[R] problem in a glm model.

[Simona Avanzo]

Hallo all, 

I have a problem in a glm model.

The glm model has, among the other variables, 4 dummy variables: 
flg.a2 (takes values 0 or 1) 
flg.d.na2 (takes values 0 or 1) 
flg.v2 (takes values 0 or 1) 
flg.cc2 (takes values 0 or 1) 
In addition to these variables, I have included their interaction, with the order:  
(flg.a2+flg.d.na2+flg.v2+flg.cc2)^2 

In the model´s summary there are not some interactions (flg.a2:flg.cc2 e flg.d.na2:flg.cc2), because there are not cases that present these interactions. 
Then, the model´s summary shows that the p-value of  the interaction flg.a2:flg.d.na2 is 0.66, and so this interaction can be eliminated. 
But when I apply the test anova between the model whit this interaction and the model without this interaction, the p-value is 0.004, and so, the interaction is significative. 
Why I obtain this difference?
Many thanks for any help, 

Simona