Statistical Power and Sample Size in PLS Path Analysis

12 Pages Posted: 30 Aug 2014

See all articles by J. Christopher Westland

J. Christopher Westland

University of Illinois at Chicago - Department of Information and Decision Sciences

Date Written: August 29, 2014

Abstract

The PLS path analysis algorithm has been used in numerous research papers in various branches of the social sciences. Prior literature has conflated the generation of coefficients without abnormally terminating, as equivalent to estimating with small samples. This research shows that: (1) PLS path estimates are biased and highly dispersed with small samples; (2) sample sizes must grow very large to control this bias and dispersion with dispersion ∝1/log (sample size) and bias ∝1/(sample size) ; and finally (3) the power of PLS hypothesis tests is low at most effect levels, leading PLS software to generate a disproportionate number of false positives.

Keywords: Econometrics, Analytical modeling, Statistical testing, Mathematical methods, Computer simulations, Statistics

JEL Classification: C39

Suggested Citation

Westland, J. Christopher, Statistical Power and Sample Size in PLS Path Analysis (August 29, 2014). Available at SSRN: https://ssrn.com/abstract=2488982 or http://dx.doi.org/10.2139/ssrn.2488982

J. Christopher Westland (Contact Author)

University of Illinois at Chicago - Department of Information and Decision Sciences ( email )

University Hall, Room 2404, M/C 294
Chicago, IL 60607-7124
United States

Do you have negative results from your research you’d like to share?

Paper statistics

Downloads
371
Abstract Views
1,844
Rank
147,090
PlumX Metrics