آیا نظریه های جرم و جنایت و یا خشونت تفاوت نژاد در بزهکاری را توضیح می دهد؟
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|38564||2008||18 صفحه PDF||سفارش دهید||10917 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Social Science Research, Volume 37, Issue 2, June 2008, Pages 624–641
Abstract We examine race differences in delinquency using the National Longitudinal Study of Adolescent Health. We use a new method that permits an examination of offense specialization. We argue that an examination of offense patterns provides an opportunity for testing theoretical explanations of race effects. If race differences in violent crime reflect race differences in serious crime, then theories of crime can explain race effects. Otherwise, theories of violence are needed to explain the phenomenon. Our results suggest that black adolescents have higher rates of violence, particularly armed violence, but they do not have higher rates of serious (or minor) property or drug crime. Race differences in violence are generally stronger for adolescents who would otherwise be at lower risk: girls and adolescents from educated and intact families. Puerto Rican adolescents also have higher rates of violence than Anglos, but other Hispanic groups do not. We conclude with a discussion of the implication of the empirical literature (including our results) for various theoretical explanations of race differences in violence
Introduction Arrest data and data from victimization surveys suggest that African-Americans have higher crime rates than White Americans (e.g., Bureau of Justice Statistics, 1995 and Hawkins et al., 2000; see Sampson and Lauritsen, 1994). While race differences can ultimately be attributed to racism and the historic oppression of African-Americans (e.g., Hawkins, 1995, McCord, 1997 and Sampson and Wilson, 1995), the more proximate causal process is unclear. In fact, we argue that it is not even clear what racial patterns in offending require explanation. In this research, we use data from the National Longitudinal Study of Adolescent Health (hereafter AddHealth) to examine racial patterns in violence and delinquency (Udry, 1998). We attempt to determine whether blacks and whites differ in their tendency to engage in violence or in their tendency to engage in serious delinquency, violent or not. AddHealth is particularly useful for examining racial patterns because it is based on a large national sample, it over-samples African-Americans, and it uses a method that yields higher frequencies of self-reported delinquency (Harris et al., 2003). As a result, this research is more likely than past research to reveal the extent to which race effects are mediated and moderated by other demographic variables. We use a method of theory testing that focuses on establishing the dependent variable rather than the introduction of mediating variables (although we do that as well). We argue that it is theoretically important to determine whether there are race differences in violent offenses or any type of serious offenses. If race is associated with violence but not other types of crimes, then one must look to theories of violence, not crime, for an explanation. On the other hand, if race is associated with all types of crime, or serious crime, then theories of crime and norm violation are likely to provide the explanation. Our goal, therefore, is to examine what group of theories is likely to explain race differences. Our methods also differ from the methods used in earlier studies. First, we rely upon a statistical method that yields a true measure of specialization and that allows us to determine exactly what types of offenses vary by race (Deane et al., 2005). This method is well-suited to the analysis of criminal behavior, since most offenders commit a variety of offenses, and offenses cannot easily be rank ordered. The versatility of many offenders, however, does not preclude the possibility that predictors might be different for different types of criminal behaviors (Nagin and Paternoster, 1993 and Horney et al., 1995).
نتیجه گیری انگلیسی
Results In Table 2, we present race effects on the nine offenses using the marginal logit regression methodology described above. The table compares race effects when mediating and control variables are left out of equations (total effects) with race effects when those variables are included in the equations (partial/net effects). In Model 1 (“unadjusted” or “total effect” model), we see statistically significant differences between blacks and whites on all types of violent criminal behavior. Race differences in drug offenses and property crimes, on the other hand, are not statistically significant. Race effects are strongest for armed violence. For example, a black adolescent is about three times more likely to commit armed violence than a white adolescent and over twice as likely to commit armed robbery. The race difference in group violence is either similar to or slightly smaller than race differences in individual violence. Table 2. Marginal logit regressions of criminal offense on race Armed violence Unarmed violence Group violence Cause serious injury Armed robbery Sell drugs Use drugs Serious property crime Minor property offenses Model 1 unadjusted effect b Race c Black 1.092 0.659 0.400 0.477 0.751 0.195 0.062 0.152 −0.067 0.109 0.058 0.071 0.072 0.135 0.112 0.064 0.083 0.061 Other race 0.645 0.352 0.484 0.191 0.595 0.167 −0.003 0.392 0.380 0.134 0.068 0.079 0.085 0.151 0.121 0.071 0.088 0.069 Intercept −2.741 −0.581 −1.446 −1.562 −3.273 −2.529 −0.842 −1.892 −0.383 0.110 0.056 0.068 0.069 0.135 0.100 0.057 0.078 0.054 Model 2 adjusted effect b Race c Black 0.772 0.481 0.225 0.282 0.488 −0.102 −0.306 −0.009 −0.068 0.134 0.075 0.086 0.092 0.144 0.133 0.081 0.100 0.075 Other race 0.440 0.111 0.277 0.051 0.365 −0.017 −0.056 0.281 0.296 0.158 0.080 0.091 0.100 0.181 0.140 0.083 0.102 0.077 Hispanic origin Mexican 0.110 0.123 0.181 −0.022 0.336 0.368 0.019 0.240 0.197 0.217 0.111 0.122 0.133 0.236 0.193 0.118 0.140 0.107 Cuban 0.408 −0.052 0.002 −0.191 −1.477 −0.668 −0.174 −0.043 0.230 0.412 0.226 0.271 0.268 0.553 0.463 0.242 0.281 0.219 Puerto Rican 0.432 0.665 0.406 0.238 0.805 0.381 0.047 −0.195 0.469 0.282 0.166 0.184 0.185 0.327 0.283 0.169 0.237 0.162 Central American −0.068 0.029 0.083 −0.136 −0.630 −0.497 −0.641 −0.515 0.109 0.371 0.190 0.212 0.257 0.527 0.368 0.215 0.259 0.182 Other Hispanic −0.770 0.066 −0.512 −0.682 −0.451 0.316 −0.245 0.285 −0.044 0.413 0.239 0.280 0.298 0.501 0.317 0.231 0.290 0.225 Male 1.261 1.066 0.493 1.098 0.743 0.804 0.130 0.435 0.515 0.103 0.048 0.056 0.061 0.115 0.091 0.049 0.062 0.047 Age 0.033 −0.110 −0.106 −0.042 −0.031 0.217 0.257 0.070 −0.055 0.027 0.014 0.016 0.016 0.030 0.024 0.015 0.17 0.013 Place of residence 0.114 0.133 0.090 0.153 0.130 0.302 0.183 0.158 0.196 0.098 0.050 0.061 0.063 0.121 0.095 0.053 0.069 0.050 Public assistance 0.433 0.142 0.047 0.254 0.159 −0.030 −0.016 0.098 0.036 0.141 0.090 0.097 0.105 0.171 0.151 0.091 0.109 0.088 Single-parent family 0.462 0.386 0.208 0.313 0.518 0.687 0.684 0.409 0.317 0.104 0.053 0.062 0.066 0.124 0.094 0.054 0.067 0.053 Parent’s education −0.033 −0.069 −0.053 −0.049 −0.000 0.045 0.026 0.006 0.026 0.016 0.009 0.010 0.010 0.019 0.017 0.009 0.011 0.009 Concentrated disadvantage 0.059 0.032 0.044 0.015 0.015 −0.001 0.030 −0.025 −0.080 0.049 0.030 0.034 0.036 0.056 0.054 0.032 0.040 0.031 Intercept −3.886 1.312 0.581 −1.079 −3.537 −7.609 −5.682 −3.559 −0.420 0.556 0.271 0.317 0.323 0.623 0.510 0.292 0.345 0.262 Model 1 reports unadjusted (total) effect of race and Model 2 reports adjusted (net of additional covariates) effect of race.a a Model 1 and Model 2 control for U.S. Census region to match stratification in AddHealth. b Regression coefficients that are statistically significant for a two-tailed test at p < .05 are shown in bold in the table and their standard errors are shown below the regression coefficients in italics. c Reference category for Race is white. Table options Race effects are reduced with the introduction of mediating and control variables—but the pattern of race differences in violent offenses observed in Model 1 is still very much evident in Model 2. Race differences in violence are still substantial for armed violence, as are race differences for armed robbery and unarmed violence. In addition, consistent with past research, blacks are less likely than whites to use drugs, but only after adjusting for other predictors of drug use. The race difference in violence, but not other crime, or other more serious crime, is not consistent with crime theories. As indicated earlier, some might question our classification of some offenses as more serious than others. For example, we classified driving a car without the owner’s permission as a serious property crime, since it involved the theft of an expensive item. One might question this classification since some of these automobiles may be returned after a “joy ride.” However, we obtained similar results when we reanalyzed the data omitting this response. We also did a separate analysis of stealing more than $50 and less than $50 since the former is clearly more serious than the latter. Race did not have a significant effect on either type of theft. Finally, we investigated our decision to include marijuana with other drugs as multiple indicators of drug use. Results are unchanged when we use marijuana as a single indicator. Very few respondents use drugs other than marijuana. 6.1. Other effects Model 2 also shows that adolescents from single-parent families are more likely to commit all types of crime. The effects are particularly strong for drug offenses. Urban residence also shows a positive effect across all criminal offenses, although the effects are not statistically significant for armed violence, group violence, and armed robbery. Urban residence has its strongest effect on selling drugs. This evidence suggests that family structure and urban residence mediate race effects on delinquency generally. Socioeconomic status has effects on violent crime but not drug or property crime. Adolescents whose parents receive public assistance are more likely to engage in armed violence and injurious violence. Adolescents whose parents are less educated are more likely to engage in violent crime, except armed robbery, but they are less likely to engage in drug-related and minor property offenses. This evidence suggests that the family’s socioeconomic status mediates race differences in violence but not drug or property crime. Age and gender differences are also observed in the table. As adolescents age, their involvement in drug and serious property crime increases while their involvement in unarmed and group violence, injurious violence, and minor property crime declines. Gender differences are stronger for violence, particularly violence involving weapons, and (relatively) weaker for property crimes and drug use. Model 2 reveals ethnic differences, primarily for adolescents from Puerto Rican backgrounds. Adolescents with Puerto Rican backgrounds are more likely than Anglos to engage in violent offenses (although not all coefficients are statistically significant) as well as minor property offenses. Adolescents with Mexican, Cuban, and Central American backgrounds, on the other hand, are no more likely than Anglos to commit any type of crime. In fact, Cubans are less likely to engage in armed robbery and Central Americans are less likely to use drugs and commit serious property crime. Note that the evidence does not support the idea that Hispanic cultures produce higher rates of violence because of an emphasis on machismo. Finally, we find no evidence that adolescents from disadvantaged neighborhoods are more likely to engage in criminal behavior net of individual-level characteristics. Only one of the coefficients attains even marginal statistical significance (using a two-tailed test) and it is not in the predicted direction. The results are inconsistent with McNulty and Bellair’s (2003b) analysis of AddHealth data which found that neighborhood disadvantage was associated with serious violence. However, their contextual measure is based on block groups while ours is based on census tracts. 6.2. Statistical interactions In Table 3, we present the interaction effects between race and the other explanatory variables (except Hispanic origin) for each criminal behavior. Each race interaction is estimated separately. For example, in the first model, we enter only a race by gender interaction in addition to the full set of explanatory variables. Using this method, the race main effect shown in Table 2 is conditioned by only one other explanatory variable. For the sake of simplicity we only report the interaction coefficients (and their standard errors) in Table 3.10 The coefficients should be interpreted as raising or lowering the main effects of race shown in Table 2. When the z-ratio associated with the coefficient is not statistically significant, the main effect of race shown in Table 2 is unchanged. 11 Table 3. Marginal logit regressions of criminal offense on full set of explanatory variables and race interactionsa,b Armed violence Unarmed violence Group violence Cause serious injury Armed robbery Sell drugs Use drugs Serious property offenses Minor property offenses Black x Male −0.783 −0.489 0.180 −0.299 −0.081 0.568 0.388 −0.141 0.083 0.235 0.118 0.137 0.147 0.238 0.131 0.161 0.121 0.126 Age −0.046 −0.078 −0.033 −0.075 −0.069 −0.148 −0.115 −0.072 −0.092 0.058 0.035 0.038 0.040 0.067 0.059 0.037 0.045 0.034 Place of residence 0.042 0.217 −0.094 0.289 −0.001 −0.132 0.049 −0.103 −0.167 0.220 0.123 0.141 0.151 0.268 0.235 0.135 0.164 0.144 Public assistance −0.350 −0.317 0.156 −0.414 0.379 0.028 0.005 0.286 0.121 0.275 0.189 0.202 0.216 0.353 0.304 0.191 0.231 0.187 Single-parent family −0.199 −0.336 0.133 −0.058 0.090 −0.365 −0.095 0.029 0.015 0.224 0.124 0.145 0.151 0.269 0.230 0.135 0.167 0.126 Parent’s Education 0.052 0.065 0.052 0.061 0.017 0.033 0.012 0.012 0.008 0.034 0.022 0.023 0.024 0.039 0.035 0.024 0.029 0.022 Concentrated disadvantage −0.073 −0.192 −0.084 −0.146 0.272 0.117 −0.113 −0.014 0.007 0.098 0.061 0.068 0.073 0.118 0.112 0.065 0.082 0.062 a For simplicity, table only reports interactions with dummy variable for Black. See Table 2 for the full list of variables. Interaction terms for Other Race were also included in the analysis but they are not reported here. b Interaction effects that are statistically significant for a two-tailed test at p < .05 are shown in bold. Standard errors are shown below the regression coefficients in italics. Table options The results suggest that race differences in violence are stronger for youth from educated families. All five coefficients involving the multiplicative effects of race and parents’ education are positive, although only three are statistically significant. The pattern is contrary to the multiple disadvantage argument. We also find evidence of interactions involving race and gender. The negative signs for the interactions indicate that race differences in armed violence, unarmed violence, and injurious violence are stronger for girls than for boys. For example, African-American boys are almost 1.8 times more likely to commit armed violence than white boys while African-American girls are almost 4 times more likely to commit armed violence than white girls. On the other hand, race differences in selling and using drugs are stronger among boys. We also observe statistical interactions involving race and age. All the coefficients are negative, although only the coefficients involving unarmed violence, selling drugs, using drugs, and minor property crimes are statistically significant. Apparently, black adolescents engage in drug-related offenses and minor property crimes at younger ages than do white adolescents. Whites catch up as they get older. We observe a few other scattered interactions in the table. Race differences in unarmed violence are weaker for youth from single parent families and from poor neighborhoods. Race differences in armed robbery are stronger in poor neighborhoods. 6.3. Effects of the marginal logit methodology With conventional methods it is difficult to estimate the effects of variables on the commission of specific offenses, since offenders typically commit more than one type of offense. As noted earlier, this dependence among multiple offenders could bias the standard errors of the regression estimates. The marginal logit method does, in fact, make a difference in our results. While the regression coefficients reported in Table 2 and Table 3 are very similar in sign and magnitude to those estimated using uncorrected binomial logistic regressions, the standard errors are often different. In general, the uncorrected standard errors are smaller than the corrected standard errors, although in some instances the reverse is true. For example, twenty statistically significant (p < .05) z-ratios in Model 2 of Table 2 drop below the critical value following GEE estimation—while one non-significant z-ratio exceeds the critical value after GEE estimation (due to the precipitous decline in its standard error).