How to choose appropriate analytical method for continous dep variable and categorical indep vas?

[dunga]

Hi Forum,

I have GPA data and socio and demographic data collected from 30 students.

My data are like this.

1. GPA - I have continuous values for 30 students such as 3.1, 3.2, 3 and so on (this means I have exact values)

2. Age - I have age categories for the 30 students like (19-22)    (>22 to 25) and so on (I don't have exact age values)  

3. Gender - (I have Male,  Female categories for 30 students)

4. Commute time - I have categories for 30 students like (10 mins to 15 mins)  (>15 mins to 30 mins) (I do not have exact commute times)

5. Number of units the student is taking (I don't have exact values but have categories for 30 students)

6. Study time (I don't have exact values but have categories for 30 students)

7. Extra curricular time (I don't have exact values but have categories for 30 students)

Question:

I wanted to analyse the impact of above socio demographic factors on GPA.

However, I am not too sure what kind of analysis I could do with a continuous dependent variable against categorical independent variables (whether correlation or regression or what?)

Could you please help me to select appropriate analytical method.

Thanks

Dunga  

[PaigeMiller]

If you are willing to assume that the errors in GPA from the predicted model are normally distributed, then PROC GLM will work, it allows class variables as independent variables as well as continuous independent variables.

[sld · Accepted solution]

As @PaigeMiller notes, you can use the GLM procedure to fit an analysis of variance (ANOVA) as long as the residuals from the model follow a normal distribution and, I will add, as long as each observation is independent of other observations (in other words, you have no clustering, repeated measures, blocking, etc.).

I will also add that if your total sample size is only 30 students, you almost surely will not be able to include all seven of your independent variables simultaneously in one model because you do not have enough data to support estimation of all the parameters. Rules of thumb for a good number of observations range from 5 to 10 to 20 or even 30 observations per parameter estimate. For each categorical independent variable, the number of parameter estimates will be equal to the number of levels minus one. Depending on your analysis objectives, you will need to use some form of model selection, for example LASSO selection in the GLMSELECT procedure, or selection based on AIC or some other information criterion.

I hope this helps.