Dear Zelig users,
I have been struggling for a while now to perform a rare events logistic
regression procedure. I have come to the point where my R code runs
smoothly, however I experience trouble interpreting the Zelig output. A few
issues I hope someone can help me with:
- Some categorical (factor) variables and intercepts that are not
significant in an ordinary logit model, receive in the relogit procedure
Pr(>|z|) values of 0.000e+00.
- The case control prior correction for the intercept is based on tau, the
ratio between positive and negative events (e.g. King and Zeng, 2001, Eq.
7). However, changing tau affects all coefficients except the intercept.
- Based on e.g. King and Zeng (2001), I understand that after prior and bias
correction, probabilities calculated according to general equation Pr(Y=1|B)
= 1/(1 + exp(xB)) lead to underestimation and still need to be corrected
for uncertainties, as described in Eq. 15-16 of King and Zeng (2001). It is
not clear however, how (or if) Zelig relogit deals with this correction.
Fitted() or predict() obviously do not apply this correction.
Any help would be appreciated - I have been trying to figure this out for
more than a week now. Please note that I am not subscribed to the mailing
list. I am not sure if answers appear on gmane.comp.lang.r.zelig, if not:
comments can be sent to hanne.debrue(a)ees.kuleuven.be.
Many thanks!!
Hanne
Show replies by date
The message is not displayed properly, so second try...
Dear Zelig users,
I have been struggling for a while now to perform a rare events logistic
regression procedure. I have come to the point where my R code runs
smoothly, however I experience trouble interpreting the Zelig output. A few
issues I hope someone can help me with:
- Some categorical (factor) variables and intercepts that are not
significant in an ordinary logit model, receive in the relogit procedure
Pr(>|z|) values of 0.000e+00.
- The case control prior correction for the intercept is based on tau, the
ratio between positive and negative events (e.g. King and Zeng, 2001, Eq.
7). However, changing tau affects all coefficients except the intercept.
- Based on e.g. King and Zeng (2001), I understand that after prior and bias
correction, probabilities calculated according to general equation Pr(Y=1|B)
= 1/(1 + exp(xB)) lead to underestimation and still need to be corrected
for uncertainties, as described in Eq. 15-16 of King and Zeng (2001). It is
not clear however, how (or if) Zelig relogit deals with this correction.
Fitted() or predict() obviously do not apply this correction.
Any help would be appreciated - I have been trying to figure this out for
more than a week now. Please note that I am not subscribed to the mailing
list. I am not sure if answers appear on gmane.comp.lang.r.zelig, if not:
comments can be sent to hanne.debrue@...
Many thanks!!
Hanne