REML between-study variance

Viechtbauer's (2005) fixed-point iteration. REML rather than ML for the same reason a sample variance divides by n - 1: ML does not account for the degree of freedom spent estimating mu and is biased DOWNWARD. The 1/sum(w) term in the numerator is exactly that correction, and dropping it recovers the ML estimator, which is returned alongside. Convergence is not guaranteed on a flat likelihood, so it is reported rather than assumed.

Usage

morie_psy_ma_reml(yi, vi, max_iter = 200L, tol = 1e-12)

Arguments

yi, vi

effects and within-study variances.

max_iter, tol

iteration controls.

Value

list: tau2, tau, mu, se, ci, tau2_ml, tau2_dl, reml_correction, converged, n_iter, Q, I2, k, method.

References

Viechtbauer (2005), JEBS 30:261-293; Viechtbauer (2010), JSS 36(3).

Examples

morie_psy_ma_reml(c(0.2, 0.5, 0.1), c(0.02, 0.03, 0.04))$tau2
#> [1] 0