Modéliser les phénomènes multiniveaux en marketing

Abstract

Comme en témoigne la littérature internationale, le recours aux modèles multiniveaux, bien que récent en marketing, connaît un fort développement. Or cette méthode, qui s'applique à des données structurées en niveaux distincts (ex.: consommateurs/magasins/enseignes), n'est pas évidente à comprendre et mettre en œuvre sans explications appropriées. Ce texte, qui s'appuie sur une illustration marketing, souligne les objectifs, l'intérêt et les limites des modèles multiniveaux.

The influence of social belonging and creativity on the immoral consumption behaviors of new adolescents: A cross-cultural study of 16–24 year olds

“…Marketing research has previously examined the country’s impact on the mechanisms underpinning adolescent consumption behaviors (Gentina et al, 2014; Yang et al, 2014; Yang and Laroche, 2011) or immoral consumption behaviors (Flurry and Swinberghe, 2016; Gentina et al, 2018). But there has been research on the joint effect of the country of residence and cultural values on the immoral consumption behaviors of adolescents, a gap also highlighted by Ray and Ray (2008) who recommend using a single model to examine variables linked to country and cultural values.…”

“…Model II shows Level 1 fixed effects to assess the relationship between Level 1 predictor and consumer unethical beliefs. Model III shows Level 1 random effects to examine the extent to which relationships between level 1 predictors and consumer unethical beliefs vary between Level 2 units. Model IV involves Model III and Level 2 fixed effects to assess the relationship between Level 1 predictors and consumer unethical beliefs. ** In their synthesis article “Multilevel modeling for marketing: a primer” published in the journal Recherche et Applications en Marketing, Ray and Ray (2008, p. 65) explain that “ The results resemble, in part, those of an ordinary regression, except that the tools used to measure the model’s goodness of fit do not include R 2 but are tools specific to maximum likelihood estimations (the value of the maximum likelihood function at its maximum and criteria derived from this value): AIC ( Akaike’s Information Criterion ), BIC or SBC ( Bayesian Information Criterion or Schwarz’s Bayesian Criterion ), and so on, with which users of Logit, Probit and other nonlinear models are quite familiar.” …”

“… ** In their synthesis article “Multilevel modeling for marketing: a primer” published in the journal Recherche et Applications en Marketing, Ray and Ray (2008, p. 65) explain that “ The results resemble, in part, those of an ordinary regression, except that the tools used to measure the model’s goodness of fit do not include R 2 but are tools specific to maximum likelihood estimations (the value of the maximum likelihood function at its maximum and criteria derived from this value): AIC ( Akaike’s Information Criterion ), BIC or SBC ( Bayesian …”

L’influence de l’appartenance sociale et de la créativité sur les comportements de consommation immoraux des nouveaux adolescents: une étude cross-culturelle auprès des 16–24 ans

“…Des travaux en marketing avaient examiné l'impact du pays sur les mécanismes sous-jacents aux comportements de consommation des adolescents (Gentina et al, 2014 ;Yang et al, 2014 ;Yang et Laroche, 2011) ou aux comportements de consommation immoraux (Flurry et Swimberghe, 2016 ;Gentina et al, 2018). Mais aucune recherche n'avait étudié l'effet conjoint du pays de résidence et des valeurs culturelles sur les comportements de consommation immoraux des adolescents, limite d'ailleurs soulignée par Ray et Ray (2008) qui préconisent d'examiner au sein d'un même modèle des variables liées au pays et aux valeurs culturelles.…”

Balanced market orientation

“…In more detail, the link between customer satisfaction and customer loyalty has been one of the hypothesis made in several studies (Anderson & Sullivan,1993;Bearden et al, 1980;Bolton & Drew, 1991;Fornell, 1992;Bei & Chiao, 2006;Ray & Ray, 2008), applied to the tourism sector, as well (e.g. Andreassen & Lindestad, 1998), showing that a non-linear relationship may also exist between the two constructs (Bowen & Chen, 2001…”