I want to explore the effect of the average prenatal maternal stress (cortisol level, continuous measure) on offspring growth during a linear growth period (monthly body size measure, N = 17 infants, 16.7+-1.3 body size measures per infant). So my basic model is
lm(BodySize ~ PrenatalStress + Age + PrenatalStress*Age)
but this one does not control for repeated measurements/dependent data. In theory, I could add infant-ID as a random slope factor, but this would take away the variance between the infants which is what I´m interested in (indeed, “PrenatalStress” has only on value per infant, and R informs me that the mixed model (lme4) is nearly unidentifiably etc.).
Therefore, I´m now looking for a method that controls for repeated measurements without taking away the between-subject variance. I was thinking about running a GEE with geepack or a GLS (package nlme) with infant-ID as “repeated statement” similar to SPSS but I´m not sure if this is what I need. All 3 methods bring up very different results although the direction of the estimated coefficients is identical. So the formulas are actually:
GEE (geepack): geeglm(formula = BodySize ~ PrenatalStress + Age + PrenatalStress*Age, data = xdata, id = ID, corstr = "ar1")
GLS (nlme): gls(BodySize ~ PrenatalStress + Age + PrenatalStress*Age, data=xdata,corr=corAR1(,form=~Age|ID))
Thank you so very much! Andreas
