Empirical Investigation of Habit, Variety-Seeking, and Satiation in Snack Consumption Using Multiple Discrete-Continuous Framework




Journal Title

Journal ISSN

Volume Title




This dissertation consists of three research papers examining the role of satiation and state-dependence, choice sets, latent segments in the context of snack consumption. The three chapters specifically examine whether habituation or variety-seeking govern snacking. In the second chapter, we explore variety seeking behavior in a richer context. By using the multiple discrete-continuous extreme value (MDCEV) framework we estimate a model that captures choice of multiple alternatives and quantity consumption. We investigate the effects of satiation and state dependence and use a rich panel data of individual snack consumption to estimate the model estimate the model. We use consumption data of individuals recorded through hand-held devices and model consumers' choices from a variety of snack categories. Using a single framework, we separate the effects of satiation, intrinsic utility, and state dependence. Our modeling approach provides evidence of greater variety seeking in consumers at a brand level than at the category level within a day across time-periods. Across days, we find that category consumption choices are driven by habituation. We find evidence of satiation or diminishing marginal utility, and that satiation varies by snack categories and by dayparts. We show that by accounting for state-dependence and unobserved heterogeneity, the fit for MDCEV model improves tremendously over the base model that doesn’t capture neither of these factors. In the third chapter, we propose a new framework for modeling consideration sets in the MDCEV choice model framework. Using a gradient boosting algorithm from machine learning literature, we predict alternatives that are most likely to be chosen by a consumer at a daypart. In doing so, we reduce the computational burden associated with consideration set enumeration. These consideration sets are constructed as a function of dayparts, prior choices and prior choices, allowing us to predict alternatives that vary across individuals and time of the day. Our modeling approach allows us to estimate bias in parameter estimates, which is an outcome observed when choice models are estimated without inclusion of consideration sets. Using a rich panel data of individual level snack consumption, a setting where multiple discreteness and quantity choices play a role, along with groups of alternatives that are usually considered by individuals based on the time of consumption, we calibrate estimate the parameters of the model. We show that the proposed method provides a superior model fit by about 50% and reduces bias in parameter estimates compared to the base model. Using the proposed approach, we conduct two thought experiments – how does calorie consumption change when the time of consumption of a snack is changed and when a snack with switched with another snack. In the fourth chapter, we uncover latent segments of consumers using their snack consumption behavior using the individual level snack consumption data. We estimate a model of choices and quantity consumption using the multiple discrete-continuous framework with latent segments. Our approach results in a three-segment structure for the snack consumers which are labeled as “old, overweight and inactive”, “male and obese” and “young and active”. Since our model captures both preference for alternatives and quantity choices, we are able to get a better picture of consumption behavior. Latent segment models relied on the multinomial logit framework to uncover segments of consumers purely based on preferences alone. A fundamental assumption of is this model is that consumers face constant marginal utility. However, consumers do face diminishing marginal utility as we consume more of an alternative. Through the MDCEV framework, we relax this assumption and the models enables us to estimate a satiation parameter that captures diminishing marginal utility, thus giving us a complete picture of consumption behavior. To our knowledge, this is the first paper in marketing to show that satiation can also be used an additional dimension for customer segmentation apart from consumer preferences. We find that category consumption is governed by habituation across days in just one of three segments. Within a day, the “male and obese” segment seeks more variety in category consumption over the other segments. We find that all three segments are brand variety-seekers within a day while habituated across days for brand choices. Preference levels for each category varies across segments, while satiation levels also differ across segments. We create profiles for the three segments and find that the calorie consumption varies significantly across the three segments varies by categories. Our results have implications for managers interested in creating optimal consumption bundles and for policymakers interested in addressing over-consumption leading to obesity among US consumers.



Stimulus satiation, Food habits, Snack foods, Food -- Caloric content, Expectation-maximization algorithms



©2020 Sriharsha Kamatham. All rights reserved.