PROC LOGISTIC using CLOGLOG | How to interpreted the results?

[MariaD]

Hi folks,

'm using, for the first time a logistic model with cloglog link. I'm doing something similar to this example: https://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/viewer.htm#statug_logistic_se...

My questions is, how could I calculate the impact on the probability when the cons variable increase? And how could I understand the importance of each variable in probability calculation?

My doubt is because we use the data expanded by day (or month in my case).  

Thanks!

[Ksharp]

For your frist question, check ODDS Ratio  by using

model y=x /  exp ;

or

odds STATEMENT.

For your second question, the abs(Estimate)  is bigger ,the variable is more important .

[MariaD]

Thanks @Ksharp  for your help.

So for the second question, I understand that for this kind of analysis I need to exclude (don't considerate) the estimated value for each day (or month in my case). in other words, for the example of the beetle analysis I only evaluate the ABS(estimate) of the variables sex and cons. Is it correct? And how about classification variables? Because in this case I have an estimated value for each dummy. 

[Ksharp]

I don't know. It obviously you take variable DAY as a class variable.

[StatDave]

When you say you want the impact of a predictor on the response, that suggests you want the marginal effect as discussed in this note. If so, you can fit the model and estimate the marginal effects using the Margins macro as discussed there. For the documentation example you refer to, this call of the macro estimates the marginal effect for the CONC predictor:

%Margins(data     = days,
         class    = day,
         freq     = freq,
         response = y,
         roptions = event='1',
         dist     = binomial,
         model    = day sex conc,
         link     = cll,
         modelopts= noint,
         effect   = conc, 
         options  = cl  )

By making SEX a CLASS variable in the model, this call provides the marginal effect of SEX as the difference in the predictive margins for each SEX level:

%Margins(data     = days,
         class    = day sex,
         freq     = freq,
         response = y,
         roptions = event='1',
         dist     = binomial,
         model    = day sex conc,
         link     = cll,
         modelopts= noint,
         margins  = sex, 
         options  = diff cl )