I've conducted an experiment in which we have registered a physiological measure (FR) of 30 participants in two different conditions. In each condition, I registered the measure in 20 time points, equally spaced in time.

To analyze the data, I'have fitted a multilevel model in R, using the nlme package. I've considered the condition as fixed factor, while time is a random factor (nested in subjects). My interest is to check whether there is an effect of condition on FR.

The first model (model 1) I fitted shows that condition is a significant predictor of the outcome. Then, as measures in each time point seem to be correlated and this correlation is weaker as time gets further apart, I included an autoregressive covariance structure in the model (corAR1), and this second model (model 2) fits better, although condition is no longer significant.

My questions are:

How can I check the homoskedasticity of the models? and the independence of residuals?

If both models show heteroskedasticity, what can I do to correct it?

If only one of them is homoskedastic and residuals are not correlated, should I interpret that the model is right, and that the covariance structure applied to the other model is wrong?

Thank you for your help!!



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.