Second-Order Formative Construct: Weight of only one of four FOCs is significant

Questions about the implementation and application of the PLS-SEM method, that are not related to the usage of the SmartPLS software.
Post Reply
fkreitne
PLS Junior User
Posts: 4
Joined: Wed Feb 13, 2019 8:13 am
Real name and title: Fabian Kreitner

Second-Order Formative Construct: Weight of only one of four FOCs is significant

Post by fkreitne »

Hi together,

I have a second-order reflective-formative construct with four dimensions. I used the repeated-indicator approach with mode B and path weighting.
The problem I face is that one of the FOCs has very high weight (0.8), while the weight of the other FOCs is low, insignificant and also negative in one case. Furthermore, the indicators of the other FOCs change signs and are all negative.
I assessed the convergent validity of my second-order formative construct using the redundancy analysis described in Hair et al. (2013). Here, the weight of that one FOC with high weight in the model, had a very low and insignificant weight. Does that mean, that it is not a dimension of the model I want to measure? How do In interpret this?

Best regards,
Fabian
MrcJnk
PLS User
Posts: 16
Joined: Thu Oct 17, 2013 10:04 am
Real name and title: Marc Janka
Location: Germany

Re: Second-Order Formative Construct: Weight of only one of four FOCs is significant

Post by MrcJnk »

Hi Fabian,
Let’s consider the latent construct "drunkenness" that is formatively measured with (1) number of beers, (2) number of wine, and (2) number of liquor. In your research model, for instance, people on average that got headache most probably had not drunken that much beer. The weight would not be significant; however, dropping item (1) may cause invalid measurement of your construct drunkenness. A nonsignificant relationship tells nothing (because of the beta-error). You can just say that within the above exemplified research model, it could not be approved that the number of beers causes drunkenness, which finally leads to headache. No more, no less. That’s the major issue with formative measures.
Best
Marc
msaradhi
PLS Junior User
Posts: 3
Joined: Mon Jun 18, 2018 12:16 am
Real name and title: Saradhi Motamarri (Mr)

Re: Second-Order Formative Construct: Weight of only one of four FOCs is significant

Post by msaradhi »

Hi,
I am also testing a model that includes a 2nd Order Reflective-Formative (RF) Model, using Mode-B.

When I test 2nd Order construct (SOC) formulation, the path coefficients to all the 1st Order constructs seem OK.

But when I extend the model to include other constructs to test the SOC's nomological validity, notice this intriguing behavior of change of coefficient values to insignificant and for some change of sign.

If we assume that the insignificant/negative dimension has no significance on the dependent construct: does it convey:
1. the formation of the 2nd order construct is adequate? or
2. these dimensions to be dropped from the 2nd order construct?

OR
3. Does this convey, a Reflective-Reflective (RR) 2nd Order Model is appropriate for this case than an RF Model?
4. Are there any measures to compare to decide on which SOC model (i.e., RR or RF) is good?

Look forward for valuable advice from the respected scholars.

Thank you and regards... Saradhi.
jmbecker
SmartPLS Developer
Posts: 1281
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: Second-Order Formative Construct: Weight of only one of four FOCs is significant

Post by jmbecker »

The choice between reflective and formative should ideally been made by theoretical considerations and not empirical. If you want to do an empirical test you can do a CTA-PLS analysis. That would work for second-order constructs with the two-stage approach.

The nomological network is important for second-order construct estimation. You have predictive relations in PLS that focus on the "good neighbor assumption", that is, they are optimized for predictive fit. If you do not have any other variables in the model except for the second-order construct all dimensions are likely to be equally important (and therefore also significant). Only if you place the second-order construct in a nomological network you will see differential weights.
Dr. Jan-Michael Becker, BI Norwegian Business School, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
Post Reply