Turn every sample of a covariance-family table into a list of matrices
Source:R/cov.R
irw_cov2list.RdA table that holds several samples, such as a meta-analytic SEM corpus
with one study per group, becomes a named list of matrices on a common set
of variables plus a named vector of sample sizes: what
metaSEM::tssem1(Cov =, n =) wants. Each matrix is built as
irw_cov2matrix() builds it, then padded to the union of the variables
across samples, in order of first appearance. A variable a sample did not
measure is NA in that sample's matrix, diagonal included, which is
metaSEM's convention for a missing variable.
A cell a sample measured but did not report (a row with an empty value)
is NA too, off the diagonal, with no warning: metaSEM takes both.
Value
A list: matrices (a named list of square matrices, named by
group exactly as the table gives it), n (a named integer vector, one
N per sample), type ("cor" or "cov", the same for every sample),
variables (the common variable set), and sd, mean and rel (named
lists of named vectors, empty vectors where a sample reports none).
Reliabilities (rel) are what a correction for attenuation needs, matched
to each study's matrix.
See also
irw_cov2matrix() for one sample.
Examples
pub <- data.frame(
stat = "cor",
var1 = c("a", "b", "b", "a", "c", "c"),
var2 = c("a", "a", "b", "a", "a", "c"),
value = c(1, 0.4, 1, 1, 0.3, 1),
n = c(100, 100, 100, 80, 80, 80),
group = c(rep("study 1", 3), rep("study 2", 3)))
x <- irw_cov2list(pub)
x$matrices
#> $`study 1`
#> a b c
#> a 1.0 0.4 NA
#> b 0.4 1.0 NA
#> c NA NA NA
#>
#> $`study 2`
#> a b c
#> a 1.0 NA 0.3
#> b NA NA NA
#> c 0.3 NA 1.0
#>
x$n
#> study 1 study 2
#> 100 80
# metaSEM::tssem1(x$matrices, x$n)