Two questions look similar and are not. Does hippocampal volume differ between groups? is an inference question — it asks about a parameter in a population. Can I tell from this scan whether someone will convert to dementia? is a prediction question — it asks about accuracy on data you have not seen.
The methods diverge accordingly. Inference cares about standard errors, confounds, and whether an effect is distinguishable from zero; it is evaluated on the data you have. Prediction cares about generalization; it is evaluated on data held out from fitting. A model can be excellent at one and useless at the other. A regression with a highly significant coefficient can predict individual outcomes barely better than chance, because a reliable group-level difference and a separable individual-level signal are different things — and in clinical neuroimaging, the gap between them is usually large.
This matters here because most of the Research Methods section is predictive while most of the clinical literature scholars read is inferential. Knowing which question you are answering determines what counts as evidence, what the reviewers will ask for, and whether a p-value or a held-out AUC belongs in your abstract.
Related¶
Multiple comparisons — the inference side
Data leakage — the prediction side
Normative modelling — sits deliberately between the two