With a view to SEM, I want to evaluate a measurement model with six items (expressing behaviour frequencies) and 596 observations. The items are heavily skewed and moderately to weakly correlated (Spearman rho 0.01-0.5). To assess model reliability I have so far used Cronbach's alpha which is easily obtained through the alpha command. However, McDonald's omega may be a more appropriate statistic. How can I calculate it, and what could be the citation to justify it? I've found a recommendation to calculate it as [(sum of factor loadings)^2 / (sum of uniquenesses + (sum of factor loadings)^2)], but I can't figure out if the sum of loadings should include only the first factor, or several factors, and I still lack a relevant citation.

Best regards,
Jan