What GLM should is use in this dataset?

#1
Hi everyone and thank you!

I'm not a student of statistics, i'm in a different but somehow related field. i'm doing this research, been reading books on advanced statistical methods, yet i'm still not sure, if what i've been doing are correct, or if i've used the appropriate methods to tackle my problem, so here it is.

I think what I need is to perform a regression analysis on a dataset with the following variables:

a. course, a multinomial categorical variable i.e. levels are Socio 101, Eng 32, etc
b. time_interval, a multinomial cat variable i.e. containing the time interva; the course was given i.e. levels 7-9AM, 9-11AM, etc.
c. grade, the grade incurred by the student in the course i.e. 1.0, 1.25, 1.5, etc, DRP, INC, ...

Now, here are my questions, what is the appropriate method to use if:

a. (confirmatory) i want to show that given a course, the time the course was given is conditionally independent to the grades incurred by the student who took that course? simply, i want to show that time does not have an effect in the performance (grade) of students in the different courses...

at the moment, what i did is to do a Chi-square test of independence on the partial tables of time_interval and grade, controlling course. at some levels of course, i find out that time and grades are really independent, and on others, found out that they are not( they are dependent). is this the best way to do this? i'm actually confused, since what i want is a single measure, that will tell me that indeed, time does not have significant effect overall.

b. (exploratory), and then i want to identify those courses which are highly associated to poor performance of the students i.e. difficult courses?

at the moment, i did a correspondence analysis on the marginal contingency table of course and grades, and actually have results. But i want to know the strength of the association that I identified i.e. Odd ratios, which would tell me that passing in course A has odds way lower than passing in any other course?

thank you really, and i really hope that someone could me on this.

thank you very much!
 
Last edited:
#2
Hi everyone, i've come across loglinear analysis, and used this to analyze the effects of each of the variables to each other, probably answering my first question...

Using SPSS, i've run loglinear model selection in the menu, and it produced the following parsimonious model.
Notice that instead of grade, I used a binary variable Passed, passed = 1 if the student passed the subject or not.

Model = Constant + course*time + course*passed.

Based on how i understand loglinear analysis, the following model tells us, that the missing time*passed interaction would mean that indeed, time does not have a significant interaction effect with Passed, hence would imply, that time and passed are independent at any levels of course. Here, it also tells us that yes, course*passed interaction is significant, implying that course and passed has actually a strong dependency.

Is my interpretation correct?

thank you!
 
Last edited: