Analysis using PARAM=EFFECT in log-linear

[Salih]

Hello,

I normally use (param=ref ref=...) in my coding so I easily find the parameters in my model but I am having difficulty finding the right parameters for log-linear saturated model when it is coded as (param=reffect)

In the attached pictures, I am looking for the parameters that are associated with blond hair and hazel eye color.(hazel is reference category) My program is below;

proc genmod data=students;
class hair (param=effect) eye (param=effect);
model ct = hair eye hair*eye/ dist=poisson link=log obstats;
run;

I am getting below which is not giving me the correct number.

ln(μ24)=3.1895(intercept)+(-0.2985) (blonde hair)=

Please help me understand what I am missing here.

[StatDave · Accepted solution]

The Class Level Information table tells you exactly how to use the parameters. For hazel, it tells you that you add in the negative of each of the Eye parameters. Similarly, for the interaction parameters, you would add in the negatives of all of the Eye parameters for blond. That gives 10 as the predicted value after exponentiating.