Critical Predictors of Homicide in
the International System
By: Oliver W. Bengle
and
Dr. Jack Vincent, UI
Introduction
This research paper attempts to create a probabilistic analysis of nation's homicide rates, using selected variables, such as GNP/per capita, and various recorded crimes such as robberies and frauds. That is, we intend to evaluate recorded nation-state's attributes, including crimes, economic, social, religious, government and various other attributes of states as they relate to homicide.
In this connection, Michael D. Maltz argues “Homicide is a crime in and of itself; but in another sense it is the fatal outcome of… many different crimes” (Maltz 1998, p.1490). For example, armed robbery, if it includes killing is also recorded as homicide. Thus, we are attempting to analyze the linkage of not only standard attribute variables, such as population size, but also to evaluate the linkage of homicide to other measures of criminal behavior.
This research evaluates the accuracy of a theory (based on clear theoretical assumptions) as to which variables appear to be critically linked to homicide rates within nation-states. In this connection, we reviewed literature relating to homicide as well as statistics relating to homicide rates. From this review it was determined that continuing research in this area could prove to be valuable by including additional variables and their corresponding assumptions which previously literature omitted
It is assumed that the prevention of crime, especially homicide, is very important for quality of life within states. Hopefully, by furthering the understanding of the possible causative factors relating to homicide, world states might utilize our findings to enact new policies to possibly minimize homicide rates.
Data Sources
Much of our data came from the Martin Archives (located at the University of Idaho). Crime data was compiled, in part, from United Nations Crime and Justice Information Network Database (UNCJIN). The UNCJIN was formed in 1989 and has been supported by the Bureau of Justice Statistics since 1990. They have been aided by the Center for International Crime Prevention, the State University of New York at Albany and the Research Foundation of the State University of New York. The UNCJIN database is located at the University of Vienna and is hosted and supported by the Institute of Applied Computer Science and Information Systems (UNCJIN, 1999). The source of the attribute data is fully explained by documentation found in the Martin Archives.
Partial explanations of criminal behavior
Some of the literature exploring the causes of crime cite subjective feelings such as "lack of happiness", or "frustration", as important attributes of people who commit crimes. The relationship of happiness and frustration is likely negatively correlated since low frustration may result in more happiness and high frustration may result in less happiness. As would be expected, much of the documented causes of frustration appear to be related to “poverty.” Evidence from David Lester shows that low GDP per capita is more likely to contribute to a high homicide rate when compared to GDP as a whole (Lester, 1996).
Connected to this possible "economic cause argument", two possible psychological justifications emerge as to why certain groups of people are more likely to commit crimes than others. One who commits such crimes might use the rationale of extreme class differences for justifying his/her actions. These extreme class differences occur in many societies. John Hagen suggests a societal class struggle of "have’s" against the "have not’s". For example, a variable such as the rate of unemployment, is a measuring stick separating “have” and “have-not” nations. He argues
“insofar as unemployment is a core component of class, and insofar as unemployment and crime form important causal components in the formation of life courses trajectories, there can be little doubt that the relationship between class and crime is a key element in criminological research” (Hagan 1992, p.8).
Another contributing factor of crime may relate to overall social degradation. Members of a society with a uniformly low GDP per capita may convince themselves that they have no other way to obtain the goods or services that they need or desireexcept by committing crimes. The general population may, in a sense, turn on itself in an attempt to provide for the basics of life. In addition, low GDP per capita as a widespread phenomenon in a society may also have a negative impact when cultural norms are first formed. That is, violence and other criminal activity may become an “acceptable norm” when few other options are left to an individual to promote their economic status. Fred E. Markowitz and Richard E. Felson appear to support this kind of interpretation in their article Social-Demographic Attitudes and Violence, when they argue: “Poor people are more likely to engage in violence than people of higher status because they are more punitive and because they are more concerned about showing courage in conflicts with others.” (Markowitz 1998, p.134).
High GDP and/or high GNP per capita may also have the positive benefit of contributing to better medical treatment. This could lead to a reduction in the number of deaths from criminally induced injuries. In this regard, Doerner points out “results indicate that medical resources do impinge on criminally induced lethality ... trauma studies show that prompt ambulance response reduces the time that the patient languishes in the field and lowers the probability of mortality by transporting the patient expeditiously to an appropriate treatment facility.” (Doerner 1988, p.1182).
It will be seen in the tables that follow that there are many additional explanations of why homicide occurs. Each variable that is investigated in this study lists these additional explanations along with their sources in the tables that follow.
Some Theoretical Concerns relating to “Cause”
Vincent argues that: "Social scientists, like 'hard science' scientists, wish to create theoretical models to 'explain' what they are researching (2002). As in the hard sciences, these models are open to revision as new tests and observations reveal deficiencies in a previous model's applications. A theoretical model, where all cases are expected to follow the model, is suggested to be a 'deterministic model.' Single case deviations from the model, then can create questions about that model's accuracy. In contrast, a theoretical model where most of the cases fit the model but some of the cases are expected to deviate from the model and do, is called a ‘probabilistic model.’ Social scientists, who use statistics to validate their models, normally assume their models are 'probabilistic.' This means that even if a considerable number of cases do not follow the model, it can still be considered a 'good' one, if most of the cases do." (Vincent, 2002)
It should be clear that a "probabilistic" approach is employed in this paper. In this connection, Vincent argues: "There is an important issue concerning 'cause' in the social sciences. Cause, however, is very difficult to deal with in both the hard sciences and the social sciences. In the later case, social scientists are seldom in a position to provide 'definitive tests' of presumed causal relationships.” (Vincent, 2002)
For example, in this paper, we will assert that population growth rate, as an annual percentage, may be a "critical" variable regarding nation’s homicide rates. We, of course, can test to see if the high homicide nations have mean population growth rates that are different than the low homicide nations. In this case, high homicide rate nations are predicted, by our theory, to have higher population growth rates than low homicide rate nations. In fact it should be clear that we are not really in a position to actually manipulate various nation's population annual growth rate percent to see if their homicide rates drop (in probabilistic terms) as their population’s annual growth rate percent go down. In this connection, if we observe, without manipulation, that changes in two variables, in fact, co-vary across time, we can not simply conclude one "causes" the other, since a third variable might be "causing" the changes observed in both. Our probabilistic theoretical model, then, may be viewed as a "plausible" (but not actually proven) explanation of what we are observing. The "policy recommendation equation" that comes later in the paper must be understood with these observations in mind.
The policy equation, presented later, is offered with the caveat these policy recommendations can be viewed as "reasonable" policy recommendations, since the observed mean differences are, in fact, consistent with our probabilistic theoretical model. Such “causal” relationships, then, are only "assumed," in regard to such mean differences, so that possible policy recommendations can be forthcoming. It should be noted that without such causal assumptions, the findings generated here could appear to be without social significance. That is, why bother to change anything if it is assumed that there are no causal connections between the variables under consideration in the research? Further, without making such causal assumptions, it then follows that a great deal of the probabilistic social science research, done to date, would wind up in the same dubious position of being viewed as disconnected to actual social policy for the same reasons. This is because, as noted above, most social scientists are not usually in a position to definitely "prove" possible posited causal connections related to their research, for the various reasons stated above.
Theoretical Model Building
Our theoretical model relating to possible causal connections) are derived from three sources:
1)Arguments developed in the literature.
2)Our own assumptions, and
3)Predictions from factor clusters.
The distinctions in this regard should become evident from the tables that follow.
Methodology
Standard statistical procedures were used in this paper, including factor analysis and discriminant function analysis. Vincent indicates:
Factor analysis refers to certain mathematical techniques, the purpose of which is to reduce a large number of indices (variables) into a smaller number of factors. Factors may be viewed as intervening variables that emerge from the pattern of intercorrelation among indices. Each indices ‘loads’ differently upon each other. These loadings or weights may be viewed as correlations of the index with the factors and may be utilized to generate ‘factor scores’ which express each subject's position on each factor. (Vincent, 2002).
This form of data analysis allowed us to develop variable clusters. That is, certain variables tend to exhibit strong associations with one another, forming a cluster, which is basically independent of other clusters. When one of the variables in the cluster is "explained" the remaining variables in the cluster are predicted to relate in the "same direction,” depending, of course, on the sign (+ or -) of the loading. This type of prediction is only done if the literature, or our own assumptions, does not suggest a contrary direction for the variable in question.
As for discriminant function analysis, it scores subjects through a linear combination of variables creating a composite score, so that the means of the groups, to which the individual nations belong, are maximally different on that composite variable. If p are the i predictors and w the i weights applied to those predictors, it is necessary to compute D using the following rules: D= or the summation index (called the discriminant function scores) are created by adding the results of multiplying all the weights times all predictors under the rule that the mean differences of the groups on the discriminant function scores are maximally different.
For example, if economically developed and economically underdeveloped states are scored on 10 predictor variables, these variables must be weighted in such a way that the means of the two groups, developed and underdeveloped, on D, are as different as possible. That is, md – mu, where md refers to the mean of the developed groups on D, and mu refers to the mean of the underdeveloped groups on D, must be as large as possible through the linear weighting procedures. These techniques are very useful when dealing with categories such as developed vs. underdeveloped, large vs. small, etc when it is desirable to examine a large number of variables in terms of their relative ability to discriminate between the categories used. (Vincent, 2002)
In tables 1 and 2 there are various categories of information built into the analyses. In table 1, the predictor variables and their mean differences and significance level are presented. For example, this table section from Table 1 indicates:
Variable(s) / MEAN #1 / MEAN#2 / SignificanceB_V1 gross national product per capita b_indicates total for l975 to l979 for b_v1 to b_v23 / 22594.18721 / 20590.0615 / 0.754
B_V2 population total / 311021216 / 45745662.72 / 0.047
Thus, we find that, descriptively, low homicide states (MEAN #1)had a somewhat higher gross national product per capita than high homicide states(MEAN #2). This mean difference, however, was not significant. For v1, GNP per capita from 1975-1979 there are seventy-five (75) chances out of one-hundred (100) that the difference was due to chance. In contrast there are less than five (5) chances in one-hundred (100) that the population difference v2, Population total is due to chance.
The second group of information one should be aware of in this table is the “code” column that indicates how we predicted the variable in question. Code 1 indicates we used a literary source, code 2 indicates that we predicted the variable by our own assumptions, while code 3 indicates that the variable was located in a unique variable cluster. There are six unique clusters used in this essay, which are derived from two separate factor cluster evaluations. One dealt specifically with those variables not dealing with crime, but all other variables (which we named Homicide), and the other dealt specifically with crime variables (which we named crime). These clusters are identified as factor cluster 1 (Homicide), factor cluster 2 (Homicide), factor cluster 3 (Homicide), factor cluster 1 (crime), factor cluster 2 (crime), and factor cluster 3 (crime). The other variables, using either code 1 or code 2, provided the basis for the direction of the predicted relationship based on the sign directions of the factor loadings. If the variable in question loaded in the same direction as the other predicted variables, the prediction for that variable, naturally, was in the same direction as for the other variables. If the variable in question loaded in the opposite direction of the other variables then the prediction, of course, was in the opposite direction.
The predicted and found columns indicate whether the prediction was accurate in descriptive terms, with “1” indicating a prediction of lower scores and “2” indicating a prediction of higher scores. The overall success of the predictions was determined by correlating the two columns (predicted and found).
The final entries (explained) provide the basis for the predictions. As noted, quotes and citations are labeled “Code 1,” our assumptions are labeled “Code 2” and a factor cluster prediction is labeled “Code 3.”
Basis for Code 2 predictions
In order to make predictions as to how our variables relate to national homicide rates, we devised a preliminary theoretical framework. This framework served as a foundation not only to predict but also to gain insight into what kinds of causal effects the variables might have. This original model was further expanded upon and revised upon examination of the actual findings. Our preliminary theory, as will be seen, is based on two main assumptions.
The first assumes the amount of homicide occurring within a nation is directly linked to the nation’s “social capital.” Social capital is defined in terms of “degree of interpersonal trust and the level of civic engagement” (Baumer 2001, p.2). In this connection, “trust” and “engagement” can be seen as mutually reinforcing and are expected to generate, over time, even more social capital. Peaceful teachings, a variable we shall incorporate in relation to social capital, should also be functional in this regard. That is, a society that advocates peaceful relations should be one more predisposed to have lower homicide rates. In contrast, a nation that has a number of prominent groups which are free to actively advocate, and practice, hatred and militant action without any restraints, will likely have a much higher homicide rate than nations not so inclined.
We, of course, cannot directly measure "social capital" or "peaceful teachings" but the variables we predicted to be associated with lower homicide rates are assumed to be functional in that regard. Variables evaluated in that regard include variables dealing with religion, participation in civic duties (e.g. voting), observation of current events through media sources, and fractionalization of ethnic linguistics.
The second main assumption is based on the concept of “frustration.” Frustration, within a society, although likely inversely related to “social capital,” may also be viewed as an independent possible causal factor in respect to homicide. Frustration may come as a result of both an individual’s “perceived” relative deprivation as well as “actual” deprivation. Frustration may relate to income inequality (which reduces the chances of an individual to achieve either social mobility or personal growth), or denial of basic human necessities. It may also relate from racial, religious, or political inequalities. Frustration, like social capitol, is a concept variable and can not be directly measured in this study. It is assumed to relate to variables that deal with populationdensity and growth, economic measures, and health and social welfare measures. For example,variables relating to crowding and competition for scarce resources can reasonably be seen as variables likely to be associated with higher degrees of frustration. Such frustration may then be the basis of higher homicide rates.It is also our assumption that, left with little opportunity to achieve social mobility or gain needed life sustaining goods, especially in a situation of inequality, violence may be perceived as a reasonable option in order to obtain mobility and overcoming other obstacles needed to sustain life. The embodiment of such acts of violence within a nation experiencing high frustration may have a lasting social psychological effect that creates a general acceptance among members of the nation to use such acts to solve disputes.