Peer Attack w/ Weapon
ITEM-SPECIFIC PEER DELINQUENCY
One limitation of using logistic regressions to predict crime-specific outcomes is that they mask all information on the seriousness and frequency of each type of offense. Sweeten (2012) argues that when using this type of regression, researchers should estimate additional models by removing the most frequently occurring (and usually the less serious) offense to see if the model is driven by the more common type of offending. Using the violence example, hitting others is much more common than attacking others with a weapon, meaning the violence models may simply capture hitting as opposed to general violence.
33 Analyses for the first two hypotheses were replicated using dichotomous measures of each peer delinquency variable and all substantive findings are identical to the main models.
127 In order to estimate models that address this limitation, additional analyses match each of the item-specific independent variables with its corresponding dependent variable (e.g., peer hitting is used to predict self-reported hitting). Each of the following three tables focus on one crime-specific outcome, meaning there are separate tables for
violence, acquisitive crime, and substance use. Within each table, the first three columns replicate earlier findings from each of the three hypotheses (designated by H1, H2, and H3). The next set of columns focus on each specific item within that crime type, thus allowing for comparison across the main analyses presented earlier and the item-specific analysis of each crime type.
Table 7.06 (see page 144) compares the main violence models presented earlier with the two specific types of violence, hitting others with the idea of hurting them and attacking others with a weapon. Support is found for the first two hypothesis across types of violence, and no support is found for the third hypothesis. There is a rather noticeable deviation in the H2 model for attack with a weapon. Here, the FTF peer measure of attacking others is not significant when controlling for online peers attacking others. This is rather surprising given the robustness of the traditional measure of peer delinquency in past research. Recall from Chapter 6 that this type of peer delinquency was more
prevalent within the cyber context and had the least amount of overlap with FTF peer delinquency. This is also the least common type of peer delinquency, meaning these results should be interpreted with caution given the limited number of respondents who have these types of peers and who also attacked others with a weapon. Other differences between the violence and attack models include the fact that race was significant for H1 and poor grades was significant across all hypotheses, meaning attacking others with a
128 weapon is associated with slightly different risk factors. Also of note, online parental monitoring was significant in the main violence model but is not significant when
looking at specific types of violence. These deviations may likely be a result of capturing two different types of violence, one being the most common type of peer delinquency (hitting) and the other the least common (attacking others).
Table 7.07 (see page 145) compares the main findings of the acquisitive crime model to a model focusing on theft below $50. Examining this individual type of acquisitive crime is especially important since it is the only peer measure of theft. Once again, support was only found for the first two hypotheses. The magnitude of the coefficients are mostly similar, although the associations between both types of peer delinquency and self-reported delinquency appears slightly stronger in the theft model. While age was significant for H1 and race for H2 in the main models, neither of these demographic characteristics were significant in the theft models. The significant effect of impulsivity and temper in the acquisitive crime model was also not evident in the theft model. These differences in coefficients between models suggest that the other types of acquisitive crime may be associated with these variables, meaning the more precise model may be the desirable way to test for the effect of peer theft.34
Finally, Table 7.08 (see page 146) presents the results of the substance use model along with item specific models, one focusing on alcohol and tobacco and the other on marijuana and other drugs. Consistent with previous analyses, the first two hypotheses
34 A total of 315 respondents reported involvement in acquisitive crime, but only 184 reported involvement in theft below $50. A crosstabulation of the items within the acquisitive crime scale reveals that among those who did not steal something less than $50, 117 reported avoiding paying for things, 11 reported theft over $50, and 12 reported going into a building to steal something.
129 are supported across types of substance use and no support was found for the moderating effect of online peer substance use. On the one hand, these results speak to the robustness of the first two hypotheses, which have been supported across all analyses. On the other hand, these item-specific analyses reveal additional discrepancies from the main model. Although males, black respondents, and those who live in single parent households were more likely to use illegal drugs in general, the gender-related findings appear to be related to alcohol and tobacco use, while race and living situation appear to be related to marijuana and other drugs. Furthermore, while the interaction between online and FTF peer substance use was not significant, the coefficient is negative for alcohol and tobacco but positive for marijuana and other drugs. Finally, the effect of online peer marijuana and other drug use appears to be stronger than that of online peer alcohol and tobacco use. A one-unit increase in online peer marijuana and other drug use is associated with a 90 percent increase in the odds of using the same drugs, whereas a one-unit increase in online peer tobacco and alcohol use is associated with a 52 percent increase in odds of using tobacco and alcohol.
As a whole, the item-specific analyses support the findings from the initial models: online peer delinquency has a statistically significant effect on self-reported delinquency that is independent of the effect of FTF peer delinquency; however, online peer delinquency does not moderate the effect of FTF peer delinquency. Given the deviations in the magnitude of coefficients and significant risk factors, these findings indicate that the item-specific measures matching each type of peer delinquency with its corresponding outcome may be the ideal approach when examining peer influence in
130 online and offline contexts. As such, the following analyses will also include additional models focusing on these item-specific types of peer delinquency.