Hello,
I am estimating a model in Smartpls 3.0 and I want to use the consistent pls procedure. I have done EFA in SPSS and the factors should be OK.
What I am a bit surprised to see is that the parameter estimates are generally higher, but the bootstrapping results are lower - i.e. lower t-values.
Also I have noticed considerable inconsistencies in the t-values - for instance, if I run the bootstrapping procedure three times in a row, then one t-value changes from 1,2 to 0,4 to 1,6.
Is there something inherently wrong with the model or the data that could explain such behavior?
Thanks for all given help,
Thorhildur
Thank you so much Theo! I have posted a question in the Smartpls forum on LinkedIn, it seems that if I run the regular bootstrap procedure first, and then the consistent one (after the consistent pls algorithms) the problem is eliminated but not sure if that is intended functionality...
Best regards,
Thorhildur
I don't like to employ the non-parametric approach in my work. I prefer to use the established parametric approach in Amos or Lisrell for my work. I thing Smart-PLS (PLS 1.0, PLS 2.0) is still at developing stages. As you said, even the Consistent PLS 3.0 is still not consistent.
I used to ask the Regent Professor from US (you can guess who) about this. He said "If you already have SEM to work with, why should you still go for lesser SEM?"
PLSc is viewed suit for confirmatory and exploratory purpose as well. However, since you just face this issue, maybe you can try CBSEM for your research.
the benefits of PLSc and other flexible elements within PLS functionality make some choices increasing easy to make for the analyst.
maybe we will see some models published in the future with more than 5 or 7 constructs in them. The world is a complex place.
introduce some interaction effects or look at a multiple moderated mediation model... Watch some software explode
how many models and research follow a strictly confirmatory approach anyways despite the write up which alludes to the it being the case.
what theo and thorhildur highlight is the importance to look at fringe cases, novel results and behaviour of the method/algorithm and have solid refereed forums that publish this work. Monte Carlo studies etc.
thanks Theo for jumping into discussions with such detail and clarity. It helps us all.
brad
theos' comments around third and fourth order moments should not be ignored.
another point that is often glossed over. Choose wls or adf estimators and look at the requisite sample sizes. The educational and economics researchers can often deal with adequate n there.
thanks again Theo
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling