Regression optimizes explanation of data at hand while machine learning optimizes out-of-sample prediction, and studies must state which aim they pursue¶

Claim · All claims
Evidence · 1 study · theoretical r2 · q1
Currency · newest study 2021 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q1 i? Standard analyses like regression optimize model fit for explanation, whereas machine-learning approaches choose coefficients to minimize prediction error on new out-of-sample data; both serve complementary goals. → van Atteveldt 2021

Evidence¶

van Atteveldt 2021¶

van Atteveldt, N., Vandermosten, M., Weeda, W. & Bonte, M. (2021). How to capture developmental brain dynamics: gaps and solutions. npj Science of Learning. https://doi.org/10.1038/s41539-021-00088-6

q1 · i? · theoretical · r2

Theoretical/methodological argument from the article's fourth identified gap. Predicting outcomes can provide insights on the need for additional support, e.g., early intervention or policy, while explanatory methods highlight underlying mechanisms that can be targeted by intervention.

"they try to choose coef ficients in a way that minimizes the prediction error of new (out-of-sample) data. The coef ficients are thus not optimized for explanation, but for prediction"

Discussion¶

Related Claims¶

• A machine learning model using 105 predictors predicts next-year standardized math achievement with out-of-sample prediction R-squared of .70–.78 — a narrower finding that bears on this claim
• Machine learning on pre-pandemic platform click data explains up to 77% of the variance in academic resilience — related
• The single-task Random Forest achieves the best average predictive performance but is optimal for only 7 of 15 concepts — related