R: Summary of a Unidimensional Confirmatory Factor Analysis

summary_ucfa {quest}

R Documentation

Summary of a Unidimensional Confirmatory Factor Analysis

Description

summary_ucfa provides a summary of a unidimensional confirmatory factor analysis on a set of variables/items. Unidimensional meaning a one-factor model where all variables/items load on that factor. The function is a wrapper for cfa and returns a list with four vectors/matrices: 1) model info, 2) fit measures, 3) factor loadings, 4) covariance/correlation residuals. For details on all the cfa arguments see lavOptions.

Usage

summary_ucfa(
  data,
  vrb.nm,
  std.ov = FALSE,
  std.lv = TRUE,
  ordered = FALSE,
  meanstructure = TRUE,
  estimator = "ML",
  se = "standard",
  test = "standard",
  missing = "fiml",
  fit.measures = c("chisq", "df", "tli", "cfi", "rmsea", "srmr"),
  std.load = TRUE,
  resid.type = "cor.bollen",
  add.class = TRUE,
  ...
)

Arguments

`data`	data.frame of data.
`vrb.nm`	character vector of colnames from `data` providing the variables/items
`std.ov`	logical vector of length 1 specifying if the variables/items should be standardized
`std.lv`	logical vector of length 1 specifying if the latent factor should be standardized resulting in all factor loadings being estimated. If FALSE, then the first variable/item in `data[vrb.nm]` is fixed to a factor loading of 1.
`ordered`	logical vector of length 1 specifying if the variables/items should be treated as ordered categorical items where polychoric correlations are used.
`meanstructure`	logical vector of length 1 specifying if the mean structure of the factor model should be estimated. This would be the variable/item intercepts (and latent factor mean if `std.lv` = FALSE). Note, this must be true to use Full Information Maximum Likelihood (FIML) to handle missing data via `missing` = "fiml".
`estimator`	character vector of length 1 specifying the estimator to use for parameter estimation. Popular options are 1) "ML" = maximum likelihood estimation based on the multivariate normal distribution, 2) "DWLS" = diagonally weighted least squares which uses the diagnonal of the weight matrix, 3) "WLS" for weighted least squares whiches uses the full weight matrix (often results in computational problems), 4) "ULS" for unweighted least squares that doesn't use a weight matrix. "DWLS", "WLS", and "ULS" can each be used with ordered categorical items when `ordered` = TRUE.
`se`	character vector of length 1 specifying how standard errors should be calculated. Popular options are 1) "standard" for conventional standard errors from inverting the information matrix, 2) "robust.sem" for robust standard errors, 3) "robust.huber.white" for sandwich standard errors.
`test`	character vector of length 1 specifying how the omnibus test statistic should be calculated. Popular options are 1) "standard" for the conventional chi-square statistic, 2) "Satorra-Bentler" for the Satorra-Bentler test statistic, 3) "Yaun.Bentler.Mplus" for the version of the Yuan-Bentler test statistic that Mplus uses, 4) "mean.var.adjusted" for a mean and variance adjusted test statistic, 5) "scaled.shifted" for the version of the mean and variance adjusted test statistic Mplus uses.
`missing`	character vector of length 1 specifying how to handle missing data. Popular options are 1) "fiml" = Full Information Maximum Likelihood (FIML), 2) "pairwise" = pairwise deletion, 3) "listwise" = listwise deletion.
`fit.measures`	character vector specifying which model fit indices to include in the return object. The default option includes the chi-square test statistic ("chisq"), degrees of freedom ("df"), tucker-lewis index ("tli"), comparative fit index ("cfi"), root mean square error of approximation ("rmsea"), and standardized root mean residual ("srmr"). Note, if using robust corrections for `se` and `test`, you will probably want to call the scaled versions of model fit indices (e.g., "chisq.scaled"). See `fitMeasures` for details.
`std.load`	logical vector of length 1 specifying whether the factor loadings included in the return object should be standardized (TRUE) or not (FALSE).
`resid.type`	character vector of length 1 specifying the type of covariance/correlation residuals to include in the return object. Popular options are 1) "raw" for conventional covariance residuals, 2) "cor.bollen" for conventional correlation residuals, 3) "cor.bentler" for correlation residuals that standardizes the model-implied covariance matrix with the observed variances, 4) "standardized" for conventional z-scores of the covariance residuals.
`add.class`	logical vector of length 1 specifying whether the lavaan classes should be added to the returned vectors/matrices (TRUE) or not (FALSE). These classes do not change the underlying vector/matrix and only affect printing.
`...`	any other named arguments available in the `cfa` function. See `lavOptions` for the list of arguments.

Value

list of vectors/matrices providing statistical information about the unidimensional confirmatory factor analysis. If add.class = TRUE, then the elements have lavaan classes which affect printing (except for the first "model_info" element which always is just an integer vector). The four elements are:

model_info: integer vector providing model information. The first element "converged" is 1 if the model converged and 0 if not. The second element "admissible" is 1 if the model is admissible (e.g., no negative variances) and 0 if not. The third element "nobs" is the number of observations used in the analysis. The fourth element "npar" is the number of parameter estimates.
fit_measures: double vector providing model fit indices. The number and names of the fit indices is determined by the fit.measures argument.
factor_load: 1-column double matrix providing factor loadings. The colname is "latent" and the rownames are the vrb.nm argument.
cov_resid: covariance/correlation residuals for the model. Note, even though the name has "cov" in it, the residuals can be "cor" if the argument resid.type = "cor.bollen" or "cor.bentler".

Examples


# types of models
dat <- psych::bfi[1:250, 16:20] # nueroticism items
summary_ucfa(data = dat, vrb.nm = names(dat)) # default
summary_ucfa(data = dat, vrb.nm = names(dat), estimator = "ML", # MLR
   se = "robust.huber.white", test = "yuan.bentler.mplus", missing = "fiml",
   fit.measures = c("chisq.scaled","df.scaled","tli.scaled","cfi.scaled",
      "rmsea.scaled","srmr"))
summary_ucfa(data = dat, vrb.nm = names(dat), estimator = "ML", # MLM
   se = "robust.sem", test = "satorra.bentler", missing = "listwise",
   fit.measures = c("chisq.scaled","df.scaled","tli.scaled","cfi.scaled",
      "rmsea.scaled","srmr"))
summary_ucfa(data = dat, vrb.nm = names(dat), ordered = TRUE, estimator = "DWLS", # WLSMV
   se = "robust", test = "scaled.shifted", missing = "listwise",
   fit.measures = c("chisq.scaled","df.scaled","tli.scaled","cfi.scaled",
      "rmsea.scaled","wrmr"))

# types of info
dat <- psych::bfi[1:250, 16:20] # nueroticism items
w <- summary_ucfa(data = dat, vrb.nm = names(dat))
x <- summary_ucfa(data = dat, vrb.nm = names(dat), add.class = FALSE)
y <- summary_ucfa(data = dat, vrb.nm = names(dat),
   std.load = FALSE, resid.type = "raw")
z <- summary_ucfa(data = dat, vrb.nm = names(dat),
   std.load = FALSE, resid.type = "raw", add.class = FALSE)
lapply(w, class)
lapply(x, class)
lapply(y, class)
lapply(z, class)