Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • dfactor and factor questions

    I am running factor and static dfactor models on the same set of variables. I was expecting the factor loadings of the former (estimated via ml) to match the coefficients of the latter, but that is not the case. Could you please explain why?

    Also, when one estimates a static factor model using dfactor, the predict command only produces non-zero values for the residuals (for factor and xbf a zero vector is computed). Is this correct?

    On a similar vein, when one runs factor with pcf and ml options, and one specifies more factors than there are eigen values above 1, factor still retains only as many factors as there are eigen values above 1. Why is this?

    Finally, could you please provide some guidance on how to interpret the lrtests for factor model using ml? The hand book (page 360[TS]) was not helpful. With the p-value of the chi_square statistic at 0.0087, the interpretation was that there was little evidence of more factors, but that there might be an additional factor.

    Many thanks,
    Carolina


  • #2
    Interesting questions. Hopefully somebody else will answer them better, but I'll give you some thoughts.
    • dfactor should differ a bit from factor, since it does take into account non-independence of observations. If the ICC were 0, then they should get the same estimates, but with a non-zero ICC, they should differ.
    • I know nothing about limitations on the predict command for dfactor. All I can say is that commands that make it into the main body of Stata have been heavily vetted.
    • Strange thing about factor. The "factors" option, despite the documentation, behaves as you describe. However, by manipulating the "mineigen" option, you can get factors with eigenvalues<1. Quirky, counter-intuitive, but functional.
    • Interpretation of the ML LR test, ignore it, see below.
    Move it to SEM! SEM is easy enough with Stata 12 and 13 that you should be able to do what you want, and more.

    Comment

    Working...
    X