topic proc nlmixed interaction in Statistical Procedures

proc nlmixed interaction

npandis — Mon, 27 Jan 2020 07:46:32 GMT

Hi,

I have two categorical variables coded 0/1

Can I add their interaction like var1*var2 in proc nlmixed?

Would the procedure correctly recognise it and treat it correctly or do I need to create dummy variables?

Thank you,

Nikos

Re: proc nlmixed interaction

Rick_SAS — Mon, 27 Jan 2020 16:46:17 GMT

Since PROC NLMIXED does not have a CLASS statement, you need to construct the interaction terms yourself. Robin High has an example in his SGF paper.

The other alternative is to use a SAS procedure to output the design matrix, which will include the interaction term as a separate column. If the column is named Var12, you can then use the interaction in the model.

Re: proc nlmixed interaction

npandis — Mon, 27 Jan 2020 20:19:20 GMT

Thank Rick.

I have seen this paper and my understanding is that the goal is to convert the categorical variables to dummy 0/1 vars. This especially important when categorical vars have more than 2 levels. My 2 categorical vars have only two levels and are already coded 0/1.

However, I am not sure if I am ok.

Best,

Nikos

Re: proc nlmixed interaction

StatDave — Mon, 27 Jan 2020 20:55:18 GMT

Since you say you only have two binary predictors, there is only 1 degree of freedom for each main effect and their interaction. So you could use a model specification like this, assuming the variables are A and B with values 'yes' and 'no':

ba*(a='yes') + bb*(b='yes') + bab*(a='yes')*(b='yes')

You can see examples of using this construction in these notes: 60240, 37228, 48506 and 44931.