The elephant in the room: Predictive performance of PLS models
Section snippets
Introduction and motivation
Business research has adopted PLS path modeling, known simply as PLSPM or just PLS, as a methodological tool that bridges exploration with explanation. The popularity of PLS stems partly from its ability to produce parameter estimates of complex models without many of the distributional and other constraints of traditional parametric methods. PLS achieves these estimates by using an iterative estimation procedure considered “predictive” in nature (Hair, Ringle, & Sarstedt, 2011). Yet, the PLS
Two key features of predictive performance in predictive analytics
As early as the 1950s, thinkers of the philosophy of science recognized the fundamental difference between explanation and prediction and the necessity of both (Dubin, 1969). Management research typically focuses on explanatory models. But with the recent popularity of Big Data and predictive analytics, some predictive modeling now appears in the management literature.
Before undertaking an examination of predictive performance in the context of PLS models, please note that in contemporary
Predictive performance of PLS models: A conceptual framework
Unlocking the full potential of PLS predictions requires understanding the full picture of what is possible with predictions in PLS, where current proposals fit into this picture, and which of these proposals supports the goal of predictive performance evaluation of a PLS model.
A procedure for evaluating predictive performance of a PLS model
The lack of established techniques and tools to generate and evaluate predictions from PLS has hindered progress on creating predictive methodology for PLS. This study proposes a set of procedures for prediction and evaluation that together create a framework for evaluating the predictive performance of PLS models. The framework is tailored to the desirable properties of predictive performance evaluation highlighted earlier: out-of-sample, item-level, and operative. Specifically, the procedure
Enhancing PLS models using theory-informed prediction
As demonstrated, PLS models can be used not only to quantify and test a causal theory but also to generate point predictions and prediction intervals for manifest items both in-sample and out-of-sample. These predictions are based on a causal model, unlike predictions typically generated by machine learning algorithms. The ability of PLS to generate predictions and evaluate their performance gives PLS a self-diagnostic capability that permits use of predicted outcomes and metrics to evaluate
Conclusions and future directions
Being able to generate empirical case-wise out-of-sample predictions from a model and to evaluate the predictive power of explanatory models is vital to theory building and evaluation. Prediction using PLS models differs from machine learning algorithms and even from regression models in that PLS allows a researcher to incorporate a theoretically justified causal structure into the prediction process. Although regression models are useful for prediction, “[they are] most effective when dealing
Acknowledgments
The authors thank Kwok-Kee Wei and Loretta Fung for their feedback and suggestions. Thanks also to the organizers and participants of the 2nd International Symposium on Partial Least Squares Path Modeling (Seville, 2015) for discussions and feedback on prediction and PLS. Special thanks (in alphabetical order) to Jean-Michel Becker, Gabriel Cepeda, Wynne Chin, Nicholas Danks, Theo Dijkstra, Jorg Henseler, Christian Ringle, Ed Rigdon, José L. Roldán, Mikko Rönkkö, and Marko Sarstedt for sharing
References (39)
Illusions in regression analysis
International Journal of Forecasting
(2012)Some comments on maximum likelihood and partial least squares methods
Journal of Econometrics
(1983)- et al.
Bootstrap prediction intervals for single period regression forecasts
International Journal of Forecasting
(2002) - et al.
PLS path modeling
Computational Statistics and Data Analysis
(2005) Responding to the severe limitations of cross-sectional surveys: commenting on Rong and Wilkinson's perspectives
Australasian Marketing Journal
(2011)Moving beyond multiple regression analysis to algorithms: calling for adoption of a paradigm shift from symmetric to asymmetric thinking in data analysis and crafting theory
Journal of Business Research
(2013)The significance of simultaneous methods of parameter estimation in econometric models
Journal of the Royal Statistical Society, Series C
(1963)- et al.
The partial least squares (PLS) approach to causal modeling, personal computer adoption and use as an illustration
Technology Studies
(1995) - et al.
The moderator–mediator variable distinction in social psychological research – conceptual, strategic, and statistical considerations
Journal of Personality and Social Psychology
(1986) - et al.
Predictive validity and formative measurement in structural equation modeling: embracing practical relevance, thirty fourth international conference on information systems
(2013)
The partial least squares approach to structural equation modelling
Bootstrap cross-validation indices for PLS path model assessment
On partial least squares variance-based SEM (VBSEM) vs. covariance-based SEM (CBSEM) for confirmatory analysis: It's all about the components and variance explained
A paradigm for developing better measures for marketing constructs
Journal of Marketing Research
The superiority of simple alternatives to regression for social science predictions
Journal of Educational and Behavioral Statistics
Guidelines for choosing between multi-item and single-item scales in construct measurement: a predictive validity perspective
Journal of the Academy of Marketing Science
Latent variables and indices: Herman Wold’s basic design and Partial Least Squares
Consistent partial least squares path modeling
MIS Quarterly
Theory building: A practical guide to the construction and testing of theoretical models
Cited by (796)
Harnessing green innovation via green transformational leadership in Italian luxury hotels: Key strategic takeaways
2024, International Journal of Hospitality ManagementWhat drives higher active customer engagement in luxury brands’ social media? Measurement and contingencies
2024, Journal of Retailing and Consumer ServicesUnveiling the link between digital entrepreneurship education and intention among university students in an emerging economy
2024, Technological Forecasting and Social ChangeUnleashing the university potential: Exploring the impact of staffing autonomy on organizational performance through strategy implementation
2024, International Journal of Educational Research OpenInvestigating pre-service teachers’ artificial intelligence perception from the perspective of planned behavior theory
2024, Computers and Education: Artificial Intelligence