I’d always found the SPSS help pretty basic, so I don’t even consider it when looking for information. However, courtesy of the UCLA Academic Computing Group, which has a bunch of the SPSS case studies on-line, I found this one on mixed models.

It really is one of the most straight-forward explanations of mixed models, fixed effects, random effects and repeated measures I have found, complete with descriptions of how to use model fit statistics.

Unfortunately, all of the examples come from business. I teach primarily education students and the design in the SPSS examples doesn’t fit most of their studies.

In most of the case studies, the examples of random effects are markets. To remind you, in case you forgot, a random effect is one selected at random from a population of “levels”, which carries with it its own variance. For example, if you select ten media markets at random, and mail people at random one of four types of coupons, then measure sales in the areas, MARKET is a random effect and TYPE OF COUPON is a fixed effect.

This is modeled fairly easily using SPSS – go to ANALYZE > MIXED MODELS > LINEAR

Pick market as your subject, sales as your dependent, type of coupon as your factor – well, hell, just follow the link above or if you have SPSS installed, under HELP, select CASE STUDIES, and pick the Advanced Statistics option and then Linear mixed models.

That’s not really the type of random effect we have in a lot of educational studies, though. Often what we have is TEACHER or CLASSROOM and unlike market, we don’t give each teacher five different interventions. That would be nice and it is a great way to control for the teacher variance. Similar with the medical study with the cross-over design. Great idea. Great way to control for the clinic or physician variation. You have four clinics and they get randomly assigned to one of four treatments the first week, and then each moves up one the second week (the one with Treatment A goes to Treatment B, the one with B goes to C and so on) At the end of four weeks, each clinic has tried each treatment.

That’s not how it works in schools, though. Teachers aren’t repeated.

In teaching my data analysis class this fall, I want to start with simpler models and work to more complex ones. I’ve plenty of plain vanilla general linear model examples and some very complex generalized mixed models and nothing in between. The SPSS ones would be perfect if they were relevant to education.

I have other work to do and have spent all of the time I can spare on mixed models tonight. Maybe this weekend I can continue my search for good examples of mixed models in education. If you have any good suggestions, hit me up.

Comments

2 Responses to “Great Resource on Mixed Models (Just not great for me)”

  1. Alex Reutter on May 24th, 2012 12:11 pm

    I wanted to thank you for the kind words about the mixed models tutorial. You see, I wrote it for SPSS years ago when MIXED was introduced in v11, and it has been part of SPSS’s online “Case Studies” (from the menus, choose: Help > Case Studies) for 10+ years. Since SPSS was bought by IBM, the whole help system is finally on the web, and this particular topic is at: http://publib.boulder.ibm.com/infocenter/spssstat/v20r0m0/topic/com.ibm.spss.statistics.cs/mixed_table.htm.

    For an educational study, would the GENLINMIXED example at:
    http://publib.boulder.ibm.com/infocenter/spssstat/v20r0m0/topic/com.ibm.spss.statistics.cs/glmm_testscores_intro.htm be more appropriate? You should be able to run essentially the same analysis using MIXED if you don’t have v19.

    Cheers,
    Alex

  2. AnnMaria on May 27th, 2012 12:52 am

    Cool!
    Well it is nice to meet you, Alex, even if only virtually. I will check out those resources. Thanks.

Leave a Reply