Understanding the self-perceived customer experience and repurchase intention in live streaming shopping: evidence from China

The rapid growth of live streaming shopping (LSS) in China, enabled by 5G and accelerated by COVID-19, has brought a large user base but also high return and cancellation rates, indicating room to optimize user experience and repurchase behavior. LSS differs from traditional e-commerce through heightened focus on streamers, real-time social interactivity (e.g., bullet screens, live video, comments), and aggressive discounting claims, which jointly shape users’ shopping experiences. Prior research has emphasized purchase intention drivers (e.g., interaction quality, IT affordances, music), but repurchase intention—the key to model maturity—remains underexplored, especially amid concerns such as false promotion and high return rates. This study investigates whether users’ current LSS experiences align with their initial expectations and how self-perceived experiences influence satisfaction and repurchase intention. Grounded in expectation confirmation theory (ECT), the study proposes and tests a model linking perceived interactivity, perceived quality, perceived discounts, and perceived value to confirmation, satisfaction, and repurchase intention, while examining gender and other demographics as controls.

The study builds on the Theory of Reasoned Action (TRA) and Expectation-Confirmation Theory (ECT). TRA explains behavioral intentions through attitudes and subjective norms and has been widely applied to purchase intention across domains (e.g., organic food, luxury fashion, sustainable clothing, functional foods, and eWOM), with evidence that attitudes and social norms shape intentions; past behavior also correlates with repurchase intention. ECT (Oliver, 1980) posits that satisfaction and continuance behavior arise from the comparison of perceived performance with prior expectations (confirmation/disconfirmation). ECT and related ECM have been used to predict repeat purchase and continuance in e-commerce and IS contexts (e.g., service quality, fintech, impulsive buying). The authors argue LSS is consumer behavior best examined via ECT (post-purchase focus) rather than ECM (IS continuance). They identify a gap in research on repurchase intention in LSS and propose an ECT-based model where self-perceived interactivity, quality, discounts, and value influence confirmation and satisfaction, which in turn affects repurchase intention. A review table summarizes related studies using TRA/ECT/ECM, their data, methods (mostly SEM/PLS-SEM), and key findings, underscoring satisfaction’s central role and the mediating role of confirmation in various contexts.

Design: Structural equation modeling (SEM) was employed to validate a conceptual model linking perceived interactivity (PI), perceived quality (PQ), perceived discounts (PD), and perceived value (PV) to confirmation (CO), satisfaction (SA), and repurchase intention (RE), with controls (gender, age, education, income, platform). Software: AMOS 22.0 and SPSS 25.0. Sampling and data collection: An online questionnaire (Questionnaire Star, https://www.wjx.cn) was distributed in China from July 6, 2022 to September 16, 2022. Two pre-surveys were conducted to refine items. Recruitment extended beyond college students via WeChat and other social media. Eligibility required prior LSS experience. Final sample N=507. Demographics: 58.97% female; 41.03% male. Age: under 18 (12.23%), 18–25 (34.12%), 26–35 (31.36%), 36–50 (17.95%), over 50 (4.34%). Education: bachelor’s or higher 76.73%. Occupation: corporate staff 55.82%. Monthly income concentrated at 5001–10000 yuan (48.72%). Usage patterns: 92.31% had used LSS less than two years; platforms most used: TikTok 27.02%, Taobao 26.63%; typical monthly LSS spend clustered at 2000–4000 yuan (28.99%) and under 500 yuan (28.60%). Measures: Multi-item Likert scales for PI, PQ, PD, PV, CO, SA, RE. All standardized factor loadings were significant (p<0.001). Reliability: Cronbach’s α>0.8 for all constructs; Composite reliability (CR)>0.883; AVE≥0.655. Convergent validity satisfied. Discriminant validity: Fornell-Larcker criterion met (square roots of AVE exceeded inter-construct correlations). HTMT ratios <0.85 confirmed discriminant validity. Common method bias: Harman’s single-factor test showed the first factor explained 43.705% (<50%). CFA models with and without a common latent factor indicated negligible improvement with CLF; thus, CMB not a major concern. Non-response bias: t-tests between early and late respondents showed no significant differences. Model fit: CFA (no CLF): χ²/df=2.224, RMSEA=0.049, GFI=0.894, AGFI=0.873, CFI=0.958, IFI=0.958. Structural model: χ²/df=2.651, RMSEA=0.057, GFI=0.877, AGFI=0.854, CFI=0.943, IFI=0.943, indicating satisfactory fit. Hypotheses: H1 SA→RE (+); H2 CO→SA (+); H3 PI→SA (+); H4 PI→CO (+); H5 PQ→SA (+); H6 PQ→CO (+); H7 PD→SA (+); H8 PD→CO (+); H9 PV→SA (+); H10 PV→CO (+). Mediation of CO between self-perceived variables and SA tested via bias-corrected bootstrapping (MacKinnon; Preacher & Hayes).

- Path results (standardized coefficients): SA→RE β=0.757, p<0.001 (H1 supported); CO→SA β=0.300, p<0.001 (H2 supported). PI→SA β=0.189, p<0.001 (H3 supported); PI→CO β=0.187, p<0.001 (H4 supported). PQ→SA β=0.158, p<0.001 (H5 supported); PQ→CO β=0.149, p<0.01 (H6 supported). PD→SA β=0.015, ns (H7 not supported); PD→CO β=0.325, p<0.001 (H8 supported). PV→SA β=0.343, p<0.001 (H9 supported); PV→CO β=0.159, p<0.01 (H10 supported). - Relative effects on confirmation: PD had the strongest effect (β=0.325), followed by PI (β=0.187), PV (β=0.159), and PQ (β=0.149). - Explained variance: RE R²=59.5%; SA R²=60.6%; CO R²=41.6%. - Controls: Gender significantly affected repurchase intention; other controls (age, education level, income, platform) showed no significant effects. - Mediation (bootstrapping): Confirmation significantly mediated effects of PI, PQ, PD, PV on satisfaction. Indirect effects: PI→CO→SA=0.056 (95% BC CI [0.025, 0.105], partial mediation); PQ→CO→SA=0.045 ([0.012, 0.091], partial); PD→CO→SA=0.097 ([0.050, 0.169], complete mediation; direct PD→SA nonsignificant); PV→CO→SA=0.048 ([0.012, 0.102], partial).

Findings show that users’ self-perceived interactivity, quality, discounts, and value significantly enhance confirmation of expectations during LSS. Confirmation then increases satisfaction, which is the strongest predictor of repurchase intention. Discounts play a prominent role in confirming expectations but do not directly improve satisfaction; their effect on satisfaction operates entirely through confirmation, suggesting users need to validate perceived price benefits to feel satisfied. Interactivity, perceived quality, and perceived value directly and positively impact satisfaction and also increase confirmation, underscoring the importance of immersive interaction, product quality assurance, and perceived cost-effectiveness in shaping attitudes and behaviors. Gender differences significantly influence repurchase intention, indicating shopping perceptions and decision processes differ by gender, whereas age, education, income, and platform do not show significant effects. These results extend ECT to LSS repurchase behavior by highlighting the mediating role of confirmation between self-perceived experience and satisfaction and emphasizing post-purchase dynamics.

This study develops and validates an ECT-based SEM framework explaining LSS users’ repurchase intention through self-perceived interactivity, quality, discounts, and value, mediated by confirmation and leading to satisfaction. Key contributions include a post-purchase focus on repurchase intention in LSS, empirical evidence that confirmation mediates the effects of self-perceived experiences—completely for perceived discounts—and strong support for satisfaction as a driver of repurchase intention. Practically, platforms and streamers should enhance interactivity, maintain credible and transparent discounting strategies, ensure high product quality, and design gender-sensitive experiences to foster satisfaction and repeat purchases. Future research should broaden to non-adopters, integrate dual-processing perspectives for multi-modal LSS stimuli, and incorporate objective behavioral and eye-tracking data, as well as real transaction records, to triangulate findings.

- Sample focuses on experienced LSS users; determinants for non-adopters were not examined. Future work should explore barriers and adoption drivers among non-users. - Cognitive processing of rich multimedia and interactive stimuli was not modeled; integrating ECT with dual-processing theory could clarify how different information streams affect confirmation and satisfaction. - Data rely on self-reported questionnaires; future studies should collect objective measures (e.g., eye-tracking during interactions and discount presentations) and real transaction data, subject to ethical and legal compliance, to validate behavioral effects.