Structural equation models with adaptive regression and construction ofan index to validate constructs used to distinguish the profiles of specialtycoffee consumers

Abstract

This work consists of presenting a new approach to Adaptive Linear Regression adapted to structural equation models and improving the index related to the Average Variance Extracted (AVE), given a plug-in approach, and replacing the error variances with the factor loadings of the estimated adaptive regressions. To do so, a Monte Carlo simulation study was performed considering scenarios with different numbers of outliers, which were generated by distributions with symmetry deviations and kurtosis excess. Sample sizes were defined as n=50, 100 and 200. In formative structural models and considering outliers generated either from symmetrical distributions or from multivariate log-normal distributions, the Adaptive Linear Regression modeling was found to be efficient in the different scenarios under analysis. Likewise, for models with specification errors, this method was proven to have low efficiency, as expected. Furthermore, constructs were elaborated with variables that could enable both the characterization and the distinction of individuals among the different groups of Brazilian specialty coffee consumers and that could provide different perspectives on the transition among them. The results made it possible to better distinguish the consumers and better characterize the proposed categories, thus contributing to the improvement and simplification of marketing strategies used by players in this market. In addition, the results also promoted the discussion on which factors stimulate the transition of an individual from an initial construct to another, and we showed that transitioning from regular consumers to enthusiasts is easier than moving from enthusiasts to specialists.