Two Approaches to Community Development

This study covers two distinct experimental interventions.

1. Long run effects of CDD: The first is a community driven development (CDD) project that was implemented as a large scale randomized trial from 2005 to 2009 in Sierra Leone. CDD is a participatory approach popular with foreign aid donors that involves communities directly in the financial management and implementation of local public goods. CDD has three aims: i) improve the stock and quality of local public goods via the provision of block grants ($5,000 per community); ii) democratize local decision-making via intensive social facilitation focused on the participation of marginalized groups; and iii) establish ties with local government that can help communities avail of external funding opportunities. In earlier work, we analyzed the medium run effects of the "GoBifo" CDD project, and found substantial positive impacts on local public goods and economic activity, stronger links between the community and local government, but no evidence for more inclusive local decision-making (see Casey, Glennerster and Miguel QJE 2012).

We now plan to revisit the 236 communities in the original study to assess long run effects. Specifically, we will evaluate whether the "hardware" effects on public goods and economic activity persist several years later; assess any late onset institutional change, which may have been triggered by follow on CDD projects that were subsequently implemented in 60 of the treated villages; and also test whether the CDD experience equipped communities to respond more effectively to new crises, with a focus on local response to the Ebola epidemic.

2. Managerial capital experiment: The second intervention is a light touch technocratic alternative to CDD. It aims to leverage two dimensions of managerial capital - the selection of project managers and the availability of management training - to bolster community ability to take advantage of public funds available for small scale infrastructure projects. Communities in our study areas will have an opportunity to enter a project challenge competition run by local government that awards US$2,500 implementation grants to the best project proposals. The intervention aims to help communities mobilize in response to the new funding opportunity by guiding them through a process that identifies high competence members, uses short tests to provide the community with information on concrete measures of their actual managerial ability, and offers training to augment the managerial capital that exists.

More specifically, the intervention involves one day visits to communities during which a field enumerator advertises the project challenge competition, explains the skills needed to submit a successful proposal, and asks a group of local leaders to deliberate and nominate 5 residents (in addition to the village headman, who is the traditional leader) who are strong on these dimensions. Enumerators then administer a short test to each of these nominees (in private), which is designed to measure their management skills. It includes questions on basic math and writing, costing materials commonly used in small scale infrastructure, and previous management experience. Next steps vary across three treatment arms. In the status quo (SQ) arm, the local traditional authority is designated to lead the project proposal process. In the manager selection arm (MS), the community is informed about which of 5 nominated individuals scored highest on the test and this person is designated to be the project proposal leader. The final manager selection plus training arm (MS+T) follows the MS arm, and further includes the offer of no cost training in basic project management skills offered in collaboration with the local government.

2016-11-17

2017-01-31

1. Long run effects of CDD: key outcomes are mean effects indices on the 12 hypotheses that we explored in our previous study of medium run impacts. These 12 are grouped into two families. The "hardware" family A includes three hypotheses that cover the creation of functional development committees, the quality and stock of local public goods, and economic welfare. The "software" family B includes 9 hypotheses that cover collective action and contributions to public goods, inclusion and participation, local systems of authority, trust, groups and networks, access to information, participation in governance, crime and conflict, and political and social attitudes.

2. Managerial capital experiment: the key outcome is community submission of a proposal to enter the project challenge competition.

All outcomes, including the construction of mean effects indices, will be pre-specified in our pre-analysis plan before any data is analyzed.

1. Long run effects of CDD: the GoBifo project was randomly assigned to 118 of 236 eligible communities in two districts of Sierra Leone. Randomization was stratified by ward, which is the smallest administrative unit of elected government. Within each ward there are 6 treated and 6 control communities, with the exception of Bonthe Island, were there are 4 treated and 4 control communities.

2. Managerial capital experiment: three treatment arms were assigned across the 236 communities stratifying by ward and GoBifo treatment status. Within each ward, we randomly assigned 2 communities to the SQ arm, 2 to the MS arm and 2 to MS+T arm among the 6 GoBifo treatment communities; and did the same (2 SQ, 2 MS, 2 MS+T) among the 6 GoBifo control communities. For Bonthe Island, we assigned 2 communities to SQ, 1 to MS and 1 to MS+T within the 4 GoBifo treatment and 4 control communities.

1. Long run effects of CDD: The main randomization covering 228 communities was done on a computer in Stata. Randomization for the 8 communities on Bonthe Island was done via a public lottery.

2. Managerial capital experiment: randomization was done on a computer in Stata.

1. Long run effects of CDD: randomization unit is the community

No

1. Long run effects of CDD: 236 communities

2. Managerial capital experiment: 236 communities

For both, the vast majority of data is collected at the community level (hence, no clustering).

1. Long run effects of CDD: 236 communities 2. Managerial capital experiment: 236 communities

1. Long run effects of CDD: 118 treatment, 118 control

2. Managerial capital experiment: 80 status quo, 78 manager selection, 78 manager selection plus training

1. Long run effects of CDD: our reference outcome is the mean effects index for "hardware" outcomes. In our medium run data, this was mean=0, sd=0.37 for controls and mean=0.31, sd=0.32 for GoBifo communities. Leveraging the panel structure of the data (corr=0.66), we could detect an effect size as small as 0.08 with alpha=0.10 and power=0.80. 2. Managerial capital experiment: calculations reference the take-up rate observed (0.53) in a 2009 building material voucher program that was part of our medium run study. As the project challenge opportunity has no co-pay and larger prizes ($2,500 vs. $300), we expect take-up to be higher, at 0.70. With alpha=0.10 and two-sided tests, we could detect an increase of 15 percentage points (to 0.85) comparing: control to GoBifo communities (n1=n2=118) with power=0.84; and status quo to technocratically selected managers (n1=156, n2=80) with power=0.80. We have slightly less power to isolate the added effect of training: with two-sided alpha=0.10, we could detect the same percentage point change on a higher base (0.75 to 0.90) for technocratic selection versus selection plus training (n1=78, n2=78) with power=0.73.

Pre-Analysis Plan

Pre-Analysis Plan Outcomes Table

Data management and access plan

PI confirmation of adherence to data management plan