bifactor indicators

[R Monica S]

Dear all,

I have a basic question regarding bifactor models.

Is it possible for some indicators to load only on the general factor?  For example..

GeneralFactor =~ A1 + A2 + B1 + B2 + C
FactorA =~ A1 + A2
FactorB =~ B1 + B2

I would be very grateful for your help!

[Chao Xu]

I would say methodologically yes because it is equivalent to a model where C loads on to factor A/B yet with its loading fixed to 0. It should be estimable. You can try.

But you would have to find a theoretical meaning for this conceptualization. Reviewer will surely be picky on this issue.

Chao

[R Monica S]

Dear Chao,

Thank you very much for your help and reply!

I am wondering whether dropping insignificant indicators (based on p-values of parameter estimates) from factors while retaining those indicators in the general factor is a usual procedure?  For example, in an established measurement, indicator C also loads onto FactorB, but in my data it is only a significant indicator for GeneralFactor.

GeneralFactor =~ A1 + A2 + B1 + B2 + C
FactorA =~ A1 + A2

FactorB =~ B1 + B2 (+ C)

In this case, shall I still keep C on FactorB, or drop it only from FactorB?
And if I do this, can I still compute bifactor indices (e.g., omega)?

It would be very helpful if you could kindly give me further suggestions.

[Chao Xu]

I assume by insignificant indicators, you mean indicators with very weak loadings. However, its loading on the general factor is very strong.

If this is the case, you must keep C onto the factor B in order to estimate coefficient omega and/or other statistical indices.

Or you could simply drop C off of both the factor B and the general factor. For me, I fail to see the reason of keeping C because it has little substantive meaning (weak loading on the substantive factor B).

Chao

[R Monica S]

Dear Chao,

Sorry for the late response and thank you for your help!  I understood that I must keep C on both of the factors to estimate coefficients indices.  Maybe the factor structure does not fit my data thus I should modify the model as you suggest.

Many thanks again!