Brief introductory guide to agent-based modeling and an illustration from urban health research

Auchincloss, Amy H.; Garcia, Leandro Martin Totaro

doi:10.1590/0102-311X00051615

Amy H. Auchincloss

School of Public Health, Drexel University, Philadelphia, U.S.A

Leandro Martin Totaro Garcia

Faculdade de Saúde Pública, Universidade de São Paulo, São Paulo, Brasil

Correspondence A. H. Auchincloss School of Public Health, Drexel University. 3215 Market Street, Philadelphia/PA – 19104, U.S.A. aha27@drexel.edu

Contributors

A. H. Auchincloss conceived the original idea of the paper, discussed, wrote and reviewed its content and approved the final version. L. M. T. Garcia discussed, wrote and reviewed the paper content and approved the final version.

Thumbnail

Figure 1
Conceptual sketch of core features to include in the model. The sketch illustrates structural feedback between households and stores and behavioral feedback within households and stores (arrows around them). Households choice depends on their income, proximity to stores and food preferences. On the other hand, stores influence households’ diet habit.

Thumbnail

Figure 2
This four-panel figure illustrates the types of displays one can get from one run of a model. The scenario shown is where poor households were segregated from wealthy households and poor households were near stores with healthier foods. Panel 2a is a snapshot of the grid (world) were agents interact. Households are squares, stores are diamonds. Colors and shading map to select agent characteristics. Panels 2b,is an example of output that can assist with verification and validadtion: average number of customers at expensive stores (top line is total customers, middle line is high income, bottom line is low income). Panel 2c is a secondary outcome: healthier and unhealthy stores (top line is total stores, middle line is unhealthy, bottom line is healthier). Panel 2d is the main outcome: average proportion of times the household shopped at a helathier food store (on left side of plot top line is low income, bottom line is high income)11.

Thumbnail

Figure 3
Example of summary data from different scenarios. The x-axis show five store behavior algorithms that used various probabilities for store move-out/move in and changes in the type of foods are sold at the stores. The purpose of this plot was to illustrate how changes in store dynamics influence diet. Each algorithm was run 60 times to obtain the distribution of outcomes and then summarized as the median (symbol in the graph) and the 5th to 95th simulation percentile (bar in the graph). The right y-axis represents the average proportion of times that households shopped at a healthy food store. The left y-axis represents the difference in proportions between diet of high- and low-income households. Diet was derived from the average proportion of times the household shopped at a healthy food store 11.

Thumbnail

Figure 4
Outline of high-level description of the model presented in Auchincloss et al. 11.

Thumbnail

Table 1
Functions for weighting and scoring the inputs for utility.

Thumbnail

Dimension	Justification and description for scoring		Weight *	High-income	Low-income
Price **	Price score. In our model, both low and high-income households preferred a cheap store – thus all households scored a cheap store as 1. To account for the fact that rich households were comparatively insensitive to price, rich households in our model scored an expensive store as 0.8 (i.e., close to 1), while poor households scored it 0.1 (i.e., close to 0)		0.2
		Prices at the food store are cheap		1.0	1.0
		Price at the food store are expensive		0.8	0.1
Distance **	Both low- and high-income households preferred a cheap store, thus, all households scored a cheap store as 1. Both high- and low-income households equally preferred a closer store but, because travel was more of an obstacle for low-income households, poor households gave a distant store a lower score than a high-income household. The distance score was 1 – (number of grid cells between residence and store) / 30 for low-income households, and 1 – (number of grid cells between residence and store) / 130 for rich households		0.5
		Distance between household and store = 0		1.0	1.0
		Distance between household and store > 0		1 - d / 130	1 - d / 30
Habit	Households remembered the store(s) they visited during the past five time steps. They gave a score of 1 (highest) to the store they last shopped at and decreased the score by 1/5 for each model step since they last visited that store		0.1
		Same store that household shopped at on previous time step		1.0	1.0
		Different store (steps since = number of time steps since the household last shopped at store)		1 - steps / 5	1 - since / 5
Food preference	Preference scores in the utility just reflected the household’s preference: households scored a healthy store with their own household preference; an unhealthy store got a score of 1 – household preference. Household food preference was a continuous attribute 0 to (preference for unhealthy food = 0, preference for healthy food = 1). Preference was either randomly assigned or assigned by household income in our “preference experiments” (unhealthy preference was assigned the lower end of the preference range [0.0-0.6] and healthy preference the upper end of the range [0.4-1.0] with some overlap between possible preferences)		0.2
		Store sells healthy food		Preference	Preference
		Store sells unhealthy food		1 - Preference	1 - Preference

[1] Correspondence A. H. Auchincloss School of Public Health, Drexel University. 3215 Market Street, Philadelphia/PA – 19104, U.S.A. aha27@drexel.edu

Brasil

Brasil

Brief introductory guide to agent-based modeling and an illustration from urban health research

Breve guia introdutório de modelagem baseada em agentes e uma ilustração a partir da pesquisa em saúde urbana

Breve guía introductoria de modelación basada en agentes y una ilustración a partir de la investigación en salud urbana

Abstract