forum.smartpls.com

Imagine, your dataset includes both a formative AND a reflective operationalization of the SAME construct (each operationalization consists of a block of 5 indicators).

Is it possible to use both blocks (that is the formative block and the reflective block) within the SAME model? Or are there any problems if the same contruct is operationalized in two ways (e.g. due to multicollinearity etc.)?

Thanks in advance for any help!!!

Chris

Hi Chris,

it seems redundant, in fact, there is the "redundancy analisys" where the LV (formative) is conected with LV(reflective).
In http://disc-nt.cba.uh.edu/chin/icis2000plstalk.pdf you could see a exemple in the slides 38-40. The explanation of this example you could find in "CHIN, W. W. The Partial least squares approach for SEM. In MARCOULIDES, George A. (Ed.) Modern Methods for business research. 1998, p.321-329".

I hope it helps.

Best regards.

Hi Chris,

But note that this type of redundancy analysis is distinct from a larger theoretical model. If you try to include both LVs within a single structural model, you'll surely run into multcollinearity issues at the LV level.

I would suggest you select the set of indicators (formative or reflective) that make the most theoretical sense within your larger model and use only that set.

John

G'day all,

I have a similar problem concerning the use of formative and reflective indicators for the same construct. The aim of my research is to develop a scale for a construct. Different qualitative techniques (expert interviews etc.) only generated formative indicators to measure the construct. In order to be able to say if the construct is measured appropriately by the formative indicators I decided to use theoretically plausible reflective indicators as well (as in the discussed redundancy analysis) to determine the error term.
The model includes only one more related construct to check for nomogical validity, hence, in my case it's not a much larger model.
Is it still problematic to use both formative and reflective indicators?

Thanks very much in advance

Sebastian

Hi,

1) If your model seems like this LV1formative => LV1reflec => LV2
Probably you will not have problems, but

2) If your model seems LV1formative => LV2 <= LV1reflec
Surely you will have problem with collinearity (correlation between independent variables - exogenous) as Prof Sailors remembered us above.

Best Regards.

Bido