Two Timescales, Two Calibers: Auditing Consumer Adaptation in Algorithmically Mediated Environments

Platform analytics often use between-consumer behavioral differences to infer longitudinal consumer adaptation, although such differences need not describe changes within the same consumers. This study examines whether a matched between–within contrast retains its substantive interpretation across defensible analytical constructions. In algorithmically mediated environments, the study builds on established within–between decomposition and embeds a matched-estimand Caliber Divergence Index (CDI) in a portability audit spanning joint uncertainty, behavioral-opportunity conditioning, temporal conditioning, and behavioral representation. The primary analysis uses a longitudinal panel of 60,423 consumers (242,771 consumer-wave observations). In the focal matched-estimand specification, the between-consumer and within-consumer associations are both negative but differ in magnitude (βB = −2.9957, βW = −3.5132; CDI = 0.5174, BCa 95% CI [0.4757, 0.5573]). Opportunity conditioning reverses the contrast (CDI = −0.4812), as does relative wave-position adjustment (−0.8077). Continuous elapsed-time adjustment reduces it to approximately zero (−0.0007), whereas integer-month fixed effects retain a positive contrast (+0.5186). Replacing top-brand share with brand HHI attenuates the contrast (+0.0565), with a confidence interval that includes zero. Competing generative regimes produce distinct signatures under the examined parameterizations, but none reproduces the focal empirical CDI. Complementary analyses of historical reviews and consumer-level behavior delineate temporal, measurement, and construct boundaries. Together, the findings show that a precisely estimated between–within contrast does not necessarily support a portable substantive interpretation across defensible analytical constructions.