12.03.2014 Views

The Effects of Sanction Intensity on Criminal Conduct - JDAI Helpdesk

The Effects of Sanction Intensity on Criminal Conduct - JDAI Helpdesk

The Effects of Sanction Intensity on Criminal Conduct - JDAI Helpdesk

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

17 OR = .98, p ≤ .08. A likelihood ratio test comparing the models with and without the squared terms also<br />

indicated that the inclusi<strong>on</strong> <str<strong>on</strong>g>of</str<strong>on</strong>g> the squared term did not improve model fit: LR χ 2 (1 d.f.) = 3.16, p ≤ .08.<br />

18 Probati<strong>on</strong>ers from the West were significantly older than those in the Northeast, and significantly more<br />

likely to be n<strong>on</strong>white and <str<strong>on</strong>g>of</str<strong>on</strong>g> lower SES. We examined several combinati<strong>on</strong>s <str<strong>on</strong>g>of</str<strong>on</strong>g> interacti<strong>on</strong> terms between<br />

regi<strong>on</strong> and these <str<strong>on</strong>g>of</str<strong>on</strong>g>fender characteristics, and found that West regi<strong>on</strong> <str<strong>on</strong>g>of</str<strong>on</strong>g>fenders at the $20,000-$29,999 SES<br />

level were significantly less likely to re<str<strong>on</strong>g>of</str<strong>on</strong>g>fend (b = –1.84, p ≤ .023). When this interacti<strong>on</strong> is c<strong>on</strong>trolled, the<br />

probability <str<strong>on</strong>g>of</str<strong>on</strong>g> recidivism is higher in the West than the Northeast, but the associati<strong>on</strong> is n<strong>on</strong>-significant (OR<br />

= 2.91, p ≤ .172).<br />

19 <str<strong>on</strong>g>The</str<strong>on</strong>g> likelihood ratio test comparing the negative binomial and Poiss<strong>on</strong> models was highly significant,<br />

suggesting that the negative binomial model fits the data better (LR χ 2 (1 d.f.) = 6913.3, p < .001). <str<strong>on</strong>g>The</str<strong>on</strong>g><br />

likelihood ratio test comparing zero-inflated negative binomial versus zero-inflated Poiss<strong>on</strong> also supports<br />

the use <str<strong>on</strong>g>of</str<strong>on</strong>g> the former model (LR χ 2 (1 d.f.) = 1930.81, p < .0001). <str<strong>on</strong>g>The</str<strong>on</strong>g> Vu<strong>on</strong>g test statistic comparing the<br />

zero-inflated and standard negative binomial models is positive and highly significant (z = 7.04, p < .0001),<br />

again suggesting that the zero-inflated negative binomial model is the most appropriate.<br />

20 <str<strong>on</strong>g>The</str<strong>on</strong>g> key assumpti<strong>on</strong> <str<strong>on</strong>g>of</str<strong>on</strong>g> Cox regressi<strong>on</strong> is that the hazard functi<strong>on</strong> for each individual follows the same<br />

form, although we do not impose any shape for the form. We used scaled Schoenfeld residuals to examine<br />

proporti<strong>on</strong>ality. <str<strong>on</strong>g>The</str<strong>on</strong>g> detailed results <str<strong>on</strong>g>of</str<strong>on</strong>g> this test are presented in Appendix G. N<strong>on</strong>-significant coefficients<br />

indicate that the proporti<strong>on</strong>al hazards assumpti<strong>on</strong> is satisfied. For this model, our assumpti<strong>on</strong> appears to be<br />

justified. One <str<strong>on</strong>g>of</str<strong>on</strong>g> the c<strong>on</strong>trol variables for jail time appeared to be n<strong>on</strong>proporti<strong>on</strong>al, but this is not a<br />

substantial cause for c<strong>on</strong>cern.<br />

21 We again tested a squared jail time term in this model, which was not statistically significant and did not<br />

improve model fit (OR = .99, p ≤ .630; LR χ 2 (1 d.f.) = .23, p ≤ .634.<br />

22 Likelihood ratio test for Poiss<strong>on</strong> vs. negative binomial: LR χ 2 (1 d.f.) = 2946.78, p < .001. Likelihood<br />

ratio test for zero-inflated Poiss<strong>on</strong> vs. zero-inflated negative binomial: LR χ 2 (1 d.f.) = 304.47, p < .0001.<br />

Vu<strong>on</strong>g test for zero-inflated vs. standard negative binomial: z = 5.68, p < .0001.<br />

23 Covariates in Figures 2.6 and 2.8 are held at the same values as they were for Figure 2.4.<br />

24 Likelihood ratio test comparing models with and without squared term: LR χ 2 (1 d.f.) = 6.62, p ≤ .010.<br />

25 Likelihood ratio test for Poiss<strong>on</strong> vs. negative binomial: LR χ 2 (1 d.f.) = 1041.73, p < .001. Likelihood<br />

ratio test for zero-inflated Poiss<strong>on</strong> vs. zero-inflated negative binomial: LR χ 2 (1 d.f.) = 90.93, p < .0001.<br />

Vu<strong>on</strong>g test for zero-inflated vs. standard negative binomial: z = 5.62, p < .0001.<br />

26 See Appendix G for test <str<strong>on</strong>g>of</str<strong>on</strong>g> the assumpti<strong>on</strong>s <str<strong>on</strong>g>of</str<strong>on</strong>g> proporti<strong>on</strong>al hazards. We proceeded with the<br />

proporti<strong>on</strong>al hazards model despite some evidence for n<strong>on</strong>proporti<strong>on</strong>ality in <strong>on</strong>e <str<strong>on</strong>g>of</str<strong>on</strong>g> the m<strong>on</strong>thly jail<br />

indicator variables.<br />

109

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!