severidad de las penas
DESCRIPTION
TRANSCRIPT
ATTITUDES TOWARDS SEVERITY OF PUNISHMENT:
A CONJOINT ANALYTIC APPROACH
MICHAELA BROCKE*, CHRISTIAN GOLDENITZ, HEINZ HOLLING andWOLFGANG BILSKY$
Westfalische Wilhelms-Universitat Munster, Psychologisches Institut IV, Fliednerstraße 21,48149 Munster, Germany
Past research suggests that attitudes towards severity of punishment are affected by crime-specific factors. Theimpact of such factors has usually been investigated by between-subjects designs. The studies reported in thispaper, however, are based on within-subjects designs, using conjoint analysis for data collection and analysis.Study 1 employs a rape scenario for investigating the impact of the victim�/offender relationship and of twovictim characteristics �/ provocative behavior and intoxication. Study 2 uses a theft and an assault scenario foranalyzing the influence of several offender and crime characteristics on sanctioning: offender’s age, readinessto confess, previous convictions, and severity of the offense. Results from both studies are reported anddiscussed in terms of utility values. These values represent the importance placed on the case characteristicsfocused upon. In addition to the general evaluation of case characteristics, inter-individual differences areanalyzed by means of hierarchical cluster analysis. Advantages of the conjoint analytic approach overconventional research methods on sanctioning behavior are discussed.
Keywords: Sentencing; Severity of Punishment; Conjoint Analysis; Within-subjects Design
INTRODUCTION
Attitudes towards sanctioning in general and towards severity of punishment in particular
have been repeatedly in the focus of research on crime and delinquency. The same applies to
actual sanctioning behavior (e.g. McFatter, 1986; Ouimet and Coyle, 1991; Pfeiffer and
Oswald, 1989). Beside differences between professionals, differences between lay persons
received considerable attention (e.g. Carroll and Payne, 1977; Gabriel and Greve, 1996).
These latter differences seem relevant with regard to public opinion (e.g. Durham, 1993).
However, the mere observation of differences is of little importance. What counts is the
knowledge of factors causing divergent orientations towards sanctioning.
Explaining laymen’s differences in the evaluation and appraisal of punishment proves to
be a difficult and complex task. This is true because people do not show a uniform and
undifferentiated orientation towards this topic. It is possible, of course, to identify
differences between individual attributions and attitudes towards crime and sanctioning
on a general level (Furnham, 1988). However, there is ample evidence that the individual’s
orientation towards (the severity of) punishment is affected by very different factors. Thus,
in order to get below the surface of a global analysis, the type of offense as well as situation-
ISSN 1068-316X print/ISSN 1477-2744 online # 2004 Taylor & Francis Ltd
DOI: 10.1080/10683160310001614793
*Corresponding author. E-mail: [email protected]$E-mail: [email protected]
Psychology, Crime & Law, June 2004, Vol. 10(2), pp. 205�/219
205
and offense-specific factors should be taken into account (Carroll and Payne, 1977; Hollin
and Howells, 1987). Similarly, offender characteristics like sex, age, socio-economic status,
or previous conviction proved to be relevant (Gabriel and Greve, 1996; Hagan, 1989).
Especially in rape cases (Krahe, 1991), two further factor groups have to be considered:
victim characteristics and victim behavior, and the offender�/victim relationship (e.g. Best
and Demmin, 1982; Hammock and Richardson, 1997; Workman and Freeburg, 1999). It is
obvious, then, that a general, undifferentiated approach for assessing attitudes towards
severity of punishment is unlikely to go far enough in explaining judgmental differences.
In our own research on attitudes towards severity of punishment, we recently conducted
two studies, varying several factors in a quasi-experimental design. These factors included
offender characteristics (prior conviction and employment) within short scenarios of theft
and assault (Reichert, 1999; Reichert and Bilsky, 2001), and victim characteristics
(intoxication, provocativeness) as well as the offender�/victim relationship in a rape scenario
(Brocke et al ., 2001; Bus, 2001). Several observer characteristics like sex (both studies), age
(first study), generalized attitude towards sanctioning, role orientation, rape myths, empathy
with victim and offender, and direct or indirect experience of victimization (second study)
were measured additionally, in order to control their influence on individual judgments.
Scenarios were presented as short vignettes in between-subject designs, as it is usually the
case in research on the impact of crime characteristics on sanctioning.
In our first study, prior conviction of the offender revealed considerable impact on
severity of punishment in an analysis of covariance, explaining 19.5% of variance in theft,
F (1,202)�/52.69, pB/0.001, and 24.4%, F (1,202)�/75.65, pB/0.001, in bodily harm. The
influence of employment was low but significant, too, resulting in F (1,202)�/5.51, pB/0.05
for theft, and F (1,202)�/4.02, pB/0.05 for bodily harm. Furthermore, a significant
interaction was identified for bodily harm, F (1,202)�/28.28, pB/0.001, revealing harder
punishment for an offender without employment and prior conviction. In our second study
using a rape scenario, multivariate analysis of covariance revealed only a marginal effect of
the victim�/offender relationship, F (2,313)�/2.95, p�/0.05. None of the remaining factors
and interactions came close to significance.
While findings from both studies seem interesting with respect to attitudes towards
severity of punishment, the between-subjects designs procedure employed in these
investigations reveals several shortcomings. A main disadvantage of this design refers to
the extensive sampling that is required when a high number of case characteristics is under
study. The higher the number of case characteristics, the more vignettes result and the more
respondents are needed. Thus, between-subjects designs are rather inefficient for investigat-
ing the impact of several crime characteristics simultaneously. Besides, between-subjects
designs require extensive data collection , since observer characteristics have to be measured
additionally. Moreover, with a between-subjects design only indirect conclusions about
factors influencing sanctioning decisions, drawn from inter-individual differences , are
provided. Thus, analyses of individual ‘‘sanctioning patterns’’, that is intra-individual
comparisons in weighting the impact of different factors, are not possible. As a consequence,
examination of inter-individual differences in sanctioning structures is unfeasible, too.
To overcome these shortcomings, in the studies to be reported here, an experimental
within-subjects design was used. This design was realized by means of conjoint analysis .
Conjoint analysis provides reliable estimates of intra-individual weights, even with complex
factorial designs, without extensive sampling as usually required for within-subjects designs.
Actually, conjoint analysis is a data collection and analysis technique, which has become
206 M. BROCKE et al .
popular in decision analysis and in market research (Green and Srinivasan, 1978, 1990;
Green et al. , 2001). In this type of research, participants are asked to judge objects
repeatedly with regard to their attractiveness. Other things being equal, these objects are
varied systematically with respect to some well-defined factors supposed to influence their
evaluation. Respondents’ answers are then decomposed by data analysis, assigning different
weights to the factors under study.
In the present research context, assessment differed in that respondents had to assign
subjectively adequate degrees of punishment to offenses presented in different crime
scenarios. These scenarios varied with respect to factors supposed to influence the perceived
seriousness of the criminal act under study. Conjoint analytic estimates gained from these
judgments, then, represent the importance placed on the different case characteristics with
respect to severity of punishment. In other words, the results reflect the relative impact of
case characteristics on preferred sentencing. To the best of our knowledge, this type of
analysis has not been used for analyzing sentencing behavior in the past. Interestingly,
however, for measuring the seriousness of different types of offenses the paired comparison
method and extensive measurement, both sharing kinship with conjoint analysis, have yet
been applied (Thurstone, 1927; Levi, 1974; Francis et al. , 2001).
In the following, two conjoint analytic studies on factors influencing the severity of
punishment are reported. Both studies are exploratory to the extent that they test the
applicability of conjoint analysis in this special research context. The first study used a rape
scenario for investigating the influence of victim characteristics and of the victim �/offender
relationship on sanctioning. The second study concentrated on the impact of offender as well
as context characteristics in an assault and a theft scenario . For each study, the design is
presented first, together with some information about former research on the variables
under study. Then, the conjoint analytic method and the results are sketched out. Since the
use of conjoint analysis is uncommon in this research context, data collection and data
analysis are described in some detail.
A RAPE SCENARIO (STUDY 1)
In our first study, the impact of two victim characteristics on the severity of punishment,
intoxication and provocative behavior , was analyzed in a rape scenario, together with the
victim �/offender relationship . These factors were combined in a 2�/2�/3 factorial design as
outlined in Table 1. While the factors intoxication and provocative behavior comprised two
levels each (present/absent or yes/no, respectively), the victim �/offender relationship was
operationalized by three factor levels: the categorical distinction between nodding
TABLE 1 Factors and factor levels of Study 1.
Factor Factor level
Behavior of victim no provocationprovocation
Intoxication victim is sobervictim is tipsy
Relation of victim and offender nodding acquaintanceex-couple, offender partedex-couple, victim parted
SEVERITY OF PUNISHMENT 207
acquaintance and former partnership was supposed to represent different degrees of
intimacy between victim and offender; this latter category was further specified by indicating
who quit the former partnership �/ the victim or the offender.
The effects of the above factors have been studied repeatedly in connection with rape (e.g.
Brocke et al ., 2001; Bus, 2001). For intoxication and provocative behavior of the victim, a
mitigating influence on sentencing has been confirmed by former research. Thus, studies
about the influence of a woman’s consumption of alcohol prior to rape unanimously show
that the offender is considered less guilty, if the victim is intoxicated. Accordingly, a lower
degree of punishment is chosen (e.g. Hammock and Richardson, 1997; Stormo et al ., 1997;
Schuller and Wall, 1998). Other studies support the assignment of a higher degree of guilt to
a victim if she is blamed for having shown provocative behavior prior to her victimization
(e.g. Acock and Ireland, 1983; Schult and Schneider, 1991); this holds true for different
operationalizations, for instance, wearing a provocative dress (e.g. Furnham and Boston,
1996; Workman and Freeburg, 1999), or showing improvident and role-discrepant behavior
(e.g. Best and Demmin, 1982).
The importance of the victim �/offender relationship has also been considered in various
studies; results, however, proved inconsistent. Thus, a friendship between victim and
offender may cause higher (e.g. Bridges, 1991; Szymanski et al ., 1993) or lower attribution of
guilt to the victim (e.g. Bolt and Caswell, 1981; Krulewitz, 1982). These discrepancies may
indicate that it is not the mere acquaintance but the degree of intimacy and trust between
offender and victim that has to be taken into consideration (Sczesny and Krauel, 1996). The
degree of closeness had been considered when designing Study 1, therefore.
Method
Procedure
Seventy-five students (62 female and 13 male), aged 19�/44 (median�/20), participated in
this study. Aside from some general information about the investigation, participants
received detailed instruction of how to deal with the conjoint analytic tasks. Since
consecutive trials of the experimental variation of factor levels were displayed on a screen
in a brief and standardized form only, the rape scenario within which to interpret this
information was presented first. Furthermore, it was stressed that offenses vary only with
respect to the variables displayed. Compliance with this instruction is of considerable
importance to guarantee the validity of the subsequent judgments.
Subsequently, the presentation of the experimental variables and the recording of
participants’ reactions were conducted in a computer-assisted form, using Alasca (Holling
et al. , 2000). In order to get acquainted with the case characteristics under study, the
participants had to rank order the levels of each factor first. Then, they worked on 25 graded
paired comparison and eight rating tasks, as described next.
Each graded paired comparison task consisted of two rape offenses differing in the factor
levels under study. Participants had to compare the seriousness of these offenses and to state
on a seven-point rating scale whether and to which extent one of them should be punished
harder than the other (cf. Figure 1).
Offenses were first described by two (10 tasks), and then by three factor levels (15 tasks)
each. Aside from the first pair, presentation of each of the following depended on the
previous judgments. This form of data collection, called adaptive conjoint analysis (Johnson,
1987), ensures that every response yields a high level of information about the influence of
208 M. BROCKE et al .
the factor levels on the degree of punishment. More concretely, numerical estimates of the
impact of the different factor levels are calculated after every response, and an algorithm is
used to determine the combination of factor levels to be presented next, in order to gain the
maximum information required for stabilizing the estimates. This is why adaptive conjoint
analysis represents a highly efficient way of data collection in a within-subjects design.
Finally, additional rating tasks were presented in an adaptive manner. In these tasks,
respondents had to rate the seriousness of eight offenses on a nine-point rating scale as
described by varying levels of the three factors under study (cf. Figure 2).
Data Analysis
In conjoint analysis, two kinds of estimates describing the influence of the independent
variables are available. One of them is related to the factor levels and termed part-worths . In
the present context, the part-worths can be interpreted as ‘‘penalty-worths’’, indicating the
extent to which the presence of a certain factor level is perceived as aggravating from a
participant’s point of view. To gain the part-worths, weights for the factor levels are
estimated from the paired comparisons by ordinary least squares (OLS) regression. This
estimation method is usually used in conjoint analysis, because it is equivalent (e.g. Wittink
and Cattin, 1981) or superior (e.g. Cattin and Bliemel, 1978) to nonmetric methods. If the
independent (categorical) variables are dummy coded, OLS regression is analogous to
analysis of variance, as typically used in between-subjects design studies.
Part-worths are calculated for every participant. To obtain inter-individual comparability
of these estimates, part-worths are standardized. This is accomplished for each factor by
setting the factor level with the highest part-worth to one, and the lowest to zero. Part-
FIGURE 1 Paired comparison task in Alasca .
SEVERITY OF PUNISHMENT 209
worths of intermediate levels are transformed accordingly. In order to aggregate results,
means of the standardized part-worths are calculated.
As aggregation may level differences in the constellation of part-worths between
participants or subgroups (Backhaus et al ., 2000), classification methods can be applied
to the data. To test for subgroups that differ in judging the seriousness of an offense, a
hierarchical cluster analysis on the individual, standardized part-worths was conducted in
the present study, using Ward’s method and squared Euclidean distances as a measure of
similarity between participants.
The second kind of estimate is called relative importance (score ). It refers to the factors
and specifies their influence on the judgment on the whole. For each factor, it is determined
by the difference between the highest and the lowest standardized part-worth. Thus, if the
part-worths of a two-level factor are similar, for example, a low relative importance score
results. This score indicates that judging the severity of an offense does not change
substantially, whether one or the other level of this factor is considered. For simplicity of
interpretation, importance scores are transformed into percentages by dividing each factor’s
range by the sum of the ranges of all factors (cf. Hair et al ., 1998).
Two methods exist for aggregating relative importance scores. First, relative importance
can be determined by the span of aggregated part-worths. This measure reveals the influence
of the factor levels on the sample as a whole and resembles the results achieved in between-
subjects designs. Second, importance can be calculated on an individual level and aggregated
afterwards. By applying this procedure, the mean influence of the factor levels on the
individual level is determined. If the constellations of individual part-worths diverge, the two
FIGURE 2 Rating task in Alasca .
210 M. BROCKE et al .
measures may differ. Since they complement each other, both are reported and compared for
a rough check of heterogeneity.
Besides the part-worths and the relative importance scores, measures of reliability and
validity can be determined in conjoint analysis, too. From the measures that exist (cf.
Bateson et al ., 1987), two commonly used are calculated here.
The first is the R2 of the OLS regressions conducted on an individual level. R2 values are
considered as a measure of internal consistency because goodness of fit is essentially
determined by the consistency of a participant’s responses. For all participants, R2 values are
checked, and participants with values lower than R2�/0.5 are excluded from further
analyses. This seems adequate, because part-worths calculated for these cases are not
supposed to be sufficiently reliable. For the remaining participants, the mean of the Fisher z
transformed individual R2 values is reported.
The second measure, the hit rate, indicates the predictive validity of the part-worths. It is
calculated from the data of the rating part of the conjoint analysis. All pairwise
combinations of the offenses presented in the rating part were inspected with respect to
possible differences in rating. For each pair differing it was checked, whether the part-worths
predict correctly to which of the two offenses a higher degree of punishment is assigned. The
relative frequency of correct predictions is transformed into a percentage value for each
participant and aggregated afterwards. As a rule of thumb, hit rates above 80% are usually
regarded as high (Teichert, 2000).
It should be noted that the values for internal consistency and predictive validity also offer
information about the independence and the additivity of the factor levels underlying the
regression model adopted. Both measures indicate the degree of concordance between the
predicted and the actual responses in the paired comparison and the rating part.
Results
Five participants showed goodness of fit values of the respective regression lower than R2�/
0.5, indicating a low level of internal consistency. They were excluded from further analysis,
therefore. For the remaining 70 participants, the mean of the goodness of fit indices was
R2�/0.87. Hit rates resulted in 80.2% on average. According to the criterion mentioned
above, this percentage indicates a high level of predictive validity.
The means of the standardized individual part-worths are depicted in Figure 3. They
illustrate both the different influence of the factors and the relations among the levels of each
FIGURE 3 Mean part-worths of the factor levels for rape.
SEVERITY OF PUNISHMENT 211
factor. On an aggregated level, provocative behavior has the strongest influence on the
assessment of punishment (79.8% of relative importance), since the difference between the
levels of this factor is higher than between the levels of the remaining factors. Here, a higher
degree of punishment is assessed to the offender, if the victim is not blamed for provocative
behavior. In other words, provocative behavior of the victim is likely to concede extenuating
circumstances to the offender. The second-highest importance (17.4% of relative impor-
tance) is attached to the victim �/offender relationship . For this factor, the part-worth for a
nodding acquaintance is higher than for a former partnership of offender and victim. Thus,
a high degree of intimacy is regarded as mitigating. Since the difference between the factor
levels of intoxication is negligible, this factor is supposed to play a minor role in the present
context (2.8% of relative importance). Nevertheless, part-worths imply the assignment of a
slightly higher degree of punishment to the offender, if the victim is intoxicated.
On an individual level, different relative importance scores result for all factors. Thus, the
influence of provocative behavior (49.3% of relative importance) diminishes. On the other
hand, higher relative importance scores result for the victim �/offender relationship (29.7% of
relative importance) and for intoxication (21.0% of relative importance).
The differences found in relative importance on the aggregated and on the individual level
indicate disagreement between participants. This is also reflected by a hierarchical cluster
analysis computed for the individual standardized part-worths. The dendrogram of this
analysis suggested a four-cluster solution. In Figure 4, profiles of the mean part-worths are
shown for each of these clusters. They illustrate the difference in weighting applied by these
subgroups. While there did not result any differences with respect to provocative behavior ,
the four clusters can be divided into two groups with regard to intoxication : respondents
belonging to clusters 1 and 2 (n�/23) attach a higher part-worth to the intoxication of the
victim, whereas those of clusters 3 and 4 (n�/47) do not. Finally, groups disagree in
weighting the different levels of the victim �/offender relationship : while rape by a nodding
acquaintance, as compared to a former partner, is regarded as more serious by the majority
of our participants (n�/60), this does not hold for the respondents of cluster 4 (n�/10).
FIGURE 4 Mean part-worths of the factor levels for subgroups determined by cluster analysis.
212 M. BROCKE et al .
Furthermore, information about who quit the former partnership influences the assignment
of punishment, too; clusters differ with respect to the weighting of the respective alternatives.
Thus, participants of clusters 2 and 4 (n�/21) punish less hard if the offender has separated,
while those of cluster 1 and 3 (n�/49) assign opposite weights.
DISCUSSION
With respect to the criteria mentioned, the results of the conjoint analysis in our first study
proved reliable and valid, as indicated by the internal consistency and the predictive validity.
In addition, these results demonstrate the appropriateness of the conjoint analytic model.
For provocative behavior , the part-worths are in accordance with the results of previous
studies using a within-subjects design. This applies to the whole sample (i.e. on an
aggregated level) and to the different subgroups.
For intoxication on an aggregated level, on the other hand, part-worths suggest a
direction opposite to former results, although, factor levels were nearly equal in height.
However, differences in weighting this latter factor could be demonstrated between two
subgroups by means of cluster analysis: the majority of the participants is likely to interpret
the intoxication of the victim in the expected sense, that is, in the sense of shared
responsibility. The second, smaller group, in contrast, tends to interpret the offender’s
behavior as taking advantage of the victim’s handicap in this special situation. Two factors
may have contributed to this difference between the present and the former results. First,
direct comparison of two factor levels, as simultaneously presented in our study, may lead to
other judgmental processes than the evaluation of one of the levels alone. Second, changing
role expectations, that tolerate more and more the consumption of alcohol by women, may
have come into effect.
Finally, for the impact of the victim �/offender relationship on punishment a high degree of
intimacy is regarded as mitigating on an aggregated level. However, results of the cluster
analysis reveal inter-individual differences for this factor, too. These differences may have
contributed to inconsistent results obtained with between-subjects designs in the past.
Assault and Theft (Study 2)
In our second study, effects of four factors on sanctioning were investigated in an assault
and a theft scenario: offender’s age , confession , severity of the offense and previous
conviction . For the latter factor, pertinence was considered by two factor levels. In both
scenarios, age and confession comprised two, severity and previous conviction three factor
levels, resulting in a 2�/2�/3�/3 factorial design. Factor levels were the same for both
scenarios, except for severity , which had to be adapted to the different types of crime (see
Table 2).
Former studies on the effect of offender’s age , severity of offense and previous conviction
have shown significant effects on sanctioning for several crimes. Thus, more lenient forms of
punishment, for instance, have been assigned to young perpetrators without previous
conviction, who committed less serious types of crime (Gabriel and Greve, 1996; Reichert,
1999). This is in line with both jurisdiction as established in law in many countries (Hagan,
1989), and with actual sentencing behavior of judges (Steffensmeier and Hebert, 1999). The
same holds for the mitigating influence of a confession on punishment.
SEVERITY OF PUNISHMENT 213
METHOD
Procedure and Data Analysis
Sixty students (45 female and 15 male), aged 19�/54 (median�/22), participated in Study 2.
Data collection was similar to that in our first study. However, because of the greater
number of factors and factor levels in the present study, pairs of offenses presented were
described by two (10 tasks), three (10 tasks), and four factor levels (five tasks), respectively,
in the paired comparison tasks. The offenses displayed in the rating part consisted of four
levels each. Furthermore, participants had to run through the conjoint analytic procedure
twice�/once for theft, and once for assault. The sequence of these offenses was balanced
between participants. Methods applied for data analysis were the same as in Study 1.
RESULTS
The goodness of fit values for the regressions were lower than R2�/0.5 for four participants.
For the remaining 56 participants, means of the goodness of fit indices were R2�/0.86 for
both offenses. Hit rates were 89.0% on average for theft and only slightly lower (88.2%) for
assault. Thus, the predictive validity of the part-worths proved extremely high.
The means of the standardized individual part-worths for theft are depicted in Figure 5.
On an aggregated level, previous conviction proved most important for sentencing (38.9% of
relative importance): the difference between no previous conviction and previous conviction
is higher than the differences between the levels of the remaining factors. A much higher
degree of punishment is assigned to an offender who has been previously convicted,
independently of pertinence or non-pertinence. For the other factors, severity (23.9% of
relative importance), confession (20.3% of importance), and age (16.8% of relative
importance), a lower influence on the assignment of punishment results. Young age,
confession, and low severity of the offense proved mitigating, as expected.
For assault a different weighting scheme results (see Figure 6). Whereas previous
conviction was of utmost importance for theft, severity of the offense had the strongest
impact on sanctioning for assault (41.9% of relative importance). Nevertheless, previous
convictions proved to be an influential factor, too (34.5% of relative importance). While the
TABLE 2 Factors and factor levels of Study 2.
Factor Factor level
Age 17 years30 years
Confession offender confessesoffender denies
Previous conviction no convictionconviction for theftconviction for assault
Severity of offence: theft 250 t loss750 t loss1250 t loss
Assault several punchespunches, knock-downpunches, knock-down, additional kicks
214 M. BROCKE et al .
direction of influence was the same for both offenses, the profile of part-worths is different:
part-worths for the levels of severity are graded uniformly for theft. For assault, however,
the level representing the highest severity is judged much more aggravating than the
remaining two. Furthermore, whereas the difference between a pertinent and a non-
pertinent conviction was of no importance for theft, a much higher part-worth results for a
pertinent in comparison to a non-pertinent conviction for assault. In sum, for these two
factors the relative importance scores as well as the profile of part-worths are different. With
respect to the two remaining factors, results resemble the findings for theft: confession
(13.8% of relative importance) and age (9.9% of relative importance) are both of minor
importance for sanctioning. However, mitigating effects of both factors are even less
pronounced for assault than for theft.
For both theft and assault, the relative importance scores calculated on the individual level
do not differ substantially from those on the aggregated level: for theft, the mean individual
relative importance scores differ from the aggregated relative importance scores by only
FIGURE 5 Mean part-worths of the factor levels for theft.
FIGURE 6 Mean part-worths of the factor levels for assault.
SEVERITY OF PUNISHMENT 215
3.6% on average. For assault, differences are even smaller, with a mean of 2.4%.
Accordingly, the part-worth profiles of subgroups identified by hierarchical cluster analysis
differ only slightly and are therefore not depicted graphically.
DISCUSSION
On the whole, the goodness of fit indices and the individual hit rates for both offenses
suggest again the appropriateness of the additive main effects models used to calculate the
part-worths. Part-worths show that the factor levels influence sanctioning uniformly in the
expected direction in both offenses: A young perpetrator with no previous conviction, who
commits a minor offense and confesses is punished less hard than a previously convicted
older person, who commits a major offense and refuses to confess. However, differences in
the profiles of the part-worths and in the relative importance of factors were found for
assault and theft.
SUMMARY AND CONCLUSIONS
For a differentiated analysis of attitudes towards severity of punishment, a number of
features of a crime has to be considered. Compared to between-subjects designs, within-
subjects designs realized by conjoint analysis represent a more accurate and parsimonious
approach in this context. We reported on two studies using conjoint analysis for data
collection and analysis. In a first study, we aimed at the description of the sanctioning
structure applied to rape; in a second study, we focused on the (comparison of) sanctioning
patterns applied to theft and assault. All in all, findings from both studies suggest that
detailed insights into the process of sanctioning can be gained by using a within-subjects
design and conjoint analysis, respectively. The various benefits of this approach in the
present research context are briefly summarized and discussed in the following.
The main advantage of the approach stems from the fact that the impact of different factor
levels on the degree of punishment is determined on an individual level . In other words, the
influence of factor levels on the dependent variable need not be inferred indirectly by
comparing experimental groups; rather, participants serve as their own control in a repeated
measurements design. This implies, that, unlike between-subjects designs, a within-subjects
design does not require recording of individual characteristics in order to control for
between-subjects variance that is undue to experimental manipulation. Furthermore,
subgroups of participants, differing in judgmental styles or patterns, can be identified.
With a between-subjects design, these inter-individual differences remain unconsidered.
However, the cluster analytic results of our first study showed that disentangling judgmental
styles may be revealing with respect to more fine-grained differences in sanctioning.
Moreover, a large number of factors and factor levels can be studied with conjoint
analysis. Complex research designs can be employed, because the evaluation of subsets of
stimuli within fractional factorial or adaptive designs still provides reliable estimates of
parameters. The main advantage of complex designs is that a trade-off of several different
features of an offense is possible, thus providing a more differentiated analysis of attitudes
towards severity of punishment.
As offenses are judged in pairs, the standards of comparison and the factor levels under
study are always evident in conjoint analysis. This is important for methodological as well as
216 M. BROCKE et al .
content reasons. From a methodological point of view, ratings in between-subjects designs
may result opposite to those in within-subjects designs because of unclear standards of
comparison. Birnbaum (1999) provides an instructive experimental illustration for mislead-
ing conclusions resulting from between-subjects designs. He also discusses this problem with
respect to studies on rape. As regards content, presenting a standard of comparison seems
reasonable, too. Thus, in a vignette approach, about one half of the respondents stated to be
unable to make judgments on the basis of the information provided (Krahe, 1991). The
qualitative prototype approach suggested by Krahe (1991), alternatively, resembles the
conjoint approach inasmuch as both rely on the comparison of features. In sum, though
unusual in the given context, conjoint analysis provides a means for studying intra-
individual preferences and patterns of sanctioning as well as inter-individual differences in
an economic manner.
Of course, in order to use this approach efficiently, the procedure has to be adapted to the
given research context. Therefore, we finally outline some recommendations for research
design and data collection in conjoint analysis as emphasized in the literature (e.g. Hair et
al ., 1998).
As regards the research design, the factors to be studied have to be well-selected, since the
relative importance scores of the factors are mutually dependent. In this respect conjoint
analysis closely resembles ipsative measurement. For example, in a conjoint analysis with
unimportant factors only, the relative impact of a single factor may be overestimated; if
more important factors were included, the impact of the first factors would certainly
diminish.
In addition, descriptions used for the factor levels have to be specified carefully, since
semantic variations may influence the results considerably. The range of the factor levels
under study has to be taken into consideration as well. A wide range may lead to higher
relative importance scores than a small range. Thus, to ensure the comparability of the
relative importance scores, it is recommended to use a similar range for all factors, and to
balance the number of levels across factors if possible.
Furthermore, considerable attention has to be paid to the experimental design used for
data collection. For example, in order to test for interaction effects of factor levels, designs
that allow for the estimation of interaction terms have to be constructed.
In view of the encouraging findings of our investigations, further studies should be
conducted to test other applications of conjoint analysis in the research on punishment and
sanctioning. Research on individual weighting schemes applied in judging the seriousness of
offenses using complex designs, for example, seem especially promising. In addition, the
relation of judgments relating to case characteristics and person variables could be
investigated on a more differentiated level, for instance the interdependence of evaluations
of different rape scenarios and rape myth acceptance (cf. Payne et al. , 1999) or rape empathy
(cf. Deitz et al ., 1982). Similarities and deviations in the weighting of factors, applied by
judges and laymen in the context of sanctioning and sentencing, are another possible field of
research. Finally, the predictive validity of factors identified in conjoint analysis needs
examination, if their usefulness is supposed to go beyond mere attitudinal research.
References
Acock, A. C. and Ireland, N. K. (1983). Attribution of blame in rape cases: the impact of norm violation,gender and sex role attitude. Sex Roles , 9, 179�/193.
SEVERITY OF PUNISHMENT 217
Backhaus, K., Erichson, B., Plinke, W. and Weiber, R. (2000). Multivariate Analysemethoden. Eineanwendungsorientierte Einfuhrung [Multivariate Data Analysis. An Application-oriented Introduction] , 9thedn. Berlin: Springer.
Bateson, J. E. G., Reibstein, D. and Boulding, W. (1987). Conjoint analysis reliability and validity: aframework for future research. In M. J. Houston (Ed.), Review of Marketing (pp. 451�/481). Chicago:American Marketing Association.
Best, J. B. and Demmin, H. S. (1982). Victim’s provocativeness and victim’s attractiveness as determinants ofblame in rape. Psychological Reports , 51, 255�/258.
Birnbaum, M. H. (1999). How to show that 9�/221: collect judgments in a between-subjects design.Psychological Methods , 1, 243�/349.
Bolt, M. and Caswell, J. (1981). Attribution of responsibility to a rape victim. Journal of Social Psychology ,114, 137�/138.
Bridges, J. S. (1991). Perceptions of date and stranger rape: a difference in sex role expectations and rapesupportive beliefs. Sex Roles , 24, 291�/307.
Brocke, M., Bus, J., Holling, H. and Bilsky, W. (2001). Deliktmerkmale und Strafzumessung bei einemVergewaltigungsdelikt: Vergleich einer Fragebogenstudie mit einer Conjoint-Analyse [Case Characteristicsand Sanctioning of Rape: Comparing a Vignette Survey and Conjoint Analysis]. In W. Bilsky and C.Kahler (Eds.), Berufsfelder der Rechtspsychologie. Dokumentation der 9. Arbeitstagung der FachgruppeRechtspsychologie in der DGPs, CD-ROM . Munster: Westfalische Wilhelms-Universitat.
Bus, J. (2001). Einfluss der Tater-Opfer-Beziehung und Einfluss von Opfervariablen auf die Strafzumessungbei einem Vergewaltigungsdelikt [Influence of the victim�/offender relationship and of victim characteristicson sanctioning rape]. Unpublished diploma thesis, Westfalische Wilhelms-Universitat, Munster.
Carroll, J. S. and Payne, J. W. (1977). Crime seriousness, recidivism risk, and causal attributions in judgmentsof prison term by students and experts. Journal of Applied Psychology , 62, 595�/602.
Cattin, P. and Bliemel, F. (1978). Metric vs. nonmetric procedures of multiattribute modeling: somesimulation results. Decision Sciences , 9, 472�/480.
Deitz, S. R., Blackwell, K. T., Daley, P. C. and Bentley, B. J. (1982). Measurement of empathy toward rapevictims and rapists. Journal of Personality and Social Psychology , 43, 372�/384.
Durham, A. M. (1993). Public opinion regarding sentences for crime: does it exist? Journal of Criminal Justice ,21, 1�/11.
Francis, B., Soothill, K. and Dittrich, R. (2001). A new approach for ranking serious offenses. The use ofpaired-comparisons methodology. British Journal of Criminology , 41, 726�/737.
Furnham, A. (1988). Lay Theories. Everyday Understanding of Problems in the Social Sciences . Oxford:Pergamon Press.
Furnham, A. and Boston, N. (1996). Theories of rape and the Just World. Psychology, Crime and Law , 2,211�/229.
Gabriel, U. and Greve, W. (1996). Strafe muß sein!’: Sanktionsbedurfnisse und strafbezogene Einstellungen:Versuch einer systematischen Annaherung [‘Discipline is necessary!’: needs and attitudes towardsanctioning: a systematic approach]. In C. Pfeiffer and W. Greve (Eds.), Forschungsthema ‘Kriminalitat’(pp. 185�/214). Baden-Baden: Nomos.
Green, P. E. and Srinivasan, V. (1978). Conjoint analysis in consumer research: issues and outlook. Journal ofConsumer Research , 5, 103�/123.
Green, P. E. and Srinivasan, V. (1990). Conjoint analysis in marketing: new developments with implicationsfor research and practice. Journal of Marketing , 54, 3�/19.
Green, P. E., Krieger, A. M. and Wind, Y. (2001). Thirty years of conjoint analysis: reflections and prospects.Interfaces , 31, 56�/73.
Hagan, J. (1989). Strafzumessungsforschung in Nord-Amerika [Research on sentencing in North America]. InC. Pfeiffer and M. E. Oswald (Eds.), Strafzumessung: Empirische Forschung und Strafrechtsdogmatik imDialog (pp. 147�/182). Stuttgart: Enke.
Hair, J. F., Jr, Anderson, R. E., Tatham, R. L. and Black, W. C. (1998). Multivariate Data Analysis , 5th edn.Upper Saddle River, NJ: Prentice Hall.
Hammock, G. S. and Richardson, D. R. (1997). Perceptions of rape: the influence of closeness of relationship,intoxication and sex of participant. Violence and Victims , 12, 237�/246.
Hollin, C. R. and Howells, K. (1987). Lay explanations of delinquency: global or offence-specific? BritishJournal of Social Psychology , 26, 203�/210.
Holling, H., Jutting, A. and Großmann, H. (2000). Alasca. Ein universelles Programmsystem zurEntscheidungs- und Nutzenanalyse [Alasca. A Universal Program System for Decision and Utility Analysis] .Gottingen: Hogrefe.
Johnson, R. M. (1987). Adaptive conjoint analysis. In M. Metegrano (Ed.), Sawtooth Software Conference onPerceptual Mapping, Conjoint Analysis, and Computer Interviewing (pp. 253�/265). Ketchum, ID: SawtoothSoftware.
Krahe, B. (1991). Social psychological issues in the study of rape. European Review of Social Psychology , 2,279�/309.
218 M. BROCKE et al .
Krulewitz, J. E. (1982). Reactions to rape victims: effects of rape circumstances, victim’s emotional response,and sex of helper. Journal of Counselling Psychology , 29, 645�/654.
Levi, A. M. (1974). Constructive, extensive measurement of preference to predict choice between sums ofoutcomes. Behavioral Science , 19, 326�/335.
McFatter, R. M. (1986). Sentencing disparity: perforce or perchance? Journal of Applied Social Psychology ,16, 150�/164.
Ouimet, M. and Coyle, E. J. (1991). Fear of crime and sentencing punitiveness: comparing the general publicand court practitioners. Canadian Journal of Criminology , 33, 149�/162.
Payne, D. L., Lonsway, K. A. and Fitzgerald, L. F. (1999). Rape myth acceptance: exploration of its structureand its measurement using the Illinois Rape Myth Acceptance Scale. Journal of Research in Personality , 33,27�/68.
Pfeiffer, C. and Oswald, M. (Eds.) (1989). Strafzumessung [Sentencing] . Stuttgart: Enke.Reichert, A. (1999). Zusammenhang zwischen Ursachenzuschreibung fur Kriminalitat und Strafharte: Eine
empirische Untersuchung an einer Stichprobe juristischer Laien [Links between causal attributions ofcriminal behaviour and severity of punishment: an empirical investigation with lay persons]. Unpublisheddiploma thesis, Westfalische Wilhelms-Universitat, Munster.
Reichert, A. and Bilsky, W. (2001). Kriminalitat�/Ursachenzuschreibung und Strafharte: Eine Untersuchungaus der Sicht juristischer Laien [Crime�/attributions and severity of punishment: studying lay persons’perspective]. In W. Bilsky and C. Kahler (Eds.), Berufsfelder der Rechtspsychologie. Dokumentation der 9.Arbeitstagung der Fachgruppe Rechtspsychologie in der DGPs, 13�/15 September 2001, Munster, CD-ROM .Munster: Westfalische Wilhelms-Universitat.
Schuller, R. A. and Wall, A.-M. (1998). The effects of defendant and complainant intoxication on mockjurors’ judgments of sexual assault. Psychology of Women Quarterly , 22, 555�/573.
Schult, D. G. and Schneider, L. J. (1991). The role of sexual provocativeness, rape history, and observergender in perceptions of blame in sexual assault. Journal of Interpersonal Violence , 6, 94�/101.
Sczesny, S. and Krauel, K. (1996). Ergebnisse psychologischer Forschung zu Vergewaltigung und ihreImplikationen fur Gerichtsverfahren [Results from research on rape and their implications for legalprocedures]. Monatsschrift fur Kriminologie und Strafrechtsreform , 79, 338�/355.
Steffensmeier, D. and Hebert, C. (1999). Women and men policymakers: does the judge’s gender affect thesentencing of criminal defendants? Social Forces , 77, 1163�/1196.
Stormo, K. J., Lang, A. R. and Stritzke, W. G. K. (1997). Attributions about acquaintance rape: the role ofalcohol and individual differences. Journal of Applied Social Psychology , 27, 279�/305.
Szymanski, L. A., Devlin, A. S., Chrisler, J. C. and Vyse, S. A. (1993). Gender role and attitudes toward rapein male and female colleges students. Sex Roles , 29, 37�/57.
Teichert, T. (2000). Auswirkungen von Verfahrensalternativen bei der Erhebung von Praferenzurteilen[Effects of alternative procedures on the elicitation of preference judgments]. Marketing ZFP , 2, 145�/159.
Thurstone, L. L. (1927). The method of paired comparisons for social values. Journal of Abnormal and SocialPsychology , 21, 384�/400.
Wittink, D. R. and Cattin, P. (1981). Alternative estimation methods for conjoint analysis: a Monte Carlostudy. Journal of Marketing Research , 18, 101�/106.
Workman, J. E. and Freeburg, E. W. (1999). An examination of date rape, victim dress, and perceivervariables within the context of attribution theory. Sex Roles , 41, 261�/277.
SEVERITY OF PUNISHMENT 219
Copyright of Psychology, Crime & Law is the property of Routledge and its content may not be copied or
emailed to multiple sites or posted to a listserv without the copyright holder's express written permission.
However, users may print, download, or email articles for individual use.