Lately, I’ve gotten a lot of questions about learning how to run models for repeated measures data that isn’t continuous.
Mostly categorical. But once in a while discrete counts.
A typical study is in linguistics or psychology where each subject is asked to answer some Yes/No question on each of many trials. Each trial contains a different stimulus, each of which represents some combination of conditions.
For example, the participant may be asked to pronounce a word on each trial. The outcome is whether or not they pronounce it correctly and each word is either common or uncommon and either long or short.
So one question I’ve been asked a lot is whether we cover these models in our upcoming Repeated Measure Workshop, and if not, how to learn them.
The direct answer is no, we don’t cover models with categorical or count responses. In order to model a repeated measured data set with a categorical response, you’re going to need to use either a GEE or a Generalized Linear Mixed Model (GLMM).
But that quick answer may not tell you the whole story.
GLMMs are more complicated than linear mixed models. First, you need to understand generalized linear models, like logistic and negative binomial regression. That means concepts like odds ratios, link functions, maximum likelihood.
For most people, that’s the easier part.
You also need to understand mixed models for repeated measures. That means concepts like random intercepts and slopes, covariance structures for G and R matrices, fixed and random factors, marginal models.
That’s the hard part.
It’s harder, mainly because these models are so flexible. Very slight differences in design or data collection can require different specifications for the model. And repeated measures add a different layer of complication.
This is the part most people get stuck on.
Pretty much everything you need to know about linear mixed models is also important for GLMMs.
It’s true that the coding is a little different in GLMM procedures than in linear mixed model procedures. This is true in pretty much every stat package.
So if you need to learn how to analyze a repeated measures logistic regression, where to start depends what you already understand.
If you already understand linear mixed models, but want to know how to code and apply those concepts to a binary or count outcome, then you are 90% of the way there. In that case, focus on learning Generalized Linear Models and coding.
But if you are getting stuck on the repeated or mixed part of the GLMM, focus on that–it’s the part with the most moving parts and that requires most of the decisions in the modeling.
I have a data for two populations
Asked both same 36 questions,
And count how many of pop.1 answers ( a ) choice for each question.
And i want to compare between two groups
Please advise me suitable test for such count data
Getachew Tekle says
I am pleased to inform you that this is the best way of addressing world’s community statistical difficulties including me. So, I am very happy and interested in your efforts made. Saying this, I have some difficulties in Bayesian approach in Win Bugs or R. May be helped, please?