Skip to contents

COVRATIO for brma Objects

Source: R/covratio.R
covratio.brma.Rd

Computes COVRATIO for a fitted brma object. COVRATIO measures the change in the determinant of the covariance matrix of the estimates when observation \(i\) is removed.

Usage

# S3 method for class 'brma'
covratio(model, type = "mods", component = NULL, parameter = NULL, ...)

Arguments

model

a fitted brma object.

type

type of parameters to be summarized. Defaults to "mods" (for the effect size and meta-regression coefficients). Use "scale" for heterogeneity and scale-regression coefficients.

component

optional parameter namespace. Use "random" with one explicitly selected semantic random-effect quantity.

parameter

semantic random-effect quantity used when component = "random".

...

additional arguments. The internal .weights argument can supply precomputed PSIS weights for callers that already extracted them.

Value

A named numeric vector of COVRATIO values, one for each observation.

Details

COVRATIO is computed using importance sampling weights to approximate the leave-one-out covariance matrices without refitting the model. Estimate-unit LOO must first be computed with model <- add_loo(model, unit = "estimate"), unless internal weights are supplied.

$$COVRATIO_i = \frac{\det(Cov(\beta)_{-i})}{\det(Cov(\beta))}$$

Values > 1 indicate that the observation improves precision (decreases variance), while values < 1 indicate that the observation decreases precision (increases variance). For brma.mv() objects, COVRATIO uses estimate-unit PSIS weights. With correlated known-V data, this is influence under conditional estimate deletion, not independent-study deletion. A requested random-effect correlation must be defined in every posterior draw. If either SD is zero in some draws, the complete influence request is unavailable; explicitly select an everywhere-defined quantity with component = "random" and parameter instead. Parameters with zero posterior variance are excluded from the covariance determinant and reported in a printed note when available. Ordered terms retain their fitted totals and omit one share per metadata-declared Dirichlet simplex. Aliases of a shared allocation source contribute its independent coordinates only once. If no estimable parameters remain, or if a full or LOO covariance determinant is zero or non-finite after exclusion, COVRATIO is undefined and values are reported as NaN.

See also

influence.brma, dffits.brma, cooks.distance.brma

Examples

if (FALSE) { # \dontrun{
if (requireNamespace("metadat", quietly = TRUE)) {
  data(dat.lehmann2018, package = "metadat")
  fit <- bPET(yi = yi, vi = vi, data = dat.lehmann2018, measure = "SMD")
  fit <- add_loo(fit)

  covratio(fit)
}
} # }