R/ROOTFIT.R

In EFA.dimensions: Exploratory Factor Analysis Functions for Assessing Dimensionality

ROOTFIT <- function (data, corkind='pearson', Ncases=NULL, extraction='PAF', verbose = 'TRUE', factormodel) {

# deprecated  
if (!missing(factormodel))  extraction <- factormodel

data <- MISSING_DROP(data)

Nvars <- ncol(data)

# the analyses will be run for 1-factor to (Nvars * .6)-factor solutions
# because problems tend to occur for very small factors
# Nroots <- floor(Nvars * .6)
Nroots <- Nvars

# set up cormat
cordat <- setupcormat(data, corkind=corkind, Ncases=Ncases)
cormat <- cordat$cormat
ctype  <- cordat$ctype
Ncases <- cordat$Ncases

 
if (extraction == 'PCA' | extraction == 'pca' | extraction == 'IMAGE' | extraction == 'image') {

	fits <- matrix(-9999,Nroots,5)

	fits[,1] <- 1:Nroots

	evals <- eigen(cormat)$values

	if (min(evals) < 0) {
	     message('\nThe correlation matrix is not positive definite, which means that')
	     message('there are eigenvalues less than zero. Expect problems.')
	}
	
	fits[,2] <- evals[1:Nroots]

	for (root in 1:Nroots) {
       
		if (extraction == 'PCA' | extraction == 'pca') 
			Output <- PCA(cormat, corkind=corkind, Nfactors=root, Ncases=Ncases, 
			              rotation='none', verbose=FALSE)
		
		if (extraction == 'IMAGE' | extraction=='image') 
			Output <- EFA(data=cormat, extraction='image', Nfactors=root, Ncases=Ncases, 
			              rotation='none', verbose=FALSE)		
		
		fits[root,3:5] <- unlist(Output$fit_coefs[c('RMSR', 'GFI', 'CAF')])		
	}
	colnames(fits) <- c('Root','Eigenvalue','RMSR','GFI','CAF')
	rownames(fits) <- matrix((''),nrow(fits),1)
}


if (extraction != 'PCA' & extraction != 'pca' & extraction != 'IMAGE' & extraction != 'image') {

	fits <- matrix(-9999,Nroots,13)

	fits[,1] <- 1:Nroots

	evals <- eigen(cormat)$values

	if (min(evals) < 0) {
	     message('\nThe correlation matrix is not positive definite, which means that')
	     message('there are eigenvalues less than zero. Expect problems.')
	}

	fits[,2] <- evals[1:Nroots]

	for (root in 1:Nroots) { 

		dof <- 0.5 * ((Nvars - root)^2 - Nvars - root) # The degrees of freedom for the model
		if (dof < 1) {
			fits <- fits[1:(root-1),]
			if (verbose == 'TRUE') {
				message('\nThe degrees of freedom for the model with ',
				    root,' factors is < 1 and so the procedure was stopped.')}
			break
		}
		
	essaye <- try(Output <- EFA(data=cormat, extraction=extraction, Nfactors=root, Ncases=Ncases, 
                                rotation='none', iterpaf=400, verbose=FALSE), silent=TRUE)		                                       
	if (inherits(essaye, "try-error")) {
		message('\nfactanal produced an error when Nfactors = ', root)
		fits <- fits[1:(root-1),]
		break
	}
				
	fits[root,3:13] <- unlist(Output$fit_coefs[c('RMSR', 'GFI', 'CAF','RMSEA','TLI',
		                                             'CFI','MFI','BIC','AIC','CAIC','SABIC')])
	}
	colnames(fits) <- c('Root','Eigenvalue','RMSR','GFI','CAF','RMSEA','TLI','CFI',
	                    'MFI','BIC','AIC','CAIC','SABIC')
	rownames(fits) <- matrix((''),nrow(fits),1)	
}


if (verbose == 'TRUE') {
	
	message('\n\nFit coefficients for N-factor solutions')

	message('\nType of correlations used in the analyses: ', ctype)

	message('\nExtraction Method: ', extraction)

	message('\nThe number of cases = ', Ncases)
	
	message('\nThe number of variables = ', Nvars,'\n') 

	print(round(fits,3))
}

return(invisible(fits))
}