forum.smartpls.com

i am investigating a model with exogenous Formative HOC that has four Reflective LOCs. One of the LOC originally measured with three indicators, yet when i run the PLS algorithm, two of the indicators showed negative loadings ( < -0,40) while the other one showed high outer loading (0,8). this result on low composite reliability and AVE.

And then, i tried to remove the positive indicator leaving the two negative one. After running the PLS calculation, the two indicator showed positive loading and result in acceptable composite reliability, convergent validity, and discriminant validity of the LOC. And then when i run the second stage to analyze the HOC, the LOC now result on negative-insignificant weight and negative-insignificant loading.

Note that there is no collinearity problem found (VIF < 5) and all the indicators have been recoded accordingly. In addition, my research is more concerned with the path from the HOC to endogenous construct.

Since opting for retaining the positive one will result in single indicator, does removing the indicator with positive loading justifiable?.

Regards,
I Putu Ariyasa

Is the orientation of the indicators in the same direction? Sometimes reversing the response-order (e.g., because the question is a reverse item) makes sense to align all indicators to have the same direction.

in the survey phase, they are not. the negative loading indicator are reverse item. But i have recoded the two negative indicator so that in the analysis phase, all three indicators have the same directions.

Based on the results you have either not done the recoding correctly or respondents have not noticed the reverse item and just answered as for the other two items. In the second case you have a big problem, because your measurement simple didn't work. Especially, if respondents have not payed close attention, I would doubt all study results.

jmbecker wrote:Based on the results you have either not done the recoding correctly or respondents have not noticed the reverse item and just answered as for the other two items. In the second case you have a big problem, because your measurement simple didn't work. Especially, if respondents have not payed close attention, I would doubt all study results.

Well then i hope it's not the second one. I'll try to recode it again carefully.

Thank you very much for your guidance.