�����������������������������������������������������������������
have variability , either based on published studies or on a standardized estimation approach .
The neighborhoods that households live in have an average income level reflecting the income composition of the neighborhood , and school quality , physical activity infrastructure , availability of good food , and value of housing , are all based on neighborhood income level , again with estimates based on the published literature .
Individual behaviors related to physical activity and dietary quality also reflect a variety of determinants , with parameters again based on available data : education level , neighborhood physical activity infrastructure , physical activity of friends , parental physical activity , and , respectively , education level , neighborhood food environment , dietary quality of friends , and parental dietary quality . Friends are made based on age and racial similarity , and physical propinquity . Social norm processes affect the strength of the influence that one ’ s friends have on physical activity and diet quality . Friendship patterns , behaviors , and residential location are updated on a yearly basis .
Finally , BMI is influenced directly by dietary quality , physical activity , cardiovascular disease status , and is modeled so that average BMI is consistent with NHIS data .
�������������e eff���������������������
This model provides an opportunity to
address a set of policy manipulations by systematically examining the dynamic changes in model output over time after a given policy , or combination of policies , is introduced ( Lempert , 2002 ). While we envision these as explorations of policy impacts , they could also be seen as tests of the role these pathways play in disparities in BMI ( Ip et al ., 2013 ). Because the model tracks characteristics such as residential location and behavior of each agent over time as their virtual world changes , the proximal and distal impact of policy introduction can be seen , as well as any differential effects on subpopulations of agents , and effects occurring over generations ( approximately 2.5 generations in the current analyses ).
The policy manipulation involves simulating changes in three characteristics of neighborhoods that might have an impact on BMI and BMI disparities — the availability of good food stores , infrastructure for physical activity , and quality of schools . Each of these is operationalized based on published data . Access to good food is proxied by relative availability of supermarkets compared to other places to purchase food , and the measure of physical activity infrastructure is based on seven features associated with local exercise opportunities ( Sallis et al ., 2009 ). School quality is based on teacher to student ratio , which while not measuring all aspects of school quality , it is related to economic returns from education and other benefits of education ( Card , 1992 ). For each policy dimension , we identify the neighborhoods in the lowest quintile on that dimension and examine the effects of improving the measure from 0 to 4 standard deviations , in five steps , for those neighborhoods . The three policies are implemented independent of each other , resulting in 125 ( 5x5x5 ) different scenarios , ranging from no change on any dimension for the neighborhoods in the bottom quintile , the ���������� , to very large improvements ( 4 standard deviations above the mean ) on all three dimensions .
��������������������������
First , any model needs to be calibrated
so that with regard to key features it mirrors observed data ( Railsback &
96