let the offense level for this defendant increase, knowing he could more easily sentence him to a below-guideline sentence. The observed guideline-conditional sentence could significantly fall, but the observed baseline-conditional and offense-level-unconditional sentences would insignificantly differ from their pre-Booker levels.

VI. CONCLUSION

This paper shows that the Supreme Court's decision in U.S. v. Booker, which rendered the federal sentencing guidelines "advisory" rather than "mandatory," resulted in significant decreases for the default group and female defendants conditional on their final guideline cells. There is also evidence that guideline-conditional sentence reductions were, compared to high school dropouts, larger for terminal high school graduates and smaller for college graduates. Weaker evidence indicates that Booker led to larger guideline-conditional sentence reductions for whites than blacks. These results suggest implicit discrimination among federal judges. Estimations on substantial assistance sentencings are largely insignificant, but show some weak evidence that they too changed post-Booker, even though Booker did not legally affect substantial assistance sentencings. This may have occurred because of judicial spillovers.

When accounting for the fact that sentencing judges exert control over offense levels, the evidence that Booker decreased sentences for the default group and for women largely disappears. This indicates that post-Booker increases in offense levels counteracted guideline-conditional sentence reductions for these groups, even though many effects of Booker on offense levels per se are insignificant. A possible explanation is that judges (or other actors who impacted offense levels), in response to Booker's loosening of the guidelines, became less concerned with manipulating offense levels to keep sentences low. Evidence that high school graduates experienced significant sentence decreases and college graduates experienced significant sentence increases remains when allowing final offense level to vary with judicial bias, though.

Note that this paper examines the effect of Booker treating the offense type of criminal conviction as exogenous. Since there is both anecdotal (Powell 2004) and statistical (Bjerk 2005) evidence that the severity of criminal charge and offense type is subject to prosecutorial discretion, it is possible that the offense type is itself a function of prosecutorial discrimination. Further research could study whether Booker, or similar changes in judicial control over sentencing, affected the type of charges prosecutors pursued against defendants. Research could also determine whether clarifications of Booker brought about by three cases in 2007 (Baron-Evans 2008) revealed any discrimination.

ABBREVIATIONS

MFCS: Monitoring Federal Criminal Sentences

SOCS: Specific Offense Characteristics

USSC: United States Sentencing Commission

doi: 10.1111/j.1465-7295.2011.00449.x

REFERENCES

Abrams, D., M. Bertrand, and S. Mullainathan. "'Do Judges Vary in Their Treatment of Race?" Working Paper, 2008.

Aigner, D. J., and G. G. Cain. "Statistical Theories of Discrimination in Labor Markets." Industrial and Labor Relations Review, 30(2), 1977, 175-87.

Albonetti, C. A. "Sentencing under the Federal Sentencing Guidelines: Effects of Defendant Characteristics, Guilty Pleas, and Departures on Sentence Outcomes for Drug Offenses, 1991-1992." Law, & Society Review, 31 (3), 1997, 789-822.

Anderson, J. M., J. R. Kling, and K. Stith. "Measuring Interjudge Sentencing Disparity: Before and After the Federal Sentencing Guidelines." Journal of Law and Economics, 42(1), 1999, 271-307.

Baron-Evans, A. "Rita, Gall and Kimbrough: A Chance for Real Sentencing Improvements." Office of Defender Services, 2008. Accessed May 11, 2008. www.fd.org.

Becker, G. The Economics of Discrimination. Chicago: The University of Chicago Press, 1957.

Bertrand, M., D. Chugh, and S. Mullainathan. "Implicit Discrimination." American Economic Review, 95(2), 2005, 94-98.

Bertrand. M., E. Duflo, and S. Mullainathan. "How Much Should We Trust Difference-in-Difference Estimates?" Quarterly Journal of Economies, 119(1), 2004, 249-75.

Bertrand, M., and S. Mullainathan. "Are Emily and Greg More Employable than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination." American Economic Review, 94(4), 2004, 991-1013.

Bjerk, D. "Making the Crime Fit the Penalty: The Role of Prosecutorial Discrimination under Mandatory Minimum Sentencing." Journal of Law and Economies, 48(2), 2005, 591-625.

Black, S. "Do Better Schools Matter? Parental Valuation of Elementary Education." Quarterly Journal of Economics, 114(2), 2003, 577-99.

Bradley, E. (correspondent). "More Than They Deserve." 60 Minutes (Television Program), January 4. 2004.

Caher, J. "Federal Judge Blasts Mandatory Minimum Sentences." The New York Law Journal, January 24, 2006.

Denniston, L. "Judges Agree to Consider Sentencing." The New York Times, August 2, 2004, p. 14.

Gustafson, P. "Judge Criticized for Comments about Drug Sentences; James Rosenbaum Drew the Ire of U.S. House Republicans for His Testimony." The Star-Tribune, November 9, 2002, p. 9B.

Leipold, A. D. "Why Are Federal Judges So Acquittal Prone?" Washington University Law Review, 83(1), 2005, 150-227.

Munson, L. "'Judge's Memoir, Case Work Offer a Look into How He'll Handle Vick Sentencing." ESPN.com, November 20, 2007.

Mustard, D. B. "Racial, Ethnic, and Gender Disparities in Sentencing: Evidence from the U.S. Federal Courts." Journal of Law and Economics, 44(1), 2001,285-314.

Parsons, C. A., J. Sulaeman, M. Yates, and D. S. Hamermesh. "Strike Three: Umpires' Demand for Discrimination."

NBER Working Paper, W13665, 2007. Powell, C. "J. Lewis Agreement Struck; Raven Would Serve No Prison Time during Season." The Washington Post October 2, 2004, E-9.

Price, J., and J. Wolfers. "Racial Discrimination among NBA Referees." Quarterly Journal of Economics, 125(4), 2010, 1859-87.

Schanzenbach, M. "Racial and Sex Disparities in Prison Sentences: The Effect of District-Level Judicial Demographics." Journal of Legal Studies, 32(2), 2005, 57 -92.

Schanzenbach, M. M., and E. H. Tiller. "Strategic Judging under the U.S. Sentencing Guidelines: Positive Political Theory and Evidence." Journal of Law, Economies, and Organization, 23(1), 2007, 24-56.

Steffensmeier, D., and S. Demuth. "Ethnicity and Sentencing Outcomes in U.S. Federal Courts: Who Is Punished More Harshly?" American Sociological Review, 65(5), 2000, 705-29.

Stith, K., and J. A. Cabranes. Fear of Judging: Sentencing Guidelines in the Federal Courts, Chicago: University of Chicago Press, 1998.

Weinstein, J. B. "A Trial Judge's Second Impression of the Federal Sentencing Guidelines." Southern California Law Review, 66, 1992, 357-66.

Weinstein, I. "Regulating the Market for Snitches." Buffalo Law Review, 47, 1999, 563-632.

(1.) A defendant in Zone A of the Sentencing Table is permitted to avoid imprisonment. One in Zone B must receive at least 1 month of imprisonment but can serve the rest of his sentence via house arrest or in a halfway house. A defendant in Zone C is required to serve at least half his sentence in prison.

(2.) The 15% reduction only applies to sentences of at least 13 months imprisonment.

(3.) This was a 5-4 decision written by Justice John Paul Stevens with Justices Ruth Bader Ginsburg, Antonin Scalia, David Souter, and Clarence Thomas concurring.

(4.) This was a separate 5-4 decision written by Justice Steven Breyer with Chief Justice William Rehnquist and Justices Ruth Bader Ginsburg. Anthony Kennedy, and Sandra Day O'Connor concurring.

(5.) Bertrand, Chugh+ and Mullainathan (2005) more thoroughly review the implicit discrimination literature.

(6.) Much of the analysis here is taken from Chapters 2-6 of the United States Sentencing Commission Guidelines Manual. 2003 edition.

(7.) A final offense level under 1 is amended to 1; one above 43 is amended to 43.

(8.) "Preponderance of the evidence" is a weaker standard than "beyond a reasonable doubt." This difference heavily influenced the Supreme Court's rejection of the sentencing guidelines in Booker. The Court ruled 5-4 that a defendant's sixth amendment right to a jury trial was violated when judges increased sentences by invoking SOCs and adjustments. Justice Steven Breyer wrote the compromise that led to the guidelines being declared "advisory." His decision was criticized in dissents by Justice John Paul Stevens, who wrote, "[T]he court's creative remedy is an exercise of legislative, rather than judicial, power," and Justice Antonin Scalia, who wrote, "In order to rescue from nullification a statutory scheme designed to eliminate discretionary sentencing, it discards the provisions that eliminate discretionary sentencing."

(9.) The USSC has the authority to create base offense levels for new criminal offenses, create new SOCs and adjustments, and alter existing base offense levels, SOCs, and adjustments. Amendments the USSC has made to the guidelines are available at ussc.gov.

(10.) The dataset used in this paper, which will be described in detail in Section III, did not contain variables in fiscal years 2004 and 2005 showing differences between the probation officer's recommended offense level and the defendant's final offense level. It also did not contain any variable dictating whether a defendant's final offense level was determined via "fact bargaining." In fiscal year 2006, it added variables dictating whether the judge agreed with or changed the probation officer's recommended guideline applications. There remains no variable regarding "fact bargaining."

(11.) Schanzenbach and Tiller (2007) write, "The criminal history category is more or less set by past judicial determinations."

(12.) Each prior prison sentence of at least 13 months is assigned 3 points. Each prior prison sentence of between 2 and 13 months, and each offense committed on parole, probation, or under court supervision, is assigned 2 points. Most other convictions are assigned 1 point.

(13.) There are 32 different offense types. The most common are drug trafficking (37.7% of observations), immigration violations (23.3%), firearms offenses (12.6%), and fraud (7.9%). No other offense makes up more than 3% of the dataset.

(14.) There are 96 federal districts covering all 50 states and territories. Each district is assigned to only one state/territory. There can be multiple districts in a state: California, New York, and Texas each have four.

(15.) Role in a conspiracy is determined by a judge at a sentencing hearing. There are seven possible rolein-the-offense levels: minimal (-4 adjustment), minimal/minor (-3), minor (-2), mid-level (0), leader/organizer/ manager/supervisor of a small criminal conspiracy (+2), manager/supervisor of a large criminal conspiracy (+3), and leader/organizer of a large criminal conspiracy (+4).

(16.) Weapons possession/use can either be a criminal conviction or a SOC.

(17.) Income and type of lawyer retained (e.g., private attorney, court-appointed counsel, federal public defender) were recorded in older versions of MFCS, but the USSC had stopped collecting this data by fiscal year 2005.

(18.) The fiscal year 2005 version of the MFCS contains a variable indicating whether a sentence was delivered before or after Booker. 1 assume all fiscal year 2004 sentences are pre-Booker.

(19.) Convicted drug traffickers account for over 86% of defendants who face a nonzero statutory minimum sentence. Weapons convictions account for another 9.3%.

(20.) Statutory minimum sentences are controlled for cubically, plus dummies for 12-month, 60-month, 180-month, 240-month, and life minimum sentences. (Defendants who face mandatory life sentences can avoid them via substantial assistance departures. The life dummy is not identified in estimations of defendants who do not receive such departures.) Statutory maximum sentences are controlled for cubically, plus dummies for every 12-month step between 12 and 60 months, every 60-month step between 60 and 720 months, and life sentences. Life minimum and maximum sentences are parametrically recorded as 0-month sentences in the cubic control.

(21.) All controls in Zim except those for statutory minimum and maximum sentences are dummy variables.

(22.) Over 90% of the dataset was sentenced according to the guidelines in effect in either 2003 or 2004. I remove all sentencings based on the guidelines from 2000 or before (4.4% of the dataset) and from 2005 (0.05% of the dataset) from the estimations.

(23.) This dilemma is somewhat similar to the dilemma over including occupation as a control when estimating the male-female wage gap (e.g., Blau and Kahn 1997). Women are typically located in lower-paying occupations than men. If occupational choice differs between men and women for reasons related to personal taste, it is exogenous to discrimination, and controlling for it appropriately helps identify the extent of discrimination against women in the labor market. If occupational choice differs because of labor market discrimination, including occupation as a control in wage equations understates the male/female wage gap resulting from discrimination.

(24.) Another explanation is that judges altered postBooker substantial assistance sentences for certain defendants in order to maintain a consistent price for their testimony against other defendants. The author has a working paper examining the market prices for coconspirator testimony.

(25.) Results of triple-differencing guideline-conditional estimations parallel to those in Tables 3 and 4 are available from the author upon request. Only one triple-difference interaction term--it involves legal aliens--is significant. The interaction term between BOOKER and the nonsubstantial-assistance dummy is always insignificant.

(26.) The correlation of base offense level and final offense level is 0.82.

(27.) MFCS explicitly records base offense level for defendants convicted of one criminal count. When a defendant is convicted of multiple counts, base offense level is determined by what the guidelines refer to as a combined adjusted offense level. This variable is also in MFCS.

(28.) The interaction of dummies for amendment year and dummies for offense type are especially critical when controlling for baseline cells, because defendants may have received different final offense levels depending on the guidelines in place when they were sentenced.

(29.) Permitting final offense levels to vary with Booker also ensures that the standard errors are larger in Table 5 than in Table 3, reducing the significance of some of the coefficients.

(30.) The sum of the coefficients on BOOKER and BOOKER x COLLEGE_GRAD is significantly positive at the 10% level in Columns 4-5 and at the 5% level in Columns 6-7, indicating that college graduates' baseline-conditional sentences increased as a result of Booker.

ANDREW W. NUTTING, The author wishes to thank seminar participants at the University of Idaho and Washington State University and anonymous referees for comments on earlier drafts. He also wishes to thank Jessica Murphy Manca and Melissa Reimer for information regarding the federal sentencing system. All remaining errors are his own.

Nutting: Department of Business, University of Idaho, 308 Albertson Building, Moscow, ID 83844. Phone 208-885-6204, Fax 208-885-5347, E-mail anutting@uidaho.edu

TABLE 1
Federal Sentencing Guidelines (All Ranges Are of Months Imprisonment)

 Criminal History Category

 I II III
 0-1 points 2-3 4-6

Final Offense Level

Zone A 1 0-6 0-6 0-6
 2 0-6 0-6 0-6
 3 0-6 0-6 0-6
 4 0-6 0-6 0-6
 5 0-6 0-6 1-7
 6 0-6 I-7 2-8
 7 0-6 2-8 4-10
 8 0-6 4-10 6-12
Zone B 9 4-10 6-12 8-14
 10 6-12 8-14 10-16
Zone C 11 8-14 10-16 12-18
 12 10-16 12-18 15-21
Zone D 13 12-18 15-21 18-24
 14 15-21 18-24 21-27
 15 18-24 21-27 24-30
 16 21-27 24-30 27-33
 17 24-30 27-33 30-37
 18 27-33 30-37 33-41
 19 30-37 33-41 37-46
 20 33-41 37-46 41-51
 21 37-46 41-51 46-57
 22 41-51 46-57 51-63
 23 46-57 51-63 57-71
 24 51-63 57-71 63-78
 25 57-71 63-78 70-87
 26 63-78 70-87 78-97
 27 70-87 78-97 87-108
 28 78-97 87-108 97-121
 29 87-108 97-121 108-135
 30 97-121 108-135 121-151
 31 108-135 121-151 135-168
 32 121-151 135-168 151-188
 33 135-168 151-188 168-210
 34 151-188 168-210 188-235
 35 168-210 188-235 210-262
 36 188-235 210-262 235-293
 37 210-262 235-293 262-327
 38 235-293 262-327 292-365
 39 262-327 292-365 324-405
 40 292-365 324-405 360-Life
 41 324-405 360-Life 360-Life
 42 360-Life 360-Life 360-Life
 43 Life Life Life

 Criminal History Category

 IV V VI
 7-9 10-12 134

Final Offense Level

Zone A 1 0-6 0-6 0-6
 2 0-6 0-6 1-7
 3 0-6 2-8 3-9
 4 2-8 4-10 6-12
 5 4-10 6-12 9-15
 6 6-12 9-15 12-18
 7 8-14 12-18 15-21
 8 10-16 15-21 18-24
Zone B 9 12-18 18-24 21-27
 10 15-21 21-27 24-30
Zone C 11 18-24 24-30 27-33
 12 21-27 27-33 30-37
Zone D 13 24-30 30-37 33-41
 14 27-33 33-41 37-46
 15 30-37 37-46 41-51
 16 33-41 41-51 46-57
 17 37-46 46-57 51-63
 18 41-51 51-63 57-71
 19 46-57 57-71 63-78
 20 51-63 63-78 70-87
 21 57-71 70-87 77-96
 22 63-78 77-96 84-105
 23 70-87 84-105 92-115
 24 77-96 92-115 100-125
 25 84-105 100-125 110-137
 26 92-115 110-137 120-150
 27 100-125 120-150 130-162
 28 110-137 130-162 140-175
 29 121-151 140-175 151-188
 30 135-168 151-188 168-210
 31 151-188 168-210 188-235
 32 168-210 188-235 210-262
 33 188-235 210-262 235-293
 34 210-262 235-293 262-327
 35 235-293 262-327 292-365
 36 262-327 292-365 324-405
 37 292-365 324-405 360-Life
 38 324-405 360-Life 360-Life
 39 360-Life 360-Life 360-Life
 40 360-Life 360-Life 360-Life
 41 360-Life 360-Life 360-Life
 42 360-Life 360-Life 360-Life
 43 Life Life Life

TABLE 2
Summary Statistics

 Non-Substantial-Assistance
 48,548 Observations

 Mean SD Min Max

Post-Booker 0.54 0.50 0 1
Total months imprisonment 46.9 56.4 0 600
Base offense level 16.8 9.6 0 43
Final offense level 17.5 8.1 1 43
Criminal history category 2.53 1.72 1 6
Statutory minimum-life
Statutory maximum-life 0.13 0.34 0 1
Statutory minimum sentence (a) 23.2 47.5 0 600
Statutory maximum sentence (b) 234.9 394.2 12 30,540
Safety valve 0.13 0.34 0 1
Age 34.2 10.3 18 88
Female 0.13 0.33 0 1
Number of dependents 1.66 1.74 0 16
Black 0.23 0.42 0 1
Hispanic 0.45 0.50 0 1
Other race 0.04 0.20 0 1
Legal alien 0.05 0.22 0 1
Illegal alien 0.31 0.46 0 1
Non-U.S. resident 0.02 0.13 0 1
High school graduate 0.30 0.46 0 1
Some college 0.14 0.35 0 1
College graduate 0.04 0.20 0 I
Weapon 0.08 0.26 0 1
Minimal role 0.02 0.14 0 1
Minimal/minor role 0.01 0.08 0 1
Minor role 0.07 0.25 0 1
Aggravating +2 role 0.02 0.13 0 1
Aggravating +3 role 0.01 0.07 0 1
Aggravating +4 role 0.01 0.08 0 1

 Substantial Assistance
 9,403 Observations

 Mean SD Min Max

Post-Booker 0.55 0.50 0 1
Total months imprisonment 50.4 53.7 0 480
Base offense level 24.9 9.6 6 43
Final offense level 24.0 7.8 2 43
Criminal history category 2.44 1.76 1 6
Statutory minimum-life 0.01 0.08 0 1
Statutory maximum-life 0.38 0.48 0 1
Statutory minimum sentence (a) 60.6 67.9 0 1,200
Statutory maximum sentence (b) 323.8 443.2 12 23,640
Safety valve 0.24 0.43 0 1
Age 33.9 10.0 18 85
Female 0.18 0.38 0 1
Number of dependents 1.62 1.71 0 15
Black 0.32 0.47 0 1
Hispanic 0.26 0.44 0 1
Other race 0.05 0.22 0 1
Legal alien 0.06 0.23 0 1
Illegal alien 0.11 0.31 0 1
Non-U.S. resident 0.02 0.14 0 1
High school graduate 0.36 0.48 0 1
Some college 0.19 0.39 0 1
College graduate 0.05 0.23 0 1
Weapon 0.16 0.36 0 1
Minimal role 0.03 0.17 0 1
Minimal/minor role 0.01 0.10 0 1
Minor role 0.10 0.30 0 1
Aggravating +2 role 0.03 0.18 0 1
Aggravating +3 role 0.02 0.14 0 1
Aggravating +4 role 0.03 0.16 0 1

(a) Only observed if Minimum is not Life.

(b) Only observed if Maximum is not Life.

TABLE 3
Results from Guideline-Conditional Estimations of Non-Substantial-
Assistance Sentencings

 Dependent Variable: Log Months
 Imprisonment

 1 2 3

BOOKER -0.075 * -0.082 ** -0.083 **
 [0.0421 [0.039] [0.0411
BOOKER x FEMALE -0.090 ** -0.114 *** -0.116 **
 [0.036] [0.040] [0.0461
BOOKER x BLACK 0.038 0.056 * 0.056
 [0.033] [0.034] [0_040]
BOOKER x HISPANIC 0.005 0.006 0.015
 [0.036] [0.035] [0.038]
BOOKER x OTHER_RACE -0.003 0.044 0.034
 [0.069] [0.074] [0.079]
BOOKER x LEGAL_ALIEN 0.066 0.095 0.116 *
 [0.0611 [0.058] [0.067]
BOOKER x ILLEGAL_ALIEN 0.021 0.028 0.012
 [0.033] [0.0341 [0.0371
BOOKER x NON_RESIDENT -0.102 -0.127 -0.101
 [0.176] [0.198] [0.221]
BOOKER x HS_GRAD -0.080 *** -0.080 *** -0.087 ***
 [0.0241 [0.025] [0.027]
BOOKER x SOME_COLLEGE -0.037 -0.049 -0.039
 [0.035] [0.042] [0.047]
BOOKER x COLLEGE_GRAD -0.047 0.014 0.029
 [0.064] [0.072] [0.085]
Observations 48,548 41,550 33,480
[R.sup.2] 0.853 0.852 0.852
Window of observations 7/2004- 8/2004- 9/2004-
 7/2005 6/2005 5/2005

 Dependent Variable: Log Months
 Imprisonment

 4 5 6

BOOKER -0.080 * -0.075 -0.129 **
 [0.0431 [0.046] [0.0511
BOOKER x FEMALE -0.122 ** -0.130 * -0.107
 [0.059] [0.0691 [0.0671
BOOKER x BLACK 0.062 0.073 0.091 *
 [0.043] [0.045] [0.054]
BOOKER x HISPANIC 0.045 -0.014 0.042
 [0.046] [0.066] [0.091]
BOOKER x OTHER_RACE 0.005 0.003 -0.039
 [0.088] [0.099] [0.118]
BOOKER x LEGAL_ALIEN 0.150 ** 0.208 *** 0.148
 [0.069] [0.0711 [0.120]
BOOKER x ILLEGAL_ALIEN -0.030 0.033 0.030
 [0.047] [0.0581 [0.069]
BOOKER x NON_RESIDENT -0.144 -0.108 -0.254
 [0.243] [0.235] [0.309]
BOOKER x HS_GRAD -0.103 *** -0.118 *** -0.075
 [0.035] [0.045] [0.0491
BOOKER x SOME_COLLEGE -0.048 -0.053 -0.004
 [0.054] [0.0611 [0.069]
BOOKER x COLLEGE_GRAD 0.131 0.152 0.234 *
 [0.098] [0.109] [0.1211
Observations 26,001 18,550 11,006
[R.sup.2] 0.852 0.853 0.858
Window of observations 10/2004- 11/2004- 12/2004-
 4/2005 3/2005 2/2005

 Dependent Variable:
 Log Months Imprisonment

 7

BOOKER -0.153 **
 [0.0611
BOOKER x FEMALE -0.124 *
 [0.073]
BOOKER x BLACK 0.076
 [0.0561
BOOKER x HISPANIC 0.059
 [0.087]
BOOKER x OTHER_RACE -0.108
 [0.128]
BOOKER x LEGAL_ALIEN 0.048
 [0.144]
BOOKER x ILLEGAL_ALIEN 0.054
 [0.081]
BOOKER x NON_RESIDENT -0.288
 [0.3061
BOOKER x HS_GRAD -0.068
 [0.049]
BOOKER x SOME_COLLEGE 0.017
 [0.070]
BOOKER x COLLEGE_GRAD 0.236 *
 [0.131]
Observations 3,798
[R.sup.2] 0.866
Window of observations 1/2005-
 1/2005

Robust standard errors in brackets.

* Significant at 10%; ** significant at 5%; *** significant at 1%.

TABLE 4
Results from Guideline-Conditional Estimations of Substantial
Assistance Sentencings

 Dependent Variable: Log Months
 Imprisonment

 1 2 3 4

BOOKER -0.140 -0.137 -0.121 -0.132
 [0.1181 10.1221 [0.1311 [0.1471
BOOKER x FEMALE -0.034 -0.059 -0.038 -0.121
 10.1351 10.1401 [0.1611 [0.1931
BOOKER x BLACK 0.181 ** 0.144 0.120 0.251 *
 [0.087] [0.0891 [0.102] [0.1291
BOOKER x HISPANIC 0.110 0.096 0.080 0.186
 [0.110] 10.1221 [0.1351 10.1751
BOOKER x OTHER_RACE -0.067 -0.019 0.181 0.051
 [0.1661 [0.186] [0.235] [0.2691
BOOKER x LEGAL_ALIEN 0.213 0.348 * 0.260 0.218
 [0.1971 [0.1851 [0.1991 [0.215]
BOOKER x ILLEGAL_ALIEN 0.118 0.109 0.101 0.079
 [0.1781 [0.1631 [0.2011 [0.2461
BOOKER x NONRESIDENT -0.009 0.004 -0.400 -0.501 *
 [0.258] [0.248] [0.2641 [0.2951
BOOKER x HS_GRAD -0.053 -0.045 -0.003 -0.086
 [0.0971 [0.104] [0.105] [0.110]
BOOKER x SOME_COLLEGE 0.004 -0.009 -0.059 0.020
 10.130] [0.145] [0.171] [0.179]
BOOKER x COLLEGE_GRAD 0.016 -0.005 -0.011 0.145
 [0.1861 [0.199] [0.233] [0.244]
Observations 9,403 8,103 6,623 5,159
[R.sup.2] 0.853 0.852 0.852 0.852
Window of observations 7/2004- 8/2004- 9/2004- 10/2004-
 7/2005 6/2005 5/2005 4/2005

 Dependent Variable: Log Months
 Imprisonment

 5 6 7

BOOKER -0.177 -0.162 -0.059
 [0.1581 [0.1951 [0.258]
BOOKER x FEMALE -0.010 -0.174 -0.282
 [0.2231 [0.2381 [0.3341
BOOKER x BLACK 0.242 0.146 0.048
 [0.152] [0.195] [0.3011
BOOKER x HISPANIC 0.108 0.167 0.212
 [0.1931 [0.2511 [0.3341
BOOKER x OTHER_RACE 0.005 0.215 -0.136
 [0.3181 [0.396] [0.680]
BOOKER x LEGAL_ALIEN 0.366 0.440 ** 0.321
 [0.227] [0.2181 [0.268]
BOOKER x ILLEGAL_ALIEN 0.288 0.282 0.511
 [0.279] [0.3511 [0.547]
BOOKER x NONRESIDENT -0.625 -0.757 * -0.891
 [0.404] [0.397] [0.686]
BOOKER x HS-GRAD -0.058 -0.013 -0.142
 [0.1391 [0.1811 [0.228]
BOOKER x SOME_COLLEGE 0.056 0.220 0.200
 [0.190] [0.2381 [0.324]
BOOKER x COLLEGE_GRAD 0.477 0.390 0.691
 [0.291] [0.374] [0.4451
Observations 3,659 2,114 721
[R.sup.2] 0.853 0.858 0.866
Window of observations 11/2004- 12/2004- 1/2005
 3/2005 2/2005 1/2005

Robust standard errors in brackets.

* Significant at 10%; ** significant at 5%; *** significant at 1%.

TABLE 5
Results from Baseline-Conditional Estimations of
Non-Substantial-Assistance Sentencings

 Dependent Variable: Log Months Imprisonment

 1 2 3 4

BOOKER -0.028 -0.038 -0.030 -0.017
 [0.0491 [0.0521 10.0561 [0.0611
BOOKER x FEMALE -0.023 -0.051 -0.078 -0.057
 [0.0541 [0.0601 10.0641 10.0721
BOOKER x BLACK -0.022 -0.019 -0.026 -0.014
 10.0471 [0.0491 [0.0531 10.0581
BOOKER x HISPANIC 0.004 0.001 -0.008 0.018
 [0.0541 10.0541 10.0531 10.0651
BOOKER x OTHER_RACE 0.088 0.166 0.101 0.065
 [0.1031 [0.1131 [0.1[21 [0.1241

BOOKER x LEGAL_ALIEN 0.074 0.107 0.146 0.125
 [0.099] [0.0971 [0.105] [0.108]
BOOKER x ILLEGAL_ALIEN -0.034 -0.006 -0.003 -0.066
 10.0471 10.0481 [0.0491 10.0571
BOOKER x NON_RESIDENT -0.073 -0.123 -0.096 -0.124
 [0.1661 [0.170] [0.1941 [0.2051
BOOKER x HS_GRAD -0.095 *** -0.098 ** -0.112 *** -0.147 ***
 [0.035] [0.0391 [0.0391 [0.0451
BOOKER x SOME_COLLEGE -0.017 -0.026 -0.001 -0.026
 [0.0431 [0.053] [0.0631 [0.0691
BOOKER x COLLEGE_GRAD -0.010 0.111 0.097 0.250 *
 [0.0971 [0.1131 [0.1161 10.1301
Observations 48,548 41,550 33,480 26,001
[R.sup.2] 0.734 0.731 0.730 0.729
Window of observations 7/2004- 8/2004- 9/2004- 10/2004-
 7/2005 6/2005 5/2005 4/2005

 Dependent Variable: Log Months
 Imprisonment

 5 6 7

BOOKER -0.010 -0.036 -0.062
 [0.065] [0.076] [0.090]
BOOKER x FEMALE -0.098 -0.163 * -0.146
 [0.0881 [0.090] [0.1031
BOOKER x BLACK -0.042 -0.043 -0.027
 [0.0671 [0.0841 [0.0951
BOOKER x HISPANIC -0.060 0.009 0.021
 [0.0811 [0.1051 [0.1171
BOOKER x OTHER_RACE 0.001 0.003 -0.042
 [0.1491 [0.1841 [0.1841
BOOKER x LEGAL_ALIEN 0.239 ** 0.279 ** 0.183
 [0.095] [0.1261 [0.171]
BOOKER x ILLEGAL_ALIEN 0.002 -0.066 -0.049
 [0.0731 10.0841 10.0961
BOOKER x NON_RESIDENT -0.022 -0.146 -0.218
 [0.2211 [0.3061 [0.307]
BOOKER x HS_GRAD -0.128 ** -0.068 -0.073
 [0.0551 [0.059] [0.072]
BOOKER x SOME_COLLEGE -0.010 0.020 -0.029
 [0.080] [0.104] [0.107]
BOOKER x COLLEGE_GRAD 0.294 * 0.518 *** 0.532 ***
 [0.151] [0.187] [0.190]
Observations 18,550 11,006 3,798
[R.sup.2] 0.732 0.738 0.746
Window of observations 11/2004- 12/2004- 1/2005-
 3/2005 2/2005 1/2005

Robust standard errors in brackets.

* Significant at 10%; ** significant at 5%: *** significant at 1%.

TABLE 6
Results from Baseline-Conditional Estimations of
Non-Substantial-Assistance Sentencings

 Dependent Variable: Final Offense Level

 1 2 3 4

BOOKER 0.214 0.187 0.227 0.228
 [0.206] [0.222] [0.226] [0.257]
BOOKER x FEMALE 0.230 0.147 0.124 0.310
 [0.192] [0.204] [0.242] [0.242]
BOOKER x BLACK -0.235 -0.276 -0.283 -0.209
 [0.167] [0.193] [0.210] [0.240]
BOOKER x HISPANIC 0.079 0.135 -0.013 -0.010
 [0.188] [0.211] [0.235] [0.312]
BOOKER x OTHER_RACE 0.734 * 1.051 ** 0.806 * 0.830 *
 [0.395] [0.410] [0.415] [0.438]
BOOKER x LEGAL_ALIEN -0.040 -0.133 -0.001 -0.274
 [0.302] [0.303] [0.377] [0.432]
BOOKER x ILLEGAL_ALIEN -0.227 -0.160 -0.022 -0.113
 [0.155] [0.171] [0.212] [0.277]
BOOKER x NON_RESIDENT 0.199 0.078 0.216 0.211
 [0.4711 [0.549] [0.619] [0.782]
BOOKER x HS_GRAD -0.149 -0.211 -0.260 -0.305
 [0.183] [0.201] [0.202] [0.215]
BOOKER x SOME_COLLEGE 0.109 0.144 0.222 0.140
 [0.1991 [0.2281 [0.2651 [0.2871
BOOKER x COLLEGE_GRAD 0.017 0.356 0.299 0.705
 [0.369] [0.432] [0.447] [0.536]
Observations 48,548 41,550 33,480 26,001
[R.sup.2] 0.827 0.825 0.824 0.824
Window of observations 7/2004- 8/2004- 9/2004- 10/2004-
 7/2005 6/2005 5/2005 4/2005

 Dependent Variable: Final
 Offense Level

 5 6 7

BOOKER 0.222 0.439 0.514
 [0.295] [0.347] [0.382]
BOOKER x FEMALE 0.187 -0.097 -0.105
 [0.298] [0.298] [0.344]
BOOKER x BLACK -0.299 -0.519 -0.423
 [0.286] [0.346] [0.397]
BOOKER x HISPANIC -0.145 -0.080 -0.031
 [0.365] [0.409] [0.473]
BOOKER x OTHER_RACE 0.404 0.609 0.556
 [0.511] [0.689] [0.744]
BOOKER x LEGAL_ALIEN -0.068 0.357 0.201
 [0.453] [0.519] [0.671]
BOOKER x ILLEGAL_ALIEN -0.033 -0.536 -0.683
 [0.306] [0.360] [0.439]
BOOKER x NON_RESIDENT 0.577 0.643 0.340
 [0.809] [1.1161 [1.1451
BOOKER x HS_GRAD -0.126 -0.138 -0.269
 [0.248] [0.286] [0.344]
BOOKER x SOME_COLLEGE 0.141 0.053 -0.226
 10.3561 10.4301 10.4771
BOOKER x COLLEGE_GRAD 0.859 1.483 ** 1.371 *
 10.538] [0.693] [0.769]
Observations 18,550 11,006 3,798
[R.sup.2] 0.828 0.829 0.840
Window of observations 11/2004- 12/2004- 1/2005-
 3/2005 2/2005 1/2005

Robust standard errors in brackets.

* Significant at 10%; ** significant at 5%; *** significant at 1%.

TABLE 7
Results from Offense-Level-Unconditional Estimations of
Non-Substantial-Assistance Sentencings

 Dependent Variable: Log Months Imprisonment

 1 2 3 4

BOOKER -0.049 -0.043 -0.029 -0.010
 [0.056] [0.0581 [0.067] [0.074]
BOOKER x FEMALE -0.028 -0.057 -0.090 -0.066
 [0.060] [0.065] [0.073] [0.082]
BOOKER x BLACK 0.018 0.018 0.016 0.001
 [0.059] [0.0641 [0.069] [0.077]
BOOKER x HISPANIC -0.008 -0.022 -0.021 -0.008
 [0.063] [0.065] [0.070] [0.082]
BOOKER x OTHER_RACE 0.106 0.151 0.099 0.074
 [0.112] [0.121] [0.124] [0.133]
BOOKER x LEGAL_ALIEN 0.118 0.147 0.190 0.116
 [0.136] [0.136] [0.146] [0.160]
BOOKER x ILLEGAL_ALIEN 0.001 0.018 0.003 -0.057
 [0.0491 [0.0501 [0.053] [0.060]
BOOKER x NON_RESIDENT -0.126 -0.187 -0.184 -0.265

 [0.1651 [0.168] [0.198] [0.211]
BOOKER x HS_GRAD -0.086 ** -0.099 ** -0.123 ** -0.142 ***
 [0.0411 [0.0441 [0.048] [0.0511
BOOKER x SOME_COLLEGE -0.047 -0.070 -0.054 -0.061
 [0.048] [0.059] [0.066] [0.073]
BOOKER x COLLEGE_GRAD -0.004 0.083 0.043 0.193
 [0.117] [0.133] [0.136] [0.1611
Observations 48,548 41,550 33,480 26,001
[R.sup.2] 0.853 0.852 0.852 0.852
Window of observations 7/2004- 8/2004- 9/2004- 10/2004-
 7/2005 6/2005 5/2005 4/2005

 Dependent Variable: Log Months
 Imprisonment

 5 6 7

BOOKER 0.000 -0.060 -0.022
 [0.085] [0.093] [0.099]
BOOKER x FEMALE -0.074 -0.107 -0.151
 [0.101] [0.105] [0.111]
BOOKER x BLACK -0.021 0.021 -0.019
 [0.087] [0.1071 [0.108]
BOOKER x HISPANIC -0.116 0.008 0.002
 [0.098] [0.124] [0.133]
BOOKER x OTHER_RACE 0.018 -0.040 -0.112
 [0.1591 [0.198] [0.198]
BOOKER x LEGAL_ALIEN 0.254 0.239 0.199
 [0.157] [0.170] [0.222]
BOOKER x ILLEGAL_ALIEN 0.051 -0.031 -0.056
 [0.077] [0.093] [0.096]
BOOKER x NON_RESIDENT -0.151 -0.262 -0.267
 [0.219] [0.269] [0.268]
BOOKER x HS_GRAD -0.141 ** -0.077 -0.114
 [0.062] [0.0711 [0.080]
BOOKER x SOME_COLLEGE -0.089 -0.028 -0.046
 [0.0801 [0.099] [0.1081
BOOKER x COLLEGE_GRAD 0.268 0.437 ** 0.462 **
 [0.181] [0.211] [0.211]
Observations 18,550 11,006 3,798
[R.sup.2] 0.853 0.858 0.866
Window of observations 11/2004- 12/2004- 1/2005-
 3/2005 2/2005 1/2005

Robust standard errors in brackets.

* Significant at 10%; ** significant at 5%; *** significant at 1%.

TABLE 8
Results from Offense-Level-Unconditional Estimations of
Non-Substantial-Assistance Sentencings

 Dependent Variable: Final Offense
 Level

 1 2 3 4

BOOKER 0.115 0.212 0.340 0.357
 [0.317] [0.328] [0.360] [0.417]
BOOKER x FEMALE 0.209 0.100 0.077 0.226
 [0.248] [0.266] [0.3111 [0.335]
BOOKER x BLACK 0.012 -0.102 -0.127 -0.250
 [0.276] [0.306] [0.332] [0.384]
BOOKER x HISPANIC -0.128 -0.188 -0.318 -0.465
 [0.308] [0.332] [0.387] [0.460]
BOOKER x OTHER_RACE 0.699 0.773 0.638 0.649
 [0.533] [0.552] [0.607] [0.602]
BOOKER x LEGAL_ALIEN 0.266 0.205 0.290 -0.164
 [0.651] [0.657] [0.749] [0.875]
BOOKER x ILLEGAL_ALIEN 0.040 0.076 0.083 0.125
 [0.211] [0.220] [0.288] [0.319]
BOOKER x NON_RESIDENT -0.235 -0.351 -0.591 -1.053
 [0.618] [0.683] [0.892] [1.084]
BOOKER x HS_GRAD -0.131 -0.268 -0.402 -0.327
 [0.231] [0.248] [0.2761 [0.277]
BOOKER x SOME_COLLEGE -0.146 -0.168 -0.240 -0.139
 [0.245] [0.276] [0.310] [0.359]
BOOKER x COLLEGE_GRAD 0.011 0.082 -0.082 0.399
 [0.514] [0.5871 [0.596] [0.771]
Observations 48,548 41,550 33,480 26,001
[R.sup.2] 0.853 0.852 0.852 0.852
Window of observations 7/2004- 8/2004- 9/2004- 10/2004-
 7/2005 6/2005 5/2005 4/2005

 Dependent Variable: Final
 Offense Level

 5 6 7

BOOKER 0.462 0.448 0.865
 [0.472] [0.520] [0.544]
BOOKER x FEMALE 0.274 0.068 -0.166
 [0.400] [0.426] [0.476]
BOOKER x BLACK -0.423 -0.287 -0.463
 [0.427] [0.436] [0.482]
BOOKER x HISPANIC -0.799 -0.386 -0.557
 [0.523] [0.513[ [0.6001
BOOKER x OTHER_RACE 0.087 -0.030 -0.316
 [0.818] [1.011] [1.020]
BOOKER x LEGAL_ALIEN 0.164 0.328 0.595
 [0.937] [0.974] [1.016]
BOOKER x ILLEGAL_ALIEN 0.342 -0.210 -0.447
 [0.368] [0.434] [0.495]
BOOKER x NON_RESIDENT -0.632 -0.580 -0.534
 [1.182] [1.334] [1.405]
BOOKER x HS_GRAD -0.234 -0.246 -0.599
 [0.317] [0.403] [0.435]
BOOKER x SOME_COLLEGE -0.394 -0.214 -0.458
 [0.379] [0.435] [0.504]
BOOKER x COLLEGE_GRAD 0.737 1.189 1.067
 [0.8141 [1.077] [1.086]
Observations 18,550 11,006 3,798
[R.sup.2] 0.853 0.858 0.866
Window of observations 11/2004- 12/2004- 1/2005-
 3/2005 2/2005 1/2005

Robust standard errors in brackets.

* Significant at 10%; ** significant at 5%; *** significant at 1%.