![Sample our Law journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5942/TOC-Subject-Sample-2021-Law.jpg"})
Abstract
Using longitudinal data from nearly 4,000 students across 113 public schools in Kentucky, we attempt to unravel the direction of the relationships between student weapon carrying and various objective and subjective school‐crime experiences, including victimization, perceived risk of school victimization, and fear of school victimization. Overall, we found little support for the idea that fear and victimization increase weapon carrying, controlling for other theoretically important predictors, including delinquent offending. While 7th‐grade victimization was modestly associated with increased non‐gun weapon carrying in 8th grade, high perceptions of individual victimization risk in 7th grade decreased both subsequent gun and non‐gun weapon carrying. Fear of criminal victimization in 7th grade did not predict either type of subsequent (8th‐grade) weapon carrying. Though fear, risk, and victimization were inconsistent predictors of gun and non‐gun weapon carrying, we found strong and consistent support for the effects of weapon carrying on subsequent fear, risk, victimization, and offending. However, contrary to the implications of fear and victimization hypotheses, both gun carrying and non‐gun weapon carrying in the 8th grade increased fear of school crime, perceived risk, and actual victimization in the 9th grade. Implications of these findings for the applicability of a “weapons” or “triggering” effect are discussed.
Acknowledgments
This research was sponsored, in part, by grants (DA‐05312 and DA‐11317, Richard R. Clayton, PI) from the National Institute on Drug Abuse. The authors would like to thank Richard R. Clayton, Graham C. Ousey, Kimberly Reeder, Shayne Jones, Michelle Campbell Augustine, and Jon Paul Bryan, for their contributions to the Rural Substance Abuse and Violence Project, which provides the data analyzed here.
Notes
1. Simon et al. (Citation1999) found that males were more likely to carry weapons in general (at school or away from school). But, among those who carried weapons, females were actually more likely to carry them to school specifically (as opposed to carrying them outside of the school context).
2. The effect of age on weapon carrying may be contextual, depending upon the age/grade structure of the school (e.g., whether the school houses 7–12 graders, 9–12 graders, or 10–12 graders).
3. Kleck and Hogan (Citation1999, p. 288) note, in reference to their weak yet significant positive association between gun ownership and homicide offending, “ The failure to control confounding factors that are known to positively affect both violent behavior and gun acquisition, however, is probably at least partly responsible for the positive guns–homicide association.”
4. Waves 1–3 were utilized because we felt three successive waves of data were most theoretically appropriate. Since Wave 4 was in the process of being collected when this paper was begun, Waves 1–3 were our only option. While we have no reason to suspect that findings would be different if Waves 2–4 were examined as opposed to Waves 1–3, future validation work will examine the extent to which the findings emerging here are “wave‐specific.”
5. Superintendents in three originally selected counties refused to participate. These three counties were replaced with another random selection within the appropriate stratum. Several of the final 30 selected counties contained individual schools that simply refused to participate while other schools in the district agreed (it was a school‐level decision after district superintendent‐level approval). Non‐response on the part of individual schools does not appear to be systematic. The nine refusals came from large, mid‐size, and small districts/schools, ranging in setting from rural to urban and from various regions of the state.
6. Once parental consent was obtained, student missing data were largely due to (1) student transferal between the time of parental consent and survey administrations and (2) student absenteeism on the day of survey administration. We attempted to obtain both types of missing data through tracking and follow‐up procedures, with such procedures allowing us to obtain what would otherwise be missing data from approximately half those missing after the first day of survey administration. Very little of the missing student data come from student refusals; fewer than 1 percent of sampled students were missing due to their refusal to participate.
7. Two of the smaller counties among the 30 in the sample utilize an elementary‐school high school configuration whereby students transition from elementary school to high school between 8th and 9th grades. Five schools in the sample utilized a 7–12 or K‐12 configuration; students from these 5 schools were thus in the same school throughout the time series.
8. These average scores correspond with the following frequencies (among the 3,968 total respondents for which we have at least one wave of data): Wave 1 gun carrying = 57; Wave 2 gun carrying = 63; Wave 1 non‐gun weapon carrying = 295; Wave 2 non‐gun weapon carrying = 344.
9. Despite the relatively low correlation between W1 and W2 gun/weapon carrying, we were concerned that controlling for W1 carrying might absorb all variation in W2 carrying. Therefore, we ran models with and without controls for W1 carrying. All results remain the same for the gun‐carrying model with or without W1 carrying controlled. For the other‐weapon carrying model, all results regarding our key variables (e.g., fear and victimization variables) were consistent regardless of whether W1 other‐weapon carrying was controlled. However, two key differences emerged regarding control variables in the other‐weapons model. In a model without W1 other‐weapon carrying controlled, peer weapon carrying’s positive effect and school attachment’ negative effect on W2 other‐weapon carrying became statistically significant. In addition, the effect of race was tempered, with the p value increasing from .03 to .07 in the non‐gun weapon carrying models.
10. Only 7 percent of the sample reported that they had been victimized 20 or more times. As a result, we truncated the responses. Respondents reporting more than 20 victimizations were recoded to 20.
11. Race was not originally measured dichotomously. Respondents were able to choose from several categories, including African American, Asian American, Hispanic American, Native American, White, White and Black, and Other. Less than 11 percent of the sample identified other than White, and race was therefore recoded into a dummy variable.
12. While household income is typically used to measure socioeconomic status, it is often the case that middle and high school kids do not know their family’s income. As such, we elected to measure family SES with parents’ education.
13. Items 4 and 5 were reverse‐coded so that values for all items ranged from low attachment to high attachment.
14. Delinquent activities included drinking alcohol, getting drunk, smoking marijuana, using inhalants, using cocaine, using speed, using crystal meth, selling marijuana or other drugs, forcing someone at school to give up money or property, forcing someone not at school to give up money or property, stealing (without force) someone’s money or property at school, stealing (without force) someone’s money or property not at school, physically attacking someone at school, physically attacking someone not at school, using a gun during a fight, using another weapon during a fight, getting arrested, driving after drinking, and vandalizing property. We felt that the inclusion of the youth’s involvement in delinquent activities in waves 1 and 3 controlled for a “lifestyle” effect that suggests that weapon carrying is one of a host of delinquent activities in which criminogenic youths are involved. Its control thus strengthens the test of the fear and loathing hypothesis.
15. The choice to dichotomize was driven by a coding issue affecting wave‐one survey data. During the first wave, respondents were asked to provide the number of friends who had carried a gun/weapon in an open‐ended fashion (provide the number on the blank provided). Rather than providing numbers of friends, a substantial number of respondents simply put a check mark on the blank line. Hence, in order to retain the cases that provided checkmarks instead of actual numbers, we had to dichotomize into 1 = any friends and 0 = no friends.
16. The full information maximum likelihood (FIML) estimation method used by AMOS to handle missing data uses all possible data points in a dataset to generate values for the data that are missing (as opposed to using only the few more plausible values in multiple imputation). FIML assumes that the missing data are missing at random (MAR). While it is possible that data‐especially those related to weapon carrying‐are not missing at random, we have several reasons to feel general confidence in the MAR assumption. First, as alluded to and reviewed in the text, our data yielded prevalence rates for gun/weapon carrying that were similar to other published studies. Second, we had small amounts of missing data on almost all the variables in any given wave (with the exception of SES and peer weapon carrying). Third, as reported throughout the paper, there were few substantive differences between the findings generated with the models that included no missing data and the FIML‐based models under study here (with the effects for race being an important exception). In the models under study here, both the gun carrying and non‐gun weapon carrying models converged in 10 iterations.
17. It should be noted that we also estimated all models in M‐plus, which better handles complex sampling designs through adjustment of standard errors. All findings regarding the key hypotheses tested here were consistent across models estimated in AMOS and those estimated in M‐Plus. There were several minor differences in levels of significance among control variables in several of the models, and quite substantively different findings for the effects of race. We, therefore, feel confident in the robustness of the effects between fear and victimization variables and gun/weapon carrying. In contrast, we are hesitant to draw conclusions regarding the effects of race on gun/weapon carrying and fear and victimization variables.
18. Though our endogenous variables consist of ordinal measures and/or ratio scales created from averaging ordinal measures, we assume they approximate ratio measures fairly well and thus feel reasonably comfortable with linear model assumptions made by SEM in AMOS. There is some skewness associated with the gun/weapon carrying measures, though the distribution is not entirely dichotomous. In addition, SEM does not handle dichotomous endogenous variables without violation of model assumptions, so collapsing response categories for those variables was viewed unfavorably. Additionally, we tested a number of models in an attempt to increase the goodness of fit of the models utilized in this study. The final models (for both gun carrying and non‐gun weapon carrying) had the following paths estimated in addition to those presented in Tables and : (a) all fully exogenous variables were estimated to be correlated with one another, and (2) error terms from all W3 endogenous variables were estimated to be correlated with one another. For brevity purposes, we do not present all the correlations included in our estimation. These are available from the authors upon request.
19. The negative effect of risk perception disappears in both the gun and other‐weapon models when estimated using a “pure” sub‐sample with listwise deletion of cases with any missing data.
20. While the effect of race had a p‐value < .05, its z‐score does not exceed 5.0 (the rule we use due to complex sampling designs). In addition, the effect of race on both gun and other‐weapon carrying is non‐significant in models using a sub‐sample of cases obtained after listwise deletion. In models estimated with M‐plus, the effect of race is actually in the opposite direction. As stated previously, we therefore are hesitant to make strong conclusions regarding the role of race in the feedback models estimated here.