10-29-2014 11:54 AM
I have a question regarding the standard errors of predicted values when using Proc Genmod and Proc PLM together. For example, I'm using a negative binomial link, and my genmod code is
proc genmod data =mydata;
model y=x1-x3/ dist=NB link=log type3 COVB pred;
from this procedure, I can get the predicted values as well as its standard errors, and I believe the error is calculated as sqrt(Var(y_pred)), where y_pred = log(y).
For example, for one of my observations,
my y=170, and log(y)=5.1358, and standard error is 0.09, so I expect the 95% confidence interval to be exp(5.1358+- 0.09*1.96)= (142.508, 202.796).
Later, I need to use PLM to get the model information, so I use
proc plm source = mymodel alpha = 0.05;
show cov Parms;
score data = mydata out=myoutput PRED stderr UCLM LCLM
The predicted values and the confidence intervals are the same as before in Genmod, but the standard errors is something other than 0.09, or exp(0.09) or log(0.09). I have no idea how this value comes up.
Can anyone help me explain how this stderr in PLM is calculated for ilink option please?
Thank you very much.