Sample Size Determination and Statistical Power Analysis in PLS Using R: An Annotated Tutorial

Abstract

Partial least squares (PLS) is one of the most popular analytical techniques employed in the information systems field. In recent years, researchers have begun to revisit commonly used rules-of-thumb about the minimum sample sizes required to obtain reliable estimates for the parameters of interest in structural research models. Of particular importance in this regard is the a priori assessment of statistical power, which provides valuable information to be used in the design and planning of research studies. Though the importance of conducting such analyses has been recognized for quite some time, a review of the empirical research employing PLS indicates that they are not regularly conducted or reported. One likely reason is the lack of software support for these analyses in popular PLS packages. In this tutorial, we address this issue by providing, in tutorial form, the steps and code necessary to easily conduct such analyses. We also provide guidance on the reporting results.

The role of innovation in the implementation of green supply chain management practices

“…Statistical power is considered a critical issue for the validity of results of SEM models (Riedl, Kaufmann, & Gaeckler, ). To ascertain if our sample size ( n = 173) is adequate to achieve the recommended statistical power of 0.8 (Riedl et al, ), we run a power analysis for each structural path using Monte Carlo simulations as recommended by Aguirre‐Urreta and Rönkkö (). As shown in Table , the statistical power of all significant structural path is greater than 0.8.…”

How IT executives create organizational benefits by translating environmental strategies into Green IS initiatives

“…In order to establish the adequacy of our sample size ( n = 118) to detect the effect sizes obtained in the PLS analysis with acceptable power (0.80), we conducted a post‐hoc power‐analysis simulation following Aguirre‐Urreta and Rönkkö's () simulation procedure using R. We obtained the factor loadings for the items measuring each of the constructs, the path coefficients and residual values from the PLS run in Figure . We used a sample size of 118, 1000 converged replications and 500 bootstrapping re‐samples for the simulation.…”

“…For both detailed views, we ran additional R simulations (Aguirre‐Urreta & Rönkkö, ) to determine the power of the sample size to detect each path shown. The sample was adequate to detect nine of the paths (Table in Appendix D) including the two significant paths illustrated in Figure and the three in Figure .…”

A new perspective on neutralization and deterrence: Predicting shadow IT usage

“…We followed the recommendations of [62] to analyze if our sample size would provide good power and reliability of our model results. We calculated statistical power by taking into account our sample size (440), the observed probability level (0.05), the number of predictors (8) and the observed R 2 (0.291).…”