I would like to ask two questions.
First, is it common to use linear mixed models with principals components? I checked some examples and found that they are quite limited.
I have the same participants (thousands of people) who experienced different conditions and I wanna see which conditions are more effective on participants behaviour (behaviour is collected as one scale variable). As I have many different conditions (numeric and scale) I first did correlation analysis to exclude most correlated results and then found principal components to define the factors that are more effective. Now as I have the same participants I use linear mixed models and add participants as random factors. Do the steps make sense?
And, as I have thousands of participants I got a warning from JMP saying "Limit of 46.340 parameters exceeded". I guess random factors are too many to do the analysis but then how can I run this data? Any ideas?
So participant is not the lowest level of randomization? That is, what is the error (residual) in your model now?
Actually, I believe it is. I select participants' Ids and add it as a random effect (cause they repeated the exercise for different conditions and I have results of same participants for different conditions) and I wanna see the overall picture and understand which condition was harder/easier.
So, for instance, as you can see in the image, one of the participants tried 6 different conditions, some tried 6 some 4. And I have all of them in the same table. So I thought about doing a mixed model and adding id's (User) as a random effect.
Sounds good - just checking. Common mistake but not in this case.
Good to hear, thanks!
Any ideas about the error "limit of 46.340 parameters"? Cannot I use all participants as random effect?
There are no labels assigned to this post.