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
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
although we should be cautious about reading too much into the stratum-specific risk<br />
ratios due to the small number <str<strong>on</strong>g>of</str<strong>on</strong>g> events, it is clear that low-intensity supervisi<strong>on</strong> was<br />
more effective than treatment as usual for <str<strong>on</strong>g>of</str<strong>on</strong>g>fenders without a prior history <str<strong>on</strong>g>of</str<strong>on</strong>g> serious<br />
<str<strong>on</strong>g>of</str<strong>on</strong>g>fending, than it was for <str<strong>on</strong>g>of</str<strong>on</strong>g>fenders who had committed a serious <str<strong>on</strong>g>of</str<strong>on</strong>g>fense.<br />
Discussi<strong>on</strong><br />
Our first three research questi<strong>on</strong>s examined the sensitivity <str<strong>on</strong>g>of</str<strong>on</strong>g> the predicti<strong>on</strong> model<br />
used by Philadelphia APPD in classifying <str<strong>on</strong>g>of</str<strong>on</strong>g>fenders by risk across several different<br />
definiti<strong>on</strong>s <str<strong>on</strong>g>of</str<strong>on</strong>g> severity. <str<strong>on</strong>g>The</str<strong>on</strong>g> model appears to successfully categorize probati<strong>on</strong>ers into<br />
low and n<strong>on</strong>-low risk. Overall, the probability that an <str<strong>on</strong>g>of</str<strong>on</strong>g>fender in our sample had been<br />
charged with a serious <str<strong>on</strong>g>of</str<strong>on</strong>g>fense (according to the model definiti<strong>on</strong>: murder, attempted<br />
murder, aggravated assault, robbery and sexual <str<strong>on</strong>g>of</str<strong>on</strong>g>fenses) was substantially lower if they<br />
received a low risk predicti<strong>on</strong> than if they did not. <str<strong>on</strong>g>The</str<strong>on</strong>g> average ‘reliability score’<br />
assigned by the model to each <str<strong>on</strong>g>of</str<strong>on</strong>g>fender across all <str<strong>on</strong>g>of</str<strong>on</strong>g> his or her probati<strong>on</strong> cases also<br />
appeared to be linearly related to the <str<strong>on</strong>g>of</str<strong>on</strong>g>fender’s likelihood <str<strong>on</strong>g>of</str<strong>on</strong>g> serious <str<strong>on</strong>g>of</str<strong>on</strong>g>fending: in<br />
general, the higher the score (higher scores represent the lowest risk levels), the less<br />
likely an <str<strong>on</strong>g>of</str<strong>on</strong>g>fender was to have been charged with a serious <str<strong>on</strong>g>of</str<strong>on</strong>g>fense.<br />
<str<strong>on</strong>g>The</str<strong>on</strong>g> threshold used to distinguish predicted low risk <str<strong>on</strong>g>of</str<strong>on</strong>g>fenders from predicted<br />
n<strong>on</strong>-low risk <str<strong>on</strong>g>of</str<strong>on</strong>g>fenders in the model, an average reliability score <str<strong>on</strong>g>of</str<strong>on</strong>g> 0.5, largely appears to<br />
be an appropriate cut-<str<strong>on</strong>g>of</str<strong>on</strong>g>f point in the present sample. However, the model performs<br />
slightly better in terms <str<strong>on</strong>g>of</str<strong>on</strong>g> avoiding the most serious errors – <str<strong>on</strong>g>of</str<strong>on</strong>g>fenders who were<br />
predicted to be low risk but committed serious <str<strong>on</strong>g>of</str<strong>on</strong>g>fenses – if the threshold is raised to<br />
163