Hello Sarah,

I think that SPSS will only apply listwise deletion (e.g., use only cases with all responses) in the reliability subprogram, so it's not as if you have a lot of choice!

You can infer an estimate for reliability via loadings from Mplus or any SEM program, or most factor analysis subprograms. See this link for specific explanation (note the section in the paper on omega coefficient): Article Estimating reliability coefficient for multidimensional meas...

You could as well just compute the average covariance among items and apply the usual coefficient alpha: alpha = [k * cbar] / [vbar + (k - 1) * cbar], where k is number of items in the scale, cbar is average inter-item covariance, and vbar is average item variance. Mplus can likely furnish the estimated variance-covariance matrix after applying its imputation.

Good luck with your work.

Hello Sarah,

David made a very good suggestion. In fact, computing the composite reliability is the more appropriate reliability measure as it does only require unidimensionality (all indicators measure the same factor) instead of essential tau-equivalence like alpha.

This paper contains the formula which can be easily programmed in Excel. It is based on the standardized loadings and error variances.

Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18, 39-50.

All the best,

--Holger

Many thanks for the answers to my questions!

If I understand this correctly, it would not be an elegant, but a possible approach to remove (listwise deletion) the affected missing cases from my data set and then calculate my analyses in Mplus.

If I were to calculate reliability (the composite reliability) in this way, would this mean that all my cases with the missing values would remain in the data set? I do not exclude any cases this way? Is that correct? I want to lose as little data as possible...

This procedure is also possible for longitudinal SEM, isn't it?