The per-observation cluster scores that a re-estimated
build_mcml_pc macro network was fitted on: each
respondent's weighted, sign-corrected score on every cluster. These
are the scores to carry into a profile analysis, a regression, or any
downstream model that needs one number per cluster per respondent.
Value
A data frame with one row per row of the input data, in input
order and with the input's row names, and one numeric column per
cluster, named by the cluster. A row whose members of a cluster are all
missing is NA in that column (items missing only in part are
averaged over the observed ones). The macro network is estimated on the
complete rows, so build_network(composites(fit), method = ...)
reproduces it.
The mcml_pc method errors with class
"nestimate_no_composites" for the descriptive aggregations
("average", "escoufier", "cancor"), which relate
clusters without ever forming a score.
See also
build_mcml_pc to create the fit,
item_loadings for the item weights behind these scores.
Examples
set.seed(1)
f <- stats::rnorm(200)
g <- stats::rnorm(200)
df <- data.frame(a1 = f + stats::rnorm(200), a2 = f + stats::rnorm(200),
a3 = f + stats::rnorm(200), b1 = g + stats::rnorm(200),
b2 = g + stats::rnorm(200), b3 = g + stats::rnorm(200))
clusters <- list(A = c("a1", "a2", "a3"), B = c("b1", "b2", "b3"))
fit <- build_mcml_pc(df, clusters, aggregation = "loadings",
method = "cor")
head(composites(fit))
#> A B
#> 1 -0.51616641 0.5188386
#> 2 0.46126437 1.9244471
#> 3 -0.34848366 0.3766971
#> 4 1.12989912 -0.8649034
#> 5 -0.03460362 -1.5703036
#> 6 -0.93775141 1.3283922