Reframing Anchoring Bias Measurement: A Logistic Regression Approach to Anchor Proximity

Research on anchoring bias has most often operationalized the phenomenon as an aggregate directional shift in numerical judgments, typically assessed through differences in mean or median estimates across experimental conditions. While this approach has repeatedly demonstrated the robustness of anchoring effects, it provides limited insight into how alternative operational definitions influence the detection and interpretation of bias at the individual level. In a between-subjects design, participants were exposed to either a low or a high numerical anchor prior to providing a population estimate and reporting confidence. Anchoring bias was operationalized categorically, as a binary indicator of whether an estimate fell within a proportional proximity window (±50%) of the anchor value, emphasizing relative closeness rather than directional displacement. Logistic regression was used to examine whether anchor magnitude and self-reported confidence predicted bias classification under this definition. Neither anchor value nor confidence significantly predicted categorical anchoring, and sensitivity analyses across multiple proportional thresholds yielded qualitatively similar results. These findings do not challenge the existence of anchoring effects documented under aggregate analyses; rather, they demonstrate that anchoring detectability is highly sensitive to operational and analytic choices. Specifically, anchor magnitude does not reliably predict individual-level categorical susceptibility when anchoring is modeled as a probabilistic classification outcome rather than as an aggregate shift in central tendency. Keywords: Anchoring bias; Bias operationalization; Individual-level judgment; Anchor proximity; Logistic regression with dynamic, networked, and heterogeneous decision-making contexts.