Mixed Effects Models - include cases without key independent variable?

[Newto Stata]

Hello,
I am learning about mixed-effects models and have a question. I am analyzing within-individual changes overtime in my analysis, but I don't know if I should included cases that do not have my key independent variable. Specifically, I am analyzing within-individual changes in grades among college students before and after they go to writing services. So my key independent variable is 'going to writing center' (0=No, 1=Yes) and my dependent variable are their grades. However, I don't know if I should include the rest of college students (and their grades) in my analysis, as these college students are not exposed to writing services. For what I have read, mixed-effects models should only include in the estimates the students who go to the writing center. I would greatly appreciate any assistance you could give me on this issue, and I would also appreciate if you could recommend me some literature about this topic. Thank you!

[Clyde Schechter]

I am analyzing within-individual changes in grades among college students before and after they go to writing services.

So it appears that you have longitudinal data on a population of students. Some use writing services, and you have grades for them from before and after they do that. Others never use writing services, and you have grades for them as well over similar time periods. You don't say either way, but I'm going to assume that the use of writing services is a matter that the students choose to do, that is, it is not a randomly-assigned treatment in an experiment.

In this setting, there are two different research questions that can be asked. You can ask how much difference in grades there is between those who choose to use writing services and those who do not. This is the between student effect of writing services. Or you can ask, among those who did use writing services, how much their grades subsequently change. This is the within student effect. These effects are different phenomena, and the estimates you get from them are not necessarily similar. Not only can they be quite different in size, they can even be of opposite signs. (That is, for example, it is possible that the students who chose to use writing services have worse grades than those who don't, but after using them their grades improve.)

So you need to be clear which you are estimating and do your analysis accordingly. In the sentence from your post that I have quoted above, you have stated that you are looking to estimate the within-student effect. For that purpose, you would exclude the students who never use writing services.

That said, it is the norm in this community that we use our real given and surnames as our username, in order to promote collegiality and professionalism. The Forum software will not allow you to change your username, but you can accomplish this by contacting the system administrator (click on CONTACT US in the lower right corner of this page) and request that he do it for you. Thank you in advance.

[Newto Stata]

Dear Professor Schechter,

I sincerely appreciate your response and your tip about using my true name. I will follow your advice.

Regards,


Sergio