Modeling Obesity through Simulation (MOTS)

Research Question
The causes of childhood obesity are both simple and complex. In a trivial sense, childhood obesity is the result of energy imbalance, or consuming more calories than are burned. However, this imbalance arises from the interaction of multiple influences that are not easily modeled using statistics. In particular, statistical models have difficulty handling feedback, whereby dependent and independent variables influence one another. Many studies of childhood obesity inadequately address feedback, in particular the interactions of children with their peers, families, schools, and neighborhoods. We are also interested at interactions across levels; for example, how family influences can mitigate or exacerbate peer influence in obesity.

Modeling Approach
We employ an agent-based modeling approach that draws its parameters from statistical models. In the first stage of our model, we identify relevant parameters. Second, we choose simulations that best fit the empirical data, ideally longitudinal data. Finally, we manipulate parameters in the model or the environment in order to simulate policy effects. For example, we ask whether stigmatizing obesogenic behaviors increases social isolation in obese children. Our work to date has focused on separating peer influence from selection; future work will incorporate other influences (families, neighborhoods, and schools). Each of these influences is first developed as a module that can be verified (the simulation actually does what we intended it to) and validated (real-world patterns of obesity can be replicated). Separate models are employed for eating behaviors, physical activity, and obesity. The models are parameterized using results reported in the literature; where needed, parameters will be based on original analyses conducted using two datasets: the National Longitudinal Study of Adolescent Health (Add Health), and the Avon Longitudinal Study of Parents and Children (ALSPAC). These datasets will also be used for model verification (e.g., comparing simulations of changes in BMI to the empirical observations).

The five levels of the model, the agents of that level, and the most relevant attributes and behaviors for that level, are shown in the following table.





Individual child


baseline BMI, race-ethnicity, age, gender, grade level, family income

eating behavior physical activity/ inactivity change in obesity

Social networks


degree, reciprocity, transitivity, homophily (on race-ethnicity, family income, race, grade level, and age)

network changes (adding and dropping ties), peer influence

Families/ Households


family income, family structure (one- or two-parent), parental occupation and working hours, # televisions in home, race-ethnicity

structured meal time, family interactions, food purchasing decisions, residential mobility



school's income distribution and racial composition, federal assistance received by school, funding per student

sale of "competitive foods," open/closed campus policy, physical activity requirements



food environment, transportation systems, recreation facilities, walkability, crime rates, income distribution, home prices, duration of residence, racial segregation

changes to built environment, government subsidies

Note that agents may simultaneously occupy several levels of the model. For example, children may be considered agents at all levels of the model; parents are part of households, but also may interact with food outlets (through purchasing decisions). So that the model does not become too complex, each of these levels is modeled as a separate module. The time scales involved in each module differ: there is a great deal of social network change over the course of a year, but households change residence on average only every 4.7 years (Census TWPS0069).

Modules will be integrated by simultaneously modeling agents existing at multiple levels, with a focus on changes in behavior (eating, physical activity) or attributes (BMI) of children in the model. As noted, each module will be functional on its own, and then represented very simply when joined in a combined model. One approach to integration is to allow modules to control changes in BMI and related behaviors over time. The process will necessarily involve tuning parameters so that results are consistent with empirical evidence.


Principal Investigator

David A. Shoham, PhD MSPH
Assistant Professor
Preventive Medicine and Epidemiology
Loyola University Chicago

Amy Luke, PhD
Associate Professor
Preventive Medicine and Epidemiology
Loyola University Chicago

Co-Investigator & Modeler

Amy H. Auchincloss, PhD
Assistant Professor
Department of Epidemiology and Biostatistics,
School of Public Health
Drexel University


Liping Tong, Lara Dugas, Jun Zhang, and Richard Cooper
Loyola University Chicago

Katherine Kaufer Christoffel
Children's Memorial Hospital (Chicago) and Northwestern University

Liping Tong, Lara Dugas, Jun Zhang, and Richard Cooper
Loyola University Chicago

George Davy Smith and Andy Ness
Bristol University (UK)