Construct validation - CFA

[hbviana]

I am using PLS in order to analyse a scale that I am validating in Brazil. Which tests I need to do?
I construct a model, run and more than 50% of my items (variables) present values less 0,50. I eliminated them and run again. Now I obtained better values, but my AVE stayed less than 0,50. My Cronbachs Alpha, is fine, and my communality reliability too. I really need to cut off every variable with value less than 0,50? My scale are too small this way.
I need sugestions....

Thanks a lot

Helena

[Diogenes]

Hi Helena,
these cutoffs are practical rules, and sometimes we have some LV with values below of these cutoffs.
In your case, the problem that I see is that you are modifying the model to adapt it to your data (explorative context), and when you want to run the model with another sample (validation sample - confirmatory context), probably the results will be worst.
At this moment I don´t have a better suggestion than to do a revision in your scale (content, validation with experts, and with people of your sample), before reaplly it again.
Best regards.
Bido

[hbviana]

Hi Professor Bido

Thanks a lot. I appreciate your answer and I intend to do what your advice me.
best regards

Helena

[rayouby]

Hi Helena,

Have you considered that your constructs (or some parts of them) might be formative constructs and should be modeled as such. This is perhaps why you have so many items needing to be dropped.

Check the last announcement message on the forums for some references to formative constructs.

Here is another paper which I found helpful:

Petter, Stacie, Detmar Straub, Arun Rai (2007). “Specifying Formative Constructs in Information Systems Research”, MIS Quarterly, (31:4) pp. 623-656, December 2007.

Best regards,
Reem

[hbviana]

Thanks a lot.
I will read the paper and I am thinking about what you said.

Best regards.

Helena