Demographic Differences in Sentencing, an Update

• Nov. 17, 2017 • Locations: United States of America • Topics: Discrimination, Statistics/Trends, Sentencing

Download original document:
Brief thumbnail

Document text

This text is machine-read, and may contain errors. Check the original document to verify accuracy.

Demographic Differences in Sentencing:
An Update to the 2012 Booker Report

UNITED STATES SENTENCING COMMISSION

United States Sentencing Commission
One Columbus Circle, N.E.
Washington, DC 20002
www.ussc.gov

William H. Pryor, Jr.
Acting Chair
Rachel E. Barkow
Commissioner
Charles R. Breyer
Commissioner
Danny C. Reeves
Commissioner
Zachary C. Bolitho
Ex Officio
J. Patricia Wilson Smoot
Ex Officio

Kenneth P. Cohen
Staff Director
Glenn R. Schmitt
Director
Office of Research and Data

November 2017

Glenn R. Schmitt, J.D., M.P.P., Director, Office of Research and Data
Louis Reedt, Sc.D., Deputy Director, Office of Research and Data
Kevin Blackwell, Senior Research Associate, Office of Research and Data

TABLE OF CONTENTS
Section I: Introduction........................................................................................................1
Key Findings.................................................................................................................................................................................2
Multivariate Analysis of Demographic Differences in Sentencing........................................................................3
Why Multivariate Regression Analysis is Helpful........................................................................................................3

Section II: The Commission’s Updated Findings..............................................................5
Summary of Findings from the Commission’s Updated Multivariate Analysis...............................................6

Results of Demographic Multivariate Regression Analysis for All Cases...........................................................8

Analysis of Differences in Sentencing by Guideline Application............................................................................9

Analysis of Likelihood of Obtaining a Below Range Sentence................................................................................14
The Effect of Adding Data on Prior Violence as a Factor in the Model...............................................................16

Limitations of Regression Analysis....................................................................................................................................17

Section III: Conclusion.......................................................................................................19
Appendices........................................................................................................................21
Appendix A: Analysis of Other Demographic Factors................................................................................................22

Appendix B: Offense Specific Analysis..............................................................................................................................26
Appendix C: Methodology......................................................................................................................................................31
Appendix D: Regression Model Output...........................................................................................................................33

Endnotes.............................................................................................................................37

Demographic Differences in Sentencing

Section I

Introduction

In 2010, the Commission published an analysis of federal
sentencing data which examined whether the length of sentences
imposed on federal offenders was correlated with demographic
characteristics of those offenders.1 That analysis found that some
demographic factors were associated with sentence length to a
statistically significant extent during some of the time periods studied.2
Among other findings, the analysis showed that Black male offenders
received longer sentences than White male offenders, and that the gap
between the sentence lengths for Black and White male offenders was
increasing.

In 2012, the Commission updated this analysis by examining
cases in which the offender was sentenced after the release of the
2010 report.3 The Commission also expanded its analyses to examine
demographic differences in sentences based on a comparison of the
position of the sentence imposed relative to the sentencing guideline
range that applied in the case; based on the type of offense committed
by the offender, including drug trafficking, fraud, and firearms; and
based on changes in sentence length for offenders of particular race
and gender pairings. These findings were released as part of the
Commission’s comprehensive report4 on sentencing practices after the
Supreme Court’s decision in United States v. Booker.5 In its 2012 report,
the Commission found that the type of demographic differences in
sentencing reported in 2010 continued to be observed in sentencings
that occurred after that report.

The Commission has once again updated its analysis by
examining cases in which the offender was sentenced during the period
following the 2012 report. This new time period, from October 1,
2011, to September 30, 2016, is referred to as the “Post-Report period”
in this publication. Also, the Commission has collected data about an
additional variable—violence in an offender’s criminal history—that
the Commission had previously noted was missing from its analysis but
that might help explain some of the differences in sentencing noted in
its work. This report presents the results observed from adding that
new data to the Commission’s analysis.

Demographic Differences in Sentencing

Key Findings
Consistent with its previous reports, the Commission found that sentence
length continues to be associated with some demographic factors. In particular,
after controlling for a wide variety of sentencing factors, the Commission found:
• Black male offenders continued to receive longer sentences than
similarly situated White male offenders. Black male offenders received
sentences on average 19.1 percent longer than similarly situated White male
offenders during the Post-Report period (fiscal years 2012-2016), as they had
for the prior four periods studied. The differences in sentence length remained
relatively unchanged compared to the Post-Gall period.

• Non-government sponsored departures and variances appear to
contribute significantly to the difference in sentence length between Black
male and White male offenders. Black male offenders were 21.2 percent
less likely than White male offenders to receive a non-government sponsored
downward departure or variance during the Post-Report period. Furthermore,
when Black male offenders did receive a non-government sponsored departure or
variance, they received sentences 16.8 percent longer than White male offenders
who received a non-government sponsored departure or variance. In contrast,
there was a 7.9 percent difference in sentence length between Black male and
White male offenders who received sentences within the applicable sentencing
guidelines range, and there was no statistically significant difference in sentence
length between Black male and White male offenders who received a substantial
assistance departure.
• Violence in an offender’s criminal history does not appear to account
for any of the demographic differences in sentencing. Black male offenders
received sentences on average 20.4 percent longer than similarly situated White
male offenders, accounting for violence in an offender’s past in fiscal year 2016,
the only year for which such data is available. This figure is almost the same as
the 20.7 percent difference without accounting for past violence. Thus, violence
in an offender’s criminal history does not appear to contribute to the sentence
imposed to any extent beyond its contribution to the offender’s criminal history
score determined under the sentencing guidelines.

• Female offenders of all races received shorter sentences than White
male offenders during the Post-Report period, as they had for the prior
four periods. The differences in sentence length decreased slightly during the
five-year period after the 2012 Booker Report for most offenders. The differences
in sentence length fluctuated across all time periods studied for White females,
Black females, Hispanic females, and Other Race female offenders.

Multivariate Analysis of Demographic Differences in
Sentencing

For this report and in its prior two reports, the Commission
has used multivariate regression analyses to explore the relationships
between demographic factors, such as race and gender, and sentencing
outcomes. These analyses were aimed at determining whether there
were demographic differences in sentencing outcomes that were
statistically significant, and whether those findings changed during the
periods studied.

Multivariate regression analysis usually begins with a decision
to examine an observed phenomenon or outcome. In the Commission’s
work, the outcome observed are the sentences imposed in federal
cases. The researcher will then develop a hypothesis as to the many
possible factors that might produce that outcome. Among the factors
considered by the Commission were the type of offense, the sentencing
range determined by the court in each case using the Commission’s
Guidelines Manual,6 and whether the offender was subject to a mandatory
minimum penalty. The hypothesis is then tested using multivariate
regression analysis to bring together data about the outcome and many
possible factors that might affect that outcome. The goal of multivariate
regression analysis is to determine whether there is an association
between the factors being studied and, if so, to measure the extent to
which each factor contributes to the observed outcome. Researchers
refer to the outcome (in this report the length of the sentence imposed)
as the “dependent variable.” The factors that might affect (and therefore
might explain) that outcome are referred to as the “independent
variables” or the “explanatory variables.”7

The principal benefit of multivariate regression analysis is
that it controls for the effect of each factor in the analysis by comparing
offenders who are similar to one another in relevant ways. By
controlling for such factors and comparing similarly situated offenders,
this multivariate regression analysis seeks to answer the question: if
two offenders are similar in certain ways, what other factors might be
associated with those two offenders receiving different sentences? In
addition, multivariate regression analysis measures the extent of the
difference in outcomes.

Multivariate regression analysis often cannot control for all
possible factors that might affect the outcome being studied, typically
because sufficient data about some factors is not readily available. For
example, in its past reports, the Commission noted some potentially
relevant factors were not included in its analyses, such as whether
the offender’s criminal history included violent criminal conduct, the
offender’s family ties, and the offender’s employment history.8 Data was
not readily available for those factors because the Commission did not
routinely extract that information from the court documents it receives.9
Therefore, for those prior analyses, the Commission could not control for
them.10 For this reason, caution should always be used when drawing
conclusions based on multivariate regression analysis.

Why Multivariate Regression Analysis is Helpful

Multivariate regression analysis can be helpful when
considering demographic differences in sentencing outcomes because
results from more simplistic data analyses that examine only selected
demographic factors and sentencing outcomes can be misleading. Such
simplistic analyses may yield results that might appear to indicate that
demographic factors correlate with sentence length, when the actual
correlation may be attributable to other, non-demographic factors,
such as the type of offense committed or whether the offense was
one for which a statutory mandatory minimum penalty applied. Most
importantly, simplistic analyses do not consider the effect of the guideline
range provided for under the sentencing guidelines, which takes into
account, among other things, the type of offense, the presence of
aggravating or mitigating factors, and the criminal history of the offender.

For example, a simplistic pairing of recent data regarding race
and gender with sentence length shows that sentences of Black male
offenders have always been longer than those of White male offenders.
As shown on figure 1, however, the average sentences for Black male
offenders were shorter during the Post-Report period (fiscal years 2012
to 2016) than during the Booker period (January 2005 to December
2007), and the Gall period (December 2007 to the end of fiscal year
2011). At the same time, the average sentence for White male offenders
increased during the Post-Report period after a slight decrease in the
Gall period. As a result, the difference in the sentences imposed on Black
male and White male offenders has substantially decreased in recent
years, from a difference of 34 months in fiscal year 2006, to nine months
in fiscal years 2016.
Demographic Differences in Sentencing

Fig. 1. Average Sentence for White Male and Black Male Offenders

Fiscal Years
1999-2016
Months

White Male

Black Male

100

SOURCE: U.S. Sentencing Commission, 1999 - 2016 Datafile, USSCFY 99 -USSCFY16.

Some commentators have pointed to this change to erroneously
assert that racial differences in sentencing are decreasing.11 However,
such an analysis ignores many important factors that affect sentence
length, such as the type of offense, criminal history, and weapon
possession. For instance, the narrowing gap between Black and
White male offender sentence lengths is due, in large part, to sizeable
reductions in penalties for crack cocaine offenses, in which Black
offenders constitute the large majority of the offenders. Despite these
apparent changes in sentencing outcomes, the Commission’s multivariate
regression analysis shows that when other relevant factors are controlled
for, the gap in the sentence length between Black male and White male
offenders did not shrink but, in fact, remained relatively stable across
these periods.

Because simplistic analyses do not control for other relevant
factors, they cannot provide an accurate estimate of the extent to which
demographic factors are associated with sentence length. For this
reason, multivariate regression analysis is necessary to explore the many
factors that may contribute to these sentencing outcomes.

Demographic Differences in Sentencing

Section II

Updated Findings

The Commission’s Updated Findings

The Commission’s findings are set forth below in five parts.
The first part summarizes the findings from the Commission’s updated
analysis. The second part provides the findings from the Commission’s
updated analysis of all cases regarding demographic differences
associated with race and gender. Findings regarding other demographic
factors are discussed in Appendix A to this report. In the third part of
this report the Commission presents the findings of the multivariate
regression analysis of cases divided into groups by the position of the
sentence imposed relative to the guideline range. The findings regarding
race and gender are discussed in the body of the report and the findings
regarding other demographic factors are discussed in Appendix A. In
the fourth part, multivariate regression analysis is used to examine the
differences among offenses in the likelihood of receiving a sentence
below the applicable guideline range. Lastly, the fifth part presents the
results of a multivariate regression analysis that incorporates data about
an offender’s violent criminal history. In Appendix B to this report the
Commission presents the results of a multivariate regression analysis
examining the sentences imposed for three types of offenses: drug
trafficking, fraud, and firearms offenses.

Summary of Findings from the Commission’s Updated
Multivariate Analysis

Consistent with its previous reports, the Commission found
that sentence length is associated with some demographic factors. The
Commission’s analysis considered race, gender, citizenship, age, and
education level.12

The Commission found that sentences of Black male offenders
were longer than those of White male offenders for all periods studied.
Black male offenders’ sentences were 19.1 percent longer than those of
White male offenders during the Post-Report period. The gap in sentence
length between these two groups was smallest during the PROTECT Act
period (5.5%) and largest during the Gall period (19.5%). Hispanic male
offenders received sentences that were 5.3 percent longer than those of
White male offenders during the Post-Report period. However, there was
no statistically significant difference between the average sentences for
the two groups during the Gall period. The differences in the sentences
imposed on Other Race male offenders compared to those imposed on
White male offenders were not statistically significant in the Post-Report
period or the Gall period.
6

Demographic Differences in Sentencing

Report Time Periods
• The Koon period, from October 1, 1998 through April 30, 2003.
This period relates to cases decided after the Supreme Court’s 1996
decision in Koon v. United States.
• The PROTECT Act period, from May 1, 2003 (the date of the
enactment of the PROTECT Act), through June 24, 2004 (the date of
the Supreme Court’s decision in Blakely v. Washington).
• The Booker period, from January 12, 2005 (the date of the
Supreme Court’s decision in United States v. Booker), through
December 10, 2007 (the date of the Supreme Court’s decisions in
Kimbrough v. United States and Gall v. United States).

• The Gall period, from December 11, 2007 through September 30,
2011.
• The Post-Report period, from October 1, 2011 through
September 30, 2016.

When examining all cases as a group, female offenders of all
races received shorter sentences than White male offenders during the
Post-Report period, as they had in the prior four periods. White female
offenders received sentences that were 28.9 percent shorter than those
of White male offenders in the Post-Report period, compared to 31.1
percent shorter during the Gall period. Black female offenders and
Other Race female offenders also received shorter sentences than White
male offenders during the Post-Report period, at 29.7 percent and 35.4
percent shorter respectively. In the Gall period these differences were
33.1 percent and 34.6 percent, respectively. Hispanic female offenders
received sentences that were 16.8 percent shorter than those of White
male offenders during the Post-Report period, compared to 18.2 percent
shorter in the Gall period.

Non-citizen offenders received longer sentences than United
States citizens during the Post-Report period, although this difference

decreased from the Gall period. No difference was found during the
PROTECT Act period, but differences in sentence length returned during
the Booker and Gall periods at levels exceeding those of the Koon period.
Offenders with at least some college education had no difference in
sentence length than offenders who did not attend college in the PostReport period; a finding that differs from the other four periods where
offenders with some college education received shorter sentences than
offenders with no college education.13

The Commission also studied whether there were demographic
differences in sentencing based on the position of the sentence imposed
relative to the sentencing guideline range that applied in the case.14 The
Commission found statistically significant differences in sentence length
during all time periods for Black male offenders, who received longer
within range sentences than White male offenders who received within
range sentences, although the difference decreased in the Post-Report
period. The differences in sentence length between Hispanic male
offenders and White male offenders who received within range sentences
varied over the five time periods studied, but Hispanic male offenders
generally received shorter sentences than White male offenders during
all but the Post-Report period. In the Post-Report period, however, there
was no statistically significant difference between the average sentences
for Hispanic and White male offenders who were sentenced within the
guideline range.

With respect to offenders who received below range sentences
based on their substantial assistance to the government,15 Black male
offenders received longer sentences than White male offenders during
the Koon period only. In all other periods, there was no statistically
significant difference between the two groups. Other Race male offenders
who received substantial assistance below range sentences received
longer sentences than White male offenders during the Booker period
only.

Comparing Black male offenders to White male offenders
who received a non-government sponsored below range sentence, the
differences in sentence length between the two groups were statistically
significant in two periods, including the Post-Report period, where Black
males received sentences that were 16.8 percent longer than those for
White males. Differences in sentence length between Hispanic male
offenders and White male offenders who received a non-government

sponsored below range sentence were statistically significant only in the
Gall and Post-Report periods, when the differences were 9.3 and 10.6
percent respectively. There were no statistically significant differences
between the sentences imposed on Other Race male offenders and White
male offenders who received a non-government sponsored below range
sentence.

Across most periods, female offenders who received within
range sentences, regardless of their race, were sentenced to terms
shorter than those of White male offenders who received within range
sentences. For most female offenders, these differences increased over
time. In the Post-Report period, White female offenders who received
within range sentences received sentences that were 24.1 percent
shorter than those of White male offenders who received within range
sentences. Black female offenders with within range sentences received
sentences that were 27.1 percent shorter than within range sentences for
White male offenders in the Post-Report period.

Sentences for Hispanic female offenders followed a different
pattern. The sentences for Hispanic females who received within range
sentences were not statistically different from those for White male
offenders during the Post-Report period. This is a change from all prior
periods studied, where Hispanic female within range sentences were
lower than those for White male offenders. The differences in sentence
length between Other Race female offenders who received within range
sentences and White male offenders have decreased in the recent period
studied. This difference was 37.2 percent during the Post-Report period.

Female offenders of all races who received substantial assistance
departures generally received shorter sentences than White male
offenders who received substantial assistance departures. During the
Post-Report period, the sentence length for White female offenders who
received substantial assistance departures was 26.0 percent shorter
than for White male offenders who received substantial assistance
departures. For Black female offenders who received substantial
assistance departures the difference was 21.1 percent shorter than White
male offenders. In the Post-report period, sentences for Hispanic female
offenders who received a substantial assistance departure were 25.5
percent shorter than for White male offenders who received a substantial
assistance departure. However, the difference for Other Race female
offenders and White male offenders was not statistically significant in the
Post-Report period.
Demographic Differences in Sentencing

immediately preceding period, the Gall period. The results for all periods
are presented in the figures that accompany this discussion.17 Also,
this portion of the report will discuss differences in sentencing that are
associated with race and gender. Analyses of sentencing differences
associated with other demographic factors can be found in Appendix A of
this report.

White and Black female offenders who received a nongovernment sponsored below range sentence received shorter sentences
than White male offenders who received a non-government sponsored
below range sentence during the Gall and Post-Report periods. Hispanic
and Other Race female offenders who received a non-government
sponsored below range sentence received shorter sentences than White
male offenders who received a non-government sponsored below range
sentence during the Booker, Gall, and Post-Report periods.

Analysis: All Cases

Figures 2 and 3 depict differences in sentence length for all
cases. Differences in sentence length for Black male offenders compared
to White male offenders remained relatively stable between the Gall
and Post-Report periods. During the Post-Report period, Black male
offenders were sentenced to 19.1 percent longer sentences than White
male offenders. In the Gall period, the sentences of Black male offenders
were 19.5 percent longer.

The Commission also conducted a multivariate regression
analysis of the likelihood that an offender would receive a nongovernment sponsored below range sentence. This analysis showed
statistically significant differences during the PROTECT Act, Booker, Gall,
and Post-Report periods.16 During these periods, Black male offenders
were between 20 to 25 percent less likely to receive a non-government
sponsored below range sentence than White male offenders. These
differences in the likelihood of receiving this type of below range
sentence may contribute to the sentence length differences between
Black and White male offenders. Hispanic male offenders also were less
likely than White male offenders to receive a non-government sponsored
below range sentence during all periods.

Hispanic male offenders received sentences that were 5.3
percent longer than those of White male offenders during the PostReport period. However, there was no statistically significant difference
between the average sentences for the two groups during the Gall period.

Finally, the Commission conducted a multivariate regression
Fig. 2. Differences in Sentence Length for Male Offenders
analysis of demographic differences in sentencing associated with the
Koon, PROTECT, Booker, Gall, and Post-Report Periods
specific type of offense committed. This analysis showed that sentencing Percentage Difference in Sentence Length
outcomes varied depending on the offense. For example, an analysis
Koon
PROTECT
25
of firearms cases revealed statistically significant differences between
the sentence length of White and Black male offenders during the Koon,
19.5 19.1
20
Booker, Gall, and Post-Report periods. In comparison, an analysis of
15.2
fraud cases showed statistically significant differences between sentence
15
11.2
length of White male and Black male offenders only in the Koon and Post10
Report periods.

Results of Demographic Multivariate Regression Analysis for
All Cases

The Commission’s multivariate regression analysis for all cases,
updated through fiscal year 2016, continues to demonstrate that some
demographic factors are associated with sentence length to a statistically
significant extent during the periods studied. For the remainder of
this report, the Commission’s analysis will focus primarily on the most
recent period studied—the Post-Report period—as compared to the
8

Demographic Differences in Sentencing

5.5

Gall

Post-Report

5.3

0
-5

Booker

-3.6

-4.4
-8.0

-10
Black Male vs. White Male

Hispanic Male vs. White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

Other Male vs. White Male

Analysis of Differences in Sentencing by Guideline
Application

The differences in the sentences imposed on Other Race
males compared to those imposed on White male offenders were not
statistically significant in the Post-Report period or the Gall period.

As depicted in Figure 3, female offenders, regardless of their
race, received sentences that were shorter, on average, than sentences
for White male offenders during the Post-Report period. This finding
is consistent with the differences observed in all prior periods. White
female offenders received sentences that were 28.9 percent shorter than
those of White male offenders in the Post-Report period, compared to
31.1 percent shorter during the Gall period. Black female offenders and
Other Race female offenders also received shorter sentences than White
male offenders during the Post-Report period, at 29.7 percent and 35.4
percent shorter respectively, compared to 33.1 percent and 34.6 percent
in the Gall period, respectively. Hispanic female offenders received
sentences that were 16.8 percent shorter than those of White male
offenders during the Post-Report period, compared to 18.2 percent in the
Gall period.

Fig. 3. Differences in Sentence Length for Female Offenders
Koon, PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

Koon

Booker

PROTECT

Gall

Post-Report

-5
-10
-15
-20
-25
-30
-35

-22.2

-19.0

-23.1

-16.8
-18.2

-17.8

-28.9
-31.1

-29.7

-34.5

-22.8

-23.1 -23.1

-24.9

-29.8

-33.1

-30.3
-34.6 -35.4

-40
-45
-50

-43.4

White Female vs.
White Male

Black Female vs.
White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

Hispanic Female vs.
White Male

Other Female vs.
White Male

The Commission conducted further analyses focusing on the
position of the sentence imposed relative to the applicable guideline
range and on specific offense types. These additional analyses were
undertaken for three reasons: (1) to identify more precisely where
demographic differences may be occurring; (2) to determine whether the
same pattern of results would be replicated when studying sentencing
outcomes other than sentence length, and when studying specific
offenses; and (3) to identify and measure the extent to which any
particular factor, such as type of offense or sentence relative to the range,
might contribute to the overall finding of demographic differences in
sentencing.

For the analysis discussed in this section of the report, cases
were divided into four groups: (1) those where the sentence was within
the sentencing guideline range; (2) those where the sentence was
above the guideline range,18 (3) those where the sentence was imposed
below the guideline range at the request of the government; and (4)
non-government sponsored below range sentences (i.e., downward
departures and variances).19 The Commission then examined the
presence of demographic differences in the sentences imposed in the
cases in each group. Also, because prior reports suggested that the
imposition of a non-government sponsored below range sentence may be
a source of sentencing disparity,20 the Commission also repeated its prior
analysis to estimate the likelihood that an offender would receive a nongovernment sponsored below range sentence. These analyses showed
statistically significant differences in sentence length associated with
various demographic factors for each of the periods studied.21
Within Range Sentences

Figure 4 on the next page depicts the results of the multivariate
analysis of sentences falling within the applicable guideline range. The
analysis found that the differences in sentence length for Black male
offenders who received within range sentences compared to White
male offenders who received within range sentences were statistically
significant during all periods, although the difference decreased during
the Post-Report period. In the Gall period this difference was 12.2
percent longer for Black male offenders, but 7.9 percent longer during
the Post-Report period.
Demographic Differences in Sentencing

The differences in sentence length between Hispanic male
offenders and White male offenders who received within range sentences
varied over the five time periods studied, but Hispanic male offenders
received shorter sentences than White male offenders during all but the
Post-Report period. In the Gall period, Hispanic male offenders who
received within range sentences received sentences that were 4.2 percent
shorter than those of White male offenders who received within range
sentences. In the Post-Report period, however, there was no statistically
significant difference between the average sentences for Hispanic and
White male offenders.

percent shorter than White male offenders who received within range
sentences. In the Gall period this difference was 27.3 percent. For Black
females, within range sentences were 27.1 percent shorter than within
range sentences for White male offenders in the Post-Report period. In
the Gall period this difference was 31.5 percent.

The differences in sentence length between Other Race female
offenders who received within range sentences and White male offenders
have decreased slightly in the recent period studied. In the Gall period,
that difference was 40.2 percent, the largest percentage difference for
all race/gender pairs for all time periods studied, but decreased to 37.2
percent during the Post-Report period.

In the Post-Report period, Other Race male offenders received
sentences that were 7.9 percent shorter than those imposed on White
male offenders. There was no statistically significant difference between
the sentences for these two groups in the Gall period.

Fig. 4. Differences in Within Range Sentences for Male Offenders
Koon, PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

Fig. 5. Differences in Within Range Sentences for Female Offenders
Koon, PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

PROTECT

Koon

Booker

Gall

Post-Report

12.2

Koon

6.6
3.8

7.9

-15
-20

-30
-5

-4.2

-5.7

-7.9

-8.1 -8.3

-10
Black Male vs. White Male

Hispanic Male vs. White Male

Other Male vs. White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

Demographic Differences in Sentencing

-11.9

-13.5

-14.2
-15.9
-17.2
-24.1
-27.3

-35

-4.4

-13.6

-12.1

-20.8
-24.3 -25.0

-24.1
-29.7

-27.1
-31.5

-37.2
-40.2

-40
-45

Post-Report

-5.5

-25

Gall

-10

3.5

Booker

PROTECT

-5

Sentences for Hispanic female offenders followed a different
pattern. The sentences for Hispanic females who received within range
sentences were not statistically different from those for White male
offenders during the Post-Report period. This is a change from all
prior periods studied, where Hispanic female within range sentences
were lower than those for White male offenders. For example, in the
Gall period, Hispanic female offenders sentenced within the guideline
range had sentences that were 5.5 percent shorter than the within range
sentences for White male offenders.

White Female vs.
White Male

Black Female vs.
White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

Hispanic Female vs.
White Male

Other Female vs.
White Male

Below Range Sentences

There were also no statistically significant differences between
White male and Hispanic male offenders who received substantial
assistance departures during any period. Similarly, in only one period
did Other Race male offenders have a statistically significant difference
from White male offenders. During the Booker period Other Race males
received sentences that were 24.4 percent longer than those for White
males.

Substantial Assistance Departure Sentences

The Commission also examined cases in which the sentence
imposed was below the applicable guideline range to determine whether
demographic differences were present. These “below range” sentences
can occur for different considerations. For statistical purposes, the
Commission groups below range sentences into two broad categories—
those sponsored by the government and those not sponsored by the
government.22 Government sponsored below range sentences are further
subdivided.23 Each will be addressed in turn.

Figure 6 presents the results of the analysis of below range
sentences based on substantial assistance departures. During the
Koon period, Black male offenders who received substantial assistance
departures received sentences 19.2 percent longer than White male
offenders who received such departures. However, there were no
statistically significant differences between the sentences of Black and
White male offenders who received substantial assistance departures
during any later period, including the Post-Report period.

Fig. 6. Differences in Substantial Assistance Departure Sentences for
Male Offenders
Koon, PROTECT, Booker, Gall, and Post-Report Periods

Black female offenders who received substantial assistance
departures also received shorter sentences than White male offenders
who received substantial assistance departures. During the Post-Report
period, Black female offenders who received substantial assistance

Fig. 7. Differences in Substantial Assistance Departure Sentences for
Female Offenders
Koon, PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

Percentage Difference in Sentence Length

Koon

Booker

PROTECT

Gall
24.4

Post-Report

Koon

PROTECT

Booker

Gall

Post-Report
*

-10
20

19.2

-20
-30

-40
10

-20.5
-26.0
-27.4

-31.5
-35.5

-21.1
-29.0

-45.1

-50

-25.5

-27.7
-31.6

-29.1
-35.6
-35.9
-41.7

-26.4
-38.9

-60

-70
0

Black Male vs. White Male

Hispanic Male vs. White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

Other Male vs. White Male

-80

-74.0
White Female vs.
White Male

Black Female vs.
White Male

Hispanic Female vs.
White Male

Other Female vs.
White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

Demographic Differences in Sentencing

departures received sentences that were 21.1 percent shorter than those
for White male offenders who received substantial assistance departures.
However, this was a decrease from the 31.6 percent difference observed
in the Gall period.

Figure 8 presents the findings of the Commission’s multivariate
analysis of government sponsored below range sentences for reasons
other than the offender’s substantial assistance.25 Cases from the Koon
period were excluded from this analysis because the Commission did not
collect data during that period in a manner that would indicate whether
a sentence below the guideline range for reasons other than substantial
assistance was initiated by the government or the court.

Sentences for Hispanic female offenders who received
substantial assistance departures followed a somewhat similar pattern.
In the Post-Report period, sentences for Hispanic female offenders
who received a substantial assistance departure were 25.5 percent
shorter than those for White male offenders who received a substantial
assistance departure. In the Gall period this difference was 35.6 percent.

The analysis found that the differences in sentence length
were statistically significant during some periods. During the PostReport period, sentence lengths for Black males who received an other
government sponsored below range sentence were 28.7 percent longer
than those for White males who received other government sponsored
below range sentences. However, no statistically significant difference
was noted during the Gall period. There were no statistically significant
differences between the sentence lengths of Other Race male offenders
who received an other government sponsored below range sentence and
White male offenders except during the Post-Report period, where Other
Race male offenders received sentences that were 17.6 percent longer
than those for White males.

The differences in sentence lengths between Other Race
female offenders and White male offenders who received a substantial
assistance departure have varied over all five time periods studied.24 In
the Gall period this difference was 26.4 percent, however, in the PostReport period no statistically significant difference was observed.

Fig. 8. Differences in Other Gov’t Sponsored Below Range Sentences for
Male Offenders
PROTECT, Booker, Gall, and Post-Report Periods

Other Government Sponsored Below Range Sentences

Fig. 9. Differences in Other Gov’t Sponsored Below Range Sentences for
Female Offenders
PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

Percentage Difference in Sentence Length

PROTECT

Booker

Gall

23.6

22.8

Gall

Post-Report
*

-10

25.9
25

29.8

28.7

Booker

PROTECT

Post-Report

-20

17.6

-26.0

-30

-27.3

-32.3

-28.7

-40

-38.6

-41.4

-50

5
*

Black Male vs.
White Male

*
Hispanic Male vs.
White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 2003 -2016 Datafiles, USSCFY03-USSCFY16.

Other Male vs.
White Male

-48.8

-60
-70

-64.3

White Female vs.
White Male

Black Female vs.
White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 2003 -2016 Datafiles, USSCFY03-USSCFY16.

Demographic Differences in Sentencing

-46.4

-48.2

Hispanic Female vs.
White Male

Other Female vs.
White Male

The differences in sentence length between Hispanic male
offenders and White male offenders who received an other government
sponsored below range sentence were statistically significant during
three periods. In the Post-Report period, Hispanic male offenders who
received an other government sponsored below range sentence received
sentences that were 25.9 percent longer than those for White male
offenders who received an other government sponsored below range
sentence. During the Gall period, the difference was 23.6 percent.

Figure 10 depicts the results of the Commission’s multivariate
analysis of non-government sponsored below range sentences, i.e.,
downward departures and variances resulting in sentences below the
guideline range for any reason not initiated by the government. Only four
time periods are included in this multivariate analysis: the PROTECT Act,
Booker, Gall, and Post-Report periods.26

The analysis found that in the Post-Report period, Black male
offenders who received a non-government sponsored below range
sentence received sentences that were 16.8 percent longer than those
for White male offenders who received that type of sentence. The only
other period in which such a difference was noted was the Booker period,
where the difference was 12.3 percent. There were no statistically
significant differences between the two groups in the other two periods.

For the periods in which there were statistically significant
differences, female offenders of all races who received an other
government sponsored below range sentence were sentenced to
shorter terms than White male offenders. In the Post-Report period, the
sentences of White female offenders who received an other government
sponsored below range sentence were 26.0 percent shorter than those of
White males who received an other government sponsored below range
sentence. The difference was a decrease from 41.4 percent during the
Gall period.

The sentences of Black female offenders who received an other
government sponsored below range sentence were 48.2 percent shorter
than those of White male offenders who received an other government
sponsored below range sentence during the Gall period, the only period
in which there were statistically significant differences. There was no
statistically significant difference between the two groups in the PostReport period.

Other Race female offenders who received an other government
sponsored below range sentence in the Post-Report period received
sentences that were 38.6 percent shorter than those for White male
offenders who received an other government sponsored below range
sentence. Although similar differences were noted in earlier periods, no
statistically significant difference was observed during the Gall period.

Sentences for Hispanic female offenders who received an
other government sponsored below range sentence were 28.7 percent
shorter in the Post-Report period. There were no statistically significant
differences between the sentences of Hispanic female offenders who
received an other government sponsored below range sentence and
White male offenders who received an other government sponsored
below range sentence during the Gall period.

Non-Government Sponsored Below Range Sentences

Similarly, the differences in sentence length between
Hispanic male offenders and White male offenders who received a
non-government sponsored below range sentence were statistically
significant in only two periods. In the Gall period, Hispanic male
offenders who received a non-government sponsored below range

Fig. 10. Differences in Non-Gov’t Sponsored Below Range Sentences for
Male Offenders
PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

PROTECT

Booker

Gall

Post-Report

16.8
15
12.3
10.6

9.3

Black Male vs. White Male

Hispanic Male vs. White Male

Other Male vs. White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 2003 -2016 Datafiles, USSCFY03-USSCFY16.

Demographic Differences in Sentencing

sentence received sentences 9.3 percent longer than those for White male
offenders who received that type of sentence. In the Post-Report period
the difference was 10.6 percent. There were no statistically significant
differences between the sentences imposed on Other Race male
offenders and White male offenders who received a non-government
sponsored below range sentence in any period.

were 17.2 percent shorter in the Post-Report period, a decrease from the
23.4 percent difference observed in the Gall period.

Sentences for Hispanic female offenders who received a nongovernment sponsored below range sentence were 17.2 percent shorter
in the Post-Report period, an increase from the 13.2 percent difference
during the Gall period. Other Race female offenders who received a nongovernment sponsored below range sentence received sentences that

There are difficulties in studying all types of below range
sentences and, in particular, the several types of below range sentences
sponsored by the government.29 For example, an analysis of the
likelihood of an offender receiving a substantial assistance departure
may be incomplete because the Commission does not have data
regarding which defendants who did not receive a substantial assistance
departure were eligible for a substantial assistance departure in the first
instance. Also, government sponsored departures due to an offender’s
willingness to participate in an Early Disposition Program (EDP)30
would be difficult to analyze, because most districts did not have an
EDP program during the periods studied. Additionally, because there
is little racial diversity in the offenders who receive EDP departures,
their inclusion might mask important findings. Finally, with respect to
the category of “other government sponsored below range sentences,”
which typically result from plea agreements, the Commission had no data
regarding which offenders were offered the opportunity to accept a plea
bargain but declined. For these reasons, the Commission’s analysis of
below range sentences is limited to non-government sponsored below
range sentences.

Among female offenders, White female offenders who received
a non-government sponsored below range sentence received sentences
that were 25.4 percent shorter in the Post-Report period compared to
21.3 percent shorter in the Gall period. No other time period evidenced
a statistically significant difference in sentence length when comparing
the two groups. Similarly, Black female offenders who received a nongovernment sponsored below range sentence received sentences that
were 21.6 percent shorter in the Post-Report period compared to 22.1
percent shorter during the Gall period.

Fig. 11. Differences in Non-Gov’t Sponsored Below Range Sentences for
Female Offenders
PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

Booker

PROTECT
*

Gall

Post-Report

-5
-10
-13.2

-15

-17.2

-20

-21.3
-25

-22.1 -21.6
-25.4

-30

White Female vs.
White Male

-26.7
Black Female vs.
White Male

Hispanic Female vs.
White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 2003 -2016 Datafiles, USSCFY03-USSCFY16.

-23.4

-24.0

Demographic Differences in Sentencing

Other Female vs.
White Male

Analysis of Likelihood of Obtaining a Below Range
Sentence

In this part of the report, the Commission presents the results
of an analysis of an offender’s likelihood of receiving a non-government
sponsored below range sentence, when judges had the discretion to
impose such a sentence.27 As discussed above, the Commission first
undertook this type of analysis as part of the 2012 Booker Report.28
At that time, the Commission found differences associated with
demographic factors in the likelihood that offenders would receive a
non-government sponsored below range sentence. These results could
explain part of the overall demographic differences in sentence lengths
that the Commission has observed in its prior reports.

The independent variables used in this analysis were the same
as those used in the regression analysis of all cases. In this analysis, the
Commission measured the likelihood that an offender would obtain a
non-government sponsored below range sentence when the sentencing
judge had the option of imposing it. That is, this analysis excluded cases

in which a below range sentence was not possible, such as when the
sentencing guideline range included an applicable statutory mandatory
minimum penalty, or in cases where some aspect of the sentencing was
otherwise outside the court’s discretion, such as when the court accepts
a binding plea agreement. The Koon period was not included in this
analysis.31

Figure 12 depicts the results of the Commission’s findings. The
analysis found that Black and Hispanic male offenders were less likely
to receive a non-government sponsored below range sentence than
White male offenders during all periods studied. In the Post-Report
period, Black male offenders were 21.2 percent less likely than White
male offenders to receive a non-government sponsored below range
sentence. In the Gall period, this difference was 25.2 percent. This
analysis indicated that Black male offenders were consistently less likely
to receive a non-government sponsored below range sentence compared
to White male offenders.

For Hispanic male offenders, the Commission found a 31.4
percent difference in the likelihood of receiving a non-government
sponsored below range sentence compared to White male offenders in
the Post-Report period. This difference was virtually unchanged from the
31.6 percent difference observed in the Gall period.

Fig. 12. Likelihood of Obtaining Non-Gov’t Sponsored Below Range
Sentences for Male Offenders
PROTECT, Booker, Gall, and Post-Report Periods
PROTECT

Booker

Gall
*

Post-Report
*

Fig. 13. Likelihood of Obtaining Non-Gov’t Sponsored Below Range
Sentences for Female Offenders
PROTECT, Booker, Gall, and Post-Report Periods
40

Booker

PROTECT

Gall

Post-Report

37.6

37.1

-10

-15
-15.7

-20
-20.9
-25.0

-21.2
-25.2

11.5

14.0

13.1
9.5

-25.3

-30
-35

When comparing Black female offenders to White male
offenders, only in the Post-Report period was there a statistically
significant difference between the two groups regarding the likelihood
of receiving a non-government sponsored below range sentence. In the
Post-Report period, Black female offenders were 9.5 percent more likely
than White male offenders to receive a non-government sponsored below
range sentence.

-5

-25

As depicted in figure 13, White female offenders were 13.1
percent more likely than White male offenders to receive a nongovernment sponsored below range sentence during the Post-Report
period. This compared to an 11.5 percent difference in the Gall period.
There were no statistically significant differences in the Booker period
when comparing the likelihood of White female offenders receiving a
non-government sponsored below range sentence to the likelihood of
White male offenders receiving such a sentence.

Likelihood of Obtaining a Non-Government Sponsored Below Range Sentence

There were no statistically significant differences in any time
period when comparing the likelihood of an Other Race male offender
receiving a non-government sponsored below range sentence compared
to the likelihood of a White male offender receiving such a sentence.

5
-31.6
Black Male vs. White Male

-31.4

Hispanic Male vs. White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 2003 -2016 Datafiles, USSCFY03-USSCFY16.

Other Male vs. White Male

*
White Female vs.
White Male

Black Female vs.
White Male

Hispanic Female vs.
White Male

Other Female vs.
White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 2003 -2016 Datafiles, USSCFY03-USSCFY16.

Demographic Differences in Sentencing

The likelihood of Hispanic females receiving a non-government
sponsored below range sentence was 14.0 percent higher in the PostReport period. There was no statistically significant difference between
the two groups during the Gall period or the Booker period, although
there was a 37.6 percent higher likelihood than that of White male
offenders during the PROTECT Act period.

determine whether adding the data regarding violence had any effect on
the analysis, the Commission first performed an analysis of cases from
fiscal year 2016 without the additional data involving violence. Then the
analysis was replicated, with data concerning prior violence added into
the analysis. The two results were then compared.

Figure 14 shows the results of those analyses. Examining
only cases from fiscal year 2016, the Commission found a 20.7 percent
difference in the sentence length for Black male offenders compared to
White male offenders, and a 9.4 percent difference between Hispanic
male offenders and White male offenders. Any difference between the
sentences for Other Race male offenders and White male offenders was
not statistically significant.

The Effect of Adding Data on Prior Violence as a Factor in
the Model

As discussed above, the Commission’s 2012 Booker Report
and its 2010 Booker Multivariate Analysis noted that judges make
sentencing decisions based on many legal factors and other legitimate
considerations, and that data regarding some of these considerations was
not included in the Commission’s regression analyses because it was not
readily available.32 One specific example concerned information about
violence in an offender’s criminal history. As the Commission explained
in 2010:
[A] judge sentencing two offenders convicted of similar
crimes with the same criminal history score under the
federal sentencing guidelines might impose a longer
sentence on the offender with a more violent criminal
past than on the offender with a less violent, or nonviolent, criminal history. Similarly, a judge sentencing
two offenders convicted of similar crimes might be
influenced by the presence of violence in one case that
was not present in the other case and was not reflected
in the final offense level for those cases as determined
under the sentencing guidelines.33

To address this issue, the Commission examined cases in which
the offender was sentenced in fiscal year 2016 and collected information
about the types of prior offenses for which the offender had been
sentenced, including both federal and state crimes.34 Using this data,
the Commission determined whether the offender had ever committed
a violent offense.35 This data was then included in the Commission’s
regression analysis to assess whether adding the data had any effect on
the demographic differences in sentencing that had been observed.

Because the data regarding violent offenses was available
only for offenders sentenced in fiscal year 2016, the Commission’s
analysis had to be limited to cases from that fiscal year. Also, in order to
16

Demographic Differences in Sentencing

In order to determine whether adding the additional data
regarding violence had any effect on the analysis, the Commission then
replicated the analysis using the data on offenders’ violent criminal
history. To do this, the Commission added an independent variable in the
model to control for that fact. Through this technique, the Commission
was able to assess the effect of data about violence in the criminal history
not only on the demographic differences in sentencing but on the overall
model itself.

Fig. 14. Differences in Sentence Length Accounting for Prior Violence for
Male Offenders
Fiscal Year 2016
Percentage Difference in Sentence Length

2016

25
20.7
20

2016 - Violence CH Variable

20.4

9.4

9.7

*
Black Male vs. White Male

Hispanic Male vs. White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 2016 Datafile, USSCFY16.

Other Male vs. White Male

As can be seen in figures 14 and 15, the addition of the variable
indicating a prior conviction for a violent offense had almost no effect
on the contribution of race and gender to the sentence of the offender
after controlling for all other factors. For example, after controlling for
violence in the offenders’ criminal history, the difference in sentence
length between Black male offenders and White male offenders was
20.4 percent, a difference of only 0.3 percentage points from the results
found without the additional data. The difference in sentence length
between Hispanic male offenders and White male offenders also changed
only slightly, from 9.4 percent to 9.7 percent. Similarly, the difference in
average sentence between the remaining offender groups was virtually
identical to what it was without the additional data included in the
analysis.

Also, the Commission found that the effect of the data about
violent criminal history on the sentences imposed was, in itself, not
statistically significant. That is, violence in an offender’s past did not
have any independent effect on the sentence imposed over and above
the effect of the other variables measured. As discussed more fully
in the 2010 Booker Multivariate Analysis, interpreting regression
analysis results depends on several factors. One primary factor is the
“significance test,” which asks whether the results of the analysis for each

Fig. 15. Differences in Sentence Length Accounting for Prior Violence for
Female Offenders
Fiscal Year 2016
Percentage Difference in Sentence Length

2016

2016 - Violence CH Variable

-5
-10
-15
-20

A second factor helpful in assessing the results of regression
analysis is the R2 (or “r-squared”). This is a numerical measure of
the extent to which the results of a regression analysis account for all
the variation in the data used in the analysis. Put another way, it is a
representation of the “degree of fit” between the regression results
and the particular data being analyzed. A high R2 means that much of
the variation in the dependent variable has been accounted for by the
independent variables in the model whereas a low R2 indicates that
there are one or more variables missing from the model that affect the
dependent variable, or the variables selected have little to no association
with the independent variables.38

In theory, the addition of data about a factor that has an effect
on a phenomenon being studied should increase the R2 of a regression
analysis. If additional data does not increase the R2, then the researcher
can conclude that the factor does not help explain the phenomenon being
studied; that is, it has no actual effect on the outcome. The inclusion of
the prior violence data in the Commission’s regression analysis had no
effect in the overall R2 of the model.39 Based on this, the Commission
concludes that violence in an offender’s criminal history has no effect on
the sentence imposed over and above the contribution that the sentence
imposed for the crime in which the violent act occurred has on the
offender’s criminal history score.40

Limitations of Regression Analysis

-25
-30

independent variable are “statistically significant.” Statistical significance
is a determination of the probability that the measured relationship
between an independent variable and the dependent variable is the
result of random chance (i.e., that the measured relationship does not,
in fact, reflect a true association).36 The Commission found that the
prior violence variable itself had no statistically significant effect on the
dependent variable (i.e., sentence length).37 That is, it cannot be said that
violence in an offender’s past has any true association with the sentence
imposed.

-27.2

-27.0

-26.4

-35
-37.7

-40
White Female vs.
White Male

Black Female vs.
White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 2016 Datafile, USSCFY16.

Hispanic Female vs.
White Male

-37.5

Other Female vs.
White Male

In its prior reports,41 the Commission noted that results from its
analyses should be taken with caution. Although regression analysis is
a tool commonly used by social scientists, as well as in a variety of legal
contexts, to examine the relationship between multiple factors, 42 it has
limitations. In particular, one or more key factors that could affect the
analysis may have been omitted from the methodologies used because
a particular factor is unknown, or because data about it is not readily
available.
Demographic Differences in Sentencing

For example, judges may consider potentially relevant
information available to them in a presentence report, such as an
offender’s employment history or family circumstances. However,
the Commission does not routinely extract this information from the
sentencing documents it receives and, therefore, data about those
factors are not controlled for in this analysis. Additionally, judges may
make decisions about sentencing offenders based on other legitimate
considerations that cannot be measured.

Because multivariate regression analysis cannot control for all of
the factors that judges may consider, the results of the analyses presented
in this report should be interpreted with caution and should not be taken
to suggest discrimination on the part of judges. Multivariate analysis
cannot explain why the observed differences in sentencing outcomes
exist, but only that they do exist.

Demographic Differences in Sentencing

Section III

Conclusion

In this report, the Commission has provided an update to its
prior reports on demographic differences in sentencing. The Commission
found that sentence length continues to be associated with some
demographic factors, in particular race and gender. After controlling for
a wide variety of sentencing factors, the Commission found that Black
male offenders continued to receive longer sentences than similarly
situated White male offenders, and that female offenders of all races
received shorter sentences than White male offenders.

Non-government sponsored departures and variances appear
to contribute significantly to the difference in sentence length between
Black male and White male offenders. Black male offenders were less
likely than White male offenders to receive a non-government sponsored
downward departure or variance during the most recent period studied.
Further, even when Black male offenders received a non-government
sponsored departure or variance, their sentences were longer than White
male offenders who received a non-government sponsored departure or
variance.

The Commission also found that prior violent crimes, as
documented in an offender’s criminal history, do not significantly
contribute to demographic differences in federal sentencing. An
offender’s past criminal violence is not a statistically significant predictor
of the sentence imposed for a federal offense to any extent beyond the
contribution it makes to the offender’s final sentence imposed through
operation of the criminal history score under the sentencing guidelines.

Demographic Differences in Sentencing

Appendices

Appendix A: Analysis of Other Demographic Factors
Results of Demographic Multivariate Regression Analysis for
All Cases

As discussed in the body of this report, the Commission’s
multivariate regression analysis for all cases, updated through fiscal
year 2016, continues to demonstrate that some demographic factors are
associated with sentence length to a statistically significant extent during
the periods studied. In this appendix, the Commission will focus on the
demographic factors of citizenship, education, and age of the offender.
Again, the analysis will focus primarily on the most recent period
studied—the Post-Report period—as compared to the immediately
preceding period, the Gall period; however, the results for all periods are
presented in the figures that accompany this discussion.43

Figure 16 depicts differences in sentence length associated with
the demographic factors of citizenship, education, and age for all cases.
Non-citizen offenders received sentences that were 5.0 percent longer
than those for United States citizens during the Post-Report period,
although this difference was lower than observed in any of the prior
periods studied. During the Gall period the difference was 10.4 percent
longer than those of United States citizens.

Fig. 16. Differences in Sentence Length in All Cases by Demographic Factors
Koon, PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

Koon

Booker

Gall

Post-Report

10.4

8.5

7.4

5.0

3.6

3.1

2.5

2.9

-5

-10

PROTECT

-3.9

-5.4

-4.6

-7.8
Non-U.S. Citizen vs.
U.S. Citizen

Any College vs.
No College

Over 25 Years of Age vs.
25 Years of Age or Younger

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

Demographic Differences in Sentencing

The differences in sentences for offenders with at least some
college education compared to offenders with no college education was
not statistically significant in the Post-Report period, although those
differences had been statistically significant in prior periods. During
the Gall period, offenders with at least some college education received
sentences that were 4.6 percent shorter than those of offenders with no
college education.

Finally, offenders over the age of 25 had slightly longer sentences
than those 25 years of age or younger, with a difference of 2.9 percent in
the Post-Report period. The differences between the sentences for these
two groups was 2.5 percent in the Gall period.

Analysis of Differences in Sentencing by Guideline
Application

For this analysis, cases were divided into four groups: (1) those
where the sentence was within the sentencing guideline range; (2) those
where the sentence was above the guideline range;44 (3) those where the
sentence was imposed below the guideline range at the request of the
government; and (4) all other below range sentences.45

Within Range Sentences

Figure 17 depicts the results of the multivariate analysis
regarding citizenship, education, and age for within range sentences.
During the Koon period, non-citizen offenders sentenced within the
guideline range received sentences that were 5.5 percent shorter than
those for United States citizens sentenced within the range. However, in
the four periods afterward, the within range sentences for non-citizens
were longer than those for United States citizens. This difference was 7.0
percent in the Gall period, and 6.3 percent in the Post-Report period.

The Commission’s prior analysis found no statistically significant
differences in the sentence length of offenders who had attended college
and received within range sentences compared to offenders who had
not attended college and received within range sentences. However, in
the Post-Report period, the Commission found that those who attended
college received sentences that were 1.9 percent longer than offenders
who did not attend college.

Similarly, the Commission’s prior analysis found no statistically
significant differences in the sentence length for offenders over the age of
25 who received within range sentences compared with the within range

Fig. 17. Differences in Within Range Sentences by Demographic Factors
Koon, PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

Koon

PROTECT

Booker

Gall

Post-Report

sentences for offenders 25 years of age or younger. In the Post-Report
period, however, the Commission found a 2.8 percent decrease in the
within range sentences of those who were over 25 years of age compared
to those who were 25 years of age and younger.
Below Range Sentences

The Commission also examined cases in which the sentence
imposed was below the applicable guideline range to determine whether
demographic differences were present. These “below range” sentences
can occur for different considerations. For statistical purposes, the
Commission groups below range sentences into two broad categories—
those sponsored by the government and those not sponsored by the
government.46 Government sponsored below range sentences are further
subdivided.47 Each will be addressed in turn.

Figure 18 depicts the findings of the analysis regarding
citizenship, education, and age for cases in which there was a substantial
assistance departure. There were no statistically significant differences
between the sentence lengths of non-citizen offenders compared to
Fig. 18. Differences in Substantial Assistance Departure Sentences by
Demographic Factors
Koon, PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

9.7

10
6.0

-2

7.0

Booker
*

Gall

Post-Report

-8

*
*

-10.5

-12

-2.8

-5

-8.4

-10

-11.8

-14

-18

Any College vs.
No College

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

Over 25 Years of Age vs.
25 Years of Age or Younger

-12.5

-14.0 -14.3

-16

-5.5

Non-U.S. Citizen vs.
U.S. Citizen

PROTECT

-6

1.9

-10

Koon
*

-4

6.3

Substantial Assistance Departure Sentences

-18.2

-20
Non-U.S. Citizen vs.
U.S. Citizen

Any College vs.
No College

Over 25 Years of Age vs.
25 Years of Age or Younger

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

Demographic Differences in Sentencing

sentence lengths of United States citizens who received substantial
assistance departures in any period.

The differences in sentences for offenders with some college
education who received substantial assistance departures compared to
offenders with no college education who received substantial assistance
departures were statistically significant in all but one time period. In the
Post-Report period, offenders with any college education who received
substantial assistance departures received sentences that were 10.5
percent shorter than offenders with no college education who received
such departures. During the Gall period this difference was 14.3 percent.

The differences between sentences of offenders over the
age of 25 who received a substantial assistance departure compared
with those 25 years of age or younger who received such a departure
were statistically significant in only the last three time periods. In
the Post-Report period, offenders over the age of 25 who received a
substantial assistance departure received sentences 12.5 percent shorter
than offenders who were 25 years of age or younger who received a
substantial assistance departure, a slight increase from the 11.8 percent
difference in the Gall period.

Fig. 19. Differences in Other Gov’t Sponsored Below Range Sentences by
Demographic Factors
PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

PROTECT

Booker

Gall

Post-Report

27.7
22.9

-10

-10.1
-15.2

-20
Non-U.S. Citizen vs.
U.S. Citizen

Any College vs.
No College

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 2003 -2016 Datafiles, USSCFY03-USSCFY16.

Demographic Differences in Sentencing

Other Government Sponsored Below Range Sentences

Figure 19 presents the findings of the Commission’s multivariate
analysis of government sponsored below range sentences for reasons
other than the offender’s substantial assistance.48 Cases from the Koon
period were excluded from this analysis because the Commission did not
collect data during that period in a manner that would indicate whether
a sentence below the guideline range for reasons other than substantial
assistance was initiated by the government or the court.

Figure 19 shows the Commission’s findings of the multivariate
analysis as to citizenship, education, and age specific to other government
sponsored below range sentences. In the Gall period, the sentence for
non-citizen offenders was 19.2 percent longer than those for U.S. citizen
offenders. This difference increased to 27.7 percent in the Post-Report
period.

There was no statistically significant difference in sentence
length between offenders with at least some college education
compared to offenders with no college education who received an other
government sponsored below range sentence in any period. In the Gall
period, offenders over the age of 25 who received an other government
sponsored below range sentence received sentences 18.5 percent shorter
than offenders who were 25 years of age or younger and received an
other government sponsored below range sentence. In the Post-Report
period this difference was 10.1 percent.

Non-Government Sponsored Below Range Sentences

Figure 20 on the next page depicts the results of the
Commission’s multivariate analysis of non-government sponsored below
range sentences, i.e., those sentences falling below the guideline range for
any reason not initiated by the government. Only four time periods are
included in this multivariate analysis: the PROTECT Act, Booker, Gall, and
Post-Report periods.49

19.2

-18.5

Over 25 Years of Age vs.
25 Years of Age or Younger

The Commission’s findings regarding citizenship, education,
and age for non-government sponsored below range sentences are
depicted in Figure 20. During the Post-Report period, the sentences for
non-citizens who received a non-government sponsored below range
sentence were 24.3 percent longer than those for United States citizens
who received a non-government sponsored below range sentence. In the
Gall period, that difference was 25.2 percent. There were no statistically

significant differences in sentence length between these two groups in
the PROTECT Act period.

Only in the Post-Report period was there a statistically
significant difference found in sentence length between offenders with at
least some college education who received a non-government sponsored
below range sentence and offenders with no college education who
received that type of sentence. Offenders who attended college received
sentences 7.0 percent higher than those who did not attend in the PostReport period. In all prior periods, no statistically significant differences
in non-government sponsored below range sentences were observed.

The differences between sentences of offenders over the age of
25 who received a non-government sponsored below range sentence
and those 25 years of age or younger who received a non-government
sponsored below range sentence were not statistically significant in the
Post-Report period. In the Gall period the difference was 8.4 percent.

Booker

Gall

Post-Report

PROTECT

24.3

20.7

22.1

20.4

18.2

23.7

Booker

Gall

Post-Report

22.5

18.1

17.2

14.1

11.2

7.0

5
*

-10

-10.3
Non-U.S. Citizen vs.
U.S. Citizen

Any College vs.
No College

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 2003 -2016 Datafiles, USSCFY03-USSCFY16.

7.5

-5

-15

In the Post-Report period, offenders with at least some college
education were 22.5 percent more likely to receive a non-government
sponsored below range sentence than offenders with no college
education. The difference was 23.7 percent in the Gall period.

Likelihood of Obtaining a Non-Government Sponsored Below Range Sentence

PROTECT

As depicted in Figure 21, during the Post-Report period, noncitizen offenders had a 20.4 percent higher likelihood of receiving a
non-government sponsored below range sentence than United States
citizen offenders. The difference was an increase from the 11.2 percent
difference observed during the Gall period.

Fig. 21. Likelihood of Obtaining Non-Gov’t Sponsored Below Range
Sentences by Demographic Factors
PROTECT, Booker, Gall, and Post-Report Periods

Percentage Difference in Sentence Length

In this part of the Appendix, the Commission presents the results
of an analysis of an offender’s likelihood of receiving a non-government
sponsored below range sentence, when judges had the discretion to
impose such a sentence, and any association with the demographic
factors of citizenship, education, and age.50

The differences between sentences of offenders over the age of
25 compared with those 25 years of age or younger were statistically

Fig. 20. Differences in Non-Gov’t Sponsored Below Range Sentences by
Demographic Factors
PROTECT, Booker, Gall, and Post-Report Periods

25.2

Analysis of Likelihood of Obtaining a Below Range
Sentence

-8.4

Over 25 Years of Age vs.
25 Years of Age or Younger

-3.7

-5
-10

Non-U.S. Citizen vs.
U.S. Citizen

Any College vs.
No College

Over 25 Years of Age vs.
25 Years of Age or Younger

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 2003 -2016 Datafiles, USSCFY03-USSCFY16.

Demographic Differences in Sentencing

significant in three time periods, but in different directions. In the PostReport period, offenders who were over 25 years of age were 3.7 percent
less likely to receive a non-government sponsored below range sentence
than offenders who were 25 years of age and younger. This was a
reverse of the relationship observed during the PROTECT Act and Booker
periods, where offenders over the age of 25 were more likely to receive
a non-government sponsored below range sentence than offenders who
were 25 years of age or younger. By the time of the Gall period, however,
no statistically significant differences were found between the two
groups.

Appendix B: Offense Specific Analysis

Analysis of Differences in Sentencing by Offense Type

For the 2012 Booker Report, the Commission expanded its
multivariate analysis to examine specific types of offenses.51 While the
Commission’s multivariate analyses always control for the type of offense
involved in the case, the additional analyses presented in the 2012
Booker Report used the regression technique to examine a single offense
more closely, by limiting the cases in each analysis to those involving that
specific offense type. The Commission performed this additional analysis

Fig. 22. Differences in Sentence Length for Male Drug Trafficking Offenders
Koon, PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

Koon

PROTECT

Booker

Gall

Post-Report

17.7

15
13.1
11.8 11.8
10.6
10

9.1

9.2

6.8
4.8

Black Male vs. White Male

4.3

Hispanic Male vs. White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

Demographic Differences in Sentencing

Other Male vs. White Male

for two reasons: first, to determine whether its findings on demographic
differences in overall sentencing outcomes would be replicated in the
major offense types; and second, to determine the extent to which
demographic differences in sentencing for different offense types
contributed to the demographic differences observed for all sentences.
In the 2012 Booker Report the Commission presented an analysis of three
major offense types for which there was a sufficient number of cases and
sufficient diversity in the demographic factors: drug trafficking, fraud,
and firearms. Immigration and child pornography offenses lacked a
sufficient number of cases with offenders of different races to support a
robust analysis.52

Consistent with the results of the analysis of all cases,
demographic factors were associated with sentence length to a
statistically significant extent during some of the time periods studied
for drug trafficking, fraud, and firearms offenses. For this report, the
Commission has updated these offense type analyses with data from the
Post-Report period.53
Drug Trafficking Offenses

Figure 22 depicts the results of the Commission’s multivariate
analysis specific to drug trafficking offenses. This analysis includes
offenders involved in trafficking of the five major drug types—powder
cocaine, crack cocaine, marijuana, methamphetamine, and heroin—while
controlling for the type of drug involved in the offense. Controlling for
drug type means, for example, that crack cocaine offenders are compared
only to other crack cocaine offenders, while methamphetamine offenders
are compared only to other methamphetamine offenders. The analysis
found that the differences in sentence length for Black male drug
offenders compared to White male drug offenders have varied over time.
In the Post-Report period, Black male drug offenders received sentences
that were 17.7 percent longer than White male drug offenders. In the
Gall period the difference between these two groups was 13.1 percent.

The differences in sentence length between Hispanic male and
White male drug offenders were statistically significant in three of the
five time periods studied. In the Post-Report period Hispanic male drug
offenders received sentences 6.8 percent longer than White male drug
offenders. This compares to 4.3 percent during the Gall period. There
were no statistically significant differences between the sentences for the
two groups during the Koon or PROTECT Act periods.

Similarly, there were no statistically significant differences in
sentence length between Other Race male and White male drug offenders
during the Koon or PROTECT Act periods. However, in both the PostReport and Gall periods, the differences between the sentences for the
two groups had increased to 11.8 percent.

Female drug offenders of all races received sentences that were
shorter than those of White male drug offenders in nearly all of time
periods studied. In the Post-Report period, White female drug offenders
received sentences 26.6 percent shorter than White male drug offenders.
This difference was 22.4 percent during the Gall period.

Sentence length for Black female drug offenders compared
to White male drug offenders followed a different pattern, with Black
female drug offenders receiving increasingly shorter sentences over the
five time periods. These differences increased steadily to the Post-Report
period, where Black female drug offenders received sentences that were
38.2 percent shorter than those for White male drug offenders.

Sentences for Hispanic female drug offenders also were shorter
than those of White male drug offenders during all periods studied. In
the Post-Report period, sentences for Hispanic female drug offenders
were 23.8 percent shorter than for White male drug offenders. This

Fig. 23. Differences in Sentence Length for Female Drug Trafficking Offenders
Koon, PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

Koon

Booker

PROTECT

Gall

* Post-Report
*

-11.9

-15

The differences in sentences for drug offenders with at least
some college education compared to drug offenders with no college
education were statistically significant in four time periods. Drug
offenders with at least some college education received sentences that
were 4.5 percent shorter than drug offenders with no college experience
in the Post-Report period, and 7.8 percent shorter in the Gall period.
Fig. 24. Differences in Sentence Length for Drug Trafficking Offenders by
Demographic Factors
Koon, PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

Koon

PROTECT

Booker

Gall

Post-Report

-22.4

-25

-26.6

-27.0

-15.0
-17.1

-15.2

-20

-30.6

-20.4-21.3
-23.9

-19.8

-20.9
-23.8

-28.5

-22.4

-5

White Female vs.
White Male

Black Female vs.
White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

*
-2.7
-6.0

Hispanic Female vs.
White Male

Other Female vs.
White Male

-15

-4.9

-6.1
-8.6

Non-U.S. Citizen vs.
U.S. Citizen

-10

7.2

5.1

4.8

-26.2

-38.2

-40
-45

Figure 24 depicts the results of the analysis as to citizenship,
education, and age for drug trafficking offenses. Non-citizen drug
offenders received shorter sentences than United States citizen drug
offenders in three of the five time periods, although there were no
statistically significant differences in sentence length between these two
groups in the Post-Report and Gall periods.

-10

-35

Other Race female drug offenders received sentences 22.4
percent shorter than White male drug offenders during the Post-Report
period. However, there was no statistically significant difference in
sentence length observed between Other Race female drug offenders and
White male drug offenders during the Gall period.

-5

-30

difference was 17.1 percent in the Gall period.

-4.5
-7.8

Any College vs.
No College

Over 25 Years of Age vs.
25 Years of Age or Younger

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

Demographic Differences in Sentencing

Finally, there were differences between sentences of drug
offenders over the age of 25 compared with those 25 years of age or
younger during three of the five periods studied. In the Post-Report
period, drug offenders over the age of 25 received sentences that were
7.2 percent longer than drug offenders 25 years of age or younger.
During the Gall period, there were no statistically significant sentencing
differences between these two groups of offenders.

There were no statistically significant differences in sentence length
between Other Race male and White male fraud offenders during any
time period.

As seen in figure 26, female fraud offenders of all races generally
were sentenced to shorter terms than White male fraud offenders. In the
Post-Report period, White female fraud offenders received sentences that
were 20.9 percent shorter than those for White male fraud offenders. In
the Gall period, that difference was 18.3 percent. During the PROTECT
Act and Booker periods, there were no statistically significant differences
in sentence length between the two groups.

Fraud Offenses

The findings of the Commission’s multivariate analysis specific
to fraud offenses are presented in Figure 25. The differences in sentence
length for Black male fraud offenders compared to White male fraud
offenders were statistically significant during the Post-Report period,
when the difference was a 10.7 percent longer sentence for Black male
offenders. There was no statistically significant difference observed
during the Gall period.

The differences in sentence length between Hispanic male and
White male fraud offenders were statistically significant in each period.
In the Post-Report period, Hispanic male fraud offenders received
sentences that were 17.3 percent longer than White male fraud offenders.
That difference was a decrease from 29.6 percent in the Gall period.

Fig. 25. Differences in Sentence Length for Male Fraud Offenders
Koon, PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

Koon

PROTECT

Booker

Gall

Post-Report

Differences in sentence length for Hispanic female fraud
offenders compared to White male fraud offenders were not statistically
significant in the Post-Report and Gall periods.
Fig. 26. Differences in Sentence Length for Female Fraud Offenders
Koon, PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

0
29.6

-15

20.8
17.3

10.7

Booker

PROTECT
*

-20
-25

12.1

-12.2

-13.0

*
Hispanic Male vs. White Male

Other Male vs. White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

-20.9

-18.6 -18.1

-19.2

Demographic Differences in Sentencing

-40

-21.5

-23.0
-25.5
-28.9
-31.6

-35

Black Male vs. White Male

Post-Report

-15.4

-18.3

-30

7.4

Gall

-10

20
15

Koon

-5

25.6

The differences in sentence length for Black female fraud
offenders compared to White male fraud offenders were statistically
significant in each period. In the Post-Report period, Black female fraud
offenders received sentences 18.1 percent shorter than White male fraud
offenders. This difference was similar to the 18.6 percent observed in
Gall period.

-35.0
White Female vs.
White Male

Black Female vs.
White Male

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

Hispanic Female vs.
White Male

Other Female vs.
White Male

Other Race female fraud offenders received sentences that were
shorter than those for White male fraud offenders in three of the five
time periods studied. The difference in sentence length between the two
groups was not statistically significant in the Post-Report period. In the
Gall period, Other Race female fraud offenders received sentences that
were 21.5 percent shorter than White male fraud offenders.

For most of the periods studied, there were no statistically
significant sentence differences between fraud offenders over the age of
25 compared to those 25 years of age or younger. However, during the
Post-Report period, fraud offenders over the age of 25 had 11.5 percent
lower sentences than those 25 years of age or younger.

The differences in sentences for fraud offenders with at least
some college education compared to fraud offenders with no college
education were statistically significant in only the two most recent
periods. In the Post-Report period, fraud offenders with at least some
college education received sentences 9.4 percent longer than fraud
offenders with no college education. This difference was 8.4 percent in
the Gall period.

In contrast, there were no statistically significant differences
between the sentences of Hispanic male and White male firearms
offenders during any time period studied. The sentences for Other Race
male firearms offenders differed from those for White male firearms
offenders only in the Koon period. There were no statistically significant
differences in the sentences of these offenders in the four later time
periods.

The results of the analyses regarding citizenship, education,
and age for fraud offenders are depicted in Figure 27. Non-citizen fraud
offenders received sentences that were 29.7 percent shorter than United
States citizen fraud offenders in the Koon period. However, from the
PROTECT Act period through the Post-Report period, non-citizen fraud
offenders received sentences increasingly longer than United States
citizen fraud offenders. This difference was 24.1 percent longer in the
Post-Report period and 34.5 percent longer in the Gall period.

Fig. 27. Differences in Sentence Length for Fraud Offenders by
Demographic Factors
Koon, PROTECT, Booker, Gall, and Post-Report Periods
40
28.4

30.1

PROTECT

The results of the Commission’s multivariate analysis of firearms
offenses are depicted in Figure 28.54 The analysis found that, in general,
Black male firearms offenders received longer sentences than White
male firearms offenders. In the Post-Report period, Black male firearms
offenders received sentences that were 19.3 percent longer than those
for White male firearms offenders. This was an increase from the 10.2
percent difference observed in the Gall period.

Fig. 28. Differences in Sentence Length for Male Firearms Offenders
Koon, PROTECT, Booker, Gall, and Post-Report Periods
Percentage Difference in Sentence Length

Percentage Difference in Sentence Length

Koon

Firearms Offenses

Booker

Gall

Post-Report

Koon

PROTECT

Booker

Gall

Post-Report

34.5

19.3

24.1

18.2

20
8.4

10
*

9.4

*
*

-10

-40

8.5
6.0

-11.5
-20
-30

10.2

5
-29.7
Non-U.S. Citizen vs.
U.S. Citizen

0
Any College vs.
No College

•Indicates that the difference between the two groups was not statistically significant.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

Over 25 Years of Age vs.
25 Years of Age or Younger

Black Male vs. White Male

Hispanic Male vs. White Male

Other Male vs. White Male

•Indicates that the difference between the two groups was not statistically significant.
Hispanic Females and “Other” Females are excluded from this analysis due to lack of firearm offenders of these races and gender.
SOURCE: U.S. Sentencing Commission, 1999 -2016 Datafiles, USSCFY99-USSCFY16.

Demographic Differences in Sentencing

In general, White female and Black female firearms offenders
received sentences that were shorter than sentences for White male
firearms offenders. In the Post-Report period, White female firearms
offenders received sentences 45.0 percent shorter than White male
firearms offenders. This difference was similar to the 44.0 percent
difference observed during the Gall period.

Differences in sentence lengths for Black female firearms
offenders compared to White male firearms offenders were statistically
significant during all five periods. During the Post-Report period, Black
female firearms offenders received sentences 45.1 percent shorter than
White male firearms offenders. However, this was a decrease from the
59.4 percent difference observed during the Gall period.

Regarding citizenship, education, and age, Figure 30 depicts the
results of these analyses specific to firearms offenders. There were no
statistically significant differences between the sentences of non-citizen
firearms offenders and those of United States citizen firearms offenders
during any time period.
Fig. 29. Differences in Sentence Length for Female Firearms Offenders
Koon, PROTECT, Booker, Gall, and Post-Report Periods
Koon

PROTECT

Booker

Gall

Differences in the sentences for offenders who were over the
age of 25 were only statistically significantly in the Post-Report period.
In that period, offenders over the age of 25 years received sentences
that were 5.7 percent longer than offenders who were 25 years of age or
younger.

Fig. 30. Differences in Sentence Length for Firearms Offenders by
Demographic Factors
Koon, PROTECT, Booker, Gall, and Post-Report Periods

Percentage Difference in Sentence Length

In contrast, there were statistically significant differences in
sentences for firearms offenders with at least some college education
compared to the sentences of firearms offenders with no college
education in all five time periods. Firearms offenders with at least some
college education received sentences 12.6 percent shorter than firearms
offenders with no college education in the Post-Report period. This
difference was 9.9 percent in the Gall period.

Post-Report

Percentage Difference in Sentence Length

Koon

PROTECT

Booker

Gall

Post-Report

-10

5.7
5

-20

-30
-40
-50

-38.5
-47.5

-45.0

-45.2

-45.1

-48.3

-8.4

-10

-10.4 -9.9

-59.4

-12.6

-15

-70

-73.5
White Female vs. White Male

Black Female vs. White Male

-5

-44.0

-60

-80

Demographic Differences in Sentencing

-20

-14.9

Non-U.S. Citizen vs.
U.S. Citizen

Any College vs.
No College

Over 25 Years of Age vs.
25 Years of Age or Younger

Appendix C: Methodology

The dependent variable used in each of the analyses was an
offender’s total sentence length, in months, which included alternatives
to imprisonment.55 Probationary sentences without conditions of
confinement are included as sentences of zero months. The independent
variables were:
•

•
•
•
•
•
•
•
•
•

The presumptive sentence, which is the bottom of the
applicable sentencing guideline range that applies in a
case (i.e., the minimum sentence, in months, to which the
offender was subject under the sentencing guidelines,
taking into account all guideline, statutory, and mandatory
minimum provisions);56
Type of offense committed (violent, sexual, pornography,
drug trafficking, white collar, immigration, or other);57

Whether a statutory mandatory minimum punishment was
applied at sentencing;58
Whether the court determined that a sentence outside the
applicable sentencing guideline range was warranted;59
Detention status (whether the offender had been released
on bail prior to sentencing);
Whether the offender pleaded guilty;

Race of the offender paired with the gender of the offender;
Citizenship of the offender (whether the offender was a
United States citizen);
Educational level of the offender; and
Age of the offender.

The multivariate analyses pair race and gender into eight distinct
groups: White males/females, Black males/females, Hispanic males/

females, and Other Race males/females. Reporting the results of the
analyses in this way identifies any differences in sentencing outcomes
associated with the offender’s race, gender, or both.

For the portion of this report regarding violence in an offender’s
criminal history, the Commission used data from a separate project
examining the criminal history of federal offenders. As part of that
project, the Commission examined the presentence investigation reports
(PSRs) for all offenders sentenced in fiscal year 2016 to collect data on all
juvenile adjudications and state and federal adult convictions. An optical
recognition program scanned the criminal history section of the PSRs and
recorded the following:

• Date of arrest;
•

Date of sentence;

•

Number of points assigned under Chapter 4 of the sentencing
guidelines;

•

State in which the court was located;

Chapter 4 guideline(s) cited;

Whether the offense was a juvenile adjudication or adult
conviction;

Whether the adjudication/conviction was in a federal, state,
or tribal court; and
Conviction charge(s).

The program assigned each adjudication or conviction to one
of 102 standardized offense categories. Those categories were further
grouped into 35 broad offense groups for analytical purposes. For all
offenders with criminal history, staff reviewed all of the information
recorded by the program to ensure accuracy, made any necessary changes
to the data, and confirmed the categorization of the adjudication or
offense into one of the 35 offense groups.
Demographic Differences in Sentencing

Assault, Simple Assault, Intimidation (of a person other than a witness),
Hit and Run with Bodily Injury, Extortion, Child Abuse, Burglary, Arson,
Rioting, and Other Violent Offense (a group of infrequently-occurring
violent offenses). All offenders were then divided into two groups: those
with any prior adjudication or conviction for a violent offense and those
with no prior adjudications or convictions for a violent offense. A binary
independent variable was created with this data and included in the
Commission’s multivariate regression analysis model.

These analyses show that some differences exist, and describe
the relative size of those differences, in the periods in which the
differences were observed. However, the fact that certain sentencing
outcomes may be correlated with demographic factors does not mean
that the demographic factors caused the outcome.60 Therefore, the
demographic differences in sentencing outcomes revealed by these
analyses should not be interpreted as a finding that demographic factors
caused those differences. Neither can the analyses presented in this
report be used to explain why the observed differences in sentencing
outcomes exist.

Although multivariate regression analysis is common in social
science research, and steps were taken to ensure its appropriateness for
these analyses, this type of analysis has limitations. One or more key
factors that could affect the analysis may have been omitted from the
methodologies used because a particular factor is unknown, or because
data concerning the factor is not readily available in the Commission’s
datasets. Such factors may include whether the offender committed
crimes not reflected in an offender’s criminal history score, and the
offender’s employment record. Accordingly, the results presented in this
report should be interpreted with caution and should not be taken to
suggest race or gender discrimination on the part of judges.

Demographic Differences in Sentencing

Appendix D: Regression Model Output
Regression model
2012-2016
The REG Procedure
Model: MODEL1
Dependent Variable: logsplit
Number of Observations Read
Number of Observations Used

322275
322275

Analysis of Variance
DF

Sum of
Squares

Mean
Square

23
322251
322274

1068533
811366
1879899

46458
2.51781

Root MSE
Dependent Mean
Coeff Var

1.58676
2.68470
59.10388

Source
Model
Error
Corrected Total

R-Square
Adj R-Sq

F Value

Pr > F

18451.7

<.0001

0.5684
0.5684

Parameter Estimates
Variable
Intercept
logmin
drugtraff
sexual2
porn
immigration
othtype
whitecoll
upward
downdep
subasst
mandmin2
NEWCNVTN
custody
whitefemale
blackmale
blackfemale
hispmale
hispfemale
othermale
otherfemale
agedummy
educ
NEWCIT

Parameter
Estimate

Standard
Error

t Value

Pr > |t|

Standardized
Estimate

1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1

0.21375
0.61526
0.18545
0.23591
0.62975
-0.03436
-0.57546
-0.20351
0.64847
-1.06743
-1.11105
0.47247
0.50739
1.34406
-0.34097
0.17520
-0.35241
0.05176
-0.18347
-0.03887
-0.43742
0.02821
-0.00270
0.04916

0.01371
0.00148
0.00959
0.02985
0.01917
0.01217
0.01733
0.01154
0.01957
0.00662
0.00992
0.00966
0.01632
0.00819
0.01444
0.00949
0.01792
0.00976
0.01478
0.01704
0.03152
0.00783
0.00765
0.00938

15.59
415.49
19.33
7.90
32.85
-2.82
-33.21
-17.63
33.13
-161.36
-111.95
48.92
31.09
164.17
-23.61
18.46
-19.66
5.30
-12.41
-2.28
-13.88
3.60
-0.35
5.24

<.0001
<.0001
<.0001
<.0001
<.0001
0.0048
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
0.0225
<.0001
0.0003
0.7241
<.0001

0
0.61552
0.03604
0.00949
0.04464
-0.00647
-0.04382
-0.03100
0.03897
-0.21512
-0.15554
0.06714
0.03693
0.24219
-0.03029
0.02790
-0.02452
0.01067
-0.01629
-0.00284
-0.01640
0.00428
-0.00045774
0.00996

Demographic Differences in Sentencing

Regression model
2016
The REG Procedure
Model: MODEL1
Dependent Variable: logsplit
Number of Observations Read
Number of Observations Used

59160
59160

Analysis of Variance
DF

Sum of
Squares

Mean
Square

23
59136
59159

206222
161604
367826

8966.18995
2.73275

Root MSE
Dependent Mean
Coeff Var

1.65310
2.58726
63.89395

Source
Model
Error
Corrected Total

R-Square
Adj R-Sq

F Value

Pr > F

3281.01

<.0001

0.5607
0.5605

Parameter
Estimate

Standard
Error

t Value

Pr > |t|

Standardized
Estimate

1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1

0.03523
0.59033
0.16276
0.32368
0.69122
-0.11592
-0.63132
-0.10795
0.61161
-1.12238
-1.16465
0.53709
0.53340
1.62716
-0.31754
0.18778
-0.30730
0.08940
-0.07408
0.00265
-0.47330
0.03591
-0.01221
0.00027532

0.03399
0.00354
0.02323
0.06596
0.04528
0.02936
0.04268
0.02833
0.04569
0.01612
0.02478
0.02369
0.04138
0.02026
0.03598
0.02358
0.04444
0.02478
0.03550
0.04147
0.07757
0.01925
0.01859
0.02259

1.04
166.77
7.01
4.91
15.27
-3.95
-14.79
-3.81
13.39
-69.64
-46.99
22.68
12.89
80.30
-8.83
7.96
-6.91
3.61
-2.09
0.06
-6.10
1.87
-0.66
0.01

0.3000
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
0.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
0.0003
0.0369
0.9490
<.0001
0.0620
0.5113
0.9903

0
0.59220
0.03050
0.01396
0.04972
-0.02114
-0.04553
-0.01567
0.03721
-0.22097
-0.15234
0.07309
0.03598
0.28125
-0.02685
0.02911
-0.02046
0.01788
-0.00662
0.00018873
-0.01701
0.00524
-0.00202
0.00005390

Demographic Differences in Sentencing

Regression model
2016
Violence variable
The REG Procedure
Model: MODEL1
Dependent Variable: logsplit
Number of Observations Read
Number of Observations Used

59160
59160

Analysis of Variance
DF

Sum of
Squares

Mean
Square

24
59135
59159

206229
161597
367826

8592.87288
2.73269

Root MSE
Dependent Mean
Coeff Var

1.65308
2.58726
63.89319

Source
Model
Error
Corrected Total

R-Square
Adj R-Sq

F Value

Pr > F

3144.48

<.0001

0.5607
0.5605

Parameter Estimates
Variable
Intercept
logmin
drugtraff
sexual2
porn
immigration
othtype
whitecoll
VIOLENCE
upward
downdep
subasst
mandmin2
NEWCNVTN
custody
whitefemale
blackmale
blackfemale
hispmale
hispfemale
othermale
otherfemale
agedummy
educ
NEWCIT

Parameter
Estimate

Standard
Error

t Value

Pr > |t|

Standardized
Estimate

1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1

0.02861
0.58962
0.16793
0.32951
0.69981
-0.11570
-0.62747
-0.10230
0.02597
0.61025
-1.12151
-1.16341
0.53720
0.53385
1.62337
-0.31463
0.18526
-0.30639
0.09135
-0.06912
0.00241
-0.47040
0.03257
-0.00952
0.00520

0.03426
0.00357
0.02347
0.06607
0.04561
0.02936
0.04276
0.02856
0.01674
0.04569
0.01613
0.02480
0.02369
0.04138
0.02041
0.03603
0.02364
0.04444
0.02481
0.03564
0.04147
0.07759
0.01937
0.01867
0.02281

0.83
165.23
7.16
4.99
15.34
-3.94
-14.68
-3.58
1.55
13.36
-69.55
-46.92
22.68
12.90
79.54
-8.73
7.84
-6.89
3.68
-1.94
0.06
-6.06
1.68
-0.51
0.23

0.4037
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
0.0003
0.1207
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001
0.0002
0.0525
0.9537
<.0001
0.0926
0.6103
0.8197

0
0.59150
0.03147
0.01421
0.05034
-0.02110
-0.04525
-0.01485
0.00470
0.03712
-0.22080
-0.15218
0.07311
0.03601
0.28060
-0.02661
0.02872
-0.02040
0.01827
-0.00618
0.00017130
-0.01691
0.00475
-0.00157
0.00102

Demographic Differences in Sentencing

Endnotes

Endnotes
1
See U.S. Sentencing Comm’n, Demographic Differences in Federal
Sentencing Practices: An Update of the Booker Report’s Multivariate
Regression Analysis (2010) [hereinafter 2010 Booker Multivariate Analysis].
2
The time periods studied in the 2010 report were as follows: the
PROTECT Act period, from May 1, 2003, the date of the enactment of the
PROTECT Act, through June 24, 2004, the date of the Supreme Court’s decision
in Blakely v. Washington, 542 U.S. 296 (2004); the Booker period, from January
12, 2005, the date of the Supreme Court’s decision in United States v. Booker, 543
U.S. 220 (2005), through December 10, 2007, the date of the Supreme Court’s
decisions in Kimbrough v. United States, 552 U.S. 85 (2007) and Gall v. United
States, 552 U.S. 38 (2007); and the Gall period, from December 11, 2007 through
September 30, 2009.
3
The Commission’s updated analysis examined cases in which the
offender was sentenced between October 1, 2009 through September 30, 2011.
In that report, the Commission expanded the Gall period to include offenders
sentenced from December 11, 2007 to September 30, 2011. An additional time
period was also included in that analysis, the Koon period, involving cases in
which the offender was sentenced between October 1, 1998 through April 30,
2003. This period relates to cases decided after the Supreme Court’s decision in
Koon v. United States, 518 U.S. 81 (1996).
4
U.S. Sentencing Comm’n, Report on the Continuing Impact of
United States v. Booker on Federal Sentencing, Part E (2012) [hereinafter
2012 Booker Report].
5

6
USSG].

United States v. Booker, 543 U.S. 220 (2005).

U.S. Sentencing Comm’n, Guidelines Manual (Nov. 2016) [hereinafter

7
For a more detailed explanation of multivariate regression analysis,
the methodology behind it, and its uses and limitations, see 2010 Booker
Multivariate Analysis, supra note 1, at 4–10. See also Federal Judicial Center,
Reference Manual on Scientific Evidence 303–587 (3d ed. 2011) [hereinafter
FJC Reference Manual] (chapters titled “Reference Guide on Multiple
Regression,” by Daniel L. Rubinfeld, Ph.D.); Neil J. Salkind, Statistics for People
Who (Think They) Hate Statistics 324 (2d ed. 2007).
8
As the Commission first stated in 2006, “The presence of violent
criminal history may lead the court to sentence higher in the prescribed range.
The Commission’s datafile does not have information on the type of criminal
38

Demographic Differences in Sentencing

history behavior. In 2002, the Commission created a datafile which took a 25%
random sample of cases sentenced in fiscal year 2000. This datafile looked more
closely at [an] offender’s criminal conduct, including detailed information on the
type of criminal history the offender had. Using this data (the Intensive Study
Sample 2000, or ISS2000), it was found that 24.4% of white offenders had violent
criminal history events, as did 43.7% of Black offenders, 18.9% of Hispanic
offenders, and 23.7% of ‘other’ offenders.” U.S. Sentencing Comm’n, Final
Report on the Impact of United States v. Booker On Federal Sentencing
(2006) 105 n.317. See also 2010 Booker Multivariate Analysis, supra note 1, at
9–10; 2012 Booker Report, supra note 4, at Part E, 8.

9
Beginning with fiscal year 2016 data, the Commission has developed a
process to record all prior criminal history of federal offenders, including the date
of sentence and the type of offense involved. This data is discussed later in this
report. See infra notes 32 to 35 and accompanying text. A discussion of how this
data was collected can be found in Appendix C.
10
In addition, the Commission also does not have ready access to data
related to prosecutorial decision making, which some commentators contend
may contribute to demographic differences in sentencing. For further discussion
of demographic differences in prosecutorial decision making, see Vera Institute
of Justice, Do Race and Ethnicity Matter in Prosecution?: A Review of
Empirical Studies (2012) (reviewing 34 studies analyzing the role of race
and ethnicity in prosecutorial decision making), https://storage.googleapis.
com/vera-web-assets/downloads/Publications/do-race-and-ethnicity-matterin-prosecution-a-review-of-empirical-studies/legacy_downloads/race-andethnicity-in-prosecution-first-edition.pdf June 2012.
11
Amy Baron-Evans and David Patton, A Response to Judge Pryor’s
Proposal to “Fix” the Guidelines: A Cure Worse than the Disease, 29 Fed.
Sentencing Rep. 104, 106 (2016–2017).

12
See the Appendix to this report for more information about these
factors and a discussion of how the Commission’s analysis was performed. The
Commission’s analysis pairs race and gender factors into eight distinct groups:
White males/females, Black males/females, Hispanic males/females, and Other
Race males/females. Conducting analyses in this way helps to identify differences
in sentencing outcomes associated with the offender’s race, gender, or both.
13
For statistical purposes, the Commission divides offenders into
four broad groups based on the level of education reported by the court in the
presentence investigation report: less than high school graduate, high school
graduate, some college education, college graduate. For this analysis, offenders
were combined into two groups—those with no college education and those with
at least some college education.

14
See infra notes 18 to 21 and accompanying text for a discussion of how
the Commission determines the position of the sentence imposed relative to the
sentencing guideline range.
15 See infra notes 22 to 24 and accompanying text for a discussion of
sentences based on an offender’s substantial assistance to the government.

16
This analysis did not include the Koon period because the Commission
did not collect data during that period in a manner that would indicate whether
the below range sentence was sponsored by the government, except in cases
involving substantial assistance motions.
17
These results are unchanged from those reported in the 2010 Booker
Multivariate Analysis and the 2012 Booker Report.
18
Due to the small number of cases in the “Above Range” group, no
further analysis was possible.

19
For additional information about the Commission’s practice of
separating cases into various subgroups according to the position of the sentence
relative to the guideline range, see generally U.S. Sentencing Comm’n, 2016
Sourcebook of Federal Sentencing Statistics (2017) (Appendix A).
20

2012 Booker Report, supra note 4, at Part E, 20–23.

21
No analysis was performed for government sponsored below range
sentences based on an offender’s participation in an early disposition program,
because there was insufficient racial diversity among the offenders in that group
to perform such an analysis.

22
See generally U.S. Sentencing Comm’n, 2016 Sourcebook of Federal
Sentencing Statistics S-51 (2017). For additional information about the
Commission’s practice of separating cases into various subgroups according to
the position of the sentence relative to the guideline range, see id. at Appendix A.

23
The three categories of government sponsored below range sentences
are: those based on an offender’s substantial assistance to the government (see
18 U.S.C. § 3553(e); USSG, supra note 6, at §5K1.1), those based on an offender’s
participation in an early disposition program (see USSG, supra note 6, at §5K3.1);
and all other government sponsored below range sentences.

24
The high fluctuations for this comparison were mainly due to the small
population of Other Race female offenders who received substantial assistance
departures.

25
As discussed above, no analysis was performed for government
sponsored below range sentences based on the offender’s participation in an
early disposition program. See supra note 21.

26
Cases from the Koon period were excluded from the analysis because
the Commission did not collect data during that period in a manner that would
indicate whether the below range sentence was sponsored by the government,
except in cases involving substantial assistance motions.

27
Such an analysis is also called “odds ratio” analysis, as it measures
the probability of an outcome occurring while controlling for independent
variables. Odds ratios are calculated from a regression analysis on a binary
outcome measure (in the case of this analysis, the binary outcome is whether the
offender received a non-government sponsored below range sentence). This type
of regression analysis is known as “logistic regression.” Similar to a regression
analysis on a continuous variable, logistic regression shows the effect of the
outcome variable, controlling for the independent variables in the model.
28

2012 Booker Report, supra note 4, at Part E, 20–23.

USSG, supra note 6, at §5K3.1.

See supra notes 8 to 10 and accompanying text.

29
For more information on the nature and prevalence of these types
of sentences see U.S. Sentencing Comm’n, 2016 Sourcebook of Federal
Sentencing Statistics S-51 (2017).
31
Cases from the Koon period were excluded from the analysis because
the Commission did not collect data during that period in a manner that would
indicate whether the below range sentence was sponsored by the government,
except in cases involving substantial assistance motions.
33
2010 Booker Multivariate Analysis, supra note 1, at 9–10 (citations
omitted).

34
While the Commission regularly collects information about the number
of prior convictions and the number of points assigned to those offenses under
the guidelines (see USSG, supra note 6, at Ch. 4), the Commission did not regularly
collect information about the nature of an offender’s prior offenses (e.g., assault,
robbery, larceny, drug trafficking) prior to fiscal year 2016. Beginning with
fiscal year 2016 data, the Commission developed a method to collect data about
all prior state and federal convictions, including the type of offense and date of
sentence. This information was extracted from the presentence investigation
report prepared in connection with the offender’s federal offense and submitted
Demographic Differences in Sentencing

to the Commission by the sentencing court. For more information on how this
data was collected, and for a list of the offense types that were determined to
involve violence, see Appendix C.

35
The offender’s instant (i.e., current) federal offense was not considered
in making this determination. However, violence associated with an offender’s
instant federal offense was separately controlled for in the Commission’s
regression model.
36
The researcher determines the statistical significance threshold that
he or she wishes to use, often called the “p value.” This decision is based on a
number of factors, including the amount of data available for analysis and the
purpose for the analysis. The Commission’s measure of statistical significance
is p < 0.01. That is, the Commission will not report an observed difference in
sentencing unless the probability that it occurred simply by random chance is
less than .01%.
37
The p value for the independent variable regarding violence in an
offender’s criminal history was 0.1207.

38
The R2 has a value of between 0 and 1 and is commonly expressed as a
percentage.

39
The R2 of the Commission’s analysis for fiscal year 2016 data was
56.1% before the addition of the data on violent criminal history and 56.1% after
the inclusion of that data.

40
Under the guidelines, offenders are assigned to one of six criminal
history categories (CHCs) based on their prior criminal history. The criminal
history score establishes the CHC for an offender. See USSG, supra note 6, at Ch.
4. The CHC, along with the final offense level, determines the sentencing range
under the guidelines. The Commission regression analysis controlled for what
it called the “presumptive sentence” in each case, which is the bottom of the
guideline range that applied in the case. This variable accounts for the fact that
the sentencing judge must properly determine this range and consider it when
imposing sentencing. This variable was statistically significant in all time periods
studied. Therefore, all of an offender’s criminal history, whether involving
violence or not, has an effect on the presumptive sentence that applied in each
case.
41
2012 Booker Report, supra note 7, at Part E, 1; 2010 Booker
Multivariate Analysis, supra note 1, at 9-10.

42
See Michael O. Finkelstein and Bruce Levin, Statistics For
Lawyers 350 (2d ed. 2001).
40

Demographic Differences in Sentencing

43
These results are unchanged from those reported in the 2010 Booker
Multivariate Analysis and the 2012 Booker Report.
44
Due to the small number of cases in the “Above Range” group, no
further analysis was possible.

45
For additional information about the Commission’s practice of
separating cases into various subgroups according to the position of the sentence
relative to the guideline range, see generally U.S. Sentencing Comm’n, 2016
Sourcebook of Federal Sentencing Statistics (2017) (Appendix A).
46
See generally U.S. Sentencing Comm’n, 2016 Sourcebook of Federal
Sentencing Statistics S-51 (2017). For additional information about the
Commission’s practice of separating cases into various subgroups according to
the position of the sentence relative to the guideline range, see id. at Appendix A.

47
The three categories of government sponsored below range sentences
are: those based on an offender’s substantial assistance to the government (see
18 U.S.C. § 3553(e); USSG, supra note 6, at §5K1.1), those based on an offender’s
participation in an early disposition program (see USSG, supra note 6, at §5K3.1);
and all other government sponsored below range sentences.
48
As discussed above, no analysis was performed for government
sponsored below range sentences based on the offender’s participation in an
early disposition program. See supra note 21.

49
Cases from the Koon period are excluded from the analysis because
the Commission did not collect data during that period in a manner that would
indicate whether the below range sentence was sponsored by the government,
except in cases involving substantial assistance motions.

50
Such an analysis is also called “odds ratio” analysis, as it measures the
probability of an outcome occurring while controlling for independent variables.
Odds ratios are calculated from a regression analysis on a binary outcome
measure (in the case of this analysis, the binary outcome is receiving a nongovernment sponsored below range sentence or not). This type of regression
analysis is known as “logistic regression.” Similar to a regression analysis on a
continuous variable, logistic regression shows the effect of the outcome variable,
controlling for the independent variables in the model.
51
For a list and description of the offense types used in Commission
analyses, see U.S. Sentencing Comm’n, 2016 Sourcebook of Federal
Sentencing Statistics S-165-68 (2017).
52

Child pornography offenders are overwhelmingly White male

offenders, and immigration offenders are overwhelmingly Hispanic male
offenders. See generally U.S. Sentencing Comm’n, 2016 Sourcebook of Federal
Sentencing Statistics S-14 (2017).
53
Cases in which the offender was sentenced between fiscal year 2012
and 2016.

54
Hispanic and Other females were excluded from this analysis because
the number of these offenders sentenced for firearms offenses was insufficient
for the Commission to use regression analysis to examine the sentences imposed
in those cases.

55
Some commentators have suggested that a period of alternative
confinement should not be included in the offender’s total sentence length.
The Commission’s analysis includes periods of alternative confinement (home
detention or community confinement) because the main independent variable
is the presumptive sentence. In Zones A, B, and C of the Sentencing Table (USSG,
supra note 6, at §5A), the presumptive sentence can be satisfied by a period of
alternative confinement. The inclusion of alternative sentences reduces the
likelihood of introducing error in the statistical analysis due to the relationship of
the presumptive sentence and overall confinement.
56
In some cases, a mandatory minimum provision limits the guideline
range. For example, in a drug trafficking case in which a ten-year mandatory
minimum applies, the guideline minimum cannot be less than 120 months
unless the offender qualifies for relief from the mandatory minimum. See USSG
§5G1.1(b) (“Where a statutorily required minimum sentence is greater than the
maximum of the applicable guideline range, the statutorily required minimum
sentence shall be the guideline sentence.”). For more information on how
the guidelines incorporate mandatory minimum penalty provisions, see U.S.
Sentencing Comm’n, An Overview of Mandatory Minimum Penalties in the
Federal Criminal Justice System 16-17 (2017).

57
The offense types (or categories) used in this analysis are broad in
order to ensure a sufficient number of cases. The seriousness of the several
crimes varies within the offense type categories as does the demographic
characteristics of the offenders convicted of those crimes. Certain crimes within
an offense type are punished more severely than others (e.g., those crimes
involving injury not accounted for under the sentencing guidelines) and offenders
of a particular demographic group may be disproportionately convicted of those
crimes. If so, the offense type variables used in this analysis may not fully account
for the effect on the sentence length imposed that is attributable to certain
crimes.
58

mandatory minimum penalty at sentencing, or whether the offender obtained
relief from the mandatory minimum penalty and therefore was not subject to a
mandatory minimum penalty at sentencing (or was never subject to a mandatory
minimum penalty because no such penalty applied to the charged offense).
59
This variable refers to whether the court imposed a sentence above or
below the guideline range.
60
Correlation and causation are different concepts. A variable that is
correlated with another may not be caused by it.
[I]n interpreting the results of a multiple regression analysis, it is
important to distinguish between correlation and causality. Two
variables are correlated when the events associated with the
variables occur more frequently together than one would expect by
chance . . . . A correlation between two variables does not imply that
one event causes the second. Therefore, in making causal inferences,
it is important to avoid spurious correlation. Spurious correlation
arises when two variables are closely related but bear no causal
relationship because both are caused by a third, unexamined variable
. . . . Causality cannot be inferred by data analysis alone; rather,
one must infer that a causal relationship exists on the basis of an
underlying causal theory that explains the relationship between the
two variables. Even when an appropriate theory has been identified,
causality can never be inferred directly. One must look for empirical
evidence that there is a causal relationship. Conversely, the fact that
two variables are correlated does not guarantee the existence of a
relationship; it could be that the model—a characterization of the
underlying theory—does not reflect the correct interplay among the
explanatory variables.

FJC Reference Manual, supra note 7, at 183–85. Judges make decisions when
sentencing offenders based on many legitimate considerations that are not or
cannot be measured. Some of these factors could be correlated with one or more
of the demographic characteristics of offenders but not be influenced by any
consideration of those characteristics.

This variable refers to whether the offender remained subject to a

Demographic Differences in Sentencing