Discussion: View Thread

Hi:

Just a small follow up on this post--whose thread I have changed to reflect a more general issue--and in particular with regard to the McIntosh, Edwards, and Antonakis paper "Reflections on Partial Least Squares Path Modeling" noted by Herman below.

Jill, I would strongly advise you to not use PLS path modeling and instead to use regular SEM or some other analytically derived method that can estimate simultaneous equations models. As we note in our paper:

1. PLS path modeling is not a SEM method; it thus cannot eliminate endogeneity bias via the instrumental variable method for models with observed or latent variables

2. PLS path modeling cannot eliminate the impact of measurement error

3. PLS path modeling cannot validate common-factor measurement models (i.e., models with latent variables)

4. It is not clear if PLS path modeling can provide valid inference for tests of coefficients

5. PLS path modeling does not necessarily have advantages of SEM in small sample size conditions

6. It is not clear if PLS path modeling can be used for exploratory modeling

Conclusion: Should you use PLS path modeling? Certainly not, particularly in cases of models having latent variables or specific endogeneity issues that require modeling procedures to eliminate the endogeneity bias. SEM has robust estimators that can be used when distributional assumptions are violated, it can model continuous, discrete, or ordered data, it has overidentification tests, it can directly estimate standard errors (or bootstrap them), it has analytical methods (Barlett-type) to correct for small sample bias in overidentification tests; in the worst case one is still better off using observed scales in the context of a 2sls (or 3sls) model, as compared to PLS parth modeling, because the former is a consistent estimator that can use the instrumental variable technique and has excellent small sample properties (as well as closed-form solutions to estimation).

Why is PLS path modeling so popular in some disciplines (e.g., MIS research?). I don't know. Here is a telling insight from Stas Kolenikov that I have parsed from a discussion on this topic on SEMNET a few years ago:

"of the big names who developed pretty much all of the mainstream methods we use routinely (Bentler, Bollen, Joreskog, Muthen, Satorra, Yuan, in an alphabetical order), nobody wrote a single line about PLS. This must tell something about how much they value the methodology." 

Same goes for the econometrics literature (which I know pretty well); I have never seen PLS ever mentioned in there (except for an article in the Journal of Econometrics, published by Dijkstra, in 1983 showing that the standard PLS estimators are not a consistent estimators; although Dijkstra is currently developing a new approach in PLS--see references in our ORM article--it is still far away from the capabilities of SEM).

Best,
J.

P.S. the instrumental variable technique is described here:

Antonakis, J., Bendahan, S., Jacquart, P., & Lalive, R. (2010). On making causal claims: A review and recommendations. The Leadership Quarterly, 21(6). 1086-1120.

For a less technical introduction see:

Antonakis, J., Bendahan, S., Jacquart, P., & Lalive, R. (2014). Causality and endogeneity: Problems and solutions. In D.V. Day (Ed.), The Oxford Handbook of Leadership and Organizations (pp. 93-117). New York: Oxford University Press.

__________________________________________  John Antonakis Professor of Organizational Behavior Director, Ph.D. Program in Management  Faculty of Business and Economics University of Lausanne Internef #618 CH-1015 Lausanne-Dorigny Switzerland Tel ++41 (0)21 692-3438 Fax ++41 (0)21 692-3305 http://www.hec.unil.ch/people/jantonakis  Associate Editor:  The Leadership Quarterly Organizational Research Methods  __________________________________________ 

On 12.04.2014 04:14, Aguinis, Herman wrote: