I am currently working with a dataset that uses two waves in a longitudinal study (wave2 & wave 3).
I am trying to look at whether optimism (optim) and the number of life stressors (life_s) interact to predict subjective ratings of health (health). Other variables entered into the model as covariates.
Level 1 variables that have been measured twice from each participant are: optim, life_s, depression (covariate)
Level 2 variables that have been measured only once from each participant are: age, sex, edu
My model is as follows:
model <- lme(health~ age + sex + edu + depression + optim * life_s, random = ~ 1|ID, data = data_long,
na.action = na.omit, control = list(opt = "optim"), method = "ML")
I have several questions.
(1) If "optim" and "life_s" are within-person Level 1 variables, should I always insert optim*life_s as random slopes?
(2) If (1) is true, then my model will not converge since there are fewer observations than random effects in all level 1 groups. How should I approach this matter?
(3) Should I group-mean center (person-mean center) "optim" and "life_s" variables? I am more interested in the overall trend of the interaction and would like to know if I could just use grand-mean centering for these two variables to look into the interactions.
I am fairly new in mixed effect models, and would really appreciate some guidance in my modelling.
Thank you in advance!
I have outlined my questions above.