I am converting SAS PROC GENMOD models over to R using glm. When I compare the output for additive models the estimates match for the treatments. However, the estimates do not match when I run interactive models. The interaction estimate is the same, but the main effect estimates are different between R and SAS.
The SAS code for the additive model is:
proc genmod descending;
class Grazed CP;
model Parasitized = Grazed CP / dist=bin link=logit type1;
run;
The R code for the additive model is:
mod3 <- glm(Parasitized ~ Grazed + CP, family = binomial(link = logit), data = dick)
The SAS output for the additive model is:
Analysis Of Maximum Likelihood Parameter Estimates
Parameter DF Estimate SE Wald Chi-Square Pr > ChiSq
Intercept 1 -0.2976 0.2006 2.20 0.1379
Grazed N 1 0.0696 0.2374 0.09 0.7695
Grazed Y 0 0.0000 0.0000 . .
CP 2 1 0.2829 0.2355 1.44 0.2298
CP 25 0 0.0000 0.0000 . .
Scale 0 1.0000 0.0000
The R summary output for the additive model is:
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 0.05486 0.19377 0.283 0.777
GrazedY -0.06956 0.23743 -0.293 0.770
CP25 -0.28289 0.23555 -1.201 0.230
Now I know the sign is different between the estimates but I know why that is already and am not concerned with it.
The SAS code for the interactive model is:
proc genmod descending;
class Grazed CP;
model Parasitized = Grazed CP Grazed*CP / dist=bin link=logit type1;
run;
The R code for the interactive model is:
mod4 <- glm(Parasitized ~ Grazed * CP, family = binomial(link = logit), data = dick)
The summary output for the SAS interactive model is:
Analysis Of Maximum Likelihood Parameter Estimates
Parameter DF Estimate SE Wald Chi-Square Pr > ChiSq
Intercept 1 -0.3321 0.2281 2.12 0.1454
Grazed N 1 0.1337 0.3105 0.19 0.6668
Grazed Y 0 0.0000 0.0000 . .
CP 2 1 0.3766 0.3755 1.01 0.3159
CP 25 0 0.0000 0.0000 . .
Grazed*CP N 2 1 -0.1546 0.4821 0.10 0.7485
Grazed*CP N 25 0 0.0000 0.0000 . .
Grazed*CP Y 2 0 0.0000 0.0000 . .
Grazed*CP Y 25 0 0.0000 0.0000 . .
Scale 0 1.0000 0.0000
The summary output for the R interactive model is:
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 0.02353 0.21695 0.108 0.914
GrazedY 0.02092 0.36878 0.057 0.955
CP25 -0.22198 0.30242 -0.734 0.463
GrazedY:CP25 -0.15460 0.48211 -0.321 0.748
I cannot figure out how to remedy this situation. The only thing I can think of is that SAS an R score the models differently but am not sure how to change the code so that the output would match.
