ROSSI, Peter H. Et Al. Beyond Crime Seriousness. Fitting Punishment To The Crime
ROSSI, Peter H. Et Al. Beyond Crime Seriousness. Fitting Punishment To The Crime
1, 1985
This paper presents an exposition of how the factorial survey approach may
enhance empirical assessments of the complex judgment principles involved in
public views of just punishments for convicted offenders. Ratings of the appropri-
ateness of sentences given across 50 typical crimes obtained from a household
sample (N =774) of the Boston SMSA and several special-interest samples in
1982 are examined in three alternative ordinary least-squares (OLS) regression
equations. These analyses show there is not a one-to-one direct relationship
between public perceptions of the seriousness of criminal acts and desired
sanctions. Crime seriousness is modified by the characteristics of the offenders
and victims and by the consequences of the crimes. Preferred punishments also
vary in severity by demographic, experiential, and attitudinal characteristics of
the persons who make the judgments.
1. I N T R O D U C T I O N
public, even though the specific details of particular instances of the crimes
are not included in the descriptions to be rated. This feature of crime
seriousness ratings has been interpreted as establishing seriousness as a
dimension along which crimes are evaluated with a fairly high degree of
consensus. To go beyond crime seriousness, as we do in this paper, is to
show that the seriousness of the general crime class does not determine
uniquely how much punishment is appropriate. A number of other principles
come into play in such judgments, as we indicate.
This paper addresses empirically the issue of how the various principles
involved in judging the fairness of punishments are used in combination
by the general public. Specifically, we report on research designed to uncover
how members of the general public combine some of the particular features
of cases of convicted criminals in judging the appropriateness of the punish-
ments for those cases. We consider the joint influences on such judgments
of the following characteristics of particular cases: (i) the seriousness of
the crime committed as conventionally measured through global ratings of
classes of crimes; (ii) the social characteristics of the particular offender,
such as prior criminal record, age, sex, socioeconomic level, race, etc.; (iii)
the extent of the injuries a n d / o r losses sustained by victims; (iv) the
relationship between the offender and the victim; (v) the social characteris-
tics of the victim in terms of age, sex, socioeconomic level, and race; and
(vi) potentially mitigating circumstances involved in the cases. This analysis
is undertaken separately for four major classes of crime: property crimes,
crimes against persons, victimless crimes, and corporate crimes.
In addition, we investigate the extent to which the principles used by
members of the general public in making judgments about appropriate
punishments for persons convicted of crimes are shared principles or
idiosyncratic in character.
Lest our intentions be misunderstood, we want to make it abundantly
clear that we do n o t advocate that the criminal codes or the practices of
the criminal justice system be altered to conform more closely to the
principles used by the public in judging appropriate punishments for crimes.
Nor do we claim that the latter are exogenous to the former: the practices
of the criminal justice system undoubtedly affect the judgments of the public
as well as the other way around.
2. R E S E A R C H D E S I G N
Sample surveys are the obvious method of choice for the study of
public opinion. We employ a variant of that approach, the factorial survey
(Rossi and Anderson, 1982), a melding of principles of experimental design
with conventional sample survey methodology. This approach is applicable
62 Rossi, Simpson, and Miller
into the vignettes to the overall judgment. These features are directly
pertinent to the main issue of this research, namely, how does each
dimension of a criminal case contribute to the overall judgment whether
fair punishment has been given in that case.
The specially constructed vignettes are given to respondents for judg-
ment according to some criterion. In this case, the vignettes are descriptions
of convicted criminals who have been sentenced (randomly) to varying
lengths of confinement in prison. The respondent is asked to judge the
appropriateness (or fairness) of the sentence imposed.
Critical to the factorial survey approach is the manner in which the
vignettes are constructed, as follows.
(i) Dimensions of the social objects are identified a priori as relevant
to the judgments under study, along with specific levels within each
dimension. Thus the crime committed by the convicted offender is a
dimension of the vignettes; levels within that dimension are a set of 57
specific offenses, e.g., homicide, robbery, forgery, etc.
In the design of any factorial survey, a critical step is the specification
of the dimensions and levels to be included. In this case, the central concern
with "fitting the punishment to the crime" immediately specified two
dimensions, punishments in the form of prison sentences and crimes, selec-
ted to vary in seriousness. In addition, we chose crimes from four major
categories of felonies, crimes against the person, crimes involving the
appropriation of property, "victimless crimes," and "white-collar c r i m e s " - -
crimes committed by business organizations (Rossi and Henry, 1982; Cullen
et al., 1982). The remaining dimensions were chosen mainly because pub-
lished literature on sentencing stressed their importance, e.g., socioeconomic
characteristics of the offender, previous record, ethnicity, age, gender
(Hagen and Bumiller, 1983). Some of the socioeconomic characteristics of
crime victims were also included, when appropriate, including many of the
same characteristics that were used to describe the offender, as well as
statements about the damages inflicted on the victim.
The dimensions and illustrative examples of levels used in this study
are shown in Table I. Note that some of the dimensions are nested. For
example, characteristics of an offender are included only for crimes in which
a person can be the offender and victim characteristics are used only for
crimes against the person.
(ii) A computer program constructs each vignette by picking randomly
one of the levels within the first dimension--in this case, a specific crime.
The program then goes on to select randomly a level within the second
dimension (if appropriate to that kind of crime), repeating this selection
process through all the dimensions that are relevant to the specific crime
that was picked in the first step. The program can skip over dimensions
64 Rossi, Simpson, and Miller
Dimension 5: Prior record of the corporation (used only in crimes with corporate offenders)
Levels: 4 degrees of previous violations
Illustrative levels: "This company has never before been in court on charges like this."
"Over the past five years, this company has been in court many times on
charges like this."
Blank text (i.e., no description of prior record of corporation)
Dimension 6: Corporate mitigating circumstances (used only for corporate crimes)
Levels: 6 statements that describe mitigating circumstances
Illustrative levels: "The company claims that all their competitors do the same."
"The company claims that it was trying to save jobs for its workers."
Blank text (i.e, no statement is included in the vignette)
Dimension 7: Sentence length (used in all crimes)
Levels: 9 sentences of varying length
Illustrative levels: "3 months in jail"
"5 years in prison"
"More than 10 years in prison"
Beyond Crime Seriousness 65
Table I (cont.)
Dimension 12: Offender labor force status (for crimes with persons as offenders)
Levels: 5 levels including employed, unemployed, housewife, and blank text
Dimension 13: Occupation of offender (for crimes with persons as offenders)
Levels: 20 occupational titles, weighted with low-status occupations
Illustrative levels: "Car washer"
"Bus driver"
"Teacher"
Dimension 14: Offender-victim relationship (used only for person offender crimes with persons
as victims)
Levels: 5 degrees of relationship
Illustrative levels: "Spouse"
"Friend"
"Stranger"
Dimension 15: Gender of victim (used only for person offender a n d victim crimes)
Levels: 35 levels consisting of 14 female and 21 male n a m e s
Dimension 16: Age of victim (used only for person offender and victim crimes)
Levels: 7 levels ranging from 20 to 55 and over
Dimension 17: Employment status of victim (used only from person offender and victim crimes)
Levels: same as for offender
Dimension 18: Occupation of victim (used only for person offender and victim crimes)
Levels: same as for offenders
Dimension 19: Prior record of offender (used only for crimes with persons as offenders)
Levels: 6 prior record statements varying in severity
Illustrative levels: "The offender has not been arrested or convicted."
"The offender has been arrested once but not sent to prison."
"The offender has been convicted twice and sent to prison once."
66 Rossi, Simpson~ and Miller
Table I (cont.)
Dimension 20: Person offender mitigating circumstances (used only for crimes in which person
are offenders)
Levels: 8 statements pertaining to excuses made by offender for committing the crime of
conviction
Illustrative levels: "The offender claims to have been taking drugs at the time."
"The offender has offered to make up for the crime by paying
damages."
that are not relevant. For example, if crimes are selected that do not have
specific personal victims, as in drug trafficking or illegal possession of
firearms, the dimensions that pertain to the characteristics of a personal
victim are skipped.
Two examples of vignettes produced by the program are shown in Fig.
1. Note that each vignette contains a different set of dimensions; that is,
each contains only those that are relevant to the type of crime committed
by the offender portrayed in the vignette. Figure 1 also shows the rating
scale that the respondent is asked to use to record his or her judgment. The
criterion used is the appropriateness of the punishment shown as given to
the convicted felon described in the vignette. The scale contains 62 intervals,
each of which has been given the value 2. Care is taken in wording the
levels constituting each dimension, with the result that, when different levels
are assembled together in the form of vignettes, the flexibility of the English
language is not overly taxed.
Figure 1 is actually a sample page of the booklet containing 50 vignettes
that is handed to each respondent to read and mark. Each booklet con-
stitutes, in effect, a self-administering questionnaire upon which the respon-
dent is asked to mark his or her judgment of each of 50 vignettes.
(iii) The computer program repeats the process of assembling vignettes
until a sample of 50 vignettes is produced and then prints the vignettes in
booklet form. The program goes on to produce as many booklets as are
needed, using the same process described earlier, for each booklet. Note
that each booklet contains a random and, thus, unbiased sample of all
possible vignettes defined by permissible combinations of levels.
The total number of unique permissible vignettes as defined in this
study is 1,047,259,295,424. This feature means, for all practical purposes,
that each respondent gets essentially a unique combination of vignettes in
his or her booklet. Each respondent's set differs only by sampling variation
from other booklets in the crimes, sentences, and other features of the
vignettes. In other words, the mean seriousness of the crimes (as well as
the central tendencies of other dimensions) in each respondent's vignette
Beyond Crime Seriousness 67
I:::F:i::~N!::: F:'. ~ ';2(D~ (-'~ EtL_Pd:_',I-:::~ AN EMf:'LC)YtE.r.) F'~M::,:KI:NG [_C)1 P~FI-E!:Nt2,~'~N'I Nh~S
COP,I',2IE:rI-:D OF: ,~1 I_H~.~I IG ME:F;:CI4~I'4DI:SE F:RE]P! rq DEF'r S'Ii:]I:;:E E:[3UNIEI::;:
N O R T H t::'F:,'E]UN r) $1 . C!I]:,C:,.
!!',l Fl'fiE L.r FIMI:2 ","E~:hR!:i, YHE: [:]FFrENDE:R IHF~S EtE:EN AF;:F;:Ecl!E:D ON{:::E.~ E!UF NO1
CQI".!VICqI-[D,, "f'l-iE OFrF::EI'qDfSF;: H A S OF:~::'.:-tPED t O bllq!:::E: UP F:QR THI_:!: E;R:[Mr~: EG,"
I::'i::P," I 1'4[:~ D(:~MA(3E:S.
A f.~(31..lkl
F',' I G F IT
..... ]: ....................... I ............... ]: ...................... ]: ................ :I: .....................! ................ I ................... ! .............. :!: .................. I ............. i . . . . . .
bl U C H F1U E:H
TD(3 L.(3N H I [:?H r[][:]
I__C]b.~ HIGH
X,: ~ ~g .t,: ~ '~ ~: X,: ~: I,: ~{( I,: ~ ~,: g ~: ~: ~: .,~ ~: ~: ~ ~: ~: ~: ~: ~: g ~ ~,: ~: ~ X, ,~ I, ~, ~ g: :4<~ ~< ~: ~: ~ :~ ~: :r ~,: ~: ~;,:~: ~: .g :~ g g ~,: :~: ~1,:~' ~: ~ ~: g ~: ~: ~I, ~: ~:
#~BOUT
IR ! (3 H ' I
I .................... ] .................. I . . . . . . . . . . . I ................ I ................ I ................ I ................... ! . . . . . . . . . . 1: ..................... ! ............... I . . . . . .
NUCH 191. J C H
TC]O LOW H I (3H 100
LON H i Gt-[
~Sc:s:-t:~.e coded a!._'; :[ s !25 ~'~J L h oath c~.,,": tl-~e~ unit!:i c:c::)unL~~d c,,~i; 2., rarlc~nt;)
f:l~C%n t2D codced Fc]! . . . . muc:h ~,:c~o ]c~,J" to:, "!" ~:~:,r "mL.tC:['l t'CiC) ~iJ@-~.
booklet has the same expected value, a feature that facilitates comparison
among individual respondents' judgments. This also means that despite the
astronomical number of possible vignettes used, each set of vignettes gener-
ated is n o t entirely different from all other vignette sets. For example, about
40% of all the possible vignettes describe property crimes, and about 40%
of any respondent's booklets are property crimes; two-thirds of all offenders
shown in the vignettes are males, and about two-thirds of the offenders in
any respondent's booklet would also be males; and so on, through the entire
list of dimensions.
The computer program at the same time also produces a coded tape
that contains a record for each of the vignette samples the program has
produced. This tape becomes an analysis tape by the addition of the 50
68 Rossi, Simpson, and Miller
3. DATA ANALYSIS
Since independently drawn random samples may be pooled to produce
larger samples that are also random, pooling vignettes across respondents
produces a very large random sample of vignettes. In this case the pooled
overall sample of vignettes is certainly large e n o u g h - - o v e r 53,000--to
constitute an adequate basis for stable estimates of the separate and indepen-
dent contributions of dimensions and levels to the fairness judgments of
respondents.
There are several alternative approaches to the analysis of these data,
three of which are shown in Table II, defined as ordinary least-squares
(OLS) regression equations. 4 The first equation (A) depicts a judgment or
rating, J, given to vignette i as a simple linear function of the dimensions
2Interviews were collected under subcontract by the SurveyResearch Center of the University
of Massachusetts, Boston.
3We are indebted to the followingfor help in data collection: Geraldine O'Donnell, Assistant
Headmaster of Madison Park High School; Timothy Moran, Associate Dean, University
College of Northeastern University; Russell Smith, WestoverJob Corps Center; and Professor
Ilene Nagel, Indiana University.
aThe use of OLS models in analyzing these data may appear, at first glance, to violate the
assumption of fixed effects. However, the effects of violating this assumption are minimal
(Kmenta, 1971) and, in any event, affectonly the extent to which findings can be generalized.
It should also be noted that analyses are not constrained only to an OLS interpretation of
the general functional form of the relation between ratings and vignette characteristics. For
example, when there is good reason to believe that the rating scale used truncates responses
artificially, logistic regression may be employed.
Beyond Crime Seriousness 69
and levels within dimensions that compose the vignette. In broader terms,
this equation shows the ratings as being primarily influenced by the content
of the vignettes, with the resulting coefficients being the weights applied to
those vignette components which best reproduce, in an OLS sense, the
ratings given to each vignette. Note that because of the approach used to
construct the vignettes, the characteristics of the vignettes are uncorrelated
with respondents, each respondent being given a separate random sample
of vignettes. Hence the computed coefficients are unbiased estimates of
those weights, unaffected by any characteristics of the respondents.
The findings resulting from the application of the OLS model in Table
IIA to all of the vignettes rated are shown in Table III. Because some of
the dimensions do not apply to all types of crimes, the analysis in Table II
is based on only those dimensions which are present in all of the vignettes,
i.e., in all types of crimes.
The dependent variable in Table III is composed of the ratings given
to vignettes, as shown in Table II, converted to a scale that ranges between
1 and 125. The midpoint, defined as a sentence or punishment that is "about
right" in the respondent's judgment, receives the score value of 63. Note
that a higher score means an increase in the desired severity of punishment
for the convicted person described in the vignette. Hence the dependent
variable has been called "severity rating" to reflect that quality. A positive
coefficient associated with an independent variable means that high values
70 Rossi, Simpson, and Miller
Dependent variable is
severity ratingsa of vignettes
Independent variable /3 SE
~A high rating indicates that the sentence shown in the vignette is regarded as too low by the
respondent. Hence ratings denote the desired severity for the convicted person described in
the vignette.
bComputed from ratings of seriousness of the crimes as gathered by Wolfgang (Center for
Studies in Criminology and Criminal Law, 1978) from a national probability sample of the
United States and Rossi et aL (1974) from a sample of Baltimore residents.
CDummy variable for crimes involving specific amounts of property loss by a victim. Omitted
category consists of crimes involving personal injury to victims.
dDummy variable for crimes involving corporate entities as offenders. Omitted category same
as above.
eN is the number of vignettes used in analysis. Vignettes were rated by household sample of
Boston SMSA and convenience samples of other populations of special interest.
*P < 0.001.
o n that characteristic are associated with high severity ratings, i n d i c a t i n g
that the p u n i s h m e n t s given t e n d to be too low in the o p i n i o n s of the
r e s p o n d e n t s . Stated otherwise, a negative coefficient, say, for female offen-
ders, indicates that r e s p o n d e n t s t e n d to be less severe or more l e n i e n t t o w a r d
female c o n v i c t e d offenders, as c o m p a r e d to male offenders.
SWe are indebted to Marvin E. Wolfgang and Robert E. Figlio for providing scores from their
national survey. The Baltimore and Wolfgang scores were integrated in the following way:
The regression of Wolfgang scores on Baltimore scores for comparable crimes was computed.
The resulting regression equation was then used to estimate Wolfgang scores for crimes
included in the Baltimore study but not in the Wolfgang study. The metric for the Wolfgang
crime seriousness scores was generated by a magnitude estimation task given to a national
sample as a supplement to the Current Population Survey in 1978.
6The zero-order correlation between the crime seriousness scores and the leniency ratings was
0.40, indicating that about 16% of the explained variance in leniency ratings was accounted
for by that feature of the vignettes.
72 Rossi, Simpson, and Miller
separation allows the regressions to include all the information for each
type of crime used in the construction of the vignettes.
At the bottom of each column of regression coefficients are descriptive
statistics pertaining to crime seriousness scores and severity ratings. Clearly
the crime types varied in their seriousness and, correspondingly, in the
average severity ratings given to them by respondents. It is no surprise that
crimes against the person were regarded as the most serious and accorded
the most severity. At the opposite ends of both seriousness and severity
were victimless crimes that involve neither specific persons as victims nor
property theft. The other two classes of crime studied lay between the first
two in terms of both crime seriousness and severity ratings.
The four regression equations each involve blocks of independent
variables specific, by intention, to the crime class involved and blocks that
are common to all crime categories. The first block, "legal variables," is
comprised of dimensions that pertain directly to the legal aspects of the
crimes, e.g., the seriousness of the crime, the sentence given upon con-
viction, and, in the case of property theft, the dollar amount of property
stolen.
Perhaps the most notable feature of the regression coefficients for this
block of variables is the wide variation in the coefficient for crime serious-
ness, ranging from +0.101 for corporation crimes to +0.038 for crimes
against persons, a range of almost 3 magnitudes. Nor is this finding an
artifact of the range of seriousness in which variables with restricted vari-
ances are characterized by smaller regression coefficients, since it is the
crime class with the highest seriousness variance that has the smallest
coefficient. The severity ratings given to crimes against the person are not
as sensitive to differences in crime seriousness scores compared to other
crimes. Perhaps the other features of crimes used in the vignettes for crimes
against persons incorporated elements of seriousness in concrete ways that
took away some of the influence of global seriousness scores. In other
words, the injuries suffered by the victim of crimes against the person were
specific manifestations of seriousness that override the global score, s
The second block of independent variables pertains to the characteris-
tics of the offender, defined for three of the crime types but undefined for
crimes in which corporations are shown as the offenders. Here several
8Recall that the seriousness scores are derived from the survey by the Center for Studies in
Criminology and Criminal Law (1978), in which descriptions of crimes were given to a
national sample of respondents to rate using a magnitude estimation approach. The scores
used in this study were derived from those scores by averaging across all the instances of a
particular type of crime. Thus there were several versions of assaults, each using slightly
different formulations (e.g., a m o u n t s of injury, relationship between offender and victim,
etc.), in the study by Wolfgang et al. Averaging across those several versions to obtain a
global score for the crime undoubtedly introduced error, some of which is recovered by the
additional dimensions included in the vignettes describing crimes against persons.
T a b l e IV. R e g r e s s i o n o f Severity R a t i n g s on All Vignette D i m e n s i o n s
Independent variable /3 SE /3 SE /3 SE /3 SE
1. " L e g a l " v a r i a b l e s
Crime s e r i o u s n e s s scores +0.0625 0.002 +0.0385 0.001 +0.0865 0.002 +0.1015 0.003
Log o f s e n t e n c e given -10.985 0.145 -9.6645 0.226 -9.4815 0.171 -9.5455 0.205
Probation dummy +15.755 0.585 +13.5545 0.909 +19.0375 0.678 +15.1565 0.831
Suspended sentence dummy +15.8375 0.579 +15.2695 0.897 +17.581:~ 0.685 +16.8705 0.831
D o l l a r p r o p e r t y loss +0.0025 0.000 . . . .
2. Offender c h a r a c t e r i s t i c s
Black d u m m y -1.081t 0.436 +0.239 0.682 -0.454 0.516
Female dummy -3.9705 0.729 -3.264# 1.196 -3.3815 0.852
Age (years) +0.0665 0.017 +0.042 0.027 -0.022 0.019
Unemployed dummy +0.092 0.458 +0.631 0.702 -0.938 0.541 m
6. Mitigatin~g circumstances
m
7. Corporation characteristics
o _ m +2.249* 0.775
Very large national c o m p a n y
m +0.602 0.766
Large c o m p a n y
m 0.768
Small c o m p a n y --0.192
_ m m 0.776
First offense -4.9305
m m + 1.320 0.772
First conviction (other charges)
m _ m
m p +8.319I: 0.770
Many previous charges
m . +3.410~ 0.82/
Competition as mitigating claim a
m m
m w +1.530 0.822
Hardship for workers ~
Law is unconstitutional d +O.565 0.945
Table IV (cont.)
I n d e p e n d e n t variable /3 SE /3 SE /3 SE /3 SE
8. Respondent characteristics
Male respondent -1.9105 0.401 -2.2565 0.624 -2.754:~ 0.472 -3.3155 0.575
Black respondent +2.4185 0.591 -3.2275 0.912 +2.8845 0.690 +5.306:~ 0.845
Age (years) -0.013 0.014 -0.0845 0.021 +0.0775 0.016 -0.0825 0.019
Education (years) -0.6305 0.088 -0.179 0.137 -1.1225 0.104 -0.290* 0.126
Household income (000's) -0.016 0.014 +0.007 0.022 -0.013 0.017 -0.0865 0.020
NA on income +2.083* 1.071 -0.570 1.690 +0.500 1.221 -1.075 1.508
Job Corps sample -5.9485 0.980 -4.005t 1.526 -2.272* 1.124 -3.81 I t 1.381
State prisoner sample -9.8205 1.172 -19.1025 1.888 +0.728 1.397 -16.2775 1.719
Roxbury High School sample -0.245 0.863 -1.873 1.345 +2.746t 1.000 -0.612 1.214
Police sample -0.908 0.631 - 1.207 0.974 -2.9405 0.746 9-4.3115 0.900
Intercept +52.2135 1.750 +58.9275 2.778 +39.4315 1.986 +49.5375 2.410
R2 0.3455 0.3925 0.3515 0.315:~
N 20,410 8552 14,689 10,113
Mean seriousness score 247 464 227 260
Min, M a x seriousness 114, 537 252, 993 13, 552 140, 468
SD 99.7 251.7 130.9 94.3
Mean severity rating 60.0 74.8 48.0 66.7
SD 33.6 35.5 33.6 32.9
5"
aVignettes with crimes involving the appropriation o f property and committed by an individual offender.
bVignettes with crimes committed by an individual but without specific victims (e.g., parking violations, income tax evasion).
cOccupational prestige scores as derived from D u n c a n SEI scores (National Opinion Research Center, 1983). =
t~
dThese are statements presumably m a d e by the convicted c o m p a n y concerning why they committed the crime a n d offered as mitigating excuses, ~'
e.g., "Claimed that all competing companies do it."
*P<0.05.
tP<0.01.
5P<0.001.
Beyond Crime Seriousness 77
findings stand out. First, the race of the offender significantly affected only
the severity ratings of property crimes. Property crimes committed by blacks
were regarded slightly more leniently than those commited by either
"whites" or Hispanics. No ethnicity differences existed for either crimes
against persons or victimless crimes. Perhaps respondents were saying that
blacks, because of their poverty, can be excused to some small extent for
committing property crimes. Second, offender sex did make a difference,
with women consistently regarded with less severity than males. The
coefficient for gender is about the same size in each of the three crime
classes. Since most research on the actual sentencing of offenders also finds
a similar gender effect, respondents may have been simply reflecting the
practice of the courts, or vice versa.
Third, an offender's age made a significant difference only for property
crimes: older property-crime offenders were assessed more severely, but
age made no difference in crimes against the person or victimless crimes.
Youthful indiscretions with other persons' property were apparently more
easily forgiven.
Fourth, social status made a difference in the judgment of appropriate
punishment only for property crimes. Higher-status (as measured by SES
scores) property-crime offenders were judged slightly less severely than
lower-status offenders. No simple or easy explanation for this socio-
economic tendency comes to mind. Note also that housewives are judged
with slightly increased severity. However, social status, employment status,
and being a housewife do not affect the leniency ratings of the other two
crime classes.
Perhaps the most outstanding feature of these findings concerning
social characteristics of offenders was how slight were their effects. The
largest and most consistent effect was that of gender. Less severity con-
sistently was accorded to female offenders in all four types of crimes, but
the gender effect was also uniformly small. The other effects were all
inconsistent and small, when significant. By and large, Boston area residents
were not notably inclined to discriminate in the severity of appropriate
punishments for offenders of different ages, races, genders, and
socioeconomic levels.
In contrast, the last set of offender characteristics, record of previous
arrests and convictions, was fairly powerful in affecting severity judgments.
Vignettes that contain statements about previous records were consistently
rated differently from those that contained no statements on that characteris-
tic (the omitted category in the dummy variable analysis). An offender's
previous record counted about the same in all crime classes, with leniency
accorded to first offenders and severity to persons with records involving
several previous convictions.
78 Rossi, Simpson, and Miller
crimes. Roxbury high-school students were not different from the general
household sample except on victimless crimes, which they judged more
harshly. Finally, the sample of police was less severe toward victimless and
white-collar crimes.
It should be noted that the linear effects of respondent characteristics
noted above, although statistically significant, were not very large. By and
large, respondents of all sorts were close to each other in the ways in which
they rated vignettes higher or lower in severity. In each of the four equations,
respondent characteristics accounted for under 5% of the total variance,
the remaining 30-35% being accounted for by the dimensions and levels
included in the vignettes.
Of course, linear effects may not be the only way to represent respondent
effects. For example, respondents may have varied in the ways in which
they took particular dimensions into account in forming their judgments:
some may have been concerned mainly with the global seriousness of the
crime committed, while others gave greater attention to the losses suffered
by victims.
In order to uncover individual differences in the ways in which
dimensions were used by respondents, we turn to the analysis model shown
in Table IIC. This analysis approach focuses upon individual respondents,
characterizing each person by summary measures of the ratings that each
respondent gave to the particular sample of vignettes he/she was asked to
evaluate.
The 50 ratings made by each respondent are regarded in this approach
in much the same way as answers to 50 items in an attitude test battery or
an IQ test. Thus, the average of the ratings given by a respondent can be
regarded as the severity proclivities of that person. Similarly if we compute
separately for each respondent the regression of his/her ratings on the
characteristics of the vignettes rated by that respondent, the resultant /3
coefficients represent the rating principles used by that respondent. Thus a
respondent's /3 coefficient for crime seriousness represents the estimated
weight given by that respondent to crime seriousness. Since there are 50
observations on each respondent (50 ratings), these coefficients can be
regarded as computed with sufficient degrees of freedom.
Of course, the vignette samples differ from respondent to respondent
according to the vagaries of sampling and the ratings given by the respondent
may vary accordingly. Hence in any such analysis it is necessary to take
the vignette sample characteristics into account, since at least part of the
interrespondent variation in response to the particular vignette set can be
82 Rossi, Simpson, and Miller
accounted for by the fact that (by design) each respondent's vignette set is
a random sample of all possible vignettes.
Table V presents three OLS regression analyses using the model in
Table IIC. The first has as its dependent variable the mean of the ratings
given by respondents. Since the mean rating may be regarded as a measure
of the respondent's general tendency to be more or less severe in his/her
ratings of appropriate punishment, this equation may be regarded as an
analysis of individual differences in severity. The second equation uses as
its dependent variable the standard deviation of each individual's ratings.
Since the standard deviation may be regarded as a measure of the extent
to which an individual discriminates among vignettes in making judgments,
this equation is an analysis of individual differences in willingness to fit the
punishment to the crime.
The last equation uses as its dependent variable the regression
coefficient for crime seriousness computed over the 50 judgments made by
each individual. I~ Since crime seriousness is positively related to severity,
the higher (i.e., more positive) the regression coefficient, the more heavily
did a respondent weigh seriousness in his/her judgments. This equation is
therefore an analysis of individual differences in using seriousness as the
basis for judgments and, hence, addresses the issue whether all respondents
are using the same information in the same way in coming to their judgments.
The bottom of Table V presents descriptive statistics for each of the
dependent variables. Note that there is considerable "variation to be
explained. Mean vignette severity ratings ranged widely, from a minimum
of 6.6 to a maximum of 100.7; standard deviations ranged from 1.0 to 57.8;
and the/3 coefficient for crime seriousness ranged from +0.210 to -0.070.
Of course, some of this variation is generated by sampling differences
among respondent vignette sets. Therefore, the first block of variables in
each of the equations is measures of sampling variability in vignette sets,
as indexed by a few measures of salient dimensions, the average seriousness
scores of the crimes depicted in those vignettes, the average log of sentences
given, and the proportion of vignettes in which the offenders were given
suspended sentences. These features were chosen primarily because they
can appear on all vignettes and also because, as we saw in Table III, these
are important overall in explaining the judgments given.
Interpretations of the findings in the first block are not substantively
important, since these variables were used primarily as controls for sampling
variations in the vignettes used. Even so, there are some interesting findings.
First, the mean seriousness of the crimes shown in the vignettes affected
1~ these computations, the/3 coefficients were multiplied by 1000 in order to avoid carrying
many decimal places: the mean, range, and standard deviation shown in the last three rows
in Table V are n o t weighted, however.
Beyond Crime Seriousness 83
all three dependent variables. Understandably, the higher the mean serious-
ness, the higher the average severity score given. Not so understandably,
the higher the mean seriousness score in a respondent's vignette sample,
the greater the variability of the ratings and the more lightly was seriousness
weighted in coming to judgments. What these last two findings may mean
is that vignette samples with high mean seriousness scores tended to have
a higher proportion of crimes against persons. Such vignettes have more
information on the consequences of the crimes, a feature, as we have seen
from the analysis shown in Table III, that tends to detract from the influence
of seriousness on the judgments made.
Second, the average log of the sentence shown as given in the vignette
sample tends to lower the mean severity ratings of respondents and lower
standard deviations and has no discernible effect on the/3 coefficient for
seriousness. Finally, the proportion of sentences shown as suspended in
the vignettes raises (understandably) the standard deviation of the ratings
but not the mean leniency ratings or the/3 coefficients for seriousness.
The second block of variables represents rating principles used by
respondents. Thus the first variable in that block, the mean rating given to
vignettes, represents the general severity tendencies of respondents. 11 The
higher the average rating given, the larger the standard deviation of the
ratings and the more heavily seriousness was weighted. Reciprocally, we
may also note that the larger the standard deviation, the lower the mean
rating given. What this means is that respondents who discriminated in
rating vignettes (i.e., varied their judgments from vignette to vignette) tended
to be less severe. 12 Conversely, those who did not vary their ratings tended
to give higher severity ratings, suggesting a kind of general punitiveness
toward persons who were convicted of any crime.
The remaining variables in the second block are all measures of how
respondents used information in the vignettes in their rating judgments.
Thus the log sentence/3 coefficient lowered severity on the average. That
is, the more heavily a respondent weighted the sentence shown, the more
likely he or she was to be lenient in her judgments. These appear to be
persons who were generally opposed to long sentences and hence paid
greater attention to the sentences given. The/3 coefficient for log sentence
also affected the standard deviation of ratings, lowering the variability of
the judgments.
The next finding is that some of the dimensions of the vignettes tended
to substitute for each other. Thus, the higher the /3 coefficient for log
11Note that since some of the sampling variation in vignettes is taken care of by the first block
of variables, these coefficients are n e t of sampling variation.
~2Since the variability of ratings is partially a consequence of the rating principles used, we
did not use the staadard deviation of ratings to predict the size of the seriousness/3 coefficient.
Table V. Regression of Respondent Rating Set Measures on Vignette Sample Parameters and Respondent Characteristics
Dependent variable is
Independent variable 13 SE /5 SE /3 SE
aSummary measures that describe the vignette sample rated by the respondent.
b Measures that characterize the ratings given by the respondent. Regression coefficients c o m p u t e d for each respondent by regressing r e s p o n d e n t ' s
ratings on vignette characteristics.
CFor each sex, race, and age group probabilities of being victimized by crime were c o m p u t e d from the national victimization surveys ( B u r e a u o f
Justice Statistics, 1980).
dAttitude scale m e a s u r i n g respondent's beliefs about the fair treatment meted out by American institutions such as business enterprises, g o v e r n m e n t
agencies, etc.
eScale measuring respondent belief in defectiveness of criminals.
/Scale o f respondent a m o u n t of personal worry about crime.
~Scale o f negative events h a p p e n i n g to respondent recently (e.g., u n e m p l o y m e n t , quarrels, death in family, etc.).
hScale of respondent's desire for punitive treatment of criminals.
~Units of analysis are respondents in Boston SMSA h o u s e h o l d sample. (Samples of convenience do not have full questionnaires and h e n c e do
not contain some of the variables used above.)
*P<0.05.
?P<O.O1.
~P<O.O01.
86 Rossi, Simpson, and Miller
sentence, the less weight was given to crime seriousness. This may mean
that respondents tended to fasten upon one dimension, using it more heavily
in making judgments and at the same time tending to ignore other
dimensions. In this instance, if one paid more attention to the sentence
given, there was also a tendency to disregard the seriousness of the crime
committed, additional evidence for an interpretation of a generalized dislike
of long sentences as penalties.
The seriousness /3 coefficient also affects both the standard deviation
and the mean rating. The more heavily seriousness was weighted, the more
severe was the average judgment and the greater was the variability of the
judgments. Respondents who had high positive crime seriousness
coefficients were persons who stressed the seriousness of the crimes shown
and varied their judgments from vignette to vignette.
The final variable in this block was used only in relation to the standard
deviation of the rating. This variable is the amount of variance in the
judgments of an individual that was accounted for by the main dimensions
of the vignettes rated by that individual. The higher the R 2 for an individual,
the greater was the variability of his/her ratings. In short the interpretation
of the standard deviation as a measure of respondent discrimination given
earlier is confirmed. Note that this predictor variable accounted for more
variation (over half) than any other variable in this equation.
The average R 2 for individual respondents was 0.44 (SD = 0.17), higher
than that of any of the equations in Table III, indicating that each
individual's judgments tended to be more structured than were the pooled
ratings of all the respondents, indicating that rating principles had some
idiosyncratic qualities. Respondents' ratings were internally more consistent
than were the pooled ratings (as shown in Table 111).13 This finding also
suggests that at least some of the unexplained variance shown in Tables
III and IV reflected idiosyncracies in the judgment processes employed by
individual respondents.
The final block of independent variables pertains to characteristics of
the individual respondents. As we saw earlier, females were more severe in
their ratings than males, not any more variable than males, and more likely
to weight crime seriousness higher. Severity neither increased nor decreased
13Although the analysis is not shown in this paper, we found that there were some small but
significant systematic differences a m o n g respondents in the extent to which their ratings
were structure (as indexed by the size of the individual R 2 computed for their ratings).
Higher R 2 values were found for whites and for persons who showed a high degree of worry
about crime. We had expected that the higher the educational levels (and hence the greater
the ability to process information), the more structured the judgments, but the coefficient
for education, although positive (and hence in the predicted direction), was not statistically
significant ( P = 0.06).
Beyond Crime Seriousness 87
with age, nor did the variability of the ratings. But older persons were Iess
likely to weight crime seriousness as heavily as younger persons.
The remaining social and demographic respondent characteristics
apparently affected rating tendencies little or not at all. Catholics weighted
crime seriousness less than other religious groups but all religious groups
were alike as far as mean ratings and ratings variability were concerned.
The higher a household's income, the more likely they were to weight
seriousness higher. Whites apparently weighed crime seriousness more
highly than other groups (mainly blacks and Hispanics). Surprisingly,
educational attainment, usually an influence on everything, counted for nil
in this context. The final personal variable is the probability of being
victimized by crime, as shown in a recent victimization survey] 4 Persons
in age, sex, and ethnic groups more subject to victimization were not different
from those with low probabilities. Note, as we found in Table III, that the
social and demographic characteristics included in these analyses were not
very strong influences on rating tendencies.
The last five variables are all scores on attitude scales, administered to
respondents in a short interview after they finished the ratings of vignettes.t5
The Institutional Fairness Scale measured how fairly the respondent
believed that major American institutions (businessess, public agencies,
banks, etc.) treated their clients (workers, customers, etc.). The more fair
these institutions were seen to be, the less severe were respondents.
Apparently, respondents who trust the major institutional structures of our
society were more inclined to be lenient in preferred punishments for
convicted felons. Such persons were also likely to be less variable in their
ratings of leniency.
The next scale measures the extent to which the respondent believes
that criminals are defective persons (i.e., incapable of making moral judg-
ments or of being reformed). Persons who subscribe to the viewpoint that
criminals are defective are more severe in their judgments and weigh the
seriousness of the crimes more heavily. Apparently, a belief that criminals
are different in some fundamental way from others led one to believe in
harsher punishment and to give greater importance to the seriousness of
the crime in question.
Respondents who worried more about crime were less lenient and more
variable in their judgments. Oddly enough, a scale of their experiences with
14Probabilities of victimization by violent crimes and by property crimes were obtained for 28
demographic groups formed by the cross classification of sex, race, and seven age brackets
from the National Victimization Surveys (Bureau of Justice Statistics, 1980). The probability
of victimization shown in the analysis is simply the s u m of the two separate probabilities.
~SScale items were written by the authors and subjected to factor analysis to improve their
structures. Copies of the interview schedules used m a y be obtained from the authors.
88 Rossi, Simpson, and Miller
4. CONCLUSIONS
We have attempted to accomplish two tasks in this paper. First, we
presented an exposition of a new technical device--factorial surveys--to
demonstrate the richness of the data provided by the use of this approach
and how flexibly the data may be analyzed. Within the scope of this article,
we could give only illustrations of the types of analyses possible. Of course,
we did not present random samples of possible analytical approaches but
some of the more interesting analyses attempted so far. But we hope that
these illustrations support strongly the general points we have made about
the technique. Fuller expositions can be found in subsequent publications
from our project.
Second, we have gone beyond measuring the seriousness values of
crime to show how that aspect of crimes is related to popular views of
appropriate punishments. We have shown that there is not a one-to-one
direct relationship between the seriousness measures of crimes and desired
sanctions but rather that seriousness is modified by the characteristics of
offenders and victims and by the consequences of the crimes in question.
We have also shown that the desired punishments for crimes vary in severity
by the characteristics of the persons who make judgments, whose past
experiences, attitudinal sets toward punishments, and general views on
criminality all make differences in what they may regard as a fair sentence
for criminals convicted.
Finally, we offered some hints about how criteria for sentences for
convicted felons can change over time without accompanying changes in
the assessments of crime seriousness. Attitudinal sets concerning worry over
crime, confidence in central social institutions, and opinions on such issues
as deterrence all affect the way in which crime seriousness is converted into
opinions about appropriate sentences for convicted felons. Going beyond
crime seriousness has meant an understanding that desired punishment is
determined by a multiplicity of considerations.
ACKNOWLEDGMENTS
This article is a revision of a paper presented at the November 1983
Meetings of the American Society of Criminology. The research reported
here was supported by Grant SES-8121745, Popular Bases for Criminal
Justice: Seriousness, Punishment and Justice Judgments, from the Law and
Society and Sociology Programs of the National Science Foundation. We
90 Rossi, Simpson, and Miller
REFERENCES
Berk, R. A., and Rossi, P. H. (1977). Prison Reform and State Elites, Ballinger Press, Cambridge.
Blumstein, A., and Cohen, J. (1980). Sentencing of convicted offenders: An analysis of the
public's view. Law Soc. Rev. 14: 223-261.
Blumstein, A., Cohen, J., Martin, S. E., and Tonry, M. E. (eds.) (1983). Research on Sentencing:
The Search for Reform, National Academy of Sciences Press, Washington, D.C.
Boydell, C. L., and Grindstaff, C. G. (1974). Public opinion towards legal sanctions for crimes
of violence. J. Crim. Law Criminol. 65:113-116.
Bureau of Justice Statistics (1980). Criminal Victimization in the United States, Government
Printing Office, Washington, D.C.
Center for Studies in Criminology and Criminal Law (1978). National Survey of Crime Severity:
Final National Level Geometric Means and Ratio Scores by Offense Stimuli Items, University
of Pennsylvania, Philadelphia.
Cullen, F. T., Link, B. G., and Polanzi, C. G. (1982). The seriousness of crime. Criminology
20: 83-102.
Gibbons, D. C. (1969). Crime and punishment: A study in social attitudes. Soc. Forces 47:
391-397.
Gross, H., and Von Hirsch, A. (1981). Sentencing, Oxford University Press, New York.
Hagan, J., and Bumiller, K. (1983). Making sense of sentencing: A review and critique of
sentencing research. In Blumstein, A., et al. (eds.), Research on Sentencing: The Search
for Reform, National Academy of Sciences Press, Washington, D.C.
Hamilton, V. L., and Rytina, S. (1980). Social consensus on norms of justice: Should the
punishment fit the crime? Am. J. Sociol. 85:1117-1144.
Kmenta, J. (1977). Elements of Econometrics, Macmillan, New York.
National Opinion Research Center (1983). Codebook for General Social Survey, NORC,
Chicago.
Rose, A. M., and Prell, W. E. (1955). Does the punishment fit the crime? A study in social
evaluation. Am. J. Sociol. 61: 247-259.
Rossi, P. H., and Anderson, A. B. (1982). The factorial survey approach: An introduction. In
Rossi, P. H., and Nock, S. (eds.), Measuring Social Judgments, Sage, Beverly Hills, Calif.
Rossi, P. H., and Henry, J. P. (1980). Seriousness: A measure for all purposes? In Klein, M.
W., and Teilmann, K. S. (eds.), Handbook of Criminal Justice Evaluation, Sage, Beverly
Hills, Calif.
Rossi, P. H., and Nock, S. (eds.) (1982). Measuring Social Judgments: The Factorial Survey
Approach, Sage, Beverly Hills, Calif.
Rossi, P. H., Waite, E., Bose, C., and Berk, R. A. (1974). The seriousness of crimes: Normative
structure and individual differences. Am. Sociol. Rev. 39: 224-237.
Sellin, T., and Wolfgang, M. E. (1964). The Measurement of Delinquency, John Wiley and
Sons, New York.
Stinchcombe, A. L. (1980). Crime and Punishment: Changing Attitudes in America, Jossey-Bass,
San Francisco.
Thomas, C. W., Cage, R. J., and Foster, S. C. (1976). Public opinion on criminal law and legal
sanctions: An examination of two conceptual models. J. Crim. Law Criminol. 67:110-116.