Rebuilds the square matrix from a table in the IRW covariance layout (from
irw_fetch(..., source = "cov") or irw_cov_compute()), with what an SEM
fit needs alongside it.
Arguments
- df
A table with columns
stat,var1,var2,value,nand, optionally,group.- group
For a table holding several samples (a
groupcolumn), which one to return. Required when there is more than one.- complete
A published matrix can hold cells the source did not report: rows with an empty
value, which becomeNAhere, with a warning. Most fitting functions (lavaan among them) refuse a matrix withNA.complete = TRUEreturns the complete part instead: the matrix after dropping the fewest variables that leaves no blank cell (ties keep the variables listed first), with the dropped variables indropped. The IRW's own record (irw_metadata(source = "cov"):n_missing_cells, andis_pdjudged on the complete part) uses the same rule.
Value
A list: matrix (square, with dimnames), type ("cor" or
"cov"), n (the largest cell n, the matrix N when n is the same
everywhere), n_pairwise (a matrix of cell n), sd, mean and rel
(named vectors of standard deviations, means and reliabilities, empty
when the table has none), dropped (the variables
complete = TRUE left out; empty otherwise), and computed (TRUE when
the matrix was computed from an IRW table rather than published, with
method, use and table then set).
Examples
df <- data.frame(id = rep(1:50, each = 3), item = rep(c("a", "b", "c"), 50),
resp = c(replicate(50, sample(1:5, 3, replace = TRUE))))
m <- irw_cov2matrix(irw_cov_compute(df))
m$matrix
#> a b c
#> a 1.00000000 -0.06804574 0.2362722
#> b -0.06804574 1.00000000 -0.1669792
#> c 0.23627216 -0.16697921 1.0000000
# lavaan::cfa("f =~ a + b + c", sample.cov = m$matrix, sample.nobs = m$n)