I came across article that argues against the appropriateness of using the effect size benchmarks of Cohen in "OB/HR" research as they overestimate the actual breakpoints between small, medium, and large effects.
The article Info: Paterson, T. A., Harms, P. D., Steel, P., & Credé, M. (2016). An assessment of the magnitude of effect sizes: Evidence from 30 years of meta-analysis in management. Journal of Leadership & Organizational Studies, 23(1), 66-81
Do you think this line of logic may be extendable to the threshold values we use to assess path coefficient and R square when using PLS-SEM?
Please note that the available rules of thumb are:
- Significant path coefficients (b) approaching + 1 indicates strong positive relationships (and vice versa for negative values)
- R square values of 0.25, 0.50, and 0.75 is be weak, moderate, and substantial respectively (Hair, Hult, Ringle, and Sarstedt, 2014).
- f square (effect size) values are that values of 0.02, 0.15, and 0.35, respectively, represent small, medium, and substantial effects.
The failure of small, medium, and large (later tacitly admitted by Cohen himself, 1968). Is that their values are dependent on the variance and that is different for different measurements. The same must be true for any research where changes in experimental or observational procedures, or materials, alter the variance.
Maybe your question is not a problem, but rather, a research opportunity (perhaps a rather challenging one). There are a number of troubling questions about PLS-SEM. As to a lead to peer reviewed literature (Wikipedia is not peer reviewed) you might see the literature cited in https://en.wikipedia.org/wiki/Partial_least_squares_path_modeling
Incidentally small, medium and large can underestimate too, The point is that they are usually very inaccurate.
Good luck, great question.
- Badges
- Science topic