The digital trace inference framework: Aligning behavioural evidence with claims in applied linguistics

Digital interaction records are increasingly used to make claims about language learners, yet the behaviour recorded by a platform is not a transparent measure of the construct researchers wish to infer. This article proposes the Digital Trace Inference Framework (DTIF), a methodological framework for determining what evidence is required to warrant claims from digital trace data. DTIF is organised around claim-evidence alignment: researchers specify the intended claim and then evaluate whether the available evidence supports the inferential distance from recorded behaviour to that claim. Five linked components structure this evaluation: construct representation, trace construction, dependence structure, generalisation boundary, and inferential use. The framework is illustrated through a reanalysis of the Cambridge Dictionary Look-Up Dataset. The illustration shows that conclusions change when raw click records are converted into labels under different rules, when dependence across learners and documents is represented, and when predictive models are evaluated on unseen learners rather than unseen documents. Calibration and consultation-capture analyses further show that predicting historical consultation does not by itself establish latent lexical need or the benefit of automated support. The article therefore distinguishes claims about observed behaviour, latent constructs, generalisation, and intervention as requiring different forms and amounts of evidence. DTIF provides a structured basis for making digital-trace claims that are explicit about what was observed, where the result is expected to hold, and what further evidence is required before prediction is translated into educational action.