Mobilizing for the Public Good: A Field Experiment on Community-Driven Development and Waste Management in El Salvador

Traditional intervention (top-down approach): A team of cleaners with a garbage truck visits each community once a month to collect garbage and clean all public spaces. Community members can also use the truck to dispose of their household waste.
Community-driven intervention (community-driven development approach): In each community, a facilitator is hired to organize workshops and activities to raise awareness about the problem and solve it through collective action. The facilitator also coordinates the implementation of clean-up campaigns by community members.

Both interventions have a duration of four months and roughly the same cost.

2022-12-01

2023-03-31

(1) Waste contamination
(2) Self-reported attitudes, behaviors and perceptions of social norms regarding solid waste contamination

Our first primary outcome is waste contamination in all public places within communities, measured through two different approaches:
(a) Image classification: Enumerators take geo-tagged photos every five steps along all streets in all communities. We will then use deep learning to classify the images and construct contamination maps for all communities. For the construction of the maps, the photo assessments will be pooled into approximately 100 cells per community.
(b) Enumerator assessment: Additionally, enumerators make a cleanliness assessment of the environment every 25 steps in all communities, allowing for the construction of a second set of contamination maps based on these assessments.The assessment is done in four categories: completely clean, a bit dirty, rather dirty and very dirty. Example photos are provided for each category.

Our second set of primary outcomes stems from a survey with about 20 villagers per community and captures people's self-reported attitudes, behaviors and perceptions of social norms regarding solid waste contamination in their communities. Outcome variables constructed based on this survey data are: (i) perceived cleanliness of the community, (ii) perceived importance of cleanliness, (iii) descriptive littering norms (share of people reporting to litter and beliefs about share of others who litter), (iv) injunctive littering norms (share of people disapproving of littering and beliefs about share of others who disapprove), (v) littering punishment (share of people reporting to punish littering and beliefs about share of others who punish), (vi) waste disposal practice within households, and (vii) reported willingness to pay for garbage removal.

(1) Social capital
(2) Trust in institutions
(3) Living conditions
(4) Safety

Our survey also covered items to construct the following secondary outcomes:
(1) Social capital: A set of questions on social capital in the community covering social ties, trust in other community members, solidarity, altruism, membership in organizations, voluntary engagement, and willingness to contribute to community projects
(2) Trust in institutions: Reported trust in community leaders, municipality officials, and central government officials
(3) Living conditions: Perceptions of living conditions in the community
(4) Safety: Perceptions of safety of the community

For this study, we selected 120 communities throughout Morazán based on (i) community size, (ii) safety, (iii) the presence of solid waste management problems, (iv) accessibility, and (v) geographical distance between communities to avoid spillover effects. Communities were then randomly assigned to our three experimental groups: (1) a traditional intervention, (2) a community-driven intervention, or (3) a control group.

In all communities, we conduct a baseline assessment prior to the interventions (wave 1), an endline assessment immediately after the interventions (wave 2), and a follow-up assessment four months later (wave 3). The assessments include (1) a contamination assessment and (2) a survey.

(1) Contamination assessment: Based on a detailed manual, enumerators take geo-tagged photos along the streets in all 120 communities. This procedure yields an average of about 540 photos per community for each of the three waves. Within a set distance along the streets, enumerators further make assessments of the overall cleanliness of the environment.

(2) Survey: In each community, a convenience sample of 20 villagers is selected to take part in all three waves of data collection. The survey has a duration of 10--15 minutes and covers questions about (i) attitudes, behaviors and perceptions of social norms regarding solid waste contamination, (ii) social capital within the community (e.g. social networks, trust, solidarity and organizations) as well as perceptions of living conditions and safety, and (iii) socio-demographic background (only baseline). The endline and follow-up assessments will further include questions about involvement in/observation of activities regarding solid waste management. Baseline surveys are conducted face-to-face in people's homes. For the endline and the follow-up survey, we conduct both phone or face-to-face interviews (depending on how each person can best be reached).

Randomization is conducted by the research team using the software R. Randomization is stratified by contamination level measured during the baseline assessment (terciles) and geographical zone (North, South, Center West, Center East).

Communities

Yes

120 communities

Contamination maps: - 65,000 photo locations, aggregated into approximately 100 raster cells per community (12,000) - 13,000 assessment locations Surveys: - 2400 community members (20 per community)

Treatment 1, Traditional intervention: 40 communities
Treatment 2, community-driven intervention: 39 communities
Control: 41 communities

Photo assessment: MDE = 0.13--0.33 standard deviations. Calculations based on the following parameter values: power = 80%; alpha (level of significance) = 0.05; rho (intra-cluster correlation) = 0.1--0.4; R2b (share of between-variance absorbed by baseline scores) = 0.3--0.6; R2w (share of within-variance absorbed by baseline scores = 0.1--0.4; J (number of clusters) = 80, P (share of control clusters) = 0.5; n (observations per cluster) = 100 Survey outcomes: MDE = 0.16--0.35 standard deviations. Calculations based on the following parameter values power = 80%; alpha (level of significance) = 0.05; rho (intra-cluster correlation) = 0.1--0.4; R2b (share of between-variance absorbed by baseline scores) = 0.3--0.6; R2w (share of within-variance absorbed by baseline scores = 0.1--0.4; J (number of clusters) = 80, P (share of control clusters) = 0.5; n (observations per cluster) = 20 Calculations based on formula by Bloom (2007)