1

I'm fitting a mixed multilevel model in R using the nlme package, and I don't understand what is the difference between the covariance structures corAR1 and corCAR1. Can someone help me?

Randel
  • 6,711
Mike_999
  • 133
  • 2
    Have you checked any of the references mentioned in the help files of the two functions? – Roman Luštrik Jun 26 '12 at 08:47
  • 2
    As a hint/amplification/followup of @RomanLuštrik's comment: see Pinheiro and Bates 2000; you can see relevant bits on Google Books (pp. 236, 243-244). This is basically the "what have you tried" comment ( http://mattgemmell.com/2008/12/08/what-have-you-tried/ ). If you give us a more specific starting point than "I don't understand" (and show evidence of effort), it will be easier/more rewarding to help ... – Ben Bolker Jun 26 '12 at 09:02

1 Answers1

8

Adding an answer because this question was never directly answered and I came here looking for the answer. Looking at the NLME documentation, corAR1 represents an autocorrelation structure of order 1. The corCAR1 argument represents an autocorrelation structure of order 1, with a continuous time covariate. In addition, Singer and Willet (2003) note the important distinction is that corCAR1 allows for unequally spaced time covariates or observations, while corAR1 requires equally spaced time intervals.

Kevin Joldersma
  • 81
  • 2
    You may also need to use corCAR1 in the case of missing data, which could cause equally spaced time to become unequally spaced. – Jeffrey Girard Oct 04 '18 at 21:35