Analysis of a longitudinal study where interventions are received at different times

[Blain Waan]

I have a data set of university students. The university has 8 different assisting programs, mostly like scholarships, for needy students to help them concentrate on their studies. Since it costs a lot of money for the university to run these programs, it opts to evaluate how effective each of these programs is.

It is very important to note that these programs were received by different students at different semesters. Also that the same program was received by different students at different semesters. However, many others didn't apply and hence didn't receive any such program.

I have some background variables which do not change over time and some other variables like credits taken per semester, and both GPA and cumulative GPA of students on each semester. Since students often had semester breaks and re-admissions, these variables have missing cells (because the data structure is like: one variable for each semester regarding credits taken, GPA and CGPA). I have data for almost 4 years on each semester.

The other problem is that a few students received help from more than one such programs.

We want to see how effective each of these programs is: effectiveness will be measured in terms of the improvement in the GPA or CGPA of the students. We will also like to see if the effect was instant but temporary or instant and durable or delayed but durable or delayed but temporary.

What kind of model can help us to find these things? Could you please suggest?

 

[noetsi]

Re: Analysis of a longitudinal study where interventions are received at different ti

If you had an objective, interval, measure used to assign students to the programs then you could try regression discontinuity design. I suspect you don't.

A starting point would be to see if you have autoregression in any regression model you run and to test the IV and DV for stationarity (a trend or unit root). If you don't have autoregression or non-stationarity your design is a lot simpler.

 

[Blain Waan]

Re: Analysis of a longitudinal study where interventions are received at different ti

Thank you noetsi. Yes you are right. I don't have any quantitative measure on how the students were assigned to the programs. The main problem that I am facing is that the same program was received at different times by the students. So, I may calculate the CGPA prior and after the program for the program receivers but how can I calculate the same for the non-receivers? Otherwise is it possible to compare the these two groups?