Abstract
Building upon and expanding the previous research into structural determinants of homicide, particularly the work of Land, McCall, and Cohen (1990), the current paper introduces a multilevel theoretical framework that outlines the influences of three major structural forces on homicidal violence. The Big Three are poverty/low education, racial composition, and the disruption of family structure. These three factors exert their effects on violence at the following levels: neighborhood/community level, family/social interpersonal level, and individual level. It is shown algebraically how individual-level and aggregate-level effects contribute to the size of regression coefficients in aggregate-level analyses. In the empirical part of the study, the presented theoretical model is tested using county-level data to estimate separate effects of each of the Big Three factors on homicide at two time periods: 1950–1960 and 1995–2005 (chosen to be as far removed from one another as the availability of data allows). All major variables typically used in homicide research are included as statistical controls. The results of analyses show that the effects of the three major structural forces—poverty/low education, race, and divorce rates—on homicide rates in US counties are remarkably strong. Moreover, the effect sizes of each of the Big Three are found to be identical for both time periods despite profound changes in the economic and social situation in the United States over the past half-century. This remarkable stability in the effect sizes implies the stability of homicidal violence in response to certain structural conditions.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
On the issue of significant differences between races in family situation dynamics, Cavan’s (1959) comparison of white and black institutionalized delinquents provides a good illustration:
Many boys had lived in several types of families during their lives (e.g., with both parents at one time, with a parent and step-parent at another, with relatives at some other time, and so on). Negro boys had a median of 2.1 different family situations, whites a median of 1.0. Half of the whites but only a fifth of the Negroes had continuously lived in the same family situation; at the other extreme, 17 per cent of the Negro boys but only 5 per cent of the white boys had lived in four or more different types of families. Moreover, almost four times as many Negro boys as white (37 and 10 per cent) had lived at some point with relatives and 18 and 4 per cent respectively had at some time resided in the homes of people to whom they were not related. (Cavan 1959, pp. 235–236).
Pettit and Western (2004) estimated that among black men born between 1965 and 1969, 20% had served time in prison by their early thirties (compared to only 3% of white men). This number—already unbelievably high—further climbs to 30% for black men without college education, and to 60% for high school dropouts among blacks.
The spatial-effects coefficient failed to reach statistical significance under the traditional p < .05 level of significance. The results of analyses are available upon request from the author.
MA and NYC exceptions for the historical dataset: for counties in the state of Massachusetts, homicide counts were available for 6 years out of 11 (1950–1952, 1955, 1959–1960); and for New York City counties (boroughs), separate homicide counts were available for 8 years out of 11 (1950, 1954–1960). No imputation procedures were used. Homicide rates were calculated using the available number of years in the denominator.
To form the index of poverty, the variables for percent of families in poverty and percent “uneducated” were summed up in their original metric (non-standardized). This has been done for several reasons: (1) to preserve the absolute zero of the original scales, (2) to let the original distributions determine which of the two variables contributes more to the index of poverty, depending on the time period; (3) for easier interpretability of results in further analyses (slopes for logged variables in OLS regressions are interpreted as a percent change impact on the dependent variable).
Ideally, “percent children in single-parent families” needs to be included along with the divorce rate to form the index of family structure. However, this measure is available for the 1995–2005 dataset but not for the 1950–1960 dataset. Thus, only divorce rate remained as a measure of family disruption in the model.
“Percent non-white” from 1950 to 1960 census data is the best approximation of “percent black” in the 2000 census data. Unfortunately, the exact same measure—percent black—was not available in the earlier census data but considering that the US had very few residents representing other racial groups than blacks and whites, “percent non-white” can be taken as a valid measure of percent black in 1950 and 1960. In addition, one of the anonymous reviewers suggested restricting analyses to counties with 1,000 or more black residents to test the sensitivity of estimates to the size of black population in the counties. The data were re-analyzed with these restrictions in place (resulting in a loss of slightly over 10% of the sample) and very similar estimates were obtained for all variables of interest.
To accommodate some recent sentiments in research literature that the West is now becoming what the South used to be in terms of the rate of violence (see Parker and Pruitt 2000), and also to measure regional peculiarities in a more comprehensive manner, regions are designated by dummy variables using traditional census divisions: South, West, and Midwest, with Northeast being the reference category.
The percent of families in poverty is calculated using the threshold of family income “less than $2000” in 1950, “less than $3000” in 1959 (for 1960), and “below poverty level in 1999” (for 2000). The official poverty level is determined by the family size and does not vary geographically (though it is regularly updated for inflation).
Low education is measured as the percentage of persons 25 years and older with “less than 5 years of education” for 1950, “less than 4 years of education” for 1960, and “less than high school education” for 2000. These thresholds reflect the availability of data (for the earlier periods of time) as well as changes occurring over time in the importance of education for increasing the prospects of well-being for individuals and families.
In fact, one of the anonymous reviewers suggested using “less than 8 years of schooling” as a measure of low education for year 2000. So, this measure was obtained from the 2000 Census data (“percentage of people 25 and older with educational attainment of less than 9th grade” since “less than 8th grade” is not available). After the new measure of low education was incorporated into the poverty index and the models re-estimated, very similar results were obtained. Moreover, the regression coefficient for the poverty index was 0.41, which is even closer to the regression coefficient of 0.43 obtained for the 1950–1960 dataset.
All variables had one added to all their values before the log transformation was applied (to eliminate zero or near-zero values since logarithm of zero does not exist). As a result of the log transformation, frequency distributions for logged variables were much closer to the shape of the normal distribution compared to the original variables and their relationships with the dependent variable straightened out as well. The only variable whose frequency distribution was not significantly improved by the logarithmic transformation was the dependent variable—homicide rate. Because of the heavy clustering of values around zero, there is no readily available way to transform this variable to bring it closer to the shape of the normal distribution. However, log transformation still seems beneficial in this case because it draws the long right tale of the original homicide rate distribution closer to the more frequent values and thus, gets rid of potential outliers.
As an additional check, the residuals in all OLS regression analyses were examined for possible heteroscedasticity that may have resulted from excess zeros in the dependent variable. Upon examination, the presence of heteroscedasticity was ruled out.
In addition, all regression analyses were re-estimated to test the robustness of the results against possible effects of the partialling fallacy as some control variables (for example, unemployment) were more highly correlated with some of the main predictors than with the outcome variable. The exclusion of such control variables from the model did not change the pattern of results and thus confirmed the stability of estimates under various conditions and specifications of the model. As noted by anonymous reviewers, the changing effects of unemployment on homicide rates, as well as the shifts in regional patterns of homicide, are worth exploring further but this topic is important enough to become a subject of a separate study. Thus it is not discussed in this paper to avoid diverting the narrative from the main subject.
If all control variables are removed from the model, with only the Big Three determinants left in, the model still explains about 65% of variance in the dependent variable for the 1950–1960 dataset, and about 55% for the 1995–2005 dataset. The inclusion of any other single control variable into the equation does not improve the explanatory power of the model by much more than about 2%. If all of the control variables are included, in addition to the Big Three, they together contribute only an extra 2–4% towards the variance explained.
Though with so few observations within each cluster (in this case, each cluster is comprised of two counties, or the same county measured twice 45 years apart) and a huge number of clusters (3,044), the robust standard errors would not be very different from the ones produced by usual OLS estimation.
Another good reason for mean-centering is that it removes the collinearity between the original independent variables and the dummy variable for the period. If there was a big change in the means of some variables from one time period to the other, then even in their logged form these variables would be highly correlated with the dummy variable for the period unless they are centered around the period mean. For example, the divorce rate has almost quadrupled since 1960—from 3 to 11% of adult population (compare Tables 7, 10 in Appendices A and B, respectively). Either in its original form or in its log-transformed metric, the divorce rate variable has a correlation of more than 0.90 with the period dummy variable, so including the period dummy and the logged divorce rate (non-centered) into the same equation would lead to hugely inflated statistics of collinearity (VIF over 10 for the divorce variable). On the other hand, if the logged divorce rate is centered around the period mean, this solves the problem.
References
Anderson AL (2002) Individual and contextual influences on delinquency: the role of the single-parent family. J Crim Justice 30:575–587
Apel R, Kaukinen C (2008) On the relationship between family structure and antisocial behavior: parental cohabitation and blended households. Criminology 46:35–70
Baller RD, Anselin L, Messner SF, Deane G, Hawkins DF (2001) Structural covariates of U.S. county homicide rates: incorporating spatial effects. Criminology 39:561–590
Bank L, Forgatch MS, Patterson GR, Fetrow RA (1993) Parenting practices of single mothers: mediators of negative contextual factors. J Marriage Fam 55:371–384
Benson ML, Wooldredge J, Thistlethwaite AB, Fox GL (2004) The correlation between race and domestic violence is confounded with community context. Soc Prob 51:326–342
Blau JR, Blau PM (1982) Cost of inequality: metropolitan structure and violent crime. Am Sociol Rev 47:114–129
Brame R, Paternoster R, Mazerolle P, Piquero A (1998) Testing for the equality of maximum-likelihood regression coefficients between two independent equations. J Quant Criminol 14:245–262
Brezina T, Tekin E, Topalli V (2009) “Might bot be a tomorrow”: a multimethods approach to anticipated early death and youth crime. Criminology 47:1091–1129
Brownridge DA, Chan KL, Hiebert-Murphy D, Ristock J, Tiwari A, Leung W-C, Santos SC (2008) The elevated risk for non-lethal post-separation violence in Canada: a comparison of separated, divorced, and married women. J Interpers Viol 23:117–135
Bruce MA (2004) Contextual complexity and violent delinquency among Black and White males. J Black Stud 35:65–98
Bullock HA (1955) Urban homicide in theory and fact. J Crim Law Criminol Police Sci 45:565–575
Cavan RS (1959) Negro family disorganization and juvenile delinquency. J Negro Ed 28:230–239
Centers for Disease Control and Prevention, National Center for Health Statistics (1950–1960). Vital statistics in the United States. 1950–1960 [data books]. Available from NCHS web site http://www.cdc.gov/nchs/products/pubs/pubd/vsus/1963/1963.htm
Centers for Disease Control and Prevention, National Center for Health Statistics (1995–2005). Compressed mortality file 1979–1998; Compressed mortality file 1999–2005—CDC WONDER On-line database. Available from CDC web site http://wonder.cdc.gov
Charles CZ (2003) The dynamics of racial residential segregation. Annu Rev Sociol 29:167–207
Demuth S, Brown S (2004) Family structure, family processes, and adolescent delinquency: the significance of parental absence versus parental gender. J Res Crime Del 41:58–81
Firebaugh G (1978) A rule for inferring individual-level relationships from aggregate data. Am Sociol Rev 43:557–572
Gastil RD (1971) Homicide and a regional culture of violence. Am Sociol Rev 36:412–427
Gennetian L (2005) One or two parents? Half or step siblings? The effect of family structure on young children’s achievement. J Pop Econ 18:415–436
Gordon RA (1967) Issues in the ecological study of delinquency. Am Sociol Rev 32:927–944
Gordon RA (1968) Issues in multiple regression. Am J Sociol 73:592–616
Hackney S (1969) Southern violence. Am Histor Rev 74:906–925
Haines MR, the Inter-university Consortium for Political and Social Research (2005) Historical, demographic, economic, and social data: the United States, 1790–2000 [data file], (ICPSR 02896-v2). Available from ICPSR web site http://www.icpsr.umich.edu/
Hardesty JL (2002) Separation assault in the context of postdivorce parenting. Viol Against Women 8:597–625
Heimer K (1997) Socioeconomic status, subcultural definitions, and violent delinquency. Soc Forces 75:799–833
Heimer K, Matsueda RL (1987) Race, family structure, and delinquency: a test of differential association and social control theories. Am Sociol Rev 52:826–840
Jargowsky P (2005) Ecological fallacy. In: Kempf-Leonard K (ed) Encyclopedia of social measurement, vol 1. Academic Press, San Diego, pp 715–722
Johnson H, Hotton T (2003) Losing control: homicide risk in estranged and intact intimate relationships. Hom Stud 7:58–84
Katz MB, Stern MJ, Fader JJ (2005) The new African American inequality. J Am Hist 92:75–108
Klebanov PK, Brooks-Gunn J, Duncan GJ (1994) Does neighborhood and family poverty affect mothers’ parenting, mental health, and social support? J Marriage Fam 56:441–455
Krivo LJ, Peterson RD (2000) The structural context of homicide: accounting for racial differences in process. Am Sociol Rev 65:547–559
Land KC, Deane G (1992) On the large-sample estimation of regression models with spatial- or network effects terms: a two-stage least squares approach. Sociol Methodol 22:221–248
Land KC, McCall PL, Cohen LE (1990) Structural covariates of homicide rates: are there any invariances across time and social space? Am J Sociol 95:922–963
Lauritsen JL, White NA (2001) Putting violence in its place: the influence of race, ethnicity, gender, and place on the risk for violence. Crim Pub Pol 1:37–59
Loftin C, Hill RH (1974) Regional subculture and homicide: an examination of the Gastil–Hackney thesis. Am Sociol Rev 39:714–724
McCall PL, Land KC, Parker KF (2010) An empirical assessment of what we know about structural covariates of homicide rates: a return to a classic 20 years later. Hom Stud 14:219–243
McNulty TL (2001) Assessing the race-violence relationship at the macrolevel: the assumption of racial invariance and the problem of restricted distributions. Criminology 39:467–489
Mencken FC, Barnett C (1999) Murder, nonnegligent manslaughter, and spatial autocorrelation in mid-South counties. J Quant Criminol 15:407–422
Miles-Doan R (1998) Violence between spouses and intimates: does neighborhood context matter? Soc Forces 77:623–645
Moses ER (1947) Differentials in crime rates between Negroes and Whites, based on comparisons of four socio-economically equated areas. Am Sociol Rev 12:411–420
Nelsen C, Corzine J, Huff-Corzine L (1994) The violent West reexamined: a research note on regional homicide rates. Criminology 32:149–161
Ousey GC (1999) Homicide, structural factors, and the racial invariance assumption. Criminology 37:405–425
Parker KF (2004) Polarized labor markets, industrial restructuring and urban violence: a dynamic model of the economic transformation and urban violence. Criminology 42:619–645
Parker KF, Pruitt MV (2000) How the West was one: explaining the similarities in race-specific homicide rates in the West and South. Soc Forces 78:148–1508
Peterson RD, Krivo LJ (2005) Macrostructural analyses of race, ethnicity, and violent crime: recent lessons and new directions for research. Annu Rev Sociol 31:331–356
Pettit B, Western B (2004) Mass imprisonment and the life course: race and class inequality in U.S. incarceration. Am Sociol Rev 69:151–169
Pinderhughes EE, Nix R, Foster EM, Jones D (2001) Parenting in context: impact of neighborhood poverty, residential stability, public services, social networks, and danger on parental behaviors. J Marriage Fam 63:941–953
Porter GR (1847) The influence of education, shown by facts recorded in the criminal tables for 1845 and 1846. J Stat Soc Lond 10:316–344
Quinney R (1965) Suicide, homicide, and economic development. Soc Forces 43:401–406
Rebellon C (2002) Reconsidering the broken homes/delinquency relationship and exploring its mediating mechanism(s). Criminology 40:103–135
Ruggles S (1997) The rise of divorce and separation in the United States, 1880–1990. Demography 34:455–466
Sampson RJ, Groves WB (1989) Community structure and crime: testing social-disorganization theory. Am J Sociol 94:774–802
Sampson RJ, Wilson WJ (1995) Toward a theory of race, crime, and urban inequality. In: Hagan J, Peterson R (eds) Crime and inequality. Stanford University Press, Stanford, pp 37–56
Sampson RJ, Morenoff JD, Raudenbush S (2005) Social anatomy of racial and ethnic disparities in violence. Am J Public Health 95:224–232
Shaw CR, McKay HD (1942) Juvenile delinquency and urban areas. University of Chicago Press, Chicago
Shihadeh ES, Ousey GC (1998) Industrial restructuring and violence: the link between entry-level jobs, economic deprivation, and black and white homicide. Soc Forces 77:185–206
Simons RL, Kuei-Hsiu L, Gordon LC, Brody GH, Murry V, Conger RD (2004) Community differences in the association between parenting practices and child conduct problems. J Marriage Fam 64:331–345
Stewart S (2005) How the birth of a child affects involvement with stepchildren. J Marriage Fam 67:461–473
Stolzenberg L, D’Alessio SJ (2007) The effect of divorce on domestic crime. Crime Delinq 53:281–302
Teachman JD, Tedrow LM, Crowder KD (2000) The changing demography of America’s families. J Marriage Fam 62:1234–1246
Tolman RM, Raphael J (2000) A review of research on welfare and domestic violence. J Soc Issues 56:655–682
U. S. Census Bureau (1960) Decennial censuses of population and housing, 1790–2000—census of population: 1960 [data books]. Available from U.S. Census Bureau web site http://www.census.gov/prod/www/abs/decennial/
U. S. Census Bureau (2000) 2000 census of population and housing: Summary file 2, summary file 3 [data file]. Available from U.S. Census Bureau web site http://factfinder.census.gov/home/saff/main.html?_lang=en
Wasserma IM, Stack S (1995) Geographic spatial autocorrelation and United States suicide patterns. Arch Suicide Res 1:121–129
White KJC, Crowder K, Tolnay SE, Adelman RM (2005) Race, gender, and marriage: destination selection during the great migration. Demography 42:215–241
Acknowledgments
This paper would not have been possible without wise and generous advice from Colin Loftin, Alan Lizotte, and David McDowall. My heartfelt thanks also go to the anonymous reviewers and editors of the Journal whose criminological expertise and meticulous attention to detail helped improve the paper tremendously through several rounds of reviews and revisions.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix A: Detailed Information on Variables Included into Initial Analyses: Historical Dataset (1950–1960)
Appendix B: Detailed Information on Variables Included into Initial Analyses: Contemporary Dataset (1995–2005)
Rights and permissions
About this article
Cite this article
Tcherni, M. Structural Determinants of Homicide: The Big Three. J Quant Criminol 27, 475–496 (2011). https://doi.org/10.1007/s10940-011-9134-x
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10940-011-9134-x