multiple regression, confused!

[reluctantPhd01]

---

Members don't see this ad.

Hi,

I'm confused in running a multiple regression. I'm running a multiple regression, with improvement in GAF score from admission to discharge as the dependent variable (as a percentage).

I want to look at how length of stay and presence of BPD (0/1 variable, either you had the diagnosis or you didn't, and nobody developed it during stay) predict change in GAF.

I'm kinda stuck because BPD is a categorical variable...but since it doesn't have multiple levels, I can put it into a regression without dummying it--right?

Also, should I do it as a GLM (ANCOVA, I guess?) with LOS and BPD as fixed factors or as a regression with both predicting GAF in the same model?

 

[ADDICTED2STATS]

Hi,

I'm confused in running a multiple regression. I'm running a multiple regression, with improvement in GAF score from admission to discharge as the dependent variable (as a percentage).

I want to look at how length of stay and presence of BPD (0/1 variable, either you had the diagnosis or you didn't, and nobody developed it during stay) predict change in GAF.

I'm kinda stuck because BPD is a categorical variable...but since it doesn't have multiple levels, I can put it into a regression without dummying it--right?

Also, should I do it as a GLM (ANCOVA, I guess?) with LOS and BPD as fixed factors or as a regression with both predicting GAF in the same model?

I would put some more thought into the DV. There are some issues in analyzing percentages. An alternative would be a repeated measures ancova or regression using a change score (or just controlling for T1 GAF, predicting T2 GAF from LOS & BPD).

On to your question. Pretty straightforward really. Yes, you can add BPD to a regression equation as is (there's no coding here). As for whether or not you should use ANCOVA or regression, they're the same (both are GLM) so it doesn't matter.

You can PM me if you have any other questions.

 

[LM02]

or just controlling for T1 GAF, predicting T2 GAF from LOS & BPD

This. Don't make it more "fancy" than you need to. It's really quite simple as-is. Your regression equation would look like this:

T2GAF = Bo + B1T1GAF + B2LOS + B3BPD + e

Just look at the simple effects to determine the percentage of variance accounted for in T2GAF predicted by each of your parameters of interest, over and above variance accounted for by T1GAF.