Process Analysis for Marketing Research: An Abstract

Abstract

Process analysis is indispensable to contemporary marketing research. It is the theorization and testing of moderation and/or mediation hypotheses to obtain theoretically and managerially relevant insights into marketing processes. This dissertation presents three essays that apply, compare and extend such process analysis methodologies. Essay 1 applies mediation in an investigation of the referral reinforcement effect: referred customers are more inclined to make referrals than non-referred customers are. Study 1 is an analysis of a field-experiment among ridesharing customers. Study 2a re-analyzes a published combination of archival and survey data of a bank’s referral reward program. Study 2b is a new survey among moviegoers. Study 3 is a controlled lab-experiment. All find support for the referral reinforcement effect. Study 4 investigates customer lay beliefs and finds that other-directed rather than self-directed or product-directed motives drive the referral reinforcement effect. Overall, these studies find support for the referral reinforcement effect and provide insights in its processes. The referral reinforcement has useful implications for managers who aim to grow a customer base. Essay 2 compares six existing moderation methods in the face of measurement error. A quantitative literature review shows that taking a product of unweighted means is predominantly used for moderation analysis, in 94% of investigated articles. The remaining methods, multi-group, factor scores, corrected means, product indicators and the latent product method are barely used. A comparison of the assumptions of the methods and follow-up Monte Carlo simulations conclude that the accessible factor scores method performs much better than the dominant means method, and equally good or better than the remaining more sophisticated methods. We recommend the use of the factor scores method for moderation analysis and advise using samples that are 60% larger than that are currently common. Essay 3 focuses on discriminant validity as a precondition for meaningful process analysis. It extends bivariate discriminant validation criteria by taking a multivariate perspective. The proposed implementation is applicable to both raw and summary statistics data, accounts for measurement error in the variables and uses statistical tests of discriminant validity rather than heuristics. Case studies and an online web-based application apply the proposed methods in important multivariate theory-testing domains, multiple mediation and multidimensional measurement. Usage of the proposed methods contributes to construct validation and meaningful substantive theory tests. In sum, we hope that this dissertation demonstrates the strengths of process analysis methodologies, fosters valid applications, and inspires future research.