FINDINGS

Out of156 models of the effects of standard measures of poverty on violent offending, 68 models (43.6%) indicate a statistically significant, positive association, while 21 (31%) are negative and significant (see Table 9.1). The number of comparisons, of course, may be misleading if a small number of studies report a large number of comparisons so we also tallied the “Preponderance of Comparisons” (PoC3) by study, as was done in previous chapters. The proportion of studies with a preponderance of statistically significant coefficients was 44.8%. Thus, the association between poverty and violence is reported at a far greater than chance level, but the existence of many negative and significant coefficients shows that the findings represent a mixture of effects.

Regarding the differential etiology of violence hypothesis, the proportion of significant coefficients was very similar to the proportion of positive and significant coefficients representing the association between poverty and nonviolent offending (43.9%). By comparison to the 44.8% of studies with a statistically significant PoC just reported for violence, the PoC was positive and significant for 11 out of 29 studies of nonviolent crime (37.9%), which is substantial, and a lower proportion than that for violent crime. We conclude that the difference between those is not conclusive.

The models that employ measures of resource deprivation or concentrated disadvantage as the measure of poverty provide a much firmer grounding than the results for standard poverty measures. Out of 63 models of the effects of concentrated disadvantage on violent crime, 52 models (82.5%) indicate a positive and statistically significant relationship, and no coefficients were in a negative direction. By contrast, of 59 models that examine nonviolent outcomes, 33 models (56.9%) indicate a positive and statistically significant relationship. While this suggests a fairly consistent association between nonviolent crime and deprivation indices, it suggests that violent crime is more consistently related with such indices. At the study level, 81% of the studies of violence had a PoC in the predicted direction, and that proportion was considerably lower (though admittedly large), 50%, for the studies of nonviolent offending. We take this difference as evidence of a differential etiology of violence.

The quick “vote count” method we present fails to acknowledge effect sizes, or differences in methodological quality. Therefore, we decided to look more closely to see if we could discern reasons why some studies find larger or more consistent effects of resource deprivation on violence than others.

To begin with, we examine the finding that the differential hypothesis is very weakly supported when standard measures of poverty are used. Several of the studies in Table 9.1 that found virtually the same frequency of positive relationships between poverty and violent crime as between poverty and nonviolent crime report standardized regression coefficients. Standardized coefficients represent the “expected shift in standard deviation units of the dependent variable that is due to a one standard deviation shift in the independent variable when other variables are held constant” (Bollen, 1989, p. 349). Comparing their

Table 9.1 Summary Table: Associations Between Poverty and Offending

Independent Variable

Study-Level Number of Studies (k) PoC X

Total

Number of studies (k)

% Studies PoC X

Comparison-Level Number of Comparisons X

Total Number of Comparisons

% Comparisons

X

Standard Measures of

Violent

13

29

44.8%

68

156

43.6%

Poverty

Nonviolent

11

29

37.9%

47

107

43.9%

Deprivation Indices

Violent

13

16

81.3%

52

63

82.5%

Nonviolent

8

16

50.0%

33

58

56.9%

PoC Significant The “preponderance of comparisons” (more than half of the comparisons) reported in the study are in the predicted direction and statistically significant

X Coefficient is in the expected direction of the poverty-violence hypothesis and is statistically significant

Table 9.2 Distinguishing Between Thugs and Thieves: Average SES and Income

Indicator of SES

Neither Theft nor Violence

Theft Only

Violence

Only

Both Theft and Violence

Estimated Income

$12,130

$12,144

$11,656

$11,880

Hollingshead

4.46

4.11

4.91

4.88

Duncan

38.6

43.8

32.8

32.4

magnitude is one form of evidence related to the strength of effects, albeit an imperfect one. Of 5 studies that utilize a standard measure of poverty (Arthur, 1991; Kposowa et al., 1995; Oh, 2005; Rosenfeld, 1984; Steffensmeier & Haynie, 2000), authors of 3 report that poverty exerts a stronger positive effect on violent offending than on nonviolent offending (Kposowa et al., 1995; Rosenfeld, 1984; Steffensmeier & Haynie, 2000). The 2 divergent studies deserve a moment of examination. Arthur (1991) reports that poverty exerts a stronger positive effect on nonviolent offending in the majority of models. However, this finding (which runs counter to our hypothesis) may be unique to Arthur’s sample; his data were confined to 13 rural counties in the state of Georgia. He also includes unemployment and the percentage of the population receiving government aid in the models, which may create a redundancy for the purposes of our evaluation (we refer the reader to our note on model “ overspecification4”). Oh (2005) relied on data from 153 US central cities. He regressed change in crime rates on change in demographic factors, change in labor structure, and change in unemployment, employment and poverty rates. In this analysis, only rape and larceny were significantly, positively associated with change in poverty rates. Analysis of changes in crime rates is more conservative than using overall rates because the change scores necessarily control for many of the causes of crime that led to the rates at the beginning of the change period. Given the limitations of the two contradictory papers, the evidence from this small number of studies leans in the direction of a stronger effect of poverty on violent crime than nonviolent crime.

In studies using concentrated disadvantage or resource deprivation measures, we conclude that effect sizes are generally larger in violent crime analyses, but several exceptions make it difficult to be sure. Three (Hannon & DeFronzo, 1998; Steffensmeier & Haynie, 2000; Stretesky & Lynch, 2004) out of five studies that employ measures of resource deprivation report positive effects of deprivation on violent offending that are larger (based on standardized Beta) than the positive effects on nonviolent offending in virtually all models. Contradictory findings were seen in 3 reports. Rosenfeld (1986) found positive effects of deprivation on both violent and nonviolent crime, in similar measure overall. Bellair (2000) reports that the positive effect of concentrated disadvantage on burglary is larger in magnitude than the effect on robbery/stranger assault in models controlling for residential stability, unsupervised teens, informal surveillance and whether the neighborhood was located “downtown.” Zimmerman (2010) reports odds ratios (ORs) estimating the impact of SES on violent and nonviolent crime. While the coefficient for violent crime is negative and the coefficient for nonviolent is positive, neither is statistically significant, and they do not vary much from one another (OR = 0.95 cf. OR = 1.02).

 
Source
< Prev   CONTENTS   Source   Next >