Skip to contents

Quantify uncertainty in one estimated common between-study treatment-contrast variance using a profiled ML likelihood.

Usage

network_heterogeneity_profile(object, level = 0.95)

Arguments

object

A joint fit with one estimated common study variance, no additional random grouping levels, and no fixed tau2.

level

Coverage for the approximate profile-likelihood interval.

Value

A one-row data frame with the ML tau2 estimate, lower/upper bounds, requested level, and method. Bounds refer to variance, not its SD.

Details

For each candidate nonnegative \(\tau^2\), build its total study covariance and maximize the Gaussian ML likelihood over the treatment means by GLS. Find the ML variance estimate and retain values satisfying \(2\{\ell(\hat\tau^2)-\ell(\tau^2)\}\leq\chi^2_{1,level}\). The calculation includes the exact zero boundary. It uses ML even if the fitted object was estimated by REML, so its returned estimate may differ from the original object's REML estimate.

tau2 is variance on the squared effect scale; its square root is the between- study SD. A broad interval indicates weak precision in heterogeneity estimation. A lower bound of zero means the approximate interval includes zero, not that homogeneity has been established. The usual one-degree-of-freedom chi-squared cutoff is approximate and can be inaccurate near a variance boundary or with few studies. Sampling covariance is treated as known. This is not an incoherence or transitivity test, and does not profile an unstructured matrix or several hierarchical variance components. Failure to bracket an upper bound raises an error.

Examples

 d <- data.frame(study=1:9,a=rep(c("A","A","B"),3),
   b=rep(c("B","C","C"),3),effect=c(.1,.5,.2,.4,.8,.1,.2,.4,.4),variance=.04)
 fit <- network_meta(d,"study","a","b","effect","variance",heterogeneity="common")
 network_heterogeneity_profile(fit)
#>       tau2     lower     upper     level
#>  0.0000000 0.0000000 0.0397372 0.9500000
#>                                                 method
#>  ML profile likelihood; approximate chi-squared cutoff