For Peace and Goodwill? Using an Experimental Game to Analyse the Effect of the Desarollo y Paz Programmes in Colombia. David PHILLIPS Institute for Fiscal Studies, London (also Orazio Attanasio and Luca Pellerano) LACEA Annual Conference Buenos Aires, 1st October 2009
For Peace and Goodwill? Using an Experimental Game to Analyse the Effect of the Desarollo y Paz Programmes in Colombia. David PHILLIPS Institute for Fiscal Studies, London (also Orazio Attanasio and Luca Pellerano) LACEA Annual Conference Buenos Aires, 1st October 2009. Introduction. - PowerPoint PPT Presentation
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
For Peace and Goodwill? Using an Experimental Game to Analyse the Effect of the Desarollo y Paz
Programmes in Colombia.
David PHILLIPSInstitute for Fiscal Studies, London
(also Orazio Attanasio and Luca Pellerano)
LACEA Annual ConferenceBuenos Aires, 1st October 2009
Introduction
• Increasing emphasis on evaluating programmes with socio-cultural aims:– values, aptitudes, perceptions, motivations, capabilities, empowerment– relationships, networks, social capital
• Rigorous impact evaluation in this domain is extremely challenging as:– Expected socio-cultural changes are often not clearly specified (Theory of Change)– They are difficult to measure– As the analysis moves in the domain of the “intangibles”, the evaluation must be
driven my a multi method and multidisciplinary approach.
• Attanasio, Pellerano and Phillips (2009):– We evaluate the effect of a Social Fund type intervention in Colombia on social capital
and conflict management using a Public Goods game.
1. The Setting and the Programme2. The Experimental Games3. The Data4. Descriptive Statistics5. The Evaluation Strategy6. Results 7. Conclusions
1. The Setting and the Programme2. The Experimental Games3. The Data4. Descriptive Statistics5. The Evaluation Strategy6. Results 7. Conclusions
The Setting. Conflict in Colombia.
• Colombia has been affected by a complex civil conflict along the whole course of its recent history.
• Two main illegal factions: Guerrilla(s) and Paramilitary groups (+ new emerging groups after process of desmovilización)
• Despite progress, violence still affects many rural areas of the country in a very clustered way.
• Strong association between the presence of illegal groups and violence levels, local political instability, inequality (Sanchez et al, 2003)
The Peace and Development Regional Programmes (PRDPs)
• Community-driven development / peace-building initiative
• Networks of grassroots organizations formed from existing civil society (labour unions, the church, private foundations) in poor and violent regions
• Claim peace is bottom up and based on development and community action.
• Support a wide range of development processes informed by the principles of protecting life (“Primero la vida”), equity, solidarity and participatory democracy
• The PRDPs work as an “umbrella fund”, financing a series of projects along three main axes:
1. Productive Development2. Institutional Development and Governance3. Human Rights and “Culture of Peace”
The Peace and Development Regional Programmes
• Started in Magdalena Medio in the early 1990s (Jesuits)
• 6 PRDPs have been supported by the UNDP, the World Bank (LIL1 and LIL2 grants, 1998-2003; Paz y Desarrollo Loan, 2004-2008) and the European Commission (Laboratorios de Paz I, II and III, 2002-2010)
• Total investment of more than 180 million USD
6 Regions9 Departments
125 Municipalities
1363 projects800 grassroots organizations
180,000 estimated beneficiaries
On average:300 beneficiaries
per project2 years project
duration120,000 USD project
value
The Peace and Development Regional Programmes
1. The Setting and the Programme2. The Experimental Games3. The Data4. Descriptive Statistics5. The Evaluation Strategy6. Results 7. Conclusions
Experimental Games
Social capital
• Competing definitions and competing measuring tools• Standard survey techniques
– Structural approach (membership) versus motivational/value based approach (trust, reciprocity)
– People tend to respond to value based survey questions according to social expectations and social norms (Glaeser et al., 2000).
• Our definition: social cooperation (act collectively for mutual gain)• Our measure: a behavioural measure of contribution to a local public
good in a field experimental setting
Experimental Games
• Based on the theoretical framework of the experimental economics: players’ decisions should reveal their preferences/motivations
• The “controlled” experiments are designed in such a way that decisions only have monetary consequences (no reputation effects)
• As the decisions imply a real monetary payoff (on average the equivalent of 3 USD) participants are expected to reveal their preferences more similarly to how they would behave in a real life situation
Experimental Games
Voluntary Contribution Mechanism – VCM (Marwell and Ames, 1979)• The experiment is a simple public good game in which subjects decide to
invest in a ‘public’ or a ‘private’ account• Played in group (40 participants on average)• Every token invested in the public account yields benefits for the whole
group, while the token invested in the private account only yields benefits to the owner of the account
• Investing in the public pot is socially optimal• However an individual is always better off by investing in the private account
(Dominant Strategy)
)(100)1(2000 n
jjii GGS
The game is played in Two Rounds: First Round. Anonymous decision Second Round. 10 minutes communication
contribution to local public goodcommunication
1. The Setting and the Programme2. The Experimental Games3. The Data4. Descriptive Statistics5. The Evaluation Strategy6. Results 7. Conclusions
The Data
Two main data sources
1. (Late) Baseline for the evaluation of PyD and LP (Nov 2006 – Feb 2007).• Treated Locations. Beneficiary and Non-Beneficiary Households• 18 months of exposure to the activities of the PRDPs on average• High variability in exposure because of the roll out of the PRDPs both within and
across municipalities
2. Second follow up for the evaluation of Familias en Acción (Nov 2005 – Apr 2006)• Control Locations• Sample for the evaluation of the famous CCT programme.• Targeted to the poor (SISBEN1), mainly women
0.1
.2.3
.4.5
Fra
ctio
n
0 50 100 150Months
PRDP Project Beneficiaries Members of Other Civic Organizations
Sample of Treated Locations
904 Non-Beneficiaries
(Control Group 2)
782 Beneficiaries
(Treatment Group)
Beneficiaries with higher exposure
Beneficiaries with lower exposure
The Data
Sample of Control Locations
Sample of 2472
(Control Group 1)
PRDP Municipalities (37)FA Municipalities (67)
FA Data PRDP Data
The Data. Experimental Game
Control
Locations
Treatment
Locations Total
Number of participants 2242 1518 3760
Number of sessions 67 45 112
Size of the group for the smaller session 21 11 11
Size of the group for the biggest session 40 90 90
Average session size 36.90 35.96 36.52
Average proportion of women in the session 0.54 0.88 0.74
Average proportion of beneficiaries in the session 0 0.45 0.18
SD of the proportion of beneficiaries in the session 0 0.11 0.23
1. The Setting and the Programme2. The Experimental Games3. The Data4. Descriptive Statistics5. The Evaluation Strategy6. Results 7. Conclusions
Descriptive Evidence
Municipality CharacteristicsAltitude 1285.766Municipality development index 37.243Coca crops extension 56.472Murder rate (per 1000 inh.) 0.926
Total population 92067.281Proportion urban population 0.543Outgoing displaced population (per 1000 inh.) 27.98Voting rate (local elections) 0.485Voting rate (presidential elections) 0.768Voting majority (local election) 0.408Per capita public expenditure 0.308Per capita public investment 0.246Primary school students (per 1000 inh.) 148.43Secundary school students (per 1000 inh.) 243.632Judiciary processes initiated (per 1000 inh.) 17.962Judiciary processes terminated (per 1000 inh.) 18.723
Descriptive Evidence
• High degree of heterogeneity in beneficiary characteristics according to project types
Individual Characteristics
Female 0.528 Membership in Juntas de Acción Comunal 0.367Age (years) 39.85 Leadership role in grassroots organizations 0.421Less than primary 0.297 Voted in last elections (0-6) 3.956Full primary 0.207 One can trust the majority of people in the comm. 0.348Some secondary 0.159Full secondary + 0.253
Household Characteristics
Female head of the household 0.204 Monthly income (COP) 368165Family members younger than 18 1.568 Monthly food expenditure (COP) 253129Urban 0.495 Value of self produced food (COP) 73619Own house 0.652 Any savings during the last year 0.111Number of rooms 2.973 Any member sufferd HHRR violation (prev.year) 0.101Water by pipe 0.711Sewage system 0.476Rubbish recollection 0.449Gas by pipe 0.485Phone (landline) 0.228Top distribution assets 0.118Displaced (self declared) 0.313Months living in the neighborhood 239Other support 0.298Laboratorios de Paz 0.422
1. The Setting and the Programme2. The Experimental Games3. The Data4. Descriptive Statistics5. The Evaluation Strategy6. Results 7. Conclusions
Evaluation Strategy
• No random assignment -> Potential endogenous programme placement (selection and self selection)– Across locations– Within locations
• No pre-programme information (apart from municipality characteristics)
• Identification strategya) within locations (beneficiaries vs. non-beneficiaries in treated locations) b) across locations (treated vs. control matched municipalities)c) within treated households (based on the intensity of exposure)
• May provide insights on the presence and magnitude of spill-over effects
• Multivariate regression models. All models are based on the “selection on observables” assumption
Evaluation Strategy. Across Municipalities
• We match treatment and control locations on a complete set of pre-programme municipality characteristics
• 2 control locations dropped because of missing information• 20 control locations and 16 treated locations are dropped because they fall outside
the common support• 2,427 households (58% of the original sample) are on the common support)
0 .2 .4 .6 .8 1Propensity Score
Untreated: Off support Untreated: On supportTreated: On support Treated: Off support
Marginal Probit. Errors are Clustered at the Session level. * significant at 10%; ** significant at 5%; *** significant at 1%
Results. Experimental Game
• The level of contribution to the public good is much higher in municipalities where the PRDPs operate, than in other municipalities in the country with comparable characteristics.
• Results hold if CS is imposed based on matching on individual characteristics
• We find some light evidence that, ceteris paribus, increased exposure to the PRDPs lead to higher contribution to the public good, particularly in the long run.
Exposure effects
Round 1 Round 2
Exposure (7-14 months) 0.045 0.010 -0.047 -0.058
[0.048] [0.047] [0.047] [0.043]
Exposure (>14 months) 0.085** 0.064 -0.014 -0.032
[0.043] [0.042] [0.048] [0.043]
Months of Exposure 0.003 0.002 -0.001 -0.001
[0.003] [0.003] [0.003] [0.003]
Months of Exposure (Municipality Max)
0.003** 0.002* -0.001 -0.001
[0.001] [0.001] [0.001] [0.001]
Round 1 Decision YES YES
Location Controls YES YES YES YES
Household Controls YES YES YES YES
Session Composition Controls YES YES
Observations 684 684 684 684
Errors are Clustered at the Session level. * significant at 10%; ** significant at 5%; *** significant at 1%
Results. Experimental Game
• Exploiting knowledge networks, social capital effects seem to propagate within treated municipality through a spill-over mechanism.
• The fact of knowing the PRDP could however be endogenous to social capital.
Examining the case for Spill Over effects.
• The previous findings suggest that there might be some contamination effect in social capital formation at the community level between treated and control households within the same municipality.
Round 1 Round 2
Knows PRDP 0.078** 0.094*** 0.001 0.013
[0.033] [0.034] [0.031] [0.032]
Treated Household -0.044 -0.048 0.03 0.016
[0.038] [0.037] [0.046] [0.048]
Round 1 Decision YES YES
Location Controls YES YES YES YES
Household Controls YES YES YES YES
Session Composition Controls YES YES
Observations 1472 1472 1472 1472
Marginal Probit. Errors are Clustered at the Session level. * significant at 10%; ** significant at 5%; *** significant at 1%
Results. Experimental Game
• Because of the sampling procedures, the proportion of beneficiaries in the session doesn’t reflect coverage rates in the municipality
• In the second round, the group effect seems to be operative for control households only
• This suggests that it is mainly driven by reputation factors• Complementary evidence suggests that PRPD beneficiaries are actually
considered more “trustworthy” than the rest of the population
Critical Mass or Conditional Cooperation? Round 1 Round 2
Proportion of Treated in the Session 0.521** 0.614*** 0.584** 0.634**