Disentangling Service Participation and Hedonic Positioning in Online Reviews: A Structural Confound in Platform Data

Abstract

Online review platforms are core infrastructure of digital commerce, yet the business-level determinants of review content remain poorly understood. We examine how service participation intensity, the degree to which a business format requires active customer involvement, relates to the affective–experiential language of reviews and to rating extremity. We analyze 6,151,996 Yelp reviews (2010–2022) with two independently validated instruments: a Business Participation Index, whose 39 attribute and 1263 category weights derive from an independent coding procedure (two large-language-model raters from different model families applying written, hypothesis-blind codebooks; Krippendorff’s α=0.88–0.96) and a 35-indicator text framework validated against a review sample dual-coded under the same procedure. Participation and hedonic–utilitarian positioning, measured independently for the first time, are negatively entangled across the platform (r=−0.27). Decomposing the two shows that positioning absorbs roughly thirty percent of the raw participation–hedonic association. The remainder survives category, price, geography, and year fixed effects with business-clustered errors (b=−0.029, standardized β=−0.084), inverse-probability weighting, and Oster bounds (δ=13.7). Extremity analysis reveals an asymmetry: higher participation predicts one-star reviews (odds ratio of 1.13) but not five-star reviews. The hedonic–analytical co-occurrence index, the joint elevation of the two language modes, adds no information beyond its components (incremental R2=0.00006) and is not advanced as a mechanism. Format-based comparisons in review analytics must condition on positioning.