Skip to contents

Fits an exploratory factor analysis (EFA) model from a synthesized correlation matrix extracted from a fitted mars object.

Usage

efa_from_synthesis(
  mars_object,
  n_factors = 1L,
  rotation = "varimax",
  num_obs = NULL,
  synthesis_method = c("model", "average", "weighted"),
  synthesis_transform = c("none", "fisher_z"),
  missing_corr = c("available", "pairwise", "em"),
  attenuation = c("none", "correct"),
  reliability = NULL,
  reliability_missing = c("error", "skip", "impute_mean", "assume_1"),
  pd_adjust = c("none", "eigen_clip", "nearpd"),
  pd_tol = 1e-08,
  se_method = c("none", "simulation", "bootstrap"),
  se_draws = 1000L,
  se_seed = NULL,
  vcov_source = c("model", "robust"),
  factor_retention = c("none", "parallel"),
  parallel_n = NULL,
  parallel_draws = 200L,
  ...
)

# S3 method for class 'efa_mars'
summary(object, ...)

# S3 method for class 'efa_mars'
print(x, digits = max(3, getOption("digits") - 3), ...)

Arguments

mars_object

A fitted object returned by mars.

n_factors

Number of latent factors for EFA.

rotation

Rotation function or function name, or "none". Rotation is applied after unrotated factanal extraction.

num_obs

Optional descriptive size; it does not enable a single-sample chi-square test for synthesized correlations.

synthesis_method

Correlation synthesis method: "model", "average", or "weighted".

synthesis_transform

Optional pooling transform for "average"/"weighted": "none" or "fisher_z".

missing_corr

Missing-correlation handling for synthesized matrices. One of "available", "pairwise", or "em".

attenuation

Attenuation-correction mode for synthesized correlations. One of "none" or "correct".

reliability

Reliability input used when attenuation = "correct".

reliability_missing

Strategy for missing reliability values. One of "error", "skip", "impute_mean", or "assume_1".

pd_adjust

Positive-definite repair for synthesized correlations: "none", "eigen_clip", or "nearpd".

pd_tol

Minimum eigenvalue tolerance used by pd_adjust = "eigen_clip".

se_method

"none" (default) reports descriptive EFA. "simulation" or "bootstrap" propagates synthesis uncertainty through repeated EFA fits.

se_draws

Number of uncertainty draws, an integer >= 50.

se_seed

Optional seed; the caller's RNG state is restored.

vcov_source

Model-based or robust pooled-correlation covariance.

factor_retention

"none" or "parallel" for optional descriptive PCA parallel-analysis sensitivity; it does not select n_factors automatically.

parallel_n

User-specified independent-unit count for the parallel-analysis reference distribution. Do not automatically substitute total participant N.

parallel_draws

Number of independent normal reference matrices, >= 50.

...

Additional arguments passed to print.factanal().

object

An object of class "efa_mars".

x

An object of class "efa_mars".

digits

Number of digits used when printing numeric results.

Value

An object of class efa_mars, with descriptive loadings, factor_correlations, communalities, implied_matrix and a fit_note. Oblique solutions use the factor correlation matrix in communalities and implied covariances. Sample-size-based chi-square tests are not available for a synthesized matrix.

Details

EFA uses maximum-likelihood extraction from stats::factanal() with no single-sample chi-square test. For oblique rotation, communalities are the diagonal of L Phi L', and the implied matrix adds uniquenesses. boundary flags uniquenesses near factanal's default lower bound; inspect the selected factor count, rotation and residual matrix when this occurs.

Loading uncertainty is conditional on the chosen count and rotation. Named bootstrap correlation draws are used when available; otherwise independent studies/clusters are resampled and the stored synthesis specification is refit. Simulation uses the matched joint normal approximation to pooled correlations. Each admissible loading matrix is aligned to the reference by optimal factor permutation (up to seven factors) and sign matching; larger solutions use greedy matching. Substantial rotational ambiguity requires sensitivity analysis. SEs and loading covariance require at least 50 successful fits and an 80 percent success rate. Alternative pooling or repaired point matrices have no matched sampling covariance, so uncertainty is withheld. Draw matrices are not repaired.

Parallel analysis compares the pooled matrix eigenvalues with the 95th percentile from independent normal reference matrices. It is a descriptive PCA sensitivity at the explicitly chosen unit count, not a calibrated factor-retention test for synthesized correlations. Assess different counts and substantive interpretability.

Examples

# \donttest{
fit <- mars(
  data = becker09,
  studyID = "ID",
  effectID = "numID",
  sample_size = "N",
  effectsize_type = "cor",
  varcov_type = "weighted",
  variable_names = c(
    "Cognitive_Performance",
    "Somatic_Performance",
    "Selfconfidence_Performance",
    "Somatic_Cognitive",
    "Selfconfidence_Cognitive",
    "Selfconfidence_Somatic"
  )
)
#> Warning: Hessian inversion failed: Hessian is singular, indefinite, or numerically unidentified.
efa_from_synthesis(fit, n_factors = 1)
#> EFA From Correlation Synthesis
#> Synthesis: model | Transform: none | PD adjust: none 
#> EFA describes the pooled matrix. A single sample size does not calibrate a factor-model chi-square test for synthesized correlations; the test is unavailable. 
#> 
#> Call:
#> stats::factanal(factors = factors, covmat = covmat, n.obs = NA,     rotation = rotation)
#> 
#> Uniquenesses:
#>    Performance      Cognitive        Somatic Selfconfidence 
#>         0.9841         0.4048         0.5154         0.6527 
#> 
#> Loadings:
#>                Factor1
#> Performance    -0.1260
#> Cognitive       0.7715
#> Somatic         0.6961
#> Selfconfidence -0.5893
#> 
#>                Factor1
#> SS loadings     1.4429
#> Proportion Var  0.3607
#> 
#> The degrees of freedom for the model is 2 and the fit was 0.0514 
# }