still formative if weights of reflexive calcul. are used?

[sandra]

Dear Smart-PLS users and experts,

I have the following problem and would be happy if someone could help me.

In a datafile I am working with - with 3 formative measurement models - bootstrapping does only work, if I do not use the fomative but reflexive models.

I have 2 questions

1. The boostrapping output does always show weights and loadings - no matter how one chooses the model specification in smart pls (invert measurement model - reflexiv or formative arrows).
So, can I use the weights and understand my model as a fomative one although calculation was with a reflexive measurement model?

2. Why does bootstrapping only work if I choose reflexive models in my datafile? I do not think that the number of indicators is too low (6 each).

I would be happy about an answer - Thank you
Sandra

[jjsailors]

So, can I use the weights and understand my model as a fomative one although calculation was with a reflexive measurement model? Sandra

No matter the specification of the model, constructs/LVs in PLS are always formative and the weight is always what is used to calculate LVs. The specification only matters with respect to how the weights are calculated.

2. Why does bootstrapping only work if I choose reflexive models in my datafile? I do not think that the number of indicators is too low (6 each).
Sandra

When you say it only works do you mean that you only get significant results when you specify the LVs as reflective? If so that is very likely due to multicollinearity among the indicators.

If you means something else, please elaborate.

John

[sandra]

Thank you very much for your reply.

By saying it only works for the reflexive model I mean that when I use a formative model the procedure stops and the following message is provided:

"A singular matrix occured during the estimation of the outer weights of a block. using more indicators could solve the problem"

Whereas with the reflexive calculation the resulting weights are fine and significant.

What is the difference regarding the calculation of outer weights in the formative vs. the reflexive specification? Can both results be seen as equally valuable and correct if one aspires to use formative measurement?

thank you
Sandra

[Diogenes]

Hi,

please see viewtopic.php?t=262

The problem that you have quoted probably was caused by two variables with the same values during bootstrapping procedure.

What is your sample size?

Best regards

Bido

[yusofahmad]

Hi Prof Bido.

I understand that in this case the singular matrix is occuring during the bootstrapping porc. I am facing a similar problem. I can compute the model coefficients using the PLS-algorithm.
However, possibly, when using bootstraps, the procedure leads to the problem of 2 identical columns within a block. Essentially, in my sample this is possible because there is high correlation between the indicators of the formatrive construct (the VIFs though, are less than 10). this means that during the bootstrap procedure the same column gets replicated very soon. Is this the reason? Would I expect the same problem using a different program for the bootstrps?

Regards,
Yusof