H24 : An agent-based model to investigate the effects of social segregation around the clock on social disparities in dietary behaviour

Clémentine COTTINEAU, CNRS, Centre Maurice Halbwachs

Julien PERRET, IGN, LaSTIG

Romain REUILLON, CNRS, Institut des Systèmes Complexes / Géographie-cités

Sébastien REY-COYREHOURCQ, Université de Rouen, IDEES

Julie VALLÉE, CNRS, Géographie-cités

EIGHTIES

slides : https://goo.gl/Nrpptu

Modelling human behaviours

Modelling territories

Spatial representation of individuals to understand their behaviours with Agent Based Model

 

Keep it Simple

Keep it Detailed

[Auchincloss et al., 2011]

[Nagel, Auxhausen et al.]

Spatial representation in ABM

Abstracting the characteristics of space

Space as an input parameter

... What about time?

ex: SpaceMatters project with population density grids as varying initial spatial conditions [Raimbault et al., 2017]

Between simplistic simplification and singularity

 

- Spatio-temporal heterogeneity

- Constraints on interaction

Neutral Landscape Model

Thomas R. Etherington

Space Matters

J. Raimbault et al.

Thematical challenge

Place /neighbourhood effects :

- matter when exploring human behaviors in health, education, voting etc.

- hard to prove because people (i) adapt to each other, (ii) sort themselves between places and (iii) only live once, which makes alternative spatial scenarios impossible to observe.

 

Interactions between people & places remain often static in statistical models

- either without including people's daily trajectories

- or with people's daily trajectories but without including place's daily trajectories

 

Need ABM to explore people's and place's daily trajectories and their impact on human behaviours

Objectives

Explore effets of segregation around the clock on extent of social disparities in human behaviours


Are social disparities in human behaviours greater...

(i) when night locations are spatially segregated rather than randomly attributed ?

(ii) when day locations are taken into consideration besides  night locations ?


Random locations Survey-based locations
Night only (ie. residence) Scenario 1 Scenario 2
Night & day Scenario 3 Scenario 4

Diet behaviours

Convergent findings about social disparities in fruit and vegetable consumption

Pionner research using ABM to explore social disparities in diet, but only in the context of (simplistic) residential segregation

J Am Diet Assoc. (2008)

J Am Diet Assoc (2010)

National nutrition policies

eg.  Programme National Nutrition Santé (PNNS) in France since 2001

SSM - Population Health (2016)

Am J Prev Med (2011)

(French) data

Diet dynamics

Spatio-temporel locations

explored with same sociodemographic groups (18)

Night (ie. residence) : 2012 census

Day & evening : 2010 origin-destination survey

Baromètres Santé Nutrition

(1996 ; 2002 ; 2008)

Sex (male ; female)

X

Age (15-29 yrs.; 30-59 yrs.; 60 yrs. and +)

X

Education (poor ;  middle ; up)

Population generation

Agent diet opinion and behaviour initially defined  according to sociodemographic category

8,16 millions of agents

Spatial grid

8540 inhabited cells (1km X 1km)

Agent

with night/day/evening cells

with sociodemographic attributes

68 %  in a 'day' cell  ≠ the 'night' cell

Ile-de-France

Initial dietary behavior

From 1996 data : 5-a-day fruit and vegetable consumption

 

Source : Baromètres Santé Nutrition (1996 ; 2002 ; 2008)

Initial dietary behavior

Social Ineq.

Day_Slice

% Healthy

Model structure

Scenarios 1 & 2:

1 time slice per day

Scenarios 3 & 4:

3 time slices per day

Spatial interactions

Modelling opinion change

Modelling behaviour change under constraints

Parameter expectations

Parameter Mecha-nism Range if min if max Influence on 5-a-day
Interaction Spatial Interaction  [ 0 ; 1 ] Observation of cell behaviour only Observation of partner opinion only
Reward Behaviour-Opinion  [ 0 ; 1 ] No Feedback from behaviour Healthy behaviour strengthens opinion
Inertia Opinion-Behaviour  [ 0 ; 1 ] Opinion depends on influences only Stable opinion
Switch Proba Opinion-Behaviour  [ 0 ; 1 ] No change of behaviour Behaviour matches opinion
Constraint Opinion-Behaviour  [ 0 ; 1 ] No constraint Constraint prevents behaviour switch

+

+

+

?

?

Evaluation protocol

A - Exploring behavior of standalone model

B - Invalidate behavior by falsification, an inverted calibration

optimisation using evolutionary algorithms (EA)

method : maximise diversity using pattern exploration algorithm

expected patterns : Consistent dynamics of the model in terms of opinion diffusion and behaviour change

space : one cell with 100 people

space : 8540 cells

method : minimise distance to data using multi-objective evolutionary algorithm (EA)

expected patterns : random scenario cannot produce coherent data and inequality between education groups

Model 50hh - 50uu

Parameter Mecha-nism Range Expected influence Significant coefficient
Interaction Spatial Interaction  [ 0 ; 1 ]
Reward Behaviour-Opinion  [ 0 ; 1 ]
Inertia Opinion-Behaviour  [ 0 ; 1 ]
Switch Proba Opinion-Behaviour  [ 0 ; 1 ]
Constraint Opinion-Behaviour  [ 0 ; 1 ]

+

+

+

?

+

?

Multiple R-squared:  0.0828

Model 50hm - 50um

Parameter Mecha-nism Range Expected influence Significant coefficient
Interaction Spatial Interaction  [ 0 ; 1 ]
Reward Behaviour-Opinion  [ 0 ; 1 ]
Intertia Opinion-Behaviour  [ 0 ; 1 ]
Switch Proba Opinion-Behaviour  [ 0 ; 1 ]
Constraint Opinion-Behaviour  [ 0 ; 1 ]

+

+

+

?

?

+

+

Multiple R-squared:  0.3023

Model 90hh - 10uu

Parameter Mecha-nism Range Expected influence Significant coefficient
Interaction Spatial Interaction  [ 0 ; 1 ]
Reward Behaviour-Opinion  [ 0 ; 1 ]
Intertia Opinion-Behaviour  [ 0 ; 1 ]
Switch Proba Opinion-Behaviour  [ 0 ; 1 ]
Constraint Opinion-Behaviour  [ 0 ; 1 ]

+

+

+

?

?

+

Multiple R-squared:  0.2611

Model 10hh - 90uu

Parameter Mecha-nism Range Expected influence Significant coefficient
Interaction Spatial Interaction  [ 0 ; 1 ]
Reward Behaviour-Opinion  [ 0 ; 1 ]
Intertia Opinion-Behaviour  [ 0 ; 1 ]
Switch Proba Opinion-Behaviour  [ 0 ; 1 ]
Constraint Opinion-Behaviour  [ 0 ; 1 ]

+

+

+

?

?

+

Multiple R-squared:  0.149

Aggregated output

Specification of a simulation run:

Observables

Obs. 1 : Distance to data

SI = \sum_{i=1}^2{\sum_{j=1}^3{ \frac{ShareHealthy_{i,j,e=3}}{ShareHealthy_{i,j,e=1}} \times \frac{N_{i,j}}{N}}}

2002

2008

steps

years

3 slices

1996

simulated

data

\%healthy_{age-sex}
\%healthy_{age-sex}

Obs. 2 : Social inequality

MSE (

)

for each category

of age ( i ) & sex ( j )

ratio between more (3) &

less (1)  educated

weighted

by sex & age category

mesure inequality between extreme education groups at equal age and sex category

2002, 2008

2008

Model Evaluation

3 objectives to minimize :

Obj 1. : MSE sum for random scenario

Obj. 3 : The difference between social inequality in mobility scenario and in simpler scenario

\Delta(SI_{random},SI_{mobility})

Exceptation ?

obj1

obj2

obj3

Obj. 2 : MSE sum for mobility scenario

\sum\nolimits_{i \in(2002,2008)}{MSE_i}

min obj. 1

min obj. 2

 high obj. 3

Preliminary results

Random locations Survey-based locations
Night only (ie. residence) Scenario 1
Night & day Scenario 4
Parameter Mechanism Value
Interaction Spatial Interaction 0,5
Reward Behaviour-Opinion 0,5
Inertia Opinion-Behaviour 0,5
Switch Proba Opinion-Behaviour 0,5
Constraint Opinion-Behaviour 0,2

Social Ineq.

Day_Slice

% Healthy

% Healthy

Day_Slice

Social Ineq.

Conclusion & Perspectives

An original approach in (health) geography

An inter-disciplinary team with challenges in each domain :

  • thematic point of view
  • data analysis : EGT, nutrition, etc.
  • in software engineering : ~ 8 millions moving agents !
  • in exploration : High Performance Computing with dedicated evolutionary algorithms
  • in epistemology : Validation of realistic model ?

Thank you!

Bonus 1: Synthetic Population Generation

Bonus 2: Formal Model

influence\_partner_{i,j,c} = opinion_{j,c}
influence\_cell_{i,c} = \frac{n\_healthy\_neighbours_c}{n\_all\_neighbours_c}
rewardedOpinion\_o_{i,t}= opinion_{i,t}
min(1, (1 + healthyDietReward) * opinion_{i,t}))

if i is unhealthy at time t

otherwise

newOpinion\_o_{i,t} = inertiaCoefficient \times rewardedOpinion + (1 - inertiaCoefficient) \times
( interpersonalInfluence \times influence\_partner_{i,c} +
(1 - interpersonalInfluence) \times influence\_cell_{i,c} )
x_i = f(y_i,o_i)=
max(0, y_i * (2 * newOpinion_i - 1))
max(0, y_i * (-2 * newOpinion_i + 1))

if i is unhealthy at time t

if i is healthy at time t

y_i= maxProbaToSwitch - n_i * constraintStrength

Free parameter

 = number of constraints of agent i

n_i

= agent

i

= interacting partner

j

= cell

c

= switch probability

x_i

Bonus 3: Switching mechanism

H24 - CIST 2018

By sebastien rey coyrehourcq

H24 - CIST 2018

  • 895

More from sebastien rey coyrehourcq