proper methode for 2nd order constructs??

[Salima]

Dear Smart-PlS-Users,

I have (one more) question regarding 2nd order constructs. I red in a PHD thesis a procedure for handling 2nd order constructs, that seems kind of unusual (but easy) to me:

She has one higher construct in her model, which contains 5 dimensions. She defined, that the 2nd order construct was formative and 4 1st order constructs were reflective (1 formative).
Then she did an exploratory factor analysis with the reflective 1st order constructs and choose the indicators with the highest loadings (1 per construct and keeps all the formative indictators). With these "new" composition of items she defined the 2nd order construct and used it for the pls-analysis.

Here is my question: is this a proper and reasonable approach to handle 2nd order constructs? To me it seems like a kind of dummy-2nd order construct.

Thanks for your opinions

Best,
Salima

[jjsailors]

She has one higher construct in her model, which contains 5 dimensions. She defined, that the 2nd order construct was formative and 4 1st order constructs were reflective (1 formative).
Then she did an exploratory factor analysis with the reflective 1st order constructs and choose the indicators with the highest loadings (1 per construct and keeps all the formative indictators). With these "new" composition of items she defined the 2nd order construct and used it for the pls-analysis.

Hi Salima,

I don't agree with this method. It suggests a view of constructs in PLS where constructs are reflective--they aren't, even under Mode A modeling. When measures do not reflect constructs, but define them, there is not compelling reason to use the single measure that correlates most highly with the construct in the second order factor rather than all of them.

Regards,
John

[Salima]

Thanks for your opinion, John!