Reading focus
Syllabus: How Do We Build and Test Psychological Theories?
Distinguish prediction on genuinely new observations from fitting the data used to develop a model.
Why you’re reading it. Add a criterion for evaluating performance while keeping explanatory claims separate.
A question to consider. If a model predicts students’ exam scores accurately, what would remain unknown about an intervention intended to improve learning?
Why this reading
A model can fit the observations used to build it and still predict new observations poorly. This methodological argument makes that gap central and asks psychologists to evaluate predictive performance directly.
Before you read
Training data are used to develop a model; test data assess it on observations withheld from development. Overfitting occurs when a model learns sample-specific patterns that do not generalize. Cross-validation rotates which observations are withheld. Statistical bias in the bias–variance discussion is different from unfair treatment of social groups.
Read for
- Distinguish accounting for variation in an existing dataset from predicting an outcome in new data.
- Follow why preprocessing, choosing variables, and tuning a model must also respect the separation between development and evaluation.
- Examine the argument for prioritizing prediction and identify what a successful prediction would still leave unexplained.
Consider afterwards
- A researcher tries many models and reports the one that performs best on a supposed test set. What has that test set become, and what further evidence would establish performance on new cases?
Editorial context
The title advances a position about research priorities, not a demonstration that causal explanation is unnecessary. The authors retain a role for causal research. Predictive success does not by itself establish a mechanism or an intervention effect. The article’s scientific examples reflect 2017; the transferable contribution is its argument about model evaluation, not a current verdict on every example.