Six Methods for Latent Moderation Analysis in Marketing Research: A Comparison and Guidelines

Abstract

Moderation analysis investigates the conditions under which variables affect outcomes. In marketing, it is common that at least one of the target moderation variables is latent and measured by multiple indicators with measurement error. This paper compares six methods for latent moderation analysis: multi-group, means, corrected means, factor scores, product indicators, and latent product. It reviews their use in marketing research, describes their assumptions, and compares their performance with Monte Carlo simulations. Several recommendations follow from the results. First, although the means method is the most frequently used method in the review (95% of articles), it should only be used when reliabilities of the moderation variables are close to one, which is rare. In that situation, all methods except the multi-group method perform similarly well. Second, the results support the use of the factor scores method and latent product method when reliabilities are smaller than one. These methods perform best with parameter and standard error bias ≤ 5% under most investigated conditions. Third, specific settings can warrant the use of the multi-group method (if the moderator is discrete), the corrected means method (if moderation variables are single-indicators) and the product indicators method (if indicators are non-normally distributed). Practical guidelines and sample code on four statistical platforms (SPSS, Stata, R and Mplus) are provided to stimulate the adoption of best practices for latent moderation analysis.

Psychological barriers moderate the attitude-behavior gap for climate change

“…Finally, our last aim was to assess if the psychological barriers moderated the influence of attitudes on the frequency of pro-environmental behaviors, thereby partly explaining the attitude-behavior gap. We performed a latent moderation analysis using a factor score procedure [ 87 ]. Although this approach neglects measurement error, given our sample size ( N = 937) estimation error for the correlations in our sample should be low [ 88 ].…”

Coercion, Consent, and Class Consciousness: How Workers Respond to Amazon’s Production Regime

“…We use a multigroup analysis to examine the moderated mediation proposed in hypothesis 4. This analysis is most appropriate when the moderating variable is a single observed binary indicator, which is the case for financial strain (Cheung et al 2021; Pieters, Pieters, and Lemmens 2022). Findings from our interviews are woven into the quantitative analysis, lending greater depth and meaning to the associations our survey results provide.…”

Pourquoi et à quelles conditions un dirigeant transformationnel stimule-t-il l’orientation marché de son entreprise ? Le cas des PME sénégalaises

“…Nous modélisons une interaction entre deux variables latentes (i.e., orientation marché du dirigeant et TFL). Pour modéliser cet effet, nous avons suivi les recommandations de Pieters et al (2022). Lorsque la fiabilité du terme d’interaction est inférieure à 1 (dans notre cas 0,815 ; Equation 6 ; Pieters et al, 2022 : 943), la méthode des scores factoriels est recommandée pour calculer le terme d’interaction entre variables latentes.…”

“…Pour modéliser cet effet, nous avons suivi les recommandations de Pieters et al (2022). Lorsque la fiabilité du terme d’interaction est inférieure à 1 (dans notre cas 0,815 ; Equation 6 ; Pieters et al, 2022 : 943), la méthode des scores factoriels est recommandée pour calculer le terme d’interaction entre variables latentes. Nous avons dès lors réalisé deux estimations : l’une par CB-SEM/ML (Covariance-Based Structural Equation Modeling), l’autre par PLS-SEM.…”

“…Nous avons dès lors réalisé deux estimations : l’une par CB-SEM/ML (Covariance-Based Structural Equation Modeling), l’autre par PLS-SEM. Nos résultats étant en effet susceptibles d’être biaisés par la violation de la condition de multi-normalité des indicateurs (Pieters et al, 2022), il était important de les cross-valider par l’emploi d’une méthode d’estimation non sensible à cette condition (i.e., PLS-SEM). Nos résultats sont détaillés dans le Tableau 2.…”