From: Fabio Zona <fabio.zona@un...> - 2010-12-13 00:33:13
I have this model, based on a survey:
- dep var = four items
- indep var_1 = single item; it is a "percentage" given by the respondent
- indep var_2 = the ratio of two other single-items; both items are "figures" given by the repondent
- other control variables.
Since I have one respondent, I have to run a confirmatory factor analysis to exclude the risk of common method variance.
1. a first method would be to load all items to one single factor,and examine both factor loadings and global fit measures.
---- do you agree that, for indep var_2, I should examine survey items separately as responded by the raters, and NOT their ratio (as calculated by me)?
---- I find that items of the dep variable have these standardized factor loadings: two f.loads. well below .7; one f.load .76; the fourth f.load .80); the other three items related to the indep vars have f.loadings of aproxx 0. Can I say that these results are in favour of NO common method variance ? (even if two f.load are just above .7)
---- the overall global fit is poor: however, not so poor as I expected! Is this related to - maybe - the fact that the single-item variables add small variance to the dep variable latent ?
2. a second method would be to compare global fit indexes for the one single factor model (THE ONE ABOVE) and a four-factor model, that is, a second model that consider all the variables separately.
If there is a significant difference in global fit indexes, than one can exclude the problem of common method variance.
-----do you ever heard about this second method of comparing differences in global fit of single vs multiple factor models
-----since in my case the indep variables are single item, should I consider them only as rectangles (observed variables) and with no error term?
-----and how this multiple-factor model would look like? I guess it is constituted by all possible correlation (double arrows) lines between the circle of the latent variable (related to items of the dep variable) and the three rectangles for the single items indep_variables. Do you agree?
-----if this model is correct, then I find a significant difference in the chi-square, but it is small, just of 5%: I expected large differences. How is this happening?
On a related, separated issue: does one know how to run a confirmatory factor analysis when the items are not normally distributed?
Thanks a lot