From Social Media Content to Purchase Decisions in Local Bakeries: Specification-Sensitive Trust Mediation in a Predominantly Student Sample
Giovanny Haro-Sosa, Paúl Roberto Pino Falconí, Telmo Marcelo Zambrano Núñez, María Belén Bastidas Aráuz, César Alfredo Cabezas Guerrero, Ana Matilde Moreno Guerra, Edwin Guillermo Antamba Anrango, Efraín Rodrigo Romero Machado, María José Andrade Albán, Adriana Margarita Morales NoriegaSocial media research rarely explains whether visual esthetics, trust, and engagement jointly influence purchase decisions through content acceptance in local food retailing. This study examines their direct and indirect associations among bakery consumers in Riobamba, Ecuador, by integrating stimulus–organism–response, information-adoption, and elaboration-likelihood perspectives. A face-to-face convenience survey yielded 300 complete responses from adults who had purchased from a bakery within the preceding six months. The sample was predominantly student-based (57.7%), which restricts population-wide generalization. PLS-SEM was applied with 10,000 bootstrap subsamples, PLSpredict, CVPAT, and an AC5-deletion sensitivity analysis. Trust and engagement positively predicted content acceptance and purchase decisions, whereas visual esthetics showed no significant direct or indirect association. Content acceptance predicted purchase decisions. Engagement mediation remained significant in both specifications. The trust indirect effect was significant in the five-item primary model but became borderline and inferentially inconclusive after AC5 was removed; H8b is therefore classified as partially/conditionally supported. The model explained 73.3% of content acceptance and 66.2% of purchase decisions and outperformed the linear predictive benchmark. Unresolved discriminant-validity concerns between engagement and acceptance further qualify the findings. Thus, the evidence identifies relational credibility and engagement as relevant correlates of bakery purchase decisions within this student-heavy convenience sample, while the cross-sectional design precludes causal and population-level inference.