I was wondering whether anyone could help me with a query about using item parcels versus summary suscale scores in a CFA model. I have a model that has 5 factors each with 2 indicators that are subscale scores of different measures (3 measures in total). These subscales have been validated in the sample that I'm using.
I know that factors need to have 3 or more indicators to make the model identified when using CFA for one scale. However, I haven’t been able to find any literature on using multiple scales within a CFA and whether this makes a difference. I'm wondering if this would be any different as there is more variance within the subscale variables compared with individual items, or does this rule of three and above also apply? Does anyone have experience with this?
The other option I have is to create item parcels instead of using the subscale scores. That way I could have 3 or more item parcels under each of the 5 factors. However, there are also some who say that using item parcels is not great and may bias the results. Would this be a better option?
Any advice on what you think is the best option for model identification would be much appreciated!
Printer-friendly version
In data from pairs of relatives, the number of factors that can be estimated equals one fewer than the number of measures. For unrelated individuals, the maximum number of factors that can be estimated follows the sequence 3, 5, 6, 8, 9, 10, 12 etc (incrementing by 2 according to a triangular number series).
I surmise that you will be able to work with 2-indicator factors as long as the factors correlate, but there is still the possibility that it is not identified. In the case of data from pairs of relatives, the fact that both factor models have the same factor loadings provides additional degrees of freedom. mxCheckIdentification() should tell you whether the model is locally identified.