Introduction to reliacoef

The reliacoef package is designed to compute and compare a wide range of unidimensional and multidimensional reliability coefficients commonly used in psychometrics and social science research.

1. Unidimensional Reliability

Unidimensional reliability is appropriate when a scale is assumed to measure a single underlying factor. The unirel() function provides a comprehensive suite of estimates, allowing researchers to compare multiple coefficients beyond the traditional Cronbach’s alpha.

Example: Comparing Estimates

You can pass a data frame or a covariance matrix to unirel().

# Using the included Graham1 dataset
result_uni <- unirel(Graham1)
result_uni

unirel() computes the following: - Coefficient Alpha: The most common but often criticized for strict assumptions. - Jöreskog’s Congeneric Reliability: Also known as Composite Reliability or Omega. - Feldt-Gilmer Coefficient: A practical alternative that maintains congeneric assumptions. - Ten Berge & Zegers’ mu series: A series of lower bounds that improve upon alpha.

2. Multidimensional Reliability

When a scale consists of several sub-dimensions, multidimensional reliability coefficients provide a more accurate picture of measurement quality. The multirel() function summarizes these estimates.

Defining the Structure (The `until` argument)

Multidimensional functions require the until vector to define the boundaries of sub-constructs. For example, if a 12-item scale has 4 sub-dimensions (3 items each), use until = c(3, 6, 9).

Summary Analysis with `multirel()`

We can use the provided Cho_multi dataset (a 12x12 covariance matrix) to see a global comparison.

# Comparing various multidimensional coefficients
result_multi <- multirel(Cho_multi, until = c(3, 6, 9))
result_multi

Individual Model Functions

You can also call specific models directly if your theory supports a particular structure:

bifactor(): Computes reliability based on a general factor and specific group factors.
second_order(): For hierarchical models where sub-factors load onto a higher-order factor.
stratified_alpha(): A weighted sum of alpha coefficients for sub-tests.
nunnally(): Uses the bottom-up approach to combine sub-test reliabilities.

# Analyzing general factor saturation via Bifactor model
bif_res <- bifactor(Cho_multi, until = c(3, 6, 9))
b_omega_h <- bif_res$omega_hierarchical

3. Testing Statistical Assumptions

Before selecting a reliability coefficient, it is critical to test whether the data fits the underlying model. The test.tauequivalence() function performs a chi-square difference test between the essential tau-equivalence model (required for alpha) and the congeneric model.

# Testing the assumption for Coefficient Alpha
test_res <- test.tauequivalence(Graham1)
test_res

If the p-value is significant, Jöreskog’s congeneric reliability is generally preferred over coefficient alpha.

4. The `reliacoef` Class

All primary functions in this package return an object of class reliacoef. These objects are designed for clarity: - Clean Output: A customized print method displays results in a formatted table. - CFA Details: For model-based coefficients, you can access detailed fit indices and parameter estimates using $fit_indices and $estimates.

# Accessing detailed CFA fit indices
bif_res$fit_indices