Transcript South African Case Study Model Report - CAVES

South African Case Study
Model Report
Shah Jamal Alam, Ruth Meyer, Scott Moss
Centre for Policy Modelling, MMU
CAVES Project Meeting March 2007 ● CPM
1
Table of Contents
Declarative Model
• Current state of the model
• Experiences with Jess
• Next steps
Dynamic Networks Analysis
• Kolmogorov-Smirnov Test
• Motifs
CAVES Project Meeting March 2007 ● CPM
2
Declarative Model – Description
Individuals, households and villages
• Distributions extracted from empirical data (RADAR)
• Household size: Normal (7, 3)
• Household head age: Normal (56.2, 12.6)
• Marital status of household head: Empirical discrete,
different for female and male heads
• Age difference between spouses: Normal (8.43, 6.576)
• Type of household member: Empirical discrete
(child 0.653, grandchild 0.238, other 0.109)
• Age of household member: Gamma (2.4, 0.086)
• Number, age and gender of migrants: Empirical discrete
Decisions / behaviour on individual and household level
• Rules for individuals
• Rules for households
CAVES Project Meeting March 2007 ● CPM
3
Declarative Model – Groups
Church
• Importance rated very high according to RADAR data
• 80% of population are member of a church
• Implementation so far based on assumptions:
• 1-4 denominations per village, 1 church / denomination
• Households randomly assigned to churches in their village
• All members of a household belong to the same church
Stokvel (ROSCA)
• Third highest in importance (if there is no other financial support like SEF)
• Provide means to save up for a particular purpose
• Social aspect important: provide social support, enhance social status
• Risk of default is low in small communities
• Defaulters are unlikely to be accepted as members into any other
associations
• Formed between groups of friends, min. 3-8
Burial society
• Second highest in importance, more formal than stokvels
• Next to be implemented
CAVES Project Meeting March 2007 ● CPM
4
Declarative Model – Household Rules
• Household economy, modelled on a monthly scale, largely based
on assumptions
• Food expenses: 120 Rand / 100 Rand / 25 Rand
• Income from state grants: 870 Rand pension / 200 Rand
child grant
• Income from jobs: 800 Rand / 200 Rand
• Income from remittances: ?
• Households buy bulk food at the beginning of each month
• Spend minimum of accumulated food expenses and available
cash
• "Rich" households offer short-term employment ("piece jobs")
• if they can afford it and
• if they need it (modelled stochastically, p = 0.15)
CAVES Project Meeting March 2007 ● CPM
5
Declarative Model – Individual Rules
Endorsements
• Every agent endorses other agents with certain "labels"
• Related to existing links
• Kinship: is-kin
• Neighbourhood: is-neighbour
• Groups like churches: same-church, same-denomination
• Related to behaviour of other agent
• Reliable, trustworthy, honest, capable, recommended
• Unreliable, untrustworthy, dishonest, incapable
• Labels are evaluated according to an individual's endorsement scheme
• Resulting endorsement value is used in decisions
Friendship
Stokvels
• Only household heads are members
• When there is enough money left, household heads express a desire to form a
stokvel and ask other household heads amongst their friends
• If there is consent between a certain number of friends, they start a stokvel
CAVES Project Meeting March 2007 ● CPM
6
Declarative Model – Networks
Multi-layer network on several levels
• Individual level
• Friendship
• Based on endorsements and tags, evolves dynamically
• Acquaintanceship
• Based on group membership
• Family (parent, child, sibling)
• Set at creation of person, based on empirical data
• Household level
• Kinship
• Based on small-world network
• Neighbourhood
• Based on spatial location within village, assigned
randomly at creation
CAVES Project Meeting March 2007 ● CPM
7
Declarative Model – Visualisations
CAVES Project Meeting March 2007 ● CPM
8
Declarative Model – Visualisations
CAVES Project Meeting March 2007 ● CPM
9
Declarative Model – Friendship Network
•
Assumptions used: Friends have
• same gender
• similar age (± 3 years for children, ± 8 years for adults)
• similar interests/character traits
• similar background (same church, neighbour…)
•
Friendship network evolves from these
• Agents evaluate all known other agents
• Compute similarity index based on tags
• Compute endorsement value based on endorsement scheme
• Agents pick highest evaluated agents as friends
• Up to a maximal number of friends
Surprising effect: very low proportion of mutual links
• Solutions tried:
• Special friendship endorsement scheme
• Higher max. number of friends
CAVES Project Meeting March 2007 ● CPM
10
Experiences with Jess
Model implementation
• Java/Repast for model framework
• Jess for all cognition and decision processes
• Java classes (Person, Household, Model…) as shadow facts
• Per time step one run of the Jess engine
 Too slow to be actually used
Problem: Re-computation of the Rete network
Solution: less Jess, more Java
• Fewer rules
• Port procedural stuff to Java
• Browse fact base from Java
• Fewer facts
• Replace facts with fields in Java classes (slots in shadow facts)
CAVES Project Meeting March 2007 ● CPM
11
Experiences with Jess – Example: Fewer facts
(defclass person Person)  has slots name, gender, age, tag… knownPersons
(deftemplate known-person (slot owner) (slot known) (slot tick))
Replace facts with
field (slot)
(defrule adult-similarity-identification
"identify others with most similar and similar tags"
)
(person (tag $?own-tag) (name ?person) (gender ?gender) (knownPersons $?known-persons)
(age ?own-age &: (> ?own-age 12)))
(model (tick ?tick))
(known-person (owner ?person) (known ?other) (tick ?t &:(<= ?t ?tick)))
(person (gender ?gender) (name ?other) &: (member$ ?other ?known-persons))
(age ?other-age
&:(and
(> ?other-age 12)
(< (abs (- ?own-age ?other-age)) 8)))
(tag $?other-tag))
(not
(similarity-index (owner ?person)(other-person ?other)(tick ?t &: (< ?t ?tick))))
=>
(bind ?similarity (number-of-common-attributes ?own-tag ?other-tag))
(assert
(similarity-index (other-person ?other) (similarity ?similarity) (owner ?person) (tick ?tick)))
CAVES Project Meeting March 2007 ● CPM
12
Next steps
Integration of further processes that influence social networks
• Burial societies
• Marriage
• Inheritance of (part of the) tags from parents
• Spread of HIV/AIDS, if possible on a more individual basis
Applying network measures
Improve visualisation and data collection
• Discuss need with case study team
CAVES Project Meeting March 2007 ● CPM
13