|
Measure
3.3.3: Equality and Fairness in Sentencing
One
application of Standard 3.3 is sentencing in criminal cases. Because the
imposition of criminal sanctions deprives individuals of their liberty, the
fairness of the process and corresponding outcomes is an important topic for
the measurement of court performance. In fact, some courts might regard
fairness in sentencing to be among the most critically important goals that
it should strive to meet. However, fairness in sentencing is understandably
very difficult to measure.24 Even the most
refined measurement will produce results more suggestive than definitive,
which is not astonishing given the difficulty of sentencing for trial
judges. Just as the trial judge must weigh, balance, and take into account
many factors, the court researcher must identify, measure, and interpret the
effects of many complex factors, including some that are difficult to
express as a precise scale of measurement.
Hence,
trial courts take on a very daunting task by attempting to measure fairness
in sentencing. Why? Because of the sensitive nature of conclusions about
fairness, a court will want to know that the conclusions are valid to the
greatest extent possible. However, sound conclusions require a rigorous
methodology, which requires a substantial commitment of time, quantitative
skills, and resources. Thus, without intending to deter courts from applying
this measure, honesty requires acknowledging the labor-intensive aspect of
the measurement process necessary to reach the kind of conclusions the court
is likely to want to draw. (Note: The same point applies equally to the
measurement of fairness in bail decisions, Measure
3.3.4.)
What
does fairness in sentencing mean? According to Standard 3.3, "trial
courts give individual attention to cases deciding them without undue
disparity among like cases and only upon legally relevant factors."
Translated into more operational terms, the standard is saying that the
imposition of punishment should not be on the basis of a defendant’s race
or gender. For example, African Americans should not receive longer
sentences than non-African Americans simply because they are African
American. Different sentences should be the product of differences in
criminal backgrounds, offense severity, circumstances surrounding the
offense, and other legally relevant factors.
Finally,
while equality and fairness are positive standards, they are observed in the
negative. Courts are urged to be equal and fair in their treatment, but
their performance is measured in terms of outcomes that are not supposed to
occur—inequality, disparity, and inconsistency.
Planning/Preparation.
Courts should consider four steps in planning to
undertake the measure. First, some familiarity with the literature on
sentencing might prove useful. The most comprehensive volume, Research on
Sentencing: The Search for Reform, is published by the National Academy
of Sciences and available in most public and college libraries. The volume
is written from the researcher’s perspective, however, and contains some
articles of a technical nature. A complementary article, "Racial
Discrimination" by Rose Matsui Ochi, which appeared in 1985 in The
Judges’ Journal, illustrates how research results are interpreted and
used by practitioners who seek to eliminate bias in sentencing. This article
is also useful because it references additional readings on the topic that
are readily available.
A
second step is for the court to examine its capacity for conducting a
rigorous measurement process. If the court lacks a staff person skilled in
quantitative analysis, it might find it helpful to ask for guidance and
assistance from a staff member of a State sentencing commission, State
administrative office of the courts, or local university to assist in
designing a plan of data collection, analysis, and interpretation.
A
third step is to set some boundaries on the scope of the measurement
process. Despite the fact that researchers construct very complex
quantitative models of sentencing, the proposed measure is intended to help
a court assess itself and not necessarily to advance the state of knowledge.
Hence, it permits the court to limit the scope and detail of its inquiry
without sacrificing the validity of the results. As an example, the court
needs to decide what aspect of sentencing is of greatest importance. Is it
more important to determine fairness in the types of sentences that
defendants receive (e.g., incarceration versus probation) or in the length
of sentences imposed (e.g., are men incarcerated for longer periods of time
than women)? Are both aspects equally important?
Finally,
before applying the measure, the court should discuss how it plans to
interpret the results. The results will be in the form of numbers called
coefficients that are based on the application of quantitative techniques to
information gathered from individual case files. There will be a coefficient
for each legally relevant (e.g., prior record and offense committed by the
offender) and each extra-legally relevant factor (e.g., race of offender).
The coefficient measures the impact of a particular factor, controlling for
the effects of all other factors. If the legally irrelevant factors are not
influencing outcomes, the numerical value of their respective coefficients
will not be statistically different from zero. For example, knowing that an
offender is a man will not predict the sentence any better than knowing that
the offender is a woman. Additionally, the coefficients of all legally
relevant factors should be significantly larger than those of irrelevant
factors. If they are, one reasonably can draw the conclusion that there is
limited bias in sentencing and that sentencing is primarily a product of
legally relevant factors. If the court knows what to look for in advance, it
will be more prepared to interpret and use the results both internally for
self-improvement and for presentation to interested groups outside the
court.
Defining
the Data Elements.
Although the exact delineation of legally relevant and legally irrelevant
factors may vary somewhat across States because of differences in
substantive and procedural law, some distinctions likely will be valid in
almost all situations. For the purposes of demonstrating the utility of the
measure, therefore, it is assumed that legally relevant factors
include offense seriousness, quality of the evidence, prior criminal record,
and current legal status. Irrelevant factors include demographic,
socioeconomic, and social stability attributes, and case processing
attributes.25 Based on that assumption, a court
meeting Standard 3.3 has sentencing outcomes that can be explained more on
the basis of those legally relevant factors than on factors deemed
irrelevant.
In
addition to identifying a set of determinants of sentencing outcomes, the
initial measurement step involves specifying the outcomes of sentencing. Two
related outcomes are especially important:
The
first outcome distinguishes between convicted offenders who are sentenced to
prison or jail and those who are given a sentence outside these
institutions. The second outcome focuses on the length of the sentence in
years, months, or days imposed on individuals sentenced to jail or prison.
Legally
relevant factors: Concerning the
seriousness of the offense, a basic judgment must be made to focus on either
a broad range of offenses or to isolate particular offense categories (e.g.,
robbery, burglary). The first option is to consider a large set of offenses
and to rank them according to severity (e.g., homicide, robbery, rape,
assault, weapons, drug sale, drug possession, burglary, forgery, and theft).26
Additionally, other indicators may be used to gauge the more specific
degrees of severity, such as the use of a dangerous weapon, the extent of
injury to the victim, the amount of property taken, and whether the offender
was a principal or accessory to the offense.
Although
some version of the first approach is highly recommended, a second option is
to focus on selected offenses separately. If particular offenses are deemed
of such importance to the court and the community that they merit special
attention, this approach may be appropriate. However, this option lacks the
representativeness of the first option, which encompasses the full range of
offenses. Hence, we generally recommend some version of the first option.
The
quality of the evidence is extremely difficult to measure and may be known
fully only by the participants involved in each individual case. As a
result, retrospective reliance on case records for information only
approximates the complete and correct picture of the strength of the
evidence. Possible indicators include the number of prosecution witnesses,
the number of expert witnesses, the number of exhibits, the submission of
laboratory tests, and so forth. A limitation to these indicators, of course,
is that they relate primarily to the few cases that go to trial.
Prior
criminal history is usually information presented to the court from State
law enforcement records. Although some law enforcement information systems
are more detailed than others, criminal history generally is measured in
terms of the number of prior adult felony convictions, the elapsed time
since the last conviction, whether the last conviction was for the same
offense as the current charge, and the current legal status of the
individual at the time of arrest (e.g., on parole or probation).
Legally
irrelevant factors: Demographic,
socioeconomic, and social stability factors are a combination of
quantitative indicators such as age (years), income (earned income per
month), education (number of years) and categories such as gender (male
versus female), race (white versus nonwhite), employment status (employed
versus unemployed), and marital status (married versus nonmarried).
The
case processing characteristics are all categories. Pretrial release status
may be divided between those offenders on bail, those detained at least part
of the time between arrest and final disposition, and those detained all of
the time. Disposition similarly can be separated among those offenders who
pled guilty, those convicted by a bench trial, and those convicted by a jury
trial.
A
final factor is the judge presiding over a sentencing decision. Each judge
need only be identified by an alphabetic character (e.g., Judge A, Judge B,
Judge C, and so forth). The measure is intended to determine if any judge
has an influence on sentencing that is greater than generally accepted legal
factors. Sentencing outcomes involve the distinction between institutional
incarceration and some alternative to incarceration. This distinction
captures the in/out decision. For the length of the sentence, a standardized
measure is the percentage of the statutory maximum imposed in the actual
sentence. Because some sentences may involve a range, the minimum of the
sentence imposed should be used in calculating the percentage. This
standardization permits different offenses to be compared despite their
differences in severity.
Data
Collection.
In most jurisdictions, virtually all of the factors and sentencing outcomes
can be measured against information contained in presentence investigation
reports and closed court case records. A court can use these sources by
drawing a random sample of approximately 1,000 closed cases and selecting
from that pool those cases in which a conviction was obtained by guilty plea
or trial. (The remaining cases should not be discarded because they can be
used as part of the data set for Measure 3.3.4,
Equality and Fairness in Bail Decisions.) Of this pool, 70 percent are
likely to involve some sort of conviction, which means that these 700 cases
can be used to examine the factors associated with the in/out sentencing
decision. Of these cases, approximately half will involve a sentence of
institutional incarceration, providing the basis for assessing the factors
associated with the length of the sentence.
The
measurement of sentencing and sentencing outcomes described above needs to
be translated into a more specific and detailed form prior to the review of
court case records and presentence investigation reports. A data collection
form should be constructed for the purpose of applying the sorts of indices
suggested for the different factors. (Please see Form
3.3.3, Illustrative Sentencing Data Collection Form.)
Data
Analysis and Report Preparation.
The question of whether legally relevant factors are more powerful
predictors of sentencing outcomes than are irrelevant factors is addressed
by the use of statistical models. These models are available in many
software computer programs that are likely to be familiar to sentencing
commission staff, court researchers in a State administrative office, or
university professors. One or more of these individuals will likely know how
to use an appropriate software program to analyze the data collected on the
data collection form. Specifically, the expert will know what particular
quantitative techniques should be applied to determine the independent
impact of each legally relevant and irrelevant factor on the two types of
sentencing decisions.
In
the case of the in/out decision, an appropriate technique is logit analysis.
Logit analysis is designed to indicate the independent effects of various
factors on different categories (e.g., a sentence of institutional
incarceration versus one of nonincarceration). The numbers generated by
logit analysis include coefficients for each factor. The sign (±) of the
coefficient indicates whether there is a positive (e.g., the more serious
the offense, the more likely the sentence will involve incarceration) or
inverse (e.g., the longer the length of time since the last conviction, the
less likely the sentence will involve incarceration) relationship between
each factor and the outcome. A comparison of the magnitude of the
coefficients will indicate the relative importance of each factor in
determining whether an offender is sentenced to prison as opposed to some
alternative sentence.
The
issue of the length of sentences for incarceration is examined appropriately
through the use of regression analysis. Regression analysis is designed to
indicate the independent effects of factors on an interval factor such as
the number of months to be served. Similar to the logit analysis,
coefficients are generated by regression analysis. They indicate if there is
a positive (e.g., the older the offender, the longer the sentence) or
inverse (e.g., the higher the offender’s level of education, the shorter
the sentence) relationship between each factor and the length of the
sentence.27
The
coefficients bear upon the central purpose of the measure in two ways.
First, if the legally irrelevant factors are not influencing outcomes, the
coefficients associated with them should not be statistically different from
zero.28 Second, the coefficients of all legally
relevant factors should be significantly larger than those of irrelevant
factors.
Looking
at the coefficients associated with the different factors, the court can
begin to assess their implications for fairness in sentencing. Do the
results signal that legally irrelevant factors are having undue influence on
the likelihood of incarceration or the length of sentences? Or do the
results signal that irrelevant factors fail to account for the court’s
decisions to sentence offenders to prison or the length of prison sentences?
In sum, do the results indicate that sentencing decisions are the product
primarily of legally relevant factors and that irrelevant factors are of
limited significance?
Depending
on what the results indicate, the court can use the information as a guide
to reviewing its sentencing policies, practices, and procedures. The results
might suggest the need for special training programs for newly appointed
judges, especially those who come from private civil practice backgrounds.
Or, the results might suggest the need for a courtwide training program on
current developments in substantive and procedural criminal law.
________________________________________
24
The measure proposed outlines a statistical approach to assessing whether
there is undue disparity and bias in a court's proceedings. However, it is
not a complete treatment of every aspect of particular techniques and their
interpretation. For this reason, the court may wish to consult outside
experts when applying the measure.
25 The definition of the data elements and the
proposed methods of data analysis reflect the input and advice of academic
sentencing experts and former staff of the U.S. Sentencing Commission. Their
opinions were solicited to achieve maximum statistical validity, although
future research is likely to use even more refined methods in this growing
area of research.
26 An offense severity scale can be developed by
assigning numerical weights to different offenses. The U.S. Sentencing
Commission has constructed such a scale.
27 For discussion of parallel applications of
this technique to case processing data, see R. Flemming, P. Nardulli, and J.
Eisenstein, "The Timing of Justice in Felony Trial Courts," Law
& Policy 9 (1987); and M. Luskin and R. Luskin, "Why So Fast,
Why So Slow: Explaining Case Processing Time," Journal of Criminal
Law & Criminology 77 (1989).
28 A coefficient may be greater but not statistically
greater than zero because the factor under consideration (e.g., race) does
not have consistent, uniform effects on what is being measured (e.g.,
sentence length). However, a statistical test performed by the software will
indicate whether each coefficient is significantly greater than zero.
Go
to Form 3.3.3
Go
to Measure 3.3.4
Go
to Standard 3.3
Go
to Table of Contents |
