High Negative Outer Weights

[Ibra]

Hello fellow SmartPLSler,

from studying this forum and PLS literature for quite a while now, I know that this topic has been addressed already. However, so far I was unable to find a sufficient answer to the following issue:

After running the software, I was exposed to unusual results for my outer weights, especially when running a smaller sample to do a group analysis.

Here are the results (example; bigger samples also lead to a negative or very low value):
sample size 48 (using the thumb rule 50 is advised for a max. of 5 connections)

Outer Weights:
a 0,510
b 0,684
c -0,775
d 0,550
e -0,029

Respective VIF:
a 3,103
b 1,514
c 3,068
d 3,031
e 3,465

From my understanding, since all VIF values are <5, the VIF values indicate no collinearity issues. Is this assumption correct? (1)

My follow up question is, how should I interpret these high negative results? (2)
From what I have read so far, a low value of an outer weight indicates a low contribution to the latent variable. So what does a negative value indicate? (4)
Do I just state that the negative weights do not contribute at all to the construct? (3)

If the VIF is sufficient and does not indicate a collinearity issue (following Hair et al. 2013), I check the significance, and the respective outer loading, if outer weight is not significant. In case, a on outer loading's value is <0.5, I should delete this indicator, right? (5)

Could the issue arise because of the sample size? Would deleting indicator c one way to solve indicator weight problem and potential problem of a too small of a sample?

Thank you, for your help and advise.

[jmbecker]

1) There exist different recommendations as to when collinearity is large. I would consider VIF above 3 as large. This situation is present in your case.
Collinearity problems are more important in very small samples. The variance inflation becomes more important and coefficients and weights might become insignificant even if theiy are important. Also the weights and coefficients can "dance around", i.e., take unexpected values.

2 - 4) Negative means that it negatively contributes to your construct. You should also check the loading. If the loading is also negative as well, the item may be reverse coded air just negatively contributes to the construct. For example, price is negatively related to sales. If you increase price, sale decreases.

In addition, 48 is not a sufficient sample size. Even if the rule of thumb may result in 50, this only means that the model can be estimated with 50 cases, but not that you will get reliable results.

[Hengkov]

1. I recommend VIF for formative indicators should be < 2.5.

2. Negative weight may occur within a few possibilities. The code that needs to be reversed (for negative question), but you do not reverse it. Or indeed an indicator construct "mode k", formative-reflective.

3 & 4. Negative weight does not mean there is no contribution, but there is a problem on your indicator.

5. You have a very small sample. But, you can not remove the indicator for both reasons.