nomo_esem() fits a measurement model as exploratory structural equation
modeling (ESEM), in which every item may load on every factor, and fits the
same factors as an independent-clusters CFA, in which each item loads on one
factor only. It compares their fit, their factor correlations, and the
cross-loadings the CFA fixes at zero.
Usage
nomo_esem(
model,
data,
rotation = c("target", "geomin"),
ordered = NULL,
estimator = NULL,
missing = NULL,
guidance = nomo_defaults()
)Arguments
- model
A measurement model in which each indicator loads on one factor: a lavaan model string or a
nomo_model()object. It defines the factors, the items, and the target.- data
A data frame with the indicators.
- rotation
"target"(default) or"geomin".- ordered
Optional character vector of ordered indicators.
- estimator, missing
Optional lavaan
estimatorandmissingoptions.- guidance
Guidance settings from
nomo_defaults(); itsefa_loading_referenceandefa_crossloading_referenceare used, and they are checked before any model is fitted.
Value
A nomo_esem object. The fields to read are:
loadings: each item's standardized loading on each factor in the ESEM, whether it is the item's main loading or a cross-loading, its standard error and p value, and the CFA's loading for main loadings.factor_correlations: each pair of factors' correlation (factor1,factor2) in the ESEM and the CFA, and their difference.models: both models' chi-square, degrees of freedom, p value, CFI, TLI, RMSEA, SRMR, AIC, and BIC. With a robust estimator, the chi-square is scaled and the indices are robust, as innomo_cfa().comparisons: the chi-square difference test of the CFA against the ESEM, withchisq_diff,df_diff, andp_value. It is the likelihood-ratio test under ML, the scaled difference test (Satorra & Bentler, 2001) under a robust ML estimator, and the scaled-and-shifted test (Satorra, 2000) under WLSMV, aslavaan::lavTestLRT()chooses.nanddata_n: the cases the models used and the rows ofdata.fits: the fittedlavaanmodels, namedESEMandCFA.engine_warnings: the warningslavaanraised, as a list with elementsESEM,CFA, andcomparison(the difference test).decision_log.
Other fields record the call and the settings used. They may change
between releases and are not part of the stable interface (see
?nomologR).
print() shows both models' fit, the difference test, the factor
correlations in each model, and the flags. summary() adds the ESEM
loadings beside the CFA's and each flag's recommendation.
Details
Why ESEM. A CFA fixes every cross-loading at zero. When items in fact have small cross-loadings, as items of related constructs usually do, the constraint pushes that shared variance into the factor correlations, which are then inflated, and into misfit (Asparouhov & Muthén, 2009). ESEM estimates the cross-loadings instead, within a structural equation model, so fit, standard errors, and further structure remain available (Marsh, Morin, Parker, & Kaur, 2014).
Rotation. With an a priori structure, Marsh et al. (2014) recommend
target rotation: each item's loading on its intended factor is free, and its
cross-loadings are targeted towards zero without being fixed there. This is
Browne's (2001) partially specified target rotation, "target" here, built
from model. Geomin rotation ("geomin"), which uses no target, is
available for a more exploratory reading. The solution depends on the
rotation, as any exploratory factor solution does.
Reading the comparison. Marsh et al. (2014) suggest preferring ESEM when it fits better and its factor correlations are lower, which shows that the CFA's zero cross-loadings distort the structure; otherwise the CFA is the more parsimonious account. Fit is compared on TLI and RMSEA, which penalize the ESEM's extra parameters, so the CFA can fit better on them. The CFA is nested in the ESEM, and their chi-square difference test is reported too, but in large samples it rejects trivial misfit. The decision log also flags cross-loadings at or above the guidance's cross-loading reference and main loadings below its loading reference.
Admissibility. Both models must converge; otherwise the function stops
and names the model that did not. The warnings lavaan raises are kept, and
a negative variance, a standardized loading above 1, or a latent correlation
above 1 in either model is recorded as a concern, as nomo_cfa() records
it. Cases lavaan did not use, such as incomplete cases without
missing = "fiml", are reported.
References
Asparouhov, T., & Muthén, B. (2009). Exploratory structural equation modeling. Structural Equation Modeling, 16(3), 397-438. doi:10.1080/10705510903008204
Browne, M. W. (2001). An overview of analytic rotation in exploratory factor analysis. Multivariate Behavioral Research, 36(1), 111-150. doi:10.1207/S15327906MBR3601_05
Marsh, H. W., Morin, A. J. S., Parker, P. D., & Kaur, G. (2014). Exploratory structural equation modeling: An integration of the best features of exploratory and confirmatory factor analysis. Annual Review of Clinical Psychology, 10, 85-110. doi:10.1146/annurev-clinpsy-032813-153700
Satorra, A. (2000). Scaled and adjusted restricted tests in multi-sample analysis of moment structures. In R. D. H. Heijmans, D. S. G. Pollock, & A. Satorra (Eds.), Innovations in multivariate statistical analysis (pp. 233-247). Springer. doi:10.1007/978-1-4615-4603-0_17
Satorra, A., & Bentler, P. M. (2001). A scaled difference chi-square test statistic for moment structure analysis. Psychometrika, 66(4), 507-514. doi:10.1007/BF02296192
Examples
model <- nomo_model(list(
Agency = paste0("ag", 1:4),
Persistence = paste0("pe", 1:4)
))
es <- nomo_esem(model, nomo_demo_network)
es
#> <nomo_esem> ESEM beside its CFA
#> Asparouhov and Muthén (2009); Marsh et al. (2014).
#> Rotation: target | Cases: 800 | Factors: 2
#>
#> Fit
#> Model Chi-square df p CFI TLI RMSEA SRMR
#> ESEM 14.93 13 .312 .999 0.998 0.014 0.010
#> CFA 18.52 19 .488 1.000 1.000 0.000 0.015
#> CFA vs. ESEM: Delta chi-square(6) = 3.58, p = .733.
#>
#> Factor correlations
#> Factors ESEM CFA Difference
#> Agency with Persistence .45 .46 .00
#>
#> ESEM = exploratory structural equation modeling; CFA = confirmatory factor
#> analysis; CFI = comparative fit index; TLI = Tucker-Lewis index; RMSEA = root
#> mean square error of approximation; SRMR = standardized root mean square
#> residual; df = degrees of freedom.
#>
#> Cross-loadings are estimated, not fixed at zero. The comparison is evidence
#> for choosing a model, not a verdict.
#>
#> See summary(x) for the loadings and each flag's recommendation and
#> nomo_table(x, "models") for every fit index.
nomo_table(es, "factor_correlations")
#> # A tibble: 1 × 5
#> factor1 factor2 esem cfa difference
#> <chr> <chr> <dbl> <dbl> <dbl>
#> 1 Agency Persistence 0.455 0.457 -0.00281