Scale (version 1.0.4)

ItemAnalysis: Reliability and Validity Analysis

Description

Performs an item analysis based on item-scale correlations, and then conducts factor analysis with one factor. Reports Cronbach alpha and single factor loadings, while it returns the original analyses from the psych package.

Usage

ItemAnalysis(prep, method="spearman", fm="gls", 
nfactors=1, rcut= 0.3, score_type="z", exclude=c())

Arguments

prep

A ScaleData object pre-processed with PreProc.

nfactors

Number of factors to be extracted in validity analysis.

rcut

Lower bound for items' correlation to scale.

score_type

Type of standard scores to calculate ("z", "t", or "sten".)

method

Method to calculate the correlation matrix. Options are: "spearman" or "polychoric".

fm

Method for factor xtraction in the validity analysis.

exclude

Items to exclude from the analysis. Indices in the original order.

Value

A list of three objects. data is the dataset, passed on for other computations, rely is the output of the reliability analysis, and valid the output of the factor analysis:

data

The dataset used.

items

The item statements. If not provided value is NULL.

rely

A list of the following elements:

..alpha

Output of the psych::alpha function.

..k

Number of items

..title

Name of analysed object.

..suggest

List of 2: low_cor Items with low correlation to the rest of the scale, and a_drop Items whose deletion may improve reliability.

valid

A list of the following elements:

..model

Output of the psych::fa function

..method

character. The factor extraction method.

..loadings

numeric. The factor loadings

..kmo

list. KMO sampling adequacy statistics.

..bartlett

list. Bartlett's test of sphericity.

..scores

numeric. Factor scores (Standardized, see Details.)

Details

This function is no more than a wrap-up for psych package alpha and fa functions. Use ?psych::alpha and psych::fa for details.

Available method for correlations are "spearman" and "polychoric". Available methods for factor extraction are "minres", "wls", "gls", "pa", "ml", "minchi".

Defining number of factors is included for sake of completeness. The intended use of the function is a quick and error-proof validity measure, and not factor model fitting. Adjusting the number of factors can only serve to see if there is a better model fit with more than one factor. Scores will be calculated for the first factor only. Of course if you need to use this function as a wrapper for psych::fa, you can always extract the object with YOUROBJECT$valid$model.

Default scoring is the sum of the standardized values times the first factor loadings. T-scores translate these to have a mean of 50 and an SD of 10, and STen scores, a mean of 5.5 and an SD of 2.

Examples

data(Depression98)
depressionScale <- Scale(data=Depression98, 
                         orders=list(
                           c(16,19,11,9,1,17,5,18,4,8,2,12,
                             20,10,14,6,3,13,15,7),
                           c(1,18,4,15,7,8,3,14,20,6,19,16,
                             12,5,10,13,2,17,11,9)),
                         orders_id=c(
                           rep(1, 49),
                           rep(2, 49)),
                         reverse=c(3,4,13,14,18,20),
                         col_names= paste('q', 1:20, sep=''))
depressionScale

depressionPre <- PreProc(depressionScale)

depressionRel <- ItemAnalysis(depressionPre)
depressionRel

depressionRel <- ItemAnalysis(depressionPre, exclude=c(1, 3, 15, 13))
depressionRel