Hi - I'm using smartpls for the first time and have a single equation model (one dependent variable with 1 manifest measure and many independent vars with 2-7 manifest vars each). N=1100. I suspect that some of the manifest measures in the reflective models either don't belong there or can be moved to different latent vars. However, when I estimate the PLS model, the loadings are almost all .75 and higher. Cronbach's for each of the latent vars in greater than .70 for all except one. And the communality score for all of the latents range from .57 to .75. Further, the low communality scores seem to be on the latents that have high Cronbach's (I would have expected the reverse). What are some good strategies for determining how to change/improve these outer models? Thanks for any advice.
Cheers,
Jeff
Using model results to evaluate/specify the outer model
-
- PLS Junior User
- Posts: 7
- Joined: Thu Aug 13, 2009 3:20 am
- Real name and title:
You should also conduct some tests of construct validity to determine the extent to which your measures correspond to the focal construct. For reflective constructs, you can look at the AVE values to establish convergent validity (should be higher than 0.5). You should also do tests of discriminant validity. One such test at the construct level is the Fornell-Larcker criterion (establishes discriminant validity if for each reflective construct the square root of the AVE is larger than the construct's correlations with other constructs). At the item level, you can test for discriminant validity by looking at the cross-loadings and making sure that for each block of indicators, the loadings on the focal construct are at least one order of magnitude higher than loadings on other constructs.
-
- PLS Junior User
- Posts: 7
- Joined: Thu Aug 13, 2009 3:20 am
- Real name and title:
Thanks. Very helpful. One question re the cross loadings test for discriminant validity: when you say the loading on the focal construct should be 1 order of magnitude higher than loadings on other constructs, what exactly does that mean in numeric terms? In common usage, 1 order of magnitude implies that the focal construct loadings should be 10 times the loadings on other constructs. But that can't be right. So, for example, I have these focal loadings on construct "GR":
GR1 - .7739
GR2 - .7763
GR3 - .8225
GR4 - .8085
GR5 - .7707
GR6 - .8138
GR7 - .7935
And the loadings of these items on other factors generally ranges from .40 to .65. So they are lower than the focal loadings but are they low enough? Or, in this example, should they really all be .38 or lower (i.e., .38 x 2 = .74 < .7707)
GR1 - .7739
GR2 - .7763
GR3 - .8225
GR4 - .8085
GR5 - .7707
GR6 - .8138
GR7 - .7935
And the loadings of these items on other factors generally ranges from .40 to .65. So they are lower than the focal loadings but are they low enough? Or, in this example, should they really all be .38 or lower (i.e., .38 x 2 = .74 < .7707)