## determining which model to use

Solved
Super Contributor
Posts: 286

# determining which model to use

I want to create a model to examine the relationship between 7 independent variables and 1 dependent variable.

The dep var is EverRetained. Values are Y/N.

The  independent vars are  gender (M/F), lunchstatus (F,R,P), GTstatus (Y/N), IEPstatus (Y/N), 504status (Y/N), LEPstatus (Y/N) and RR (Y/N).

I know the type of variables determine how I should proceed but I'm sure if a regression model is appropriate. I conducted a chi square comparing the dependent var to each indep var separately. 5 of the 7 tests returned a p-value less than 0.05

Accepted Solutions
Solution
‎12-09-2016 12:51 PM
Posts: 2,655

## Re: determining which model to use

A first approach that ignores any interactions would look like:

``````proc logistic data=yourdata;
class gender lunchstatus gtstatus iepstatus 504status rr;
model everretained(event='y') = gender lunchstatus gtstatus iepstatus 504status rr/clodds=both;
run;``````

However, I am sure that interactions between the independent variables exist, and should be modeled.  Use your prior knowledge about which interactions are of interest.

Steve Denham

All Replies
Solution
‎12-09-2016 12:51 PM
Posts: 2,655

## Re: determining which model to use

A first approach that ignores any interactions would look like:

``````proc logistic data=yourdata;
class gender lunchstatus gtstatus iepstatus 504status rr;
model everretained(event='y') = gender lunchstatus gtstatus iepstatus 504status rr/clodds=both;
run;``````

However, I am sure that interactions between the independent variables exist, and should be modeled.  Use your prior knowledge about which interactions are of interest.

Steve Denham

Super Contributor
Posts: 286

## Re: determining which model to use

Ok, now I'm looking for the part of the output (odds ratios?) so that I can conclude something like "males are 1.5 times more likely to be retained than females."

Posts: 2,655

## Re: determining which model to use

That would come from the odds ratios, yes.

Steve Denham

☑ This topic is solved.