Abstract
Gottfredson and Hirschi's self-control theory is one of the most empirically tested explanations of criminal behavior. Yet questions remain about the operationalization of self-control. Researchers have examined the relationship between self-control and crime predominantly with the Grasmick et al. (Citation1993) scale. However, research investigating the measurement properties of that scale has relied primarily on statistical techniques that may underestimate or fail to detect potential problems with the scale. In addition, Hirschi (2004) recently offered a revised operational definition of self-control, which he argued is more faithful to the original theory. To date, research using Hirschi's revised self-control measure is sparse. We contribute to the literature by using Rasch measurement and negative binomial models to examine the measurement properties and construct validity of the Grasmick et al. (Citation1993) and Hirschi (2004) self-control scales. The findings indicate that while there are some differences, and revisions are necessary, both scales demonstrate adequate measurement properties and construct validity.
Acknowledgments
This research was supported by the National Institute of Justice, Office of Justice Programs, U.S. Department of Justice Award No. 2006-IJ-CX-0045. The opinions, findings, conclusions, and recommendations expressed in this article are those of the authors and do not reflect the views of the Department of Justice.
Notes
1Response rates were 32.1% for the large city, 98.1% for the medium-size city, and 91.9% and 88.6% for the two small towns, yielding an overall response rate of 68%. Although the response rate for the large city was lower than expected, it is generally consistent with response rates when active consent is required for participation (Esbensen et al. Citation1999). Still, the implications of this response rate are discussed in the “Summary and Discussion” section below. In addition, we re-estimated the Rasch analyses and negative binomial models for only the large city and only the small/medium town sample separately. Although there are some small differences, the results were substantively similar across city size and consistent with the results presented below. We also re-ran the negative binomial regressions controlling for city size; again, the results were substantively unchanged.
2Forty students did not have any information on either the Grasmick et al. (Citation1993) scale or the Hirschi (2004) scale and are excluded from the Rasch analysis. The only significant difference between the Rasch model sample and the full U.S. sample pertains to gender, as 28 of the 40 individuals lost to data attrition are males.
3Significant differences between the abbreviated samples used in the construct validity analysis and the full U.S. sample pertain to gender and family affluence: the missing participants are disproportionately male and have a lower level of family affluence.
4Previous work has also included a measure of costs of criminal acts (see Piquero and Bouffard Citation2007), which were interacted with bonds. These tests were meant to determine if bonds work through costs to impact crime. We do not include costs here because we are not concerned with the pathway of self-control to crime, but rather its measurement properties.
a Mean square fit statistics.
a Eigenvalue of the first component.
5The sample size for reliability is 2,287 for the Grasmick et al. (Citation1993) scale and 2,328 for the Hirschi (2004) scale.
*p < .05; **p < .01.
a .