predicted probability

noetsi

Fortran must die
#1
Does anyone know how to generate this in Proc Mixed and Proc Glimmix?

When I see this done with an intercept only model I notice the examples are taking the parameter and reversing the sign in generating the following formula to generate predicted probability.

pp = e to the parameter value [they actually use the negative parameter so if its .2 they use e to the negative .2 power]/1+ e to the negative parameter value.

Not sure if I am getting this right.
 

fed2

Active Member
#2
I think it depends on the 'link' function doesn't it? There is an 'ilink' option in glimmix output statement that outputs predictions on natural scale. Not sure if that applies to mixed. I always just try both exp(+) and exp(-) and take the one that looks correct. This is easier and more cost efficient than thinking.