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.

equati<strong>on</strong>s. <str<strong>on</strong>g>The</str<strong>on</strong>g> first stage equati<strong>on</strong> uses the instrument variables to predict treatment<br />

take-up (the endogenous variable):<br />

T^<br />

= " 0<br />

+ " TA<br />

+ " REGION<br />

+ " GENDER<br />

+ " RACE<br />

+ " AGE<br />

+ " SES<br />

+ " PRIORS<br />

+ " TAR<br />

!<br />

!<br />

!<br />

where<br />

+" TA<br />

* " REGION<br />

+ " TA<br />

* " GENDER<br />

+ " TA<br />

* " RACE<br />

+ " TA<br />

* " AGE<br />

+ " TA<br />

* " SES<br />

+ " TA<br />

* " PRIORS<br />

T^<br />

is the endogenous variable (predicted treatment take-up), " 0<br />

is the intercept,<br />

" TA<br />

is the instrument for assigned treatment, the interacti<strong>on</strong>s between " TA<br />

and the<br />

!<br />

subgroup variables are the additi<strong>on</strong>al instruments, and " TAR<br />

is a c<strong>on</strong>trol (exogenous)<br />

!<br />

variable for post-RA time at risk. Note that <strong>on</strong>ly the interacti<strong>on</strong>s and not the main effects<br />

variables for our subgroups are used as instruments.<br />

!<br />

We hypothesize that race, gender,<br />

etc. predict treatment take-up through their associati<strong>on</strong> with treatment assignment. In the<br />

sec<strong>on</strong>d stage equati<strong>on</strong>, we replace the instruments with the predicted treatment take-up<br />

from the first stage ( "T^ ) to predict the crime outcome Y:<br />

Y = " 0<br />

+ "T^ + " TAR<br />

+ " REGION<br />

+ " GENDER<br />

+ " RACE<br />

+ " AGE<br />

+ " SES<br />

+ " PRIORS<br />

!<br />

<str<strong>on</strong>g>The</str<strong>on</strong>g> main effects for the subgroups remain in the sec<strong>on</strong>d stage model as c<strong>on</strong>trols for any<br />

direct ! variati<strong>on</strong> in outcomes by subgroup. Another important part <str<strong>on</strong>g>of</str<strong>on</strong>g> the 2SLS approach<br />

is the estimati<strong>on</strong> <str<strong>on</strong>g>of</str<strong>on</strong>g> the ‘reduced form,’ which is simply the OLS estimate <str<strong>on</strong>g>of</str<strong>on</strong>g> the ITT<br />

effect <str<strong>on</strong>g>of</str<strong>on</strong>g> the instrument and exogenous covariates <strong>on</strong> crime. Angrist (2006) notes that it<br />

is acceptable to use OLS even if the outcome variable is dichotomous.<br />

<str<strong>on</strong>g>The</str<strong>on</strong>g> coefficient for<br />

"T^ tells us the actual effect <str<strong>on</strong>g>of</str<strong>on</strong>g> treatment received, or local<br />

average treatment effect (LATE) <strong>on</strong> the probability <str<strong>on</strong>g>of</str<strong>on</strong>g> re<str<strong>on</strong>g>of</str<strong>on</strong>g>fending. It can be compared<br />

with the coefficient ! for assigned treatment in the reduced form model (or the outcomes<br />

from our logistic participati<strong>on</strong> model, as described above) to assess whether the estimated<br />

81

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

Saved successfully!

Ooh no, something went wrong!