![Sample our Behavioral Sciences journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/110801/TOC-Subject-Sample-2021-Behavioral-Sciences.jpg"})
Abstract
Interactional theory argues that theoretical variables and delinquency have reciprocal causal relationships. While empirical support for the reciprocal relationships has been found, the impact of delinquency on later changes in the variables, including attenuated attachment to family, needs more elaboration. Labeling theory may offer a constructive extension to interactional theory, given their common emphases on theoretical integration, age-varying effects, and reciprocal relationships. The present study suggests an extended interactional model with the labeling dimension. Using structural equation modeling, both the original and extended models were tested with longitudinal data from a nationally representative sample. The present study tested statistical and substantive significance of the paths hypothesized by each model. Findings lend support for the extended interactional model; providing as much as a 48.2% increase in its explanatory power when compared to the original interactional model. The extended interactional model incorporating labeling theory may contribute to both interactional theory and labeling theory for juvenile delinquency.
Notes
1Some scholars address Thornberry's theory as integrative, because it appears to combine social bonding and social learning theories. However, he denies that assertion and claims that his theory is more of elaboration of social control theory (Thornberry Citation2005), rather than theoretical integration.
2Two conflicting perspectives exist; a person with a high stake in conformity is more vulnerable to a deviant label due to the loss of reputation and social status, but the same person may also be less vulnerable because of the possession of relatively more resources to effectively mitigate the label.
3Although their multiple regression analyses suggested that reflected appraisal of self as a drug user by parents was not significantly associated with later drug use, a possible strong correlation between self and reflected appraisal was not addressed.
4While the present study attempts to replicate the original interactional model, it does not hypothesize an effect of delinquent values on family attachment because interactional theory argues that it is weak and valid only at middle adolescence.
5Originally, the first five waves of NYS were considered, but the large number of items in the measurement model relative to the sample size produced problems in the (asymptotic) covariance matrix. We thus excluded the first wave from the present study. Only the first five waves of the NYSFS, with five contiguous years of data, are appropriate to the present analysis, which requires a minimum of three contiguous years of data. Later waves of the NYSFS are separated by three-year or longer intervals between data collection. One reviewer has raised the question whether these older data, collected in the late 1970s, are really appropriate, given the historical changes that have occurred since then. The concern is well considered, and other work comparing tests of theory for different time periods does suggest that while the results do tend to be similar, there may be some differences regarding specific aspects of the theory (e.g., with the present data, the test of strain theory in Johnson et al. Citation2012). Still, as noted in Johnson and colleagues (Citation2012), whether there are changes in these relationships over time is not something to be assumed, but an empirical question, and in this light, the present results contribute both a test of the extended interactional theory and a basis for comparing results with data collected at a later (or, potentially, earlier) time to see whether there are, in fact, variations over time. Beyond this, the NYSFS data are particularly well suited to address the issue of extending interactional theory by incorporating labeling theory because they do contain the control, learning, and labeling variables necessary for the test of the extended theory.
a inverse coded;
b fixed coefficient. All estimates were significant at p = .05.
6Separation between measurement and structural models was necessary in order to calculate and include interaction terms for age and the theoretical variables. Also, given the more than 70 items used in the study, we concluded it would produce imprecise parameter estimates if we ran the two models simultaneously (see Schumacker and Lomax Citation2010 for details). It is also a practice that is fairly common and often explicitly recommended (e.g., Kline Citation2011:265–268).
7Self-report of behaviors in the NYS were collected for the calendar year prior to survey administration. For example, attitudinal measures at wave 2 come later than behavioral measures at wave 2. This temporal order led us to exclude behavioral measures (i.e., delinquency and delinquent peer association) at wave 2, and start the modeling from attitudinal variables at wave 2.
8The suggested formula for error variance is: (1- reliability coefficient)*(variance). Specific to the reliability coefficient for a single indicator, Schumacker and Lomax (Citation2010) recommend using test–retest reliability score across different time points as a measure of internal consistency. Huizinga and Elliott (Citation1986) reported the average test–retest reliability coefficient for general delinquency in NYS as 0.74.
9Age was centered at its mean value prior to the computation of age-variable interaction terms.
10Since behavioral variables in NYS data reflect activities during the past 12 months, behavioral variables at wave 4 cover activities of participants from age 13 to 19, while attitudinal variables at wave 4 cover attitudes and perceptions of those from age 14 to 20. Consequently, behavioral variables at wave 4 reflect activities at early, middle, and later adolescence altogether.
11Due to the use of multiple waves and including the lagged effect, the models contain a recursive system in which the prior value is included in the structural equation as a predictor for the current value of the same variable. Squared multiple correlations, as explained variances for each endogenous variable, are reported in the LISREL output. For example, explained variance for general delinquency at wave 5 in the extended model was estimated to be 57.7%.
*p < .05; **p < .01. The estimates for the theoretically hypothesized lagged effects are in parentheses; the estimates for the structural paths are bold; and covariance estimates of contemporaneous relationships are italicized.
*p < .05; **p < .01. The estimates for the theoretically hypothesized lagged effects are in parentheses; the estimates for the structural paths are bold; and covariance estimates of contemporaneous relationships are italicized.
12We used a statistical test for the equality of regression coefficients suggested by Paternoster and colleagues (Citation1998). The formula is: .
13Because the mediating effect was not complete, a statistical test seemed necessary for us to conclude whether labeling mediated the effect of prior delinquency on later family attachment. We adopted a protocol recommended by Preacher and Hayes (Citation2008), and used prior delinquency and delinquent peer association as independent variables, negative labeling as a mediator, age and prior family attachment as covariates, and family attachment as a dependent variable. Their approach may be superior to the normal theory test (i.e., Sobel test) given the lack of normality in our variables.