Interpretable machine learning for psychological research: Opportunities and pitfalls

Abstract

In recent years, machine learning methods have become increasingly popular prediction methods in psychology. At the same time, psychological researchers are typically not only interested in making predictions about the dependent variable, but also in learning which predictor variables are relevant, how they influence the dependent variable, and which predictors interact with each other. However, most machine learning methods are not directly interpretable. Interpretation techniques that support researchers in describing how the machine learning technique came to its prediction may be a means to this end. We present a variety of interpretation techniques and illustrate the opportunities they provide for interpreting the results of two widely used black box machine learning methods that serve as our examples: random forests and neural networks. At the same time, we illustrate potential pitfalls and risks of misinterpretation that may occur in certain data settings. We show in which way correlated predictors impact interpretations with regard to the relevance or shape of predictor effects and in which situations interaction effects may or may not be detected. We use simulated didactic examples throughout the article, as well as an empirical data set for illustrating an approach to objectify the interpretation of visualizations. We conclude that, when critically reflected, interpretable machine learning techniques may provide useful tools when describing complex psychological relationships.

Responsible research assessment in the area of methodological or quantitative research: A comment on Gärtner et al. (2022)

“…Besides many others, current research areas include machine learning based approaches for RESPONSIBLE QUANTITATIVE METHODS RESEARCH 4 social sciences that are interpretable (Henninger, Debelak, Rothacher, & Strobl, 2022), Bayesian modeling and model selection (Heck et al, 2022), or new approaches for intensive longitudinal or other complex data (e.g., Nestler & Humberg, 2022;Orzek & Voelkle, 2023).…”

Developmental data science: How machine learning can advance theory formation in Developmental Psychology

“…The literature on interpretable machine learning suggests several ways to derive insights from machine learning models (Henninger, Debelak, Rothacher, & Strobl, 2022).…”

“…The literature on interpretable machine learning suggests several ways to derive insights from machine learning models (Henninger et al, 2022). One approach is to compute variable importance metrics, which indicate how important each variable is in predicting the outcome.…”

“…The literature on interpretable machine learning suggests several ways to derive insights from machine learning models (Henninger et al, 2022) Two-way interaction effects can be further explored using the H-statistic (Friedman & Popescu, 2008). The insights resulting from this interpretative approach to machine learning-informed theory formation can be used to amend or construct verbal theories.…”

Best Practices in Supervised Machine Learning: A Tutorial for Psychologists

“…Note also that effect plots do not necessarily indicate that the feature of interest has any causal effect on the target.While ALE plots are often discussed as an improvement on PD, this does not seem to be without controversy because both estimate slightly different concepts Molnar (2019). andHenninger et al (2022) provide a clear introduction on both methods.…”