Development of Model Influence on Consumer Behavior in U.S. e-commerce and Digital Marketing

Abstract

This study addresses the practical and theoretical problem that, although recommender systems are now embedded across U.S. e-commerce and digital marketing touchpoints, organizations still lack clear quantitative evidence on which perceived recommender attributes most strongly drive consumer behavioral outcomes and how privacy concerns constrain those effects. The purpose was to test a perception-driven influence model using a quantitative, cross-sectional, case-based design anchored in enterprise-scale, cloud-deployed e-commerce recommendation environments (multiple recommendation “surfaces” such as home-page personalization, product-page similar items, cart cross-sells, and email or push recommendations). Data were collected from N = 312 eligible consumers with recent recommender exposure, where 71.5% reported interacting with recommendation carousels at least weekly; key stimulus and organism variables included Personalization Quality (PQ), Perceived Relevance (PR), Transparency/Explainability (TRNSP), Trust (TR), Privacy Concern (PVC), and two study-specific indices, Recommendation Exposure and Interaction Intensity (REI²) and Algorithm Aversion–Appreciation (AAAT), while response variables were Purchase Intention (PI), Satisfaction (SAT), and Loyalty/Repurchase Intention (LOY). The analysis plan applied 5-point Likert scale measurement, reliability testing (Cronbach’s α range .81–.90 across constructs), descriptive statistics, Pearson correlations, and multiple regression models with multicollinearity diagnostics (VIF approximately 1.28–2.34). Descriptives indicated above-midpoint perceptions for PR (M = 4.01, SD = 0.66) and PQ (M = 3.88, SD = 0.72) with moderate TRNSP (M = 3.46, SD = 0.81) and PVC (M = 3.21, SD = 0.84), alongside strong PI (M = 3.97, SD = 0.70) and SAT (M = 3.90, SD = 0.68). Correlations showed trust as a central mechanism (TR with PI r = .61, p < .001; TR with SAT r = .55, p < .001), while privacy concern reduced trust (PVC with TR r = −.34, p < .001). In the main PI regression, the model explained substantial variance (R² = .54, p < .001), with the strongest predictors being PR (β = .31, p < .001) and TR (β = .29, p < .001); PQ (β = .18, p = .003), TRNSP (β = .12, p = .019), REI² (β = .15, p = .006), and AAAT (β = .11, p = .022) added significant positive effects, while PVC showed a smaller negative effect (β = −.09, p = .041). Exposure intensity also produced clear practical differences: high REI² users reported higher PI (M = 4.18) than low REI² users (M = 3.62), indicating an interpretable engagement-linked uplift. These findings imply that enterprise e-commerce teams should prioritize perceived relevance and trust-building transparency (explanations and control cues) while implementing privacy-assurance and preference-editing features to reduce trust erosion and improve purchase and loyalty outcomes.