Factor analysis in complex surveys (experimental).
This function fits a factor analysis model or SEM, by maximum weighted likelihood.
svyfactanal(formula, design, factors,
n = c("none", "sample", "degf","effective", "min.effective"), ...)formula |
Model formula specifying the variables to use |
design |
Survey design object |
factors |
Number of factors to estimate |
n |
Sample size to be used for testing: see below |
... |
Other arguments to pass to |
Although fitting these models requires only the estimated covariance
matrix, inference requires a sample size. With n="sample", the sample size is taken to be
the number of observations; with n="degf", the survey degrees of
freedom as returned by degf. Using "sample"
corresponds to standardizing weights to have mean 1, and is known to
result in anti-conservative tests.
The other two methods estimate an effective sample size for each
variable as the sample size where the standard error of a variance of a
Normal distribution would match the design-based standard error
estimated by svyvar. With n="min.effective" the
minimum sample size across the variables is used; with
n="effective" the harmonic mean is used. For svyfactanal
the test of model adequacy is optional, and the default choice,
n="none", does not do the test.
An object of class factanal
.
The lavaan.survey package fits structural equation models to complex samples using similar techniques.
data(api) dclus1<-svydesign(id=~dnum, weights=~pw, data=apiclus1, fpc=~fpc) svyfactanal(~api99+api00+hsg+meals+ell+emer, design=dclus1, factors=2) svyfactanal(~api99+api00+hsg+meals+ell+emer, design=dclus1, factors=2, n="effective") ##Population dat for comparison factanal(~api99+api00+hsg+meals+ell+emer, data=apipop, factors=2)
Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.