Formative: Sign. neg. weight + small Loading (Cenfetelli 09)

[Doberinho]

Hello,

I was told to post my question here since the interpretation of my results might be also interesting for you.
I have four latent exogenous constructs, all formative. My study is exploratory.
Every latent construct has about 4-6 formative indicators.
One construct (5 indicators) has 3 singificant positive indicators, 1 indicator that is neither absolutely nor relatively important/sign. but remains as the literature says that this indicator should actually be important and ONE SIGN. NEGATIVE INDIKATOR WEIGHT WITH A LOW INDICATOR LOADING.
(ALL VIF values are really low (<2))
However, I found a paper from Cenfetelli et al 2009 which suggests that this might be due to suppression.
"(4) An indicator with a statistically significant negative weight that otherwise has a positive bivariate correlation with the formatively measured construct should be interpreted as an indicator having a negative effect when controlling for the effects of other indicators"

TO MY QUESTION: CAN I INTERPRET MY RESULTS THE WAY AS IT IS SUGGESTED IN CENFETELLI? EVEN THOUGH MY LOADINGS ARE REALLY LOW (0,012)?
AND DO I HAVE TO CONSIDER THE SIGNIFICANCE OF THE LOADING WHEN INTERPRETING THE RESULTS OR IS IN THIS CASE THE LOADING THE ONLY IMPORTANT "THING"?

FYI: When I remove the neither absolute nor relative important or e.g. one of the significant indicators the magnitude of the loading (connected to the negative weight) increases.
As this is the "bivariate correlation" I assume that removing these indicators reduces the latent constructs "value" "weight" "score" ??? and thus improves the correlation, right?

I am looking forward to hearing from you.

Dirk Doberinho

[cringle]

Thanks. Sing. weight + small loading sounds very strange. Please double click on each formative construct and select Mode A. Then, estimate the model again. Do the results look better?

Kind regards
Christian Ringle

[Doberinho]

Hello Prof. Ringle,

when I do this; some of the significant weights get non-significant. However, the loadings are indeed bigger (0,12-0,2)
Do I then have to report the weights from the formative calculation and the loadings from the reflective calculation?
Or should I then restart my project and re-evaluate the indicators such as suggested in Hair et al 2017 (for the formative indicators)?
And what loadings are to consider for the reflective endogeneous construct?

And whats the trick/the story behind switiching from formative to reflective?

Sorry for all these questions but the results are supposed to get pubslied, thus I really want and have to know what I do/did...


Kind regards

Doberinho

[cringle]

If you consider formative constructs, just look at the weights. However, you can do a Mode A (correlations weights) or Mode B (regressions weights) estimation of formative constructs.

Take a look at this paper: http://www.sciencedirect.com/science/ar ... 6316304404

Best
CR

[Doberinho]

Dear Prof. Ringle,
thank you for your time and effort regarding my case. My sample size is indeed neither medium nor really small (200) and the expected R2 of 0,2-0,3 also suggests using Mode A.
Can correlation weights and regression weights be interpreted the same way, meaning individually? A positive correlation weight of an indiacator + a positive path coefficient => positive effect on the dependent variable?
However, you told me to just look at the formative weights (lets assume I stay with Mode B). Does this mean I can forget about the small loadings?

Regarding the calculation with Mode A: From my dependent variable, which consists of two reflective indicators, one loading decreased (now 0.675) whereas the other increased (now 0.94). Do you consider a loading of 0.675 as still okay for an exploratory approach/do you agree with Gaskination

Ideally, for reflective factors, the loadings of the indicators should AVERAGE out to be around 0.700 or more. This way the AVE ends up being greater than the required threshold of 0.500. So, if you have some that are higher and some that are lower than 0.700, that is fine as long as they average out above.

(viewtopic.php?f=12&t=2699&p=8480&hilit= ... tors#p8480)

Sorry for bothering you with my problems and thank you again for your effort!

Kind regards

Dirk