rho_OLS = (e' W e) / (e' W^2 e), with e the OLS residual vector.
Unlike the SAR case, this estimator is CONSISTENT for a one-parameter
CAR model, so it is a principled estimate in its own right as well as
a starting value for maximum likelihood.
Usage
car_rho_ols(Z, W, X = NULL)
Arguments
- Z
Response, length n.
- W
Symmetric adjacency weights (n by n).
- X
Covariates (n by p); an intercept when NULL.
References
Schabenberger & Gotway (2005), p. 340, citing
Haining (1990), p. 130.
Examples
car_rho_ols(Z = 5L, W = 5L)
#> [1] 0
