Dear Benjamin

You should delete item loadings lower than 0.40. When the item loading is between 0.40 and 0.70, you only delete the item when it causes an improvement in AVE. If AVE does't improves, you should delete the item. Finally, you have to maintain items with loadings higher than 0.70.

As I can see in your model, you should delete all the items < 0.40, stepwise. First, the lowest item loading, then you run the pls algorithm again, and delete again the item with the lowest loading. At the end, all the items should have loadings higher than 0.70 (or 0.40-0.70 if AVE improves).

Thus, for LV1, only delete WD3 (0.444) if AVE improves after this. If not, you can keep the item. This depends on you.

This could help: Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2017). A Primer on Partial Least Squares Structural Equation Modeling.2nd Ed. Thousand Oaks: Sage.

Regards

Hi Alejandro,

Can i ask what if the loading is below 0.700 but the t-value is more than 1.96 should i keep the item?

Dear benjamin

Since all your LV are reflective, you only have to check the indicators loadings. As I said, you maintain loadings > 0.70 and decide what to do with indicator loadings between 0.40-0.70. If deleting this indicators causes AVE improvement, then, you should delete them.

With respect to t-statistics, you check t-value > 1.96 when using formative indicators. With reflective items, as in your case, check the loadings.

Regards

Thank you Alejandro for your comprehensive response. I have a final two questions to ask. 

1) If the 0.400-0.700 does not increase the AVE much, i can keep it? What if the AVE is between 0.45-0.5 what can i deduce?

2) For the path coefficient, will t-value be necessary to deduce that A->B  is significant?

Dear Benjamin

1) In this case, I think I will drop the item from the model. With respect to AVE, it  must be equal or higher to 0.50 to confirm convergent validity of the scale.

2) You check both path coefficient and t-statistic. Usually, path coefficients greater than 0.20 are statistically significant. Report always the beta coefficient and the t-statistic (or p-value) for the significance level.

I recommend you this paper: 

https://www.researchgate.net/publication/236033273_PLS-sem_Indeed_a_silver_bullet

Regards

Article PLS-sem: Indeed a silver bullet

Thank you Alejandro! You helped me a lot!

Just to confirm, my variable only have 3 items deleting 1 will left me with 2 items to explain the latent variable is it alright in this case? Can i delete until i left with 1 item to explain the latent variable which makes my AVE >0.500?

As a general rule, it is better that latent variables have, at least, three items. However, if your results are ok, you can show a LV with 2 items. In fact, there are variables (observed) with only one item (single-item scales). However, they offer less information than multi-item scales.

In any case, some articles show latent variables consisting of one or two items. This may be caused by item deletion (low loadings) or because the original scale is formed by one/two items.

With respect to AVE, the criterion to follow is > 0.500.

There is a forum which helped me a lot. Link: http://forum.smartpls.com/

Regards

OK. First delete items with loadings