Measuring The Effects Of Marketing Actions: The Role Of Matching Methodologies

Matching methods are used to provide observational studies with characteristics similar to randomised experiment, promoting the selection of subsamples of the treatment and control groups to be balanced on observed covariates.We describe the main steps of the Matching process implementation and the balance measures used in the evaluation of the resulting matched samples. We illustrate the use of a quasi-experimental design in a Marketing application referring to the effect of an in-store campaign of a new fragrance. Matching allows to conclude whether there is a significant difference in awareness of the new fragrance between clients that entered the stores with the product's promotion and those who entered stores with no campaign. The use of Matching provides a fair comparison between the two groups of clients, contradicting preliminary results obtained with the Pearson Chi-square test of independence.Two methods for Matching are applied: univariate Propensity Score (PS) Matching and multivariate Genetic Matching, based on available covariates plus PS. Propensity Score estimation resorts to Logistic regression and Random Forest. The use of Genetic Matching overcomes the need of successive trial and error iterations — to adjust the PS model, perform matching, check balance and return to a different specification of PS model in order to improve balance.