[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: RE: counterfactuals
you need to generate the predicted probabilities, replacing the black=1 variable to black=0 for black observations. to do this, something along the following lines is needed:
1. run your logit
2. generate predicted indexes, not probabilities (i.e., predict yhat, xb)
3. replace yhat=yhat-_b[black] if black==1.
4. generate the probabilities as exp(yhat)/(1+exp(yhat)).
i think this is correct, but double check my formulas. you should get the idea.
From: Traci A Schlesinger [mailto:tracias@Princeton.EDU]
Sent: Wed 10/2/2002 8:01 PM
Subject: st: counterfactuals
i tried to post this yesterday, but never saw it posted. i apologize if
this is a double post.
I have county level data on criminal processing that includes whether
the defendant is incarcerated pretrial and / or post-trial, as well as
several legal variables regarding the type of offense the individual was
arrested for and details of their prior record, and extra-legal
variables such as race and age.
i have used xtlogit, with county level fixed effects, to analyze this
data, with specific interest in the effect of race and ethnicity on
criminal processing. i find that race and ethnicity affects both
pretrial release decisions, and the decision to incarcerate given
as a next step, i wanted to create counterfactuals that would look at
what the predicted probability of release would be for the black and
latino defendants if they were white (e.g. what would be the probability
of being granted pretrial release if one had all of the legal
characteristics of the black defendants but was white), or put
differently, what would be the predicted probability of release for
black defendants if the black coefficient equaled zero.
i understand how to use the predict command to get predicted
probabilities, but don't know how to get predicted values for this type
of counterfactual situation. i have read the manual and the FAQs, but i
am still lost.
any help would be greatly appreciated!!
thanks in advance,
traci a schlesinger
* For searches and help try: