I am conducting a CFA with Lisrel, I have sample of 127 participants and 19 items with 4-point Likert scales that should load on 3 factors. I have a very bad CFA fits :( I went back to EFA to learn about the structure (with a different sample of 169 subjects). EFA results clearly suggested an oblique three-factor solution, and also the parallel analysis method confirmed the three-factor structure. The CFA factor loadings are highly significant, but not the overall fit. I have also tried alternative solutions (one-factor, two factors, and orthogonal three-factor model) with worst fits. Now my question is: Can I use the item parceling approach? I have created 4 parcels for the first latent factor, 3 parcels for the second factor, and 2 parcels for the third factor, by randomly averaging two items scores for each parcel. What do you think about it? Thanks, Tiziana
I have been working on a similar issue, but more particularly with a non normal distribution and small sample size (n=56). Item parceling can be done, but caution should be taken with two indicator approach (2 parcels). It is recommended that if you parcel, try to have at least 3, but as many as possible indicators per factor. Also, parceled items should be relatively unidimensional (smaller correlation between factors). How many items do you have loading onto the third factor? I would suggest give this article a read:
Hau, K. T., & Marsh, H. W. (2004). The use of item parcels in structural equation modelling: Non‐normal data and small sample sizes. British Journal of Mathematical and Statistical Psychology, 57(2), 327-351.