I am searching for correlations between symptoms severity and the parameters of neural activity in a disease model. As predictors, I have two (slightly collinear) measures of disease severity on each side of the body that may differ (sometimes substantially) on the right and left body sides. Also, in each brain hemisphere, I have several recordings (4-16) from the structure involved in pathology – and for each recording, we have several estimated parameters of activity. And I want to study how these parameters correlate with symptoms severity.

So, I am wondering what is the best way to analyze the data that would account for all these conditions. Now I am thinking about using mixed linear model function for my task (nlme package):

fit <- lme(parameterA ~ measure1 + measure2 + random=~1|animal/side)

Here I wish to account for a subset of parameters derived from the given hemisphere. All the parameters in the hemisphere may be seen as repeated measures. But two subsets of recordings from the right and left hemisphere may also be dependent as they come from one animal.

Is my approach correct? Do I need to account for 'measure1 * measure2' instead of 'measure1 + measure2'? Do the variables for random effect estimates (animal and side) appear in the right order? Some of the neural parameters I want to study also have collinearity – what is the best way to account for it? I am going to analyze them in separate models and then adjust the estimates for multiple comparisons. Am I wrong?

If you know a more appropriate way to analyze such data please let me know.

  • $\begingroup$ As I understand this, you have 2 different measures of disease severity and each is assessed on both sides of the body. Is that correct? If so, when you specify side in your formula do you mean the side of the body on which the severity measure was assessed, or the side (hemisphere) of the brain from which the activity recordings were made? More information about what you mean by the "parameters of activity" might help with an answer, too. That information might help find a way to cut down on the number of models and thus on the magnitude of any multiple-comparison corrections. $\endgroup$
    – EdM
    Apr 18 '20 at 22:09

I understand that measure 1 and 2 are nested within side, and animal is randomly take from the population. Assuming that parameterA is a numeric response, it could be

fit <- lme(parameterA ~ side + side:measure1 + side:measure2, data= yourData, random=~1|animal)

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.