You are on page 1of 16

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/5822307

Correlates of the Victim-Offender Relationship in


Homicide

Article  in  International Journal of Offender Therapy and Comparative Criminology · December 2007


DOI: 10.1177/0306624X07308671 · Source: PubMed

CITATIONS READS

19 2,815

3 authors, including:

Liqun Cao Charles Hou


Ontario Tech University STB
102 PUBLICATIONS   3,765 CITATIONS    2 PUBLICATIONS   157 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Chinese drug problems View project

All content following this page was uploaded by Liqun Cao on 22 May 2014.

The user has requested enhancement of the downloaded file.


International Journal of
Offender Therapy and
Comparative Criminology
Volume XX Number X
Month XXXX xx-xx
Correlates of the Victim-Offender © 2007 Sage Publications
10.1177/0306624X07308671
Relationship in Homicide http://ijo.sagepub.com
hosted at
http://online.sagepub.com
Liqun Cao
Eastern Michigan University, Ypsilanti
Charles Hou
National Taipei University, Taiwan
Bu Huang
University of Washington, Seattle

Using a classification of homicides based on the victim-offender relationship, this


research analyzes individual-level data from a local prosecutor’s office in Taiwan with
multinomial logistic regression to locate the more precise correlates of three different
homicide relationship types. The results of the analyses provide further support for the
hypothesis that such partitioning of homicides is fruitful in revealing the relationships
otherwise obscured. They indicate that both sociodemographic variables and situational
variables are important correlates of three different homicide relationships, but their
strengths vary based on the particular homicide relationship type. Age and crime
premises correlate with homicide differently based on the victim-offender relationship.
Premeditation is related to acquaintance homicide but not to intimate homicide. In con-
trast, previous conviction is associated with intimate homicide but not with acquaintance
homicide. The implication of the findings is discussed within the limitation of the data.

Keywords: homicide in Taiwan; homicide classification; multinomial logistic regression;


marginal effects; victim-offender relationship

H omicide, without doubt, is the most heinous crime among the index crimes.
Theoretically, studies (Decker, 1993; Wolfgang, 1958) have suggested that
homicide may be qualitatively distinguished on the basis of the victim-offender rela-
tionship. Empirically, early examination of homicide has, for the most part, lumped
together different types of homicide, even though there is little reason to expect that
the patterns of the different types of homicide are necessarily the same. In the past
quarter of the century, many studies have disaggregated homicide along gender
(Browne & Williams, 1995) and racial lines (Messner & Golden, 1992) and along
the motivations of instrumental and expressive homicide (Miethe & Drass, 1999).

Authors’ Note: Please address correspondence to Liqun Cao, Department of Sociology, Anthropology,
and Criminology, Eastern Michigan University, Ypsilanti, MI 48197; e-mail: liqun.cao@emich.edu.

1
2 International Journal of Offender Therapy and Comparative Criminology

Far more commonly, studies have disaggregated different types of homicide according
to the victim-offender relationship (Kovandzic, Vieraitis, & Yeisley, 1998; Kubrin,
2003; Messner & Tardiff, 1985; Parker, 1989; Parker & Smith, 1979; Williams &
Flewelling, 1988). They have found that indeed homicide is not a homogeneous phe-
nomenon and that the relationships between victims and offenders reveal certain
structural covariates otherwise obscured. These analyses of disaggregated homicide
rates, however, were conducted only at the macrolevel.
Analyzing disaggregated homicide at the individual level is rare in the crimino-
logical literature. Built on previous studies in this area, Decker (1993) proposed a
classification of homicide according to the strength of the victim-offender relation-
ship. The five categories were homicide between strangers, between acquaintances,
between friends, between romantic links, and between relatives (not romantic). His
study showed the utility of this classification. The more specified models resulted in
a more detailed understanding of the correlates of each type of homicide. The clas-
sification based on the strength of the victim-offender relationship worked well with
the theoretical prediction of homicide along instrumental and expressive lines
(Block, 1981; Riedel, 1987). Instrumental crime involves an offender who seeks to
maximize gain and to minimize risk of apprehension, whereas expressive crime fails
to include the rationality of such cost-benefit consideration. Expressive crime is
more emotional and more spontaneous. Later, Decker (1996) proposed another clas-
sification scheme: deviant homicide versus nondeviant homicide. Deviant homicide
is so labeled because it happens (a) when the bonds or insularity provided by
primary relationships are broken to commit instrumental offenses and/or (b) when
expressive homicides result within secondary relationships.
Prior research, therefore, has devoted inadequate attention to the empirical study
of the social correlates of different relationship types of homicide at the individual
level. Given what is known about the covariates at the macrolevel, it may be hypoth-
esized that different victim-offender homicide relationships may have different cor-
relates at the microlevel too. The utility of finer distinctions in defining types of
homicide may enable us to distinguish conceptually and empirically social correlates
of different relationship types of homicide. The study of the microenvironment of
different homicides may advance our understanding of the nature of the phenome-
non (Block, 1981; Kubrin, 2003). The microenvironment refers to “the immediate
network of events and structures surrounding the crime, such as the relationship of
the victim and offender,” the crime premises, and the weapons used in the crime
(Block, 1981, p. 745). This study attempts to extend efforts in this regard in a mul-
tivariate investigation of the correlates of different relationship types of homicide.
The main goal is to examine how sociodemographic factors and situational factors
simultaneously affect the probabilities of different relationship types of homicide. In
so doing, it is expected to promote the utility of the multivariate logistical model in
doing criminological research.
Cao et al. / Victim-Offender Relationship 3

Literature Review

An understanding of homicide requires knowledge of the cultural, structural, and


situational components of social life. Homicide is not independent of social influ-
ence, as it is nested in and emerges out of the social activities of individuals. It is not
a homogeneous concept but may encompass multiple subcategories that respond dif-
ferently to various stimuli. The extant literature shows that various measures of
inequality are consistent predictors of various types of homicide (He, Cao, Wells, &
Maguire, 2003; Kovandzic et al., 1998; Kubrin, 2003; Miles-Doan, 1998; Parker,
1989; Williams & Flewelling, 1988). Beyond this correlation, the structural deter-
minants of homicide rates differ based on the types of homicide. Another relatively
persistent predictor of homicide in the American criminological literature is the sub-
culture of violence (Cao, 2004; Kovandzic et al., 1998; Kubrin, 2003; Parker, 1989;
Williams & Flewelling, 1988). Parker (1989) suggested that family homicide would
be more closely related to socioeconomic factors, whereas acquaintance homicides
would be influenced by subculture factors. Stranger homicide would be affected by
both economic and subcultural factors, depending on the circumstances. Neither the
effect of inequality nor the effect of a subculture of violence could be readily applic-
able to the explanation of homicide at the microlevel.
The extant literature also shows that men are more likely than women to commit
homicide, but there are differences in the gender breakdowns of victims and offend-
ers. Men are far more likely to be murdered by strangers or acquaintances, whereas
women are at a far greater risk of being victims of intimate partner murder (Gallup-
Black, 2005). Similarly, Decker’s (1993) study in St. Louis indicates that murder
suspects are predominantly male and that women are more likely to be involved in
homicides with romantic links. The regression analysis of Chicago homicide data
reveals that men are more likely both to commit deviant homicide and to be the
victims of deviant homicide (Varano & Cancino, 2001). The effect of age on differ-
ent relationship types of homicide at the macrolevel, on the other hand, is largely
insignificant (Messner & Golden, 1992; Miles-Doan, 1998; Parker, 1989) and incon-
sistent (Parker & Smith, 1979), and sometimes it is not controlled for (Williams &
Flewelling, 1988). In studying deviant homicide in Chicago, neither the age of
offender nor the age of victim is significant (Varano & Cancino, 2001). Educational
attainment seems to explain family and partner homicide better (Peterson & Krivo,
1993). We argue that gender, age, and education play different roles in the different
relationship types of homicide, and their effects on homicide vary based on the
victim-offender relationship.
Previous studies show that outcomes of a homicidal interaction are not predeter-
mined solely by the characteristics of the participants. They are also in part situa-
tionally determined, as routine activity theory suggests (Cohen & Felson, 1979).
Higher rates of gun ownership, for example, are found to be related to homicide rates
4 International Journal of Offender Therapy and Comparative Criminology

(Hoskin, 2001), and weapons used vary by the intensity of the victim-offender
relationship at the individual level. Because of the forethought involved with the
gun, its use doubles in homicide involved with friends, romantically linked persons,
and relatives (Decker, 1993). Non-gun weapons within one’s grasp, such as blunt
objects or bare hands, indicate more spontaneous actions.
The crime location is another important variable that may vary based on the homi-
cide relationship type. The routine activity approach (Messner & Tardiff, 1985) and
social disorganization theory (Sampson, 1987) suggest that homicides between
strangers are more likely to happen outside the home. Block (1981) finds that impul-
sive homicides among relatives are more likely to happen in the home. Decker’s (1993)
study indicates that the greater the intensity of the relationship (from strangers to rel-
atives), the more likely it is that a homicide takes place inside the home.
In summary, the utility of more specified models of homicide based on victim-
offender relationships has been demonstrated by research, but little quantitative
research has been directed at uncovering the social correlates of the different rela-
tionship types of homicide at the individual level. Finer distinctions in defining the
victim-offender relationship in homicide and in delineating their correlates may
increase our understanding of the likelihood that fatal interactions will occur.
Decker’s (1993) study is limited because the effects of independent variables are not
simultaneously assessed in a multivariate model. It is not clear, therefore, whether
some of the significant bicorrelations become insignificant in the multiple variant
analyses. It is also limited because data are from a single city in a single nation. It is
not clear whether the relationship types have significance beyond national borders
(Stack, Cao, & Adamczyk, 2007). With a data set from Taiwan, the current study
evaluates the effects of sociodemographic and situational factors on disaggregated
homicide based on the victim-offender relationship.
Building on Riedel (1987), this study uses a homicide classification with three
categories: homicides between strangers, homicides between acquaintances/friends,
and homicides between intimates (family members or romantic links). This is both
more consistent with the extant literature on the types of homicide (Kovandzic et al.,
1998; Kubrin, 2003; Messner & Tardiff, 1985; Williams & Flewelling, 1988) and
more easily defendable in achieving the goal of a conceptual scheme of exhaustive-
ness, exclusivity, and explanatory relevance. The classification of five homicide cat-
egories based on the victim-offender relationship is less defensible because even
Decker (1993) admitted that it is very difficult to separate friends from acquain-
tances, and many times such a separation is arbitrary. In addition, we propose that
the strengths of relationships with relatives and with romantic links are similar, and
their correlations should be similar too. It was thus decided to merge them into one
category. Decker’s (1996) later classification of three categories put friends in the
same category as romantic links and relatives. We argue that friends and acquain-
tances resemble each other more than they resemble intimate relationships like
romantic links and relatives, and thus friends and acquaintances share one category.
Cao et al. / Victim-Offender Relationship 5

In this study, stranger homicide refers to the situation in which there is an absence
of any prior relationship between the victim and the offender. Acquaintance/friend
homicide refers to the relationship that may exist between persons who recognize each
other for a number of years or have some shared history of interaction. For intimate
homicide, there are strong and long-term mutual bonds, including husbands and wives,
ex-husbands and ex-wives, current and previous romantic links, in-laws, and relatives.
Specifically, this research investigates the effects of age, gender, education, and
marriage on various relationship types of homicide. It is expected that age correlates
with homicide differently based on the victim-offender relationship. Younger men
are more likely to involve themselves in all types of homicide, but especially
stranger homicide, whereas older men are less likely to be involved with stranger
homicide. The effect of marriage is complex, and it is expected that married people
are less likely to be involved in stranger homicide but more likely to be involved in
acquaintance and family homicide. The variable income is not included in the analy-
sis because of missing values. Almost two thirds of the respondents marked their
incomes as unknown. As a result, this variable is not usable. Instead, the educational
achievement of homicide suspects is used as a proxy for socioeconomic status.
For the situational variables, it is expected that previous convictions and premedi-
tation reduce the probability that one kills another family member but may increase the
probability of acquaintance homicide and stranger homicide. Weapons are expected to
be used more frequently in acquaintance/friend homicide than in stranger homicide
or in intimate homicide; friend/acquaintance homicide and intimate homicide are
more likely to happen in homes or in cars than is stranger homicide, which is more
likely to take place outside, in public, such as on streets or at schools. Finally, the
effect of crime time is controlled for.
Data came from all records of homicide charges taking place between May 1994
and March 1998 in the Prosecutor’s Office in the District of Ban-ciao Court, Taiwan.
Taiwan is a newly democratic society, and the quality of crime data is quite accurate
because of the long tradition of record keeping (Cao & Dai, 2006). Homicide rates
in Taiwan were about 3 per 100,000 annually during the 1990s (Ministry of Justice,
2002). Because of the definitional differences, it is problematic to compare homicide
rates between different political entities (He et al., 2003). As a crude comparison,
homicide rates in the United States were down in the 1990s compared with those in
the 1980s but remained above 6 per 100,000 annually (Federal Bureau of
Investigation, 1998). In Hong Kong, the homicide rate was about 1.2 per 100,000
between 1989 and 1997 (Broadhurst, 2004). In South Korea, the homicide rate was
1.6 per 100,000 in 1994, and in Japan, it was below 1 (Finch, 2001). In China, homi-
cide rates were below 3 per 100,000 in the entire 1990s according to the Law
Yearbook of China, 1987-1998 (1999).
In summary, this study attempts to extend routine activity theory in another cul-
ture and to test the hypothesis that different relationship types of homicide have cor-
relates that vary in terms of offenders’ demographic and social circumstances.
6 International Journal of Offender Therapy and Comparative Criminology

Research Strategy

Because the outcome variable of interest involves three mutually exclusive cate-
gories, multinomial logistic regression is used as our analytic procedure (Cao, Cullen,
& Link, 1997; Greene, 1993). In this procedure, estimation of the probability of
belonging to one category of homicide in comparison with the probability of belong-
ing to the other categories is done simultaneously. The procedure uses all the infor-
mation simultaneously, thereby producing the most efficient estimates with the
integration of a number of choice probabilities within an efficient closed form. The
estimation of the multinomial logit model imposes the condition that the total of var-
ious choice probabilities must sum to 1. The interpretation of the logit coefficients,
however, is not straightforward. In analyzing the dependent variable with two out-
comes, it is acceptable to interpret the results directly from regression coefficients or
in odds ratios: with only 1 degree of freedom left, the variation in the ratio of two
choice probabilities reflects the changes in the probabilities per se. Applying the same
approach for the multinomial logit model, however, is implausible and can be mis-
leading. With the presence of more than two alternatives, the regression coefficients
may not necessarily bear any relationship to the changes in the probabilities (Greene,
1993). Because the probabilities for all choices always sum to unit, the simultaneous
structural variation in the proportional distribution of more than two alternatives is
beyond what the changes in a set of log odds can capture. That is why an additional
procedure is used to facilitate the understanding of the coefficients of the multinomial
logistic regression. The marginal effects of each independent variable on each choice
of the dependent variable are calculated. These effects not only provide more reliable
signs for the effects of each independent variable on the dependent variable but also
reveal the magnitude of each independent variable on the dependent variable.

Method

Data
Data were collected from all records of homicide taking place between May 1994
and March 1998 in the Prosecutor’s Office in the District of Ban-ciao Court, Taiwan.
Altogether there are 308 homicide cases during the period. Taiwan’s official name is
the Republic of China, and it has been a de facto political entity since 1949. The island
country has a population of approximately 23 million, and its capital city is Taipei (Cao
& Dai, 2006). The District of Ban-ciao is part of the greater Taipei metropolitan area.
Within its jurisdiction are 7 cities and 6 towns or villages with a total population of
approximately 3 million in 1990. About 80% of the population in Taipei County lives
in this district, and there are many immigrants from central and southern parts of
Taiwan. In general, the homicide rate in Ban-ciao is slightly higher than the homicide
rates in other cities in Taiwan. It is approximately 3 per 100,000 residents.
Cao et al. / Victim-Offender Relationship 7

All the variable coding is based on the descriptive data in these cases. Compared
with the police data, the advantage of the prosecutor’s data is that the category of
unknown deaths is largely eliminated. All cases in the prosecutor’s office are con-
sidered cleared by the police. The overall quality of data from Taiwan is as good as
the quality of data in the United States and Britain. It is better than many cities of
the United States.

Dependent Variable
Homicide refers to the legal definition in the criminal laws of the Republic of
China. It includes attempted murder, planned murder, and completed murder. It does
not include homicide from negligence. This definition is slightly different from the
homicide definition used by the FBI, in which murder and non-negligent manslaugh-
ter are defined as “willful (nonnegligent) killing of one human being by another”
(Federal Bureau of Investigation, 1998, p. 15). Attempts to kill and assaults to kill
are excluded in the homicide definition in the United States,1 whereas they are
included in the homicide definition in Taiwan. All case files are carefully reviewed,
and the homicide relationship types are coded into a three-category classification
based on the victim-offender relationship: homicide between strangers, between
acquaintances/friends, and between intimates (family members or romantic links).
Stranger relationship serves as the reference group in all comparisons.

Correlates of Homicide
There are two groups of correlates in this study: sociodemographic correlates and
situational correlates. The sociodemographic correlates include age, education, gender,
and marriage.2 Age of the homicide suspect is measured as the respondent’s actual age
in years at the time of the crime. Education is an ordinal variable where elementary
education =1, middle school = 2, high school = 3, and some college and above = 4.
Gender is a dummy variable, with male coded 1 and female as the reference group.
Marriage is a binary variable, with those who are married or widowed coded as 1 and
all other types of family structure or nonstructure as the reference group.
The situational correlates include previous convictions, premeditation, weapons,
crime premises, and crime time. Previous conviction is coded as the number of convic-
tions each person had prior to the current arrest. It varies from 0 to 14 times.
Premeditation is a dummy variable, with those who have been identified with premed-
itation coded as 1 and those who have not coded as the reference group. Weapon is a
three-category variable: (a) no weapon or no sharp weapons, (b) guns or fire, and (c)
knives. There are 21 cases (6.8%) where fire is used as a weapon and 34 cases (10.4%)
where a gun is used as a weapon. Because of the lethal nature of both, we decided to
merge the two. The category of knives serves as the reference group. The crime
premises variable is a dummy variable that codes outside homes (in school, public, or
streets) as 1 and other (home or car) as the reference group. Finally, crime time is a
8 International Journal of Offender Therapy and Comparative Criminology

binary variable where daytime (7 a.m. to 6 p.m.) is coded as 1, and nighttime is coded
as the reference group (7 p.m. to 6 a.m.). The SPSS computer software program is used
for all the analyses of data, but the marginal effects are calculated with Excel.

Results

Table 1 presents the frequency distribution of all variables in our sample. For the
continuous variables, means are reported. Almost half of all homicide suspects
(48.4%) are in the category of killing acquaintances and friends. This is quite simi-
lar to Riedel’s (1987) data in the United States from 1976 to 1985, where acquain-
tance/friend homicide rates varied between 35 to 54 per 100,000. Nearly 20% of the
suspects (19.8 percent) are in the category of intimate homicide and the rest (31.8%)
are homicides between strangers.
For the sample demographics, the homicide suspects are predominantly male
(93%), less well educated (86% did not receive education beyond middle school),
unmarried (83%), and relatively young (31 years old). The majority of suspects have
at least one previous conviction (1.28 times), almost half of the suspected planned their
actions (46%), knives are the most frequently used weapons in homicides (67%), and
most homicides take place outside homes and cars (65%) and during the night (71%).
It is interesting to note that the percentage of guns used in homicides (16%) is much
lower in Taiwan than those reported in the United States. Decker (1993) reported that
St. Louis’s above-national-average rate of gun use was 70% and guns were used in a
majority of homicides in every category of victim-offender relationship.
Table 2 presents the results of multinomial logistic regression that includes all
independent variables. The second column of the table reports the logit coefficients
of acquaintance homicide versus stranger homicide; the third column reports the
logit coefficients of intimate (family members or romantic links) homicide versus
stranger homicide.
A central issue in previous research is whether social correlates of these different
types of homicide are the same or different. The data in Table 2 suggest that both
views have a measure of credence. Age is positively related to both relationship
types of homicide, meaning older murder suspects are significantly more likely to be
involved in homicides to kill acquaintances and to kill intimate partners (family
members or romantic links) than in homicides to kill strangers. Married murder sus-
pects are more likely to kill their intimate partners than they are to kill strangers,
whereas marriage makes no significant difference between acquaintance homicide
and stranger homicide. Comparing homicides between strangers and intimates, mur-
der suspects with previous convictions are more likely to kill strangers than to kill
family members, whereas previous convictions do not affect homicide between
acquaintances. Premeditation is more likely to happen in acquaintance/friend homi-
cide than in stranger homicide, whereas intimate homicide and stranger homicide
Cao et al. / Victim-Offender Relationship 9

Table 1
Descriptive Sample Characteristics
Characteristics Percentage or Mean Number of Cases

Dependent variable
Victim-offender relationship
Acquaintance/friend 48.4 149
Intimate 19.8 61
Stranger 31.8 98
Independent Variables
Age (M) 30.6 years 308
Education
Elementary school 23.4 72
Middle school 62.7 193
High school 12.0 37
College 1.9 6
Gender (male = 1) 93.3 288
Marriage (yes = 1) 17.5 54
Previous convictions (M) 1.28 times
Premeditation (yes = 1) 46.1 142
Weapons
No weapon/no sharp weapon 17.2 53
Gun or fire 15.9 49
Knife 66.9 206
Crime premises (outside = 1) 64.6 203
Time (day = 1) 28.6 88

Note: Due to rounding errors, some percentages may not total to 100.

seem equally spontaneous. Homicides between acquaintances and between intimates


are more likely to take place in homes or in cars, whereas stranger homicide is more
likely to occur outside homes and in public domains. The other independent
variables—education, gender, guns, knives, crime time—are not statistically signif-
icantly related to the types of homicide. Only about 7% of murder suspects are
women, and 15% of all murder cases are gun or fire related.
These coefficient results, however, need to be interpreted with caution. Greene
(1993) warned that the coefficients of multinomial logistic regression might not nec-
essarily bear relationship with the changes in the probabilities because the simulta-
neous structural variation in the proportional distribution of more than two
alternatives was beyond what a set of log odds could capture. The additional proce-
dure of calculating the marginal effects was thus necessary to facilitate the correct
understanding of the coefficients of the multinomial logistic regression.
Table 3 presents the reduced multinomial logit model for victim-offender homi-
cide with only significant predictors. All the statistically significant variables in Table
2 are reassessed in the reduced model, and they all remain statistically significant in
10 International Journal of Offender Therapy and Comparative Criminology

Table 2
Multinomial Logit Model of Victim-Offender Relationships
(logit coefficients reported)
Acquaintance Homicide Intimate Homicide

Independent Variable ln (p1/p3) ln (p2/p3)


Age 0.07* 0.08*
Education –0.03 0.28
Gender (male = 1) 1.35 –0.78
Marriage (yes = 1) 0.15 1.45*
Previous convictions –0.11 –0.40*
Premeditation (yes = 1) 1.24* 0.07
No weapon (yes = 1) 0.22 0.27
Gun (yes = 1) –0.23 0.77
Crime premises (outside = 1) –2.05* –4.50*
Time (day = 1) 0.36 –0.06
Intercept –1.62 –0.43
–2 log likelihood ratio 416.677

Note: Number of cases is 308. p1 = probability of acquaintance homicide; p2 = probability of intimate


homicide; p3 = probability of stranger homicide.
*p ≤ .05.

the final model. In addition, the marginal effects of each predictor are calculated
with the help of Excel. These marginal effects tell us (a) the predicted probabilities
of belonging to one of the categories in the dependent variable and (b) the predicted
changes in probabilities resulting from changes in independent variables (Petersen,
1985). The second and third columns of Table 3 report the logit coefficients of the
reduced models, and the next three columns represent the marginal effects.
From reading the sample estimates, we learn that the estimated probability for homi-
cide taking place between acquaintances is 65%, between family members is 11%, and
between strangers is 25% with all values of independent variables fixed at the sample
means. The sum of these probabilities should equal 1 (.6446 + .105 + .2505 = 1).
As indicated in Table 2, age affects acquaintance homicide and intimate homicide
positively. From reading the marginal effects, it is shown that as a killer ages 1 year,
the probability that the victim is an acquaintance increases by 1%, the probability
that the victim is an intimate increases by 0.2%, but the probability that the victim
is a stranger decreases by 1%. All these effects are statistically significant. Although
these numbers may look small, we should remember that age is a continuous vari-
able in our sample that varies from 14 to 83. If we increase or decrease age by 10
years, the effects will look much stronger. The sum of these probability terms should
equal zero (.0105 + .0019 + [–.0124] = 0).
Married people who commit murder are 17% more likely to kill another family
member, but they are also 6% less likely to kill strangers. Note also that the former effect
is much stronger than the latter effect. For those who have previous convictions, each
Cao et al. / Victim-Offender Relationship 11

Table 3
Reduced Multinomial Logit Model of Victim-Offender
Relationships (coefficients and probability reported)
Estimated Logit Coefficients

Acquaintance Intimate
Homicide Homicide Effects on Probability

Independent Variables ln (p1/p3) ln (p2/p3) p1 p2 p3

Age 0.067* 0.069* 0.0105 0.0019 –0.0124


Marriage (yes = 1) 0.107 1.357* –0.1029 0.1659 –0.0630
Previous convictions –0.089 –0.459* 0.2896 –0.0912 –0.1984
Premeditation (yes = 1) 1.244* –0.001 0.2742 –0.0810 –0.1931
Crime premises (outside = 1) –1.999* –4.396* –0.0027 –0.3675 0.3702
Intercepts –0.302 0.170
Sample estimatea 0.6446 0.1050 0.2505
–2 log likelihood ratio 437.571
Sample size 298

Note: Number of cases is 308. p1 = probability of acquaintance homicide; p2 = probability of intimate


homicide; p3 = probability of stranger homicide.
a. Sample estimate refers to the estimated probabilities for those who belong to acquaintance homicide, to inti-
mate homicide, and to stranger homicide, with the values of independent variables fixed at the sample means.
*p ≤ .05.

increase in previous conviction decreases the probability that a murder suspect will kill
an intimate partner or family member by 9% and the probability that he or she will kill
a stranger by almost 20%. For premeditation, those with premeditation are 27% more
likely to kill an acquaintance/friend and 19% less likely to kill a stranger. Finally, being
outside one’s home reduces the probability of acquaintance homicide only by 0.3% and
the probability of intimate homicide by almost 37% but increases the probability of
stranger homicides by 37%. Although all of these effects are statistically significant, the
effects of intimate homicide and stranger homicide are much more strongly associated
with the crime premises than homicide between acquaintances is.

Discussion

The purpose of this research is to demonstrate more accurate correlates of the dif-
ferent relationship types of homicide and to apply routine activity theory to explain
homicide in another culture. All homicides are events surrounded by a unique com-
bination of various microenvironment factors. With the help of marginal effects, the
results of multilogit analyses are more easily interpretable than logit coefficients
themselves. As expected, the partitioning of homicides into victim-offender types
has revealed relationships otherwise obscured (Block, 1981; Decker, 1993; Kubrin,
12 International Journal of Offender Therapy and Comparative Criminology

2003; Parker, 1989). The analysis suggests that sociodemographic variables and sit-
uational variables play different roles in the probability of a particular relationship
type of homicide. Age and marriage are found to differentially correlate with each
relationship type of homicide. With increased age, one is significantly more likely to
be involved in acquaintance homicide and family homicide but also is significantly
less likely to be involved in stranger homicide. Marriage reduces the probability of
homicide between strangers but greatly increases the probability of intimate homi-
cide. Both findings are consistent with the routine activity approach’s expectation
(Messner & Tardiff, 1985) and with the argument of social disorganization theory
(Sampson, 1987). The effect of marriage, however, should be interpreted with cau-
tion because of the large missing values of this variable.
For the situational variables, the number of previous convictions decreases the
probability of intimate homicide. It is possible that people who know each other well
are more careful in avoiding certain explosive situations, whereas strangers tend to
deal with a situation based on the general principle (i.e., unprepared for the unusual
violent tendency of this particular person), which may result in their murders. Those
with premeditation are more likely to kill their acquaintances/friends than to kill
strangers. Consistent with Decker’s (1993) finding, acquaintance/friend homicide
often has an instrumental motive, which is indicated by the increased probability of
premeditation in acquaintance homicide. Finally, it is interesting to notice that crime
premises play such an important role in the different relationship types of homicides:
outside locations reduce the probability of killing between acquaintances/friends and
between intimates (family members or romantic links) but increase the probability
of killing between strangers. This finding is consistent with Decker’s (1993) bivari-
ate analyses. It confirms that place matters in criminal events (Gallup-Black, 2005)
and that it also matters in the type of homicide based on the victim-offender rela-
tionship and in a different culture. Note, further, that the magnitude of this correlate
is much stronger for homicide between intimates (family members/romantic links:
–36.75%) and for homicide between strangers (37.02%) than for homicide between
acquaintances/friends (0.27%).
The results of this study must be regarded as tentative for several reasons. First,
it must be remembered that one is not guilty until proven by the court. People in our
analysis are only charged with homicide. Whether they are guilty as charged remains
to be decided later by the court. Second, some salient variables are not included in
our statistical model. For example, we use education as a proxy for one’s socioeco-
nomic status, but it would be better if we could measure the concept more ade-
quately. Third, our data are from Taiwan, although the theory we rely on was
developed in the United States. Some culture-specific variables are either not signif-
icant or are not included in the current study. Homicide, as a complex and heinous
event, is nested in one’s broader cultural milieu, sometimes beyond an individual’s
immediate microenvironments. Variables that do not relate to homicide differentially
in one society do not mean that they do not relate to it differentially in another. For
Cao et al. / Victim-Offender Relationship 13

example, guns are much more frequently used in homicide in the United States
(Decker, 1993) but are infrequently used in our sample in Taiwan. Fire is rarely used
as a weapon in homicide in the West but is an important weapon in Taiwan. The
insignificant effects of guns or fire in our analysis may partially reflect the infrequent
use of guns in Taiwan, where gun ownership is minimum in the population. Another
prominent variable in the study of U.S. homicide is the role of illegal drugs (Decker,
1996; Varano & Cancino, 2001). In our sample, there are only 10 cases in which the
offenders admitted the use of illegal drugs. Our analyses did not include this variable
because of its small percentage (less than 3.2%). Finally, we want to emphasize that
although multivariate analysis implies causation, we use it only for correlation. We
simply evaluate homicide’s social correlates and argue that these conditions are nec-
essary but may not be sufficient for a homicide to happen.
These limitations aside, this study has advanced the notion that homicide is not a
homogeneous phenomenon, and its relationship with other variables along a number
of sociodemographic and microenvironmental dimensions varies based on different
victim-offender relationship types. Indeed, qualitative distinctions in homicides can be
captured quantitatively with appropriate techniques of analysis, and they are important
in studies attempting to evaluate both the magnitude of social correlates and their
direction. Future studies should replicate the analysis with U.S. data and should com-
pare the current results with those found in the United States to see the extent of sim-
ilarity and difference in cross-national research on the social correlates of homicide.

Notes
1. Murder and non-negligent manslaughter refer to the willful (non-negligent) killing of one human
being by another. Deaths caused by negligence, attempts to kill, assaults to kill, suicides, and accidental
deaths are excluded. Justifiable homicide, such as the killing of a felon by a peace officer in the line of
duty or the killing of a felon by a private citizen, is excluded, but a killing is not counted as justifiable or
excusable solely on the basis of self-defense or the action of a coroner, prosecutor, grand jury, or court.
Traffic fatalities are excluded (Federal Bureau of Investigation, 1998, adapted from pp. 15-64).
2. Education and marriage both have a large portion of missing data. We tested the possibility of
whether those who do not report their education and/or their marital statuses are different from those who
reported. The results indicate that those who reported their education are not different from those who did
not, but those who did not report their marital statuses differ significantly from those who reported their
marital statuses. As such, readers should be cautious in interpreting the effects of marriage on various
types of homicide.

References
Block, R. (1981). Victim-offender dynamics in violent crime. Journal of Criminal Law and Criminology,
72, 743-761.
Broadhurst, R. (2004). Crime trends in Hong Kong: Another look at the safe city. In R. G. Broadhoust
(Ed.), Crime and its control in the People’s Republic of China (pp. 133-148). Hong Kong: Central
Printing Press.
14 International Journal of Offender Therapy and Comparative Criminology

Browne, A., & Williams, K. R. (1995). Exploring the effect of resource availability and the likelihood of
female-perpetrated homicides. Law and Society Review, 23, 75-94.
Cao, L. (2004). Major criminological theories: Concepts and measurement. Belmont, CA: Wadsworth.
Cao, L., Cullen, F. T., & Link, B. G. (1997). The social determinants of gun ownership: Self-protection
in an urban environment. Criminology, 35(4), 629-657.
Cao, L., & Dai, M. (2006). Confidence in the police: Where does Taiwan rank in the world? Asian Journal
of Criminology, 1, 71-84.
Cohen, L. E., & Felson, M. (1979). Social change and crime rate trends: A routine activity approach.
American Sociological Review, 44, 588-608.
Decker, S. H. (1993). Exploring victim-offender relationships in homicide: The role of individual and
event characteristics. Justice Quarterly, 10, 585-612.
Decker, S. H. (1996). Deviant homicide: A new look at the role of motives and victim-offender relation-
ships. Homicide Studies, 33, 427-449.
Federal Bureau of Investigation. (1998). Uniform crime reporting handbook. Washington, DC: U. S.
Department of Justice.
Finch, A. (2001). Homicide in contemporary Japan. British Journal of Criminology, 41, 219-235.
Gallup-Black, A. (2005). Twenty years of rural and urban trends in family and intimate partner homicide.
Homicide Studies, 9, 149-173.
Greene, W. H. (1993). Econometric analysis. Englewood Cliffs, NJ: Prentice Hall.
He, N., Cao, L., Wells, W., & Maguire, E. R. (2003). Forces of production and direction: A test of an
expanded model of suicide and homicide. Homicide Studies, 7, 36-57.
Hoskin, A. W. (2001). Armed Americans: The impact of firearm availability on national homicide rates.
Justice Quarterly, 18, 569-592.
Kovandzic, T. V., Vieraitis, L. M., & Yeisley, M. R. (1998). The structural covariates of urban homicide:
Reassessing the impact of income inequality and poverty in the post-Reagan era. Criminology, 36,
569-599.
Kubrin, C. E. (2003). Structural covariates of homicide rates: Does type of homicide matter? Journal of
Research in Crime and Delinquency, 40, 139-170.
Law Yearbook of China, 1987-1998. (1999). Beijing: Press of Law Yearbook of China.
Messner, S. F., & Golden, R. M. (1992). Racial inequality and racially disaggregated homicide rates.
Criminology, 30, 421-445.
Messner, S. F., & Tardiff, K. (1985). The social ecology of urban homicide. Criminology, 23, 241-267.
Miethe, T. D., & Drass, K A. (1999). Exploring the social context of instrumental and expressive homicides:
An application of qualitative comparative analysis. Journal of Quantitative Criminology, 15, 1-21.
Miles-Doan, R. (1998). Violence between spouses and intimates: Does the neighborhood context matter?
Social Forces, 77, 623-645.
Ministry of Justice. (2002). Crime statistics and analysis. Taipei, Taiwan: Author.
Parker, R. N. (1989). Poverty, subculture of violence, and type of homicide. Social Forces, 67, 983-1007.
Parker, R. N., & Smith, M. D. (1979). Deterrence, poverty and type of homicide. American Journal of
Sociology, 85, 614-624.
Petersen, T. (1985). A comment on presenting results from logit and probit models. American
Sociological Review, 50, 130-131.
Peterson, R. D., & Krivo, L. J. (1993). Racial segregation and Black urban homicide. Social Forces, 71,
1001-1026.
Riedel, M. (1987). Symposium: Stranger violence: Perspectives, issues and problems. Journal of
Criminal Law and Criminology, 78, 223-258.
Sampson, R. J. (1987). Personal violence by strangers: An extension and test of the opportunity model of
predatory victimization. Journal of Criminal Law and Criminology, 78, 327-356.
Stack, S., Cao, L., & Adamczyk, A. (2007). Crime volume and law and order culture. Justice Quarterly,
24, 291-308.
Cao et al. / Victim-Offender Relationship 15

Varano, S., & Cancino, J. M. (2001). An empirical analysis of deviant homicides in Chicago. Homicide
Studies, 5, 5-29.
Williams, K., & Flewelling, R. (1988). The social production of criminal homicide: A comparative study
of disaggregated rates in American cities. American Sociological Review, 53, 421-431.
Wolfgang, M. (1958). Patterns in criminal homicide. Philadelphia: University of Pennsylvania Press.

View publication stats

You might also like