Journal on Policy & Complex Systems Volume 1, Number 2, Fall 2014 | Page 97

��������������������������
An Example : Modeling Social Determinants of Body Mass Index

An ABM effort needs to first begin

with a conceptual model sometimes referred to as a logic model , a step that is not unique to studying complex systems . Such a model , based on a mixture of theory , extant data , and speculation is an abstraction that attempts to portray the important forces at play that are thought to impact the health outcome in question . Figure 4 presents a portion of a larger model that colleagues and we have developed , in this case focusing on some of the determinants of BMI that extend beyond the usual energy balance formulation . We briefly describe here an attempt at simulating the impact of policies that improve the quality of education , the food environment , and the physical activity environment , and the effect of such policies on BMI and racial / ethnic disparities in BMI . The arrows indicate possible feedback between different determinants , and it is such patterns of feedback , particularly positive feedback , that characterizes complex systems and the dynamics of such systems .
Thus , we are trying to portray the possible ways in which individual characteristics like level of education , amount of physical activity , and dietary quality are influenced by neighborhood characteristics such as school quality , neighborhood SEP , availability of healthy food and places to engage in leisure-time physical activity , and social networks . They all in turn , and in interaction , influence BMI and disparities in BMI .
In order to move from this schematic representation to an agent-based model , we start from the “ bottom-up ,” populating this artificial world with several thousand non-Hispanic Black or White “ agents ” of various ages , who may be in school or have left / graduated , may or may not smoke , have some level of physical activity and diet quality , and a network of friends . The agents start out in a particular spatial location , live in households of 1 – 4 persons in 1280 housing units ( out of a possible 1600 housing units ) that are distributed over 64 neighborhoods , and there is residential mobility . The age , race / ethnicity , and income distributions of this population over the neighborhoods are based on U . S . Census data for the 100 largest US metropolitan statistical areas , supplemented by data from the Panel Study of Income Dynamics ( McGonagle , Schoeni , Sastry , & Freedman , 2012 ).
Agents enter school at age 6 , get jobs after they leave school , retire at age 65 , have one child at age 25 , and die at age-specific rates reflective of the US population . Other agent behaviors are , for the most part , based on information available from published studies , with some simplifications . The school that they attend and how long they attend is based on where they live , the quality of the school , parental income , and age , and the average number of years of education . Black / White differences in years of attendance reflect the patterns seen in the 2007 National Health Interview Survey ( NHIS ) ( Centers for Disease Control and Prevention National Center for Health Statistics National Health Interview Survey , 2008 ). Individual incomes have trajectories over time related to level of completed education and are modeled to reflect national data ( Day & Newburger 2002 ; He , Sengupta , Velkoff , & DeBarros , 2005 ). Households periodically move based on mismatch between their income and median household income of their neighborhood , and their decision to move is based on the availability of housing units they can afford , and preferences for racial composition in potential destination neighborhoods . All parameters
95