I'm looking for references/ published work supporting as well as rejecting a 2-item factor (latent construct). It's well known that 3 item per factor is considered reliable, at minimum.
"Because the constructs’ measurement properties are less restrictive with PLS‑SEM, constructs with fewer items (e.g., one or two) can be used than those that CB‑SEM requires." (Hair et al., 2011, p.140).
Hair, J.F., Ringle, C.M. and Sarstedt, M.(2011) PLS-SEM: Indeed a Silver Bullet. Journal of Marketing Theory and Practice, vol. 19, no. 2 (spring 2011), pp. 139–151.
I actually use PLS-SEM for one project. A 'by-product' of the study challenges conventional wisdom on a latent construct for where 5 items were largely adopted as a single construct. My exploratory factor analysis (EFA) shows they are 2 distinct constructs. This was further confirmed at confirmatory (CFA) stage. Of course, this has also been validated at measurement model assessment.
In the light of this issue, some scholars in the field are concerned with the item split because of one construct is defined by 3 items while the other is measured by 2, despite some earlier studies indicate this possibility.
Timo's and your contribution is a treasure. Very helpful to address their concerns.
If you are doing CFA and your model only has 1 factor with 2 items the model would not be identified. If your model has more than one factor then the factor with only 2 items might be identified.
Google model identification. The mplus website is a good place to start: http://www.statmodel.com/download/Topic%201-v11.pdf
I believe the issue of identification applicable to covariance-based SEM such that provided by Mplus and AMOS. This may not be applicable with variance-based SEM (PLS-SEM). Nevertheless, your suggestion is good in this discussion.
Agreed to Derya that Variance-based SEM / PLS-SEM supports the use of 1-item factor (latent construct). Moreover, Variance-based SEM can support both either formative or reflective item whreas Coveriance-based SEM support only reflective item.
In case you are interested to know the differences between Variance-based SEM vs Covriance- based SEM you can refer to side 12 of this presentation that I'd done recently:
This is a long shot, but maybe I can get some help. I have a question regarding the same topic. I had a variable with 4 items and after doing factor analysis(twice), it is now a construct with only 2 items. My supervisor has asked me to upload all 4 items..is there a way I can do that?