AI Contextual Measurement for Recovering Individual and Group-Level Effects: Validation Against Survey Measures and an Occupational Application

The provided search context does not mention a framework called AICOME.

The context does, however, describe two relevant AI-assisted measurement frameworks:

• AI-Assisted Economic Measurement Framework: This framework uses large language models to create respondent-level subdimension scores from survey instruments. It validates these measures through out-of-sample incremental validity tests and discriminant validity diagnostics to ensure they recover stable economic mechanisms.
• Platform-Derived Occupational Exposure Framework: This framework addresses bias in measuring AI exposure by using workforce-representative reweighting. It treats baseline platform-derived estimates and reweighted estimates as endpoints of a partially identified set to recover more accurate structural employment elasticities.

While these frameworks aim to recover individual and group-level effects, the specific AICOME framework and its validation against survey measures are not present in the provided information.

The material introduces AICOME, a framework for evaluating whether AI-generated respondent-level contextual measurements can support recovery of both individual-level and group-level effects in contextual statistical models. Its central concern is a measurement problem common in social-science and policy research: contextual influences are often difficult to observe directly, and traditional survey-based measures may be costly, coarse, or limited in coverage. AICOME positions AI-derived contextual indicators as potential scalable alternatives, but treats their validity not as assumed and instead asks whether they can preserve the distinct causal or associative structure of individual and contextual effects when used in multilevel or contextual models.

A key contribution is the paper’s emphasis on effect recovery rather than mere predictive performance. By validating AI measures against established survey measures and applying the framework in an occupational setting, the work offers a benchmark for assessing whether AI-based contextualization can substitute for, complement, or improve upon conventional contextual variables. This matters because the distinction between individual effects and group/contextual effects is central to inference in fields such as labor economics, sociology, public policy, and organizational research. If AI measures can faithfully recover these effects, they could expand the scope of contextual analysis; if they cannot, the framework provides a diagnostic for identifying where measurement error, aggregation bias, or contextual misspecification is likely to arise.