This function does an EFA with either PAF, ML, ULS/MINRES,
or DWLS with or without subsequent rotation.
Estimation and rotation are controlled through the control objects built by
estimate_control() and rotate_control(); each accepts a type
("EFAtools", "SPSS", "psych", or "none") that fills in its remaining settings.
efa_fit(
x,
n_factors,
N = NA,
estimator = c("PAF", "ML", "ULS", "MINRES", "DWLS"),
rotation = c("none", "varimax", "equamax", "quartimax", "geominT", "bentlerT",
"bifactorT", "promax", "oblimin", "quartimin", "simplimax", "bentlerQ", "geominQ",
"bifactorQ"),
se = c("none", "information", "sandwich", "np-boot"),
cor_method = c("pearson", "spearman", "kendall", "poly", "tetra", "fiml"),
use = c("pairwise.complete.obs", "all.obs", "complete.obs", "everything",
"na.or.complete"),
estimate_control = NULL,
rotate_control = NULL,
b_boot = 1000,
ci = 0.95,
seed = NULL,
...
)A list of class c("efa", "EFA") containing (a subset of) the following:
Original correlation matrix.
Initial communality estimates from PAF.
Final communality estimates from the unrotated solution.
Eigen values of the original correlation matrix.
Initial eigenvalues, obtained from the correlation matrix with the initial communality estimates as diagonal in PAF.
Eigenvalues obtained from the correlation matrix with the final communality estimates as diagonal.
For PAF, the number of iterations until convergence. For ML, ULS, and DWLS, the number of objective-function evaluations used by the optimiser (not the number of optimiser iterations).
Integer convergence code (0 = converged), using the codes of
stats::optim(). For ML and ULS it is the code from the bounded
optimiser; for DWLS the fit runs a bounded warm start followed by an
unconstrained polish, and the reported code is from the final polish. For PAF
it is 1 if the maximum number of iterations was reached
without meeting the convergence criterion and 0 otherwise. A non-zero code is
also reported with a warning.
A named integer vector indicating which variables have a Heywood (improper) case in the unrotated solution; empty if there are none.
Loading matrix containing the final unrotated loadings.
Matrix of explained variances and sums of squared loadings. Based
on the unrotated loadings. Its rows are "SS loadings" and "Prop Tot Var", followed by
"Cum Prop Tot Var", "Prop Comm Var", and "Cum Prop Comm Var"; those last three are
omitted for a single-factor solution, where they would only repeat the two above them. So
code that indexes a row by name (for example ["Prop Comm Var", ]) must allow for the
two-row form at n_factors = 1.
A named list of fit indices computed from the unrotated
loadings. ML and ULS report the full set: the model Chi Square (with its
p-value and df), CFI, TLI, RMSEA with its 90% confidence interval, AIC, BIC,
ECVI, RMSR, SRMR, and CAF. PAF and DWLS report only RMSR, SRMR, CAF, and df;
the Chi-Square-based indices are NA there, unless DWLS uses
se = "sandwich", which fills the full block from a scaled Chi Square
instead. RMSR and SRMR are both included, but the print and summary methods
display only SRMR (see Fit indices in Details). Whenever the Chi Square is
scaled (se = "sandwich", or a cor_method = "fiml" fit whose correction
could be formed), AIC, BIC, and ECVI are NA, and a few extra
scaled-statistic fields are appended for advanced diagnostics (see
Fit indices in Details). The list also carries Fm, the estimator's own
objective value at the solution -- on its own scale, not the discrepancy
the Chi Square is built from -- and the independence-model baseline
chi_null, df_null, and p_null that CFI and TLI are computed from.
The model implied correlation matrix.
Residual correlations, i.e., orig_R - model_implied_R
Residual correlations standardized by their
bootstrap standard errors. Only returned, if se = "np-boot".
Loading matrix containing the final rotated loadings. For an oblique rotation this is the pattern matrix -- each variable's unique contribution from each factor, with the other factors partialled out -- and is the matrix normally interpreted (see Rotations).
The factor intercorrelations (only for oblique rotations).
The structure matrix rot_loadings %*% Phi, holding the plain
variable-factor correlations, which are inflated by the factor intercorrelations (only
for oblique rotations).
The rotation matrix. The rotated loadings are recovered from the
unrotated loadings as unrot_loadings %*% rotmat for orthogonal rotations and
for promax, and as unrot_loadings %*% t(solve(rotmat)) for the other oblique
rotations.
Matrix of explained variances and sums of squared
loadings. Based on rotated loadings and, for oblique rotations, the factor
intercorrelations. Same rows as vars_accounted; it is returned only when a rotation
was actually applied, which never happens for a single-factor solution, so it always
has the five-row form.
A list of the settings used, including seed (NULL when none was
supplied), input_type ("raw" or "correlation", what x was), and
cor_method_used (the correlation method that actually ran, NA_character_ for a
correlation-matrix input, which consumes none). cor_method keeps the requested value
whether or not it was used. For the criterion rotations fitted by gradient projection
it additionally carries rotation_diagnostics, a list summarising the multi-start run:
converged: whether the start whose solution is returned reached the convergence
tolerance (reported separately from the counts below because the returned solution
is the one with the lowest criterion value, which need not be a converged one).
n_starts_total: the random_starts random starts plus the rational start.
n_optimized: how many of those starts were actually optimized -- fewer than
n_starts_total whenever the solver screens the random starts and optimizes only
the most promising ones.
n_converged: how many optimized starts reached the convergence tolerance.
n_distinct_minima: how many distinct local optima those converged starts found;
more than one means the criterion is multimodal on these data.
criterion_spread: the range of the criterion values they attained.
criterion_best: the criterion value of the returned solution.
When normalize = TRUE, criterion_best and criterion_spread are evaluated on
the Kaiser-normalized loadings the criterion is optimized on, not on the returned
rot_loadings, so they are not directly comparable to a criterion recomputed from
the returned loadings.
Diagnostics of the FIML correlation's EM estimation, present only for
cor_method = "fiml": whether it converged before fiml_max_iter, how many
iterations it took (iter), the number of distinct missingness patterns
(n_patterns), and the number of cases used (n, the reported N). A converged
of FALSE means the analysed correlation is the EM's last iterate, not the true
FIML estimate; this is flagged in the printed output. Set
fiml_max_iter and fiml_tol through estimate_control().
A named list of standard error matrices, returned only when se is not
"none" (see Standard errors in Details for what each method assumes and when a
component comes back NA). For se = "np-boot": bootstrap SDs of the loadings
(rotated too, if a rotation was applied), the residuals, and the fit indices, plus
-- for oblique rotations -- Phi and the structure coefficients; valid_replicates
(and, when rotated, valid_target_rotations) records how many bootstrap replicates
each of those is based on. For se = "information" and se = "sandwich": Wald-type
SEs for the loadings, the uniquenesses, and the communalities -- identical to each
other, since communality is just 1 - uniqueness -- plus Phi and the structure
coefficients for oblique rotations. "sandwich" stays valid under non-normality;
"information" assumes the model is correctly specified.
A named list of confidence intervals of width ci, matching the
components of SE: percentile intervals for se = "np-boot", Wald intervals for
se = "information" and se = "sandwich". Only returned if se is not "none".
A named list of raw bootstrap replicate arrays -- the aligned
loadings, Phi, structure coefficients, residuals, and fit indices behind SE and
CI -- with one replicate per array's last dimension (first dimension for
fit_indices). Failed replicates are left NA. Populated only for
se = "np-boot"; NULL otherwise.
The full unrotated-loading covariance matrix behind
SE$unrot_loadings: a p * n_factors by p * n_factors matrix in column-major
vec(Lambda) order, with rows and columns labelled "<variable>_<factor>".
Populated for se = "information" and se = "sandwich" -- always the unrotated
block, even when a rotation is applied -- and NA-filled if the analytic covariance
is unreliable (a Heywood case or a singular information matrix); NULL for
se = "np-boot" and se = "none". It can be populated even when
SE$unrot_loadings is NA: a weakly determined rotational orientation invalidates
only the marginal SEs, not the underlying covariance.
The asymptotic covariance of the off-diagonal sample correlations --
the meat of the robust sandwich SEs. A p (p - 1) / 2 by p (p - 1) / 2 matrix,
rows and columns ordered by utils::combn() over the column pairs and labelled
"<var_i>-<var_j>". Populated for se = "sandwich" on the polychoric/tetrachoric
and Pearson paths; NULL otherwise (including under cor_method = "fiml", whose
meat is not returned). It typically dominates a sandwich fit's size (about 11 MB at
p = 50) and is kept because efa_mi() needs it from each per-imputation fit to
build the pooled covariance.
data.frame or matrix. Dataframe or matrix of raw data or matrix with correlations. If raw data is entered, the correlation matrix is found from the data.
numeric. Number of factors to extract. Must be at least 1 and
smaller than the number of variables (the common factor model is not identified
otherwise). Use efa_retain() to decide on a value.
numeric. The number of observations. Needs only be specified if a
correlation matrix is used; with raw data, N is found from the data instead.
With N = NA, not all fit indices can be computed; a positive N that is very
small relative to the number of variables leaves the chi-square-derived indices
unavailable as well, with a warning.
With raw data, N is the number of cases the correlation matrix was actually
computed from -- see use for the general rule and how missing values change it.
Under cor_method = "fiml", use is ignored and N is instead the number of
cases carrying at least one observed value.
character. The estimator used to fit the EFA: "PAF" (principal axis factoring), "ML" (maximum likelihood), "ULS" (unweighted least squares; "MINRES" is an accepted alias returning identical results), or "DWLS" (diagonally weighted least squares, for ordinal data). See the Estimators section in Details for their properties and data requirements. Lower-case versions (e.g., "paf") are also accepted.
character. Either perform no rotation ("none"; default), an orthogonal rotation ("varimax", "equamax", "quartimax", "geominT", "bentlerT", or "bifactorT"), or an oblique rotation ("promax", "oblimin", "quartimin", "simplimax", "bentlerQ", "geominQ", or "bifactorQ"). See the Rotations section in Details for their properties and known issues.
character. Whether and how to compute standard errors (and matching
confidence intervals): "none" (default), "information" (analytic
standard errors from the expected Fisher information of the ML solution), "sandwich"
(robust "sandwich" standard errors from raw data, which stay reliable under
non-normality or a misspecified estimator weight), or "np-boot" (non-parametric
bootstrap). The
methods differ in their assumptions, their data requirements, and which estimator,
rotation, and cor_method combinations they support; see the Standard errors
section in Details.
character. How the correlation is computed from raw data:
"pearson", "spearman", or "kendall" (passed to stats::cor()); "poly" /
"tetra" for polychoric / tetrachoric correlations of ordinal / binary data; or
"fiml" for a two-stage full-information maximum-likelihood correlation from raw data
with missing values. See the Correlation methods section in Details for their
properties and the combinations they support. Default is "pearson".
character. Passed to stats::cor() if raw data
is given as input. Default is "pairwise.complete.obs". It is ignored when
cor_method = "fiml" (which handles the missingness itself, so every case
contributes), and it is overridden to listwise deletion whenever an asymptotic
covariance is required (the "DWLS" estimator, or se = "sandwich"), because the
covariance must describe the same cases as the correlation matrix.
a control object from estimate_control() bundling the
estimation control arguments: the type preset; the principal-axis-factoring iteration
settings init_comm, criterion, criterion_type, max_iter, and abs_eigen; and
the maximum-likelihood start_method. Defaults to estimate_control(), which uses
the "EFAtools" type. See estimate_control().
a control object from rotate_control() bundling the rotation
control arguments: the type preset; Kaiser normalize; the convergence precision; the
factor order_type; the varimax/promax settings varimax_type and p_type; the
simplimax/promax k; and random_starts. Defaults to rotate_control(), which uses
the "EFAtools" type. The estimation and rotation presets
are independent. See rotate_control() for details.
numeric. The number of bootstrap samples to draw. Default is 1000.
Must be at least 2, the smallest number from which a standard error is defined.
Under cor_method = "fiml" each bootstrap sample re-runs the EM moment
estimation, so a smaller value may be advisable.
numeric. The level of the confidence intervals: the percentile intervals from
the bootstrap samples under se = "np-boot", and the analytic Wald intervals under
se = "information" and se = "sandwich", the corrected two-stage intervals of
cor_method = "fiml" included. Must be greater than 0 and smaller than 1. Default is
.95 for 95% CIs.
numeric. An optional seed for the random-number generator.
Additional arguments forwarded to the rotation engine (usually not needed); an unrecognized one is an error, not a setting that is silently ignored.
The estimator is chosen with estimator.
PAF (principal axis factoring) iteratively estimates the communalities and makes no distributional assumptions, which makes it robust and a good general-purpose default. Because it minimises no likelihood or weighted discrepancy it provides no model chi-square, and hence no chi-square-based fit indices (see Fit indices).
ML (maximum likelihood) maximises the normal-theory likelihood. It yields the
full set of fit indices and is the only estimator with analytic expected-information
standard errors (se = "information"), but it assumes multivariate normality and is
the most prone to Heywood (improper) cases. Its starting values are set by
start_method.
ULS (unweighted least squares) minimises the sum of squared correlation residuals. "MINRES" (minimum residual) is the same estimator under a different name and returns identical results. It makes no normality assumption, is robust to mild non-normality, and yields the full set of fit indices.
DWLS (diagonally weighted least squares) is the recommended estimator for ordinal
data. It weights each off-diagonal correlation residual by the inverse asymptotic
variance of the corresponding polychoric correlation (Muthén, du Toit, & Spisic, 1997),
reproducing the loadings of a diagonally weighted least squares fit. It therefore
requires raw ordinal data with cor_method = "poly" or "tetra". Because the
weighting follows the polychoric
asymptotic covariance, the matrix and the weights are estimated on the
listwise-complete cases. Its fit-index behaviour is described under Fit indices.
When raw data are supplied, cor_method selects how the correlation matrix is computed
(it is ignored when a correlation matrix is entered directly).
"pearson" (default), "spearman", and "kendall" are passed to
stats::cor(). The factor model assumes a Pearson correlation, but a rank
correlation is analysed on its own scale, not converted to a Pearson-equivalent
value; Kendall's tau in particular gives the most attenuated loadings of the three.
For ordinal items prefer "poly" / "tetra" below, which estimate the correlation
of the underlying continuous variables.
"poly" / "tetra" compute polychoric / tetrachoric correlations for ordinal /
binary data, assuming an underlying bivariate-normal latent variable. They use a
two-step estimator. The polychoric asymptotic covariance that
underlies both the DWLS weights and the scaled (sandwich) statistic relies on
large-sample theory that degrades for empty or near-empty response-category
combinations; with very sparse cells the resulting weights and standard errors can be
unreliable (a warning is issued when empty cells are present), so interpret them with
caution and consider collapsing rare categories.
Each of the p(p - 1)/2 variable pairs is a separate numerical optimisation, so a
polychoric matrix takes much longer to compute than a Pearson one, and the difference grows
quadratically in the number of variables; with se = "np-boot" the whole matrix is
re-estimated for every bootstrap replicate.
"fiml" estimates a two-stage full-information maximum-likelihood correlation. The
saturated multivariate-normal mean and covariance are estimated from raw data with
missing values by an EM algorithm assuming the data are missing at random (Yuan,
Marshall, & Bentler, 2002; Little & Rubin, 2002), and the standardized covariance is
then analysed. The model fit indices are corrected
two-stage statistics wherever the correction can be formed (see Fit indices).
"fiml" uses every case and handles the
missingness itself, so use is ignored; it supplies a continuous (Pearson-type)
correlation only and is therefore not compatible with estimator = "DWLS". Standard
errors are available analytically for estimator = "ML" or "ULS" and, for any estimator,
by the non-parametric bootstrap (see Standard errors). For multiply imputed data,
efa_mi() is the alternative route to handling missingness. Both routes assume
the values are missing at random (MAR), and which one to prefer is largely
practical: FIML is a single, efficient fit and is the simpler default when the
analysis model is the whole story, whereas multiple imputation is more flexible when
the imputation model should draw on auxiliary variables not in the factor model, or
when the same imputations feed several downstream analyses.
A rotation transforms the unrotated loadings toward a simpler, more interpretable
pattern; all rotations are performed by rotation engines built into the package.
Orthogonal rotations keep the factors uncorrelated, whereas oblique rotations let them
correlate (returning a pattern matrix, a structure matrix, and the factor
intercorrelations Phi) and are usually more realistic for psychological constructs.
For an oblique solution the pattern matrix (rot_loadings) holds each variable's
unique contribution from each factor, with the other factors partialled out; it is what
is normally interpreted and reported. The structure matrix
(Structure = rot_loadings %*% Phi) holds the plain variable-factor correlations, which
are inflated by the factor intercorrelations. The two coincide only when Phi is the
identity, which is why an orthogonal rotation returns rot_loadings alone.
Orthogonal rotations:
varimax maximises the variance of the squared loadings within each factor (column simplicity). It is the most widely used orthogonal rotation and spreads variance across factors rather than concentrating it in a general factor.
quartimax simplifies the variables (rows) so that each loads mainly on one factor; it tends to produce a strong general factor.
equamax is a Crawford-Ferguson compromise between varimax (column) and quartimax (row) simplicity.
geominT uses a geometric-mean criterion that rewards a sparse pattern and
tolerates variables with cross-loadings; a smaller offset delta gives a sparser
solution but sharper local minima.
bentlerT uses Bentler's invariant pattern simplicity criterion.
bifactorT is the Jennrich-Bentler orthogonal bifactor criterion: a general factor
plus group factors (bifactor simple structure). It needs at least two group factors
(n_factors >= 3): with two factors the criterion is identically zero, so no rotation
is performed and the unrotated loadings are returned with a warning.
Oblique rotations:
promax is a fast two-step rotation: a varimax solution is raised to a power
(controlled by k and p_type) to form a target that is then fitted obliquely.
quartimin simplifies the variables (rows), like quartimax, so each loads mainly on one factor, but factors are allowed to correlate.
oblimin is quartimin with a tunable argument, gam (default 0, i.e. quartimin
itself); turning it up trades some of that row simplicity for more strongly
correlated factors, and can drive the solution toward factor collapse, so inspect
Phi before interpreting a fit with gam > 0.
simplimax drives the k smallest loadings toward zero. Its criterion is only
piecewise smooth and strongly multimodal, which makes it by far the most
start-dependent rotation offered here: different random seeds can reach noticeably
different solutions. Because every start is fully optimised (there is no screening
stage), raising random_starts to several hundred costs proportionally more time but
buys a better optimum.
bentlerQ is the oblique Bentler invariant pattern simplicity criterion.
geominQ is the oblique geomin criterion; it handles complex (cross-loading)
structure well but is multimodal, so it benefits from more random_starts (and uses a
more thorough multi-start search internally).
bifactorQ is the oblique (correlated) Jennrich-Bentler bifactor criterion, with the same two-group-factor requirement as bifactorT.
Prefer an oblique rotation unless there is a substantive reason to force the factors to
be uncorrelated: if Phi comes back near zero the oblique solution is essentially the
orthogonal one anyway, whereas imposing orthogonality on genuinely correlated factors
distorts the pattern.
The criterion-based rotations (all except varimax and promax) are fitted by gradient
projection with random_starts random starts to guard against local minima; the
complexity criteria (simplimax and geominQ in particular) are the most multimodal. The
starts are drawn from the random-number generator, so different starts can reach
genuinely different optima and such a fit is reproducible only when the generator is
controlled: pass seed, or call base::set.seed() beforehand.
se selects whether and how standard errors (and matching confidence intervals) are
computed. Which quantities they cover depends on the method. The analytic methods
("information" and "sandwich") cover the unrotated loadings, the uniquenesses and
the communalities and, when a rotation is applied, the rotated loadings and -- for
oblique rotations -- the factor correlations and the structure coefficients. The
bootstrap ("np-boot") covers the unrotated loadings, the residuals, and the fit
indices and, when a rotation is applied, the rotated loadings and -- for oblique
rotations -- the factor correlations and the structure coefficients; it reports no
uniqueness or communality standard errors.
"none" (default) computes no standard errors.
"information" returns analytic standard errors from the expected (Fisher)
information matrix of the maximum-likelihood solution, and therefore requires
estimator = "ML" and cor_method = "pearson" (or "fiml", see below). The rotated
standard errors are obtained by propagating the unrotated-loading covariance through
the rotation (Jennrich, 1973); because rotated quantities do not depend on how the
unrotated solution happens to be oriented, they are directly comparable across
programs. Unlike the bootstrap it also works from a correlation matrix as long as N
is supplied. These standard errors assume multivariate normality and a correctly
specified model; under heavy-tailed data or model misfit they can understate the
sampling variability, where "sandwich" or "np-boot" are more robust.
The rotated loadings, Phi, the structure coefficients, the uniquenesses, and the
communalities do not depend on how the unrotated solution happens to be oriented, and
so are comparable across programs. The unrotated loading standard errors are not:
a program using a different orientation will report different unrotated loading
standard errors for the same fit. That orientation can also fail to be well defined --
for example when two factors are of near-equal strength -- in which case the
unrotated loadings have no well-defined sampling distribution and their standard
errors diverge. efa_fit() detects this and returns NA for the unrotated loading
standard errors with a warning, rather than reporting a number that looks like a
standard error but is an artefact of the orientation; the rotated loadings, Phi, the
structure coefficients, the uniquenesses, and the communalities are unaffected and
still reported. Use those, or se = "np-boot", when the unrotated loadings
themselves are the quantity of interest. This detection applies to every analytic
route (the Pearson and polychoric paths and the two-stage cor_method = "fiml"
sandwich alike). A Heywood case (a uniqueness at its lower boundary) is
separate and more severe: there
the whole SE/CI block comes back NA with a warning, because the standard-error
approximation fails for every parameter. The scaled chi-square (from "sandwich", or
from cor_method = "fiml") does not rely on that approximation and is still
reported, so the fit indices are unaffected.
"sandwich" returns robust ("sandwich") standard errors estimated from raw data,
combining the estimator weight with a distribution-free covariance of the
correlations, so it stays valid under non-normality and weight misspecification
(Browne, 1984; Satorra & Bentler, 1994). It is available either for ordinal data with
cor_method = "poly" or "tetra" and estimator one of "ML", "ULS", or
"DWLS", or for continuous data with cor_method = "pearson" and
estimator = "ML" or "ULS". It reports the same coverage as "information" and
additionally fills the model fit's chi-square block with a scaled chi-square (see
Fit indices); the statistic reported as chi is always the scaled-and-shifted one
(flagged by chi_scaled_type), and a mean-adjusted alternative is returned alongside
it as chi_mean_adjusted.
Because the covariance must describe the same cases as the correlation matrix, the
sandwich (like estimator = "DWLS") is computed on the listwise-complete cases; on
data with missing values the reported N, the correlation matrix, and the point
estimate therefore reflect the complete cases regardless of use.
"np-boot" draws a non-parametric (case-resampling) bootstrap and needs raw data.
A correlation matrix carries no cases to resample; alone among the unsupported
combinations this one does not error but warns and downgrades se to "none", so
the fit is returned without an SE slot.
It is the most general method -- available for any estimator, rotation, and
cor_method -- and the most robust to non-normality and misfit, at the cost of speed;
its intervals are bootstrap percentile intervals. The replicate fits run across
replicates with the future framework; by default they run sequentially, but
registering a plan with future::plan() runs them in parallel instead, as the
examples show. With a fixed seed the bootstrap is reproducible and yields the same
result regardless of the number of workers. Under cor_method = "fiml" each resample
also re-runs the EM moment estimation and is therefore slow, so a smaller b_boot
may be advisable.
The percentile intervals are centred on the point estimate for the loadings, the
factor correlations, the structure coefficients, and the residuals, but not for
the indices derived from the chi-square (RMSEA, AIC, BIC, and ECVI): a
resample already carries the model's own misfit, so those intervals ride upward and
can even place the point estimate below their own lower bound. Read them as a spread
rather than a range for the point estimate; correcting that shift needs resampling
from a population transformed to fit the model (Bollen & Stine, 1992), which is not
what this bootstrap does. CFI and TLI are not affected, being ratios in which the
baseline chi-square shifts along with the model one.
The analytic methods ("information" and "sandwich") are not available with the
"promax" or "simplimax" rotations, which have no supported analytic route for the
rotated standard errors; use "np-boot" there. Under cor_method = "fiml",
"information" and "sandwich" instead return, for estimator = "ML" or "ULS",
corrected two-stage sandwich standard errors (Yuan & Bentler, 2000; Savalei & Bentler,
2009). estimator = "PAF" carries no Stage-2 weight to build the sandwich from, so use
se = "np-boot" there instead.
For ML and ULS, efa_fit() returns the model chi-square (with its p-value and degrees of
freedom), the Comparative Fit Index (CFI; Bentler, 1990), the Tucker-Lewis Index (TLI,
also called the non-normed fit index; Tucker & Lewis, 1973), the Root Mean Square Error
of Approximation (RMSEA) with its 90% confidence interval (Browne & Cudeck, 1992), the
Akaike and Bayesian Information Criteria (AIC, BIC), the Expected Cross-Validation Index
(ECVI; Browne & Cudeck, 1989), the Root Mean Squared Residual (RMSR), the Standardized
Root Mean Squared Residual (SRMR; Bentler, 1995), and the common-part-accounted-for
(CAF) index (Lorenzo-Seva, Timmerman, & Kiers, 2011). They come with the
independence-baseline statistics chi_null, df_null, and p_null. On the unscaled ML
and ULS paths chi_null is Bartlett's test of sphericity; a scaled chi-square, and the
two-stage statistic of a cor_method = "fiml" fit, each carry their own baseline
instead.
The degrees of freedom depend on the number of variables and factors; the baseline
degrees of freedom df_null are \(p(p - 1)/2\) for \(p\) variables.
RMSR is the root mean square of the off-diagonal residuals; SRMR rescales it by a
fixed factor that depends only on the number of variables. The print and summary
methods show SRMR, not RMSR; RMSR remains in the returned object for backward
compatibility. Both are computed over the same residuals, so an unavailable residual
leaves both NA. (psych's rms uses a different divisor and equals
RMSR\(/\sqrt{2}\); the two are not directly comparable.)
The model chi-square is the Bartlett-corrected discrepancy (matching
stats::factanal() for ML). For ULS it is the same maximum-likelihood discrepancy,
evaluated at the ULS-fitted solution, as in psych::fa() -- not the least-squares
criterion lavaan reports as its standard ULS test statistic, so the two are not
comparable. AIC and BIC are built on this chi-square and can therefore be negative;
ECVI (built on the same chi-square plus a non-negative penalty) cannot. Because of the
Bartlett correction, ECVI differs slightly from the uncorrected Browne-Cudeck form
reported by lavaan and Mplus (their AIC/BIC use an unrelated log-likelihood-based
formula, so they are not comparable). On the unscaled ML/ULS path, CFI and
TLI are computed on a slightly different discrepancy scale than the reported
chi-square, so you cannot recompute one from the other by hand there; on the scaled
(sandwich) and FIML paths, the reported chi and chi_null are exactly the pair the
indices use.
Which indices are reported depends on the estimator:
ML and ULS compute the full set above.
PAF returns only the descriptive residual indices (RMSR, SRMR, CAF) and df; the
printed model-fit block shows CAF and SRMR. The chi-square-based indices are NA,
because PAF minimises no discrepancy.
DWLS by default returns only RMSR, SRMR, CAF, and df, because the ordinary
maximum-likelihood discrepancy is not its fit function. When se = "sandwich", a
scaled chi-square and the CFI, TLI, and RMSEA derived from it are reported (AIC and
BIC remain NA); that statistic is a two-stage correction applied to the polychoric
correlation residuals (Browne, 1984), not identical to the full WLSMV test of
lavaan or Mplus, which also corrects for the response thresholds.
cor_method = "fiml" (with ML or ULS) reports two-stage-corrected statistics
(Yuan, Marshall, & Bentler, 2002); AIC, BIC, and ECVI are left NA, as for any
scaled chi-square. The correction itself can be degenerate -- typically with a small
sample, a high proportion of missing values, or near-collinear variables -- in which
case an uncorrected likelihood-ratio statistic is reported in its place with a
warning, and the print methods label that line uncorrected rather than scaled;
read its p-value and the CFI, TLI, and RMSEA derived from it as indicative only.
Beyond the estimator, the chi-square and everything derived from it are NA whenever
the statistic itself is undefined: when N is not supplied, when the model is
underidentified (a negative df), and when a positive N is too small relative to the
number of variables and factors for the small-sample correction to remain valid. Each
case raises its own warning. The residual summaries (RMSR, SRMR, CAF) and the degrees
of freedom are still returned there, but residual size does not establish that a model
is identified: below zero degrees of freedom a near-zero residual is an artefact of
over-parameterisation, not close fit.
Whenever the chi-square is a scaled one (se = "sandwich", or a cor_method = "fiml"
fit whose correction could be formed), AIC, BIC, and ECVI are NA; see the
fit_indices entry in Value for the additional components then returned. AIC, BIC,
and ECVI are NA on every cor_method = "fiml" fit, including the uncorrected
fallback above. Lorenzo-Seva, Timmerman, and Kiers (2011) describe CAF as ranging from
0 to 1, with values near 1 indicating close fit; that does not hold here, where a
well-fitting model produces a CAF near 0.5, not near 1. Read it as a relative rather
than an absolute measure.
Not every estimator, rotation, standard-error, and correlation method can be combined:
Estimator and correlation method. estimator = "DWLS" requires ordinal data with
cor_method = "poly" or "tetra". cor_method = "fiml" works with PAF, ML, and ULS
(not DWLS) and needs raw data with missing values.
Standard errors. se = "information" requires estimator = "ML" and
cor_method = "pearson" or "fiml", and can be computed from a correlation matrix
when N is supplied. se = "sandwich" requires raw data,
with either a polychoric/tetrachoric cor_method (ML, ULS, or DWLS) or a Pearson
cor_method (ML or ULS); it is not available for PAF. Under cor_method = "fiml",
"information" and "sandwich" are available for ML and ULS only and both return the
corrected two-stage sandwich. se = "np-boot" requires raw data and works with any
estimator, rotation, and correlation method. Neither "information" nor "sandwich"
is available with the "promax" or "simplimax" rotations.
Fit indices. The chi-square-based indices are available for ML and ULS (and, as
scaled statistics, for cor_method = "fiml" and for DWLS with se = "sandwich"); PAF
and DWLS otherwise report only the descriptive residual indices.
estimate_control() and rotate_control() for the estimation and rotation
tuning knobs. efa_retain() for choosing n_factors, and efa_scores(),
efa_reliability(), efa_schmid_leiman(), and efa_compare() for working with the
fitted solution.
Other factor analysis:
efa_average(),
efa_group(),
efa_mi(),
plot.efa_group(),
print.efa_group()
# Principal axis factoring with oblimin rotation
mod_oblimin <- efa_fit(test_models$baseline$cormat, n_factors = 3, N = 500,
rotation = "oblimin")
mod_oblimin
summary(mod_oblimin)
# ML estimation with oblimin rotation
mod_oblimin <- efa_fit(test_models$baseline$cormat, n_factors = 3, N = 500,
estimator = "ML", rotation = "oblimin")
mod_oblimin
summary(mod_oblimin)
# Tuning knobs are supplied through the control objects. Here the SPSS preset is
# used for the estimation and rotation, with the maximum PAF iterations raised.
mod_spss <- efa_fit(test_models$baseline$cormat, n_factors = 3, N = 500,
rotation = "promax",
estimate_control = estimate_control(type = "SPSS", max_iter = 500),
rotate_control = rotate_control(type = "SPSS"))
mod_spss
# Analytic (expected-information) standard errors for the above
ML_info <- efa_fit(test_models$baseline$cormat, n_factors = 3, N = 500,
estimator = "ML", rotation = "oblimin", se = "information")
ML_info
summary(ML_info)
# \donttest{
# Robust (sandwich) standard errors and a scaled chi-square for ordinal raw data.
# These need a polychoric/tetrachoric correlation method and estimator ML, ULS, or DWLS.
DWLS_rob <- efa_fit(DOSPERT_raw, n_factors = 6, cor_method = "poly",
estimator = "DWLS", rotation = "oblimin", se = "sandwich")
DWLS_rob
summary(DWLS_rob)
# The same robust SEs and scaled chi-square for continuous data: a Pearson
# correlation with estimator ML or ULS (the fourth-moment ADF covariance).
ML_rob <- efa_fit(GRiPS_raw, n_factors = 1, cor_method = "pearson",
estimator = "ML", rotation = "none", se = "sandwich")
ML_rob
summary(ML_rob)
# }
# \donttest{
# Two-stage FIML correlations from raw data with missing values: the saturated
# multivariate-normal moments are EM-estimated (assuming the data are missing at
# random) and the standardized covariance is analysed.
x_miss <- GRiPS_raw
x_miss[cbind(1:20, 1)] <- NA
efa_fiml <- efa_fit(x_miss, n_factors = 1, estimator = "ML", cor_method = "fiml")
efa_fiml
# }
if (FALSE) {
# Bootstrap standard errors from raw data, reproducible via a fixed seed and run
# in parallel across replicates. future::plan() returns the plan it replaces, so
# on.exit() puts the session back as it was -- also if the fit fails.
efa_boot <- local({
old_plan <- future::plan(future::multisession, workers = 2)
on.exit(future::plan(old_plan), add = TRUE)
efa_fit(GRiPS_raw, n_factors = 1, estimator = "PAF", rotation = "none",
se = "np-boot", b_boot = 1000, seed = 42)
})
}
Run the code above in your browser using DataLab