Bridges to Prosperity – Trailbridge Building in Rwanda (May 2022)

Summary

In May 2022, GiveWell recommended a grant of $3.4 million to Bridges to Prosperity (B2P) to fund trailbridge building as part of an ongoing randomized controlled trial (RCT) in Rwanda.

We had previously deprioritized B2P because of the lack of high-quality evidence of the effectiveness of trailbridges in East Africa, where B2P plans to prioritize building in the future, and because there is an ongoing, large-scale RCT of B2P trailbridges in Rwanda that we think will fill this gap in evidence. However, due to COVID-19 disruptions, B2P has had to limit the number of bridges in that trial.

We’re recommending this grant because:

  • We think bridges funded with this grant are likely to benefit households through increases in income that last several years.
  • While our best guess is that bridges are below the range of cost-effectiveness of programs we would recommend funding, we think there’s a reasonable chance the findings of the RCT update us toward believing this program is above our bar. By roughly doubling the number of bridges in the RCT, we think our grant will enhance the quality of the trial and, in turn, make it modestly more likely that we update our view of the program’s impact and direct more funding to B2P in the future based on the results.
  • Given both the direct impact of the bridges themselves and the information value of the additional bridges to the RCT, we estimate this is a cost-effective use of funding.
  • The findings of this study may contribute to increasing the currently nascent high-quality evidence of trailbridge impact and thus could be informative to policymakers and researchers. Additionally, our funding could improve how informative the study is to them.

Our main reservations are:

  • The information value of this grant is limited by the fact that our grant is providing only marginal improvements to the trial, the risk that other funders could support these same improvements in the near future, and the potentially limited generalizability of the findings beyond Rwanda. We’ve made adjustments to account for these factors, but we’re uncertain about the appropriate value for these adjustments and have high uncertainty about our value-of-information model in general.
  • We have engaged less with B2P than with our top charities or with organizations receiving larger grants because this is, relative to other funding recommendations in our portfolio, a smaller grant. As a result, we don't yet have a comprehensive picture of B2P as an organization and have uncertainties about several of the key parameters we have used in the cost-effectiveness analysis.

Published: September 2022

Table of Contents

The organization

B2P is a nonprofit that builds trailbridges in rural communities. These bridges are intended to connect isolated households to schools, health clinics, and markets.1

We had previously deprioritized B2P because of a lack of high-quality evidence of the effectiveness of trailbridges in East Africa, where B2P plans to prioritize building in the future, and because there is an ongoing, large-scale RCT of B2P in Rwanda that we think will fill this gap in evidence.

However, B2P has told us that the COVID-19 pandemic led to a drop in funding both from government partners, which pay a portion of the bridge costs, and philanthropic partners. This has left a gap in funding for the Rwanda RCT.2

The intervention

We model the primary benefits of B2P as operating through increases in household income. This results from higher off-farm employment and higher wages, as well as increased farm investment, which leads to increases in farm profits.3

We roughly estimate that B2P is slightly below the range of cost-effectiveness of programs we would recommend funding (~4x as cost-effective as cash transfers from GiveDirectly).4 However, we are sufficiently uncertain about our cost-effectiveness estimate that we think additional evidence could update us toward believing the program is within the range of cost-effectiveness of programs we would recommend funding.

Our major uncertainties are:

  • What is the effect of bridges on household income in Rwanda and other settings where B2P plans to build in the future?
  • How long do the benefits of bridges persist?
  • How many households are “affected” per bridge, and how do effects vary with distance to bridges?
  • How can we generalize results from Rwanda to other settings where B2P plans to build in the future?
  • Are there additional benefits or offsetting negative effects that would make a substantial difference in cost-effectiveness? This could include positive effects on health or education outcomes or downside effects of those on the “less affected” side of the bridge (i.e., those who were less isolated initially), as well as general equilibrium effects.

The Rwanda RCT

The Rwanda study is a stepped-wedge RCT.5 The current plan for the trial includes 147 potential sites, with 97 bridges in the treatment group and 50 sites as “long-term” controls.6 Under this plan, bridges will be phased in over three years.7

The researchers plan to conduct surveys in 471 villages and 15,735 households.8 This includes three villages close to each site and an additional 30 “distant” villages that are farther away (one each at 20 treatment sites and 10 control sites).9

The trial will measure the effect of trailbridges on income, consumption, health, savings, agricultural productivity, market prices, transportation time and costs, health outcomes, and education.10

In addition to measuring the average effect of trailbridges on outcomes, the researchers will measure:

  • How effects vary based on households’ distance from bridges11
  • How effects change over time12
  • How effects vary with climate and geological characteristics13
  • Mechanisms driving effects on income and consumption14

The investigators are:15

  • Evan Thomas, University of Colorado Boulder (Principal Investigator)
  • Denis Macharia, University of Colorado Boulder
  • Laura MacDonald, University of Colorado Boulder
  • Lambert Mugabo, Amazi Yego Ltd. in Rwanda
  • Kevin Donovan, Yale School of Management
  • Sorenie Gudissa, Yale School of Management
  • Wyatt Brooks, Arizona State University
  • Katie Dickinson, Colorado School of Public Health
  • Abbie Noriega, B2P
  • Christina Barstow, B2P

More details are available in Macharia et al. 2022.

Grant activities and budget

The budget for this grant is $3,390,138.16 A breakdown is below:

  • B2P Rwanda Personnel (41%): $1,389,956
  • Local Personnel (9%): $305,112
  • Bridge Materials (38%): $1,288,252
  • Storage (1%): $33,901
  • Material Transportation (1%): $33,901
  • Safety Equipment (10%): $339,01417

This will enable B2P to complete building roughly 46 of the remaining bridges in the “treatment” group.18

Case for the grant

  • We think trailbridges are likely to benefit households through increases in income that last several years. Our best guess is that each trailbridge leads to a 4% increase in household income across 500 households and that these benefits persist for 15 years. Bridges cost roughly $110,000 each, or ~$220 per household.19 Given these effects, B2P is slightly below the range of cost-effectiveness of grants we would recommend funding (see here for more information).
  • We think there’s a reasonable chance the findings of the RCT update us toward believing this program is above our bar. This is because:
    • There is limited high-quality evidence for the effect of trailbridges in Rwanda and other countries where B2P plans to build in the future.
    • We view the Rwanda trial as having high internal validity because it is an RCT, and randomization appears to have been successful;20 it has reasonable power to detect the outcomes we are interested in;21 we have limited concerns about contamination,22 and the researchers are publishing a pre-analysis plan.23
    • We guess the Rwanda trial will inform our ability to estimate the effects of bridges in other settings where B2P plans to build in the future, though we have substantial uncertainty about this.

    We think there is a 30% chance we conclude B2P is higher than 6x as cost-effective as cash transfers and a 10% chance we conclude it’s higher than 10x after seeing the results of the Rwanda RCT.

  • We think our funding will modestly improve the ability of the RCT to detect effects on households and, in turn, make it modestly more likely we update our view of the program’s impact based on the results. We think providing additional bridges to the RCT sample will slightly improve our ability to detect the following parameters. However, we view these individually as fairly small updates and are unsure how much to value them collectively.
    • Average effects of trailbridges on household income and consumption. The main outcome in our cost-effectiveness model is the effect on household income and consumption. Increasing the number of bridges from 50 to 97 reduces the minimum detectable effect (MDE) on income by 10% and consumption by 12%.24
    • How effects decay with distance and, in turn, how many households are affected by trailbridges. We are highly uncertain about how many households see income benefits as a result of trailbridges. We intend to use effects on income and consumption by distance to bridge sites, combined with data on number of households within a given distance to bridge sites, to estimate the “catchment area” for the impact of bridges. Increasing the number of bridges from 50 to 97 reduces the MDE for the “interaction” between bridge distance and income by 10% and consumption by 12%.25
    • How the benefits of trailbridges change over time. Funding for bridges will increase the precision of estimates of how effects change over time. This will let us see if effects persist, at least over the medium-term. It will also let us see if there is a “ramp up” period before effects begin to level off to resemble a longer-term effect.26 Increasing the number of bridges from 50 to 97 reduces the MDE for differences in effects between Year 1 and Year 3 by 55%.27 However, the precision for these differences, relative to the MDE for the pooled effect, is low.28 If GiveWell or another funder were to provide funding for a fourth year of surveys, this could improve precision further.29
    • Average effects in the long-term. We assume that the ability to detect slightly smaller increases in income and consumption also applies to estimating any longer-term follow-up effects (e.g., 5 years, 10 years, or beyond).
    • Effect of B2P at larger scale. Since we sometimes have reservations about declines in quality as a program scales, doubling the number of bridges in the trial could tell us about the effectiveness of B2P’s program at a larger scale.
    • Average effects of trailbridges on other outcomes. Increasing the number of bridges also improves precision for other impacts, such as effect on visiting clinics (12% reduction in MDE) or mid-upper arm circumference (MUAC), which is an indicator of nutritional status (20% reduction in MDE).30
    • Drivers of heterogeneity in effects. The research team plans to identify measures that correlate with the impact of bridges (e.g., electronic counting of bridge crossings). These “process” measures, such as electronic counting, could be used to predict whether bridges built in other settings will have the same effects, without having to run another full-scale RCT.31 (This is similar to another GiveWell-funded program, New Incentives, for example.32 ) We’re not sure how much additional sensitivity we’ll get for these effects.
    • Extent of spillovers and general equilibrium effects. Cost-effectiveness could be impacted by effects on wages and market prices in areas newly connected by bridges.33 The research team has indicated that additional bridges improve the ability to measure these effects, though we have not yet tried to model these or understand fully the extent to which we would update our cost-effectiveness model based on these effects.
  • Given the direct impact of the bridges themselves and the information value of the bridges and research study, we estimate this is a cost-effective use of funding. Any improvements in the quality of the trial are valuable because they increase the chance that we update towards believing B2P is above our cost-effectiveness bar and, in turn, allocate funding away from less cost-effective opportunities toward B2P in the future. We have built a back-of-the-envelope model that attempts to estimate this information value of the trial, in addition to the direct benefit of the bridges themselves. This estimate relies on several uncertain and speculative parameters, including how much this grant improves the quality of the RCT and GiveWell’s ability to update based on it, how likely it is that other funders would fund these same RCT enhancements in our absence, how much additional funding B2P could absorb in the future, and how much the results of the Rwanda RCT generalize to other settings. We estimate that the grant overall is roughly 10x as cost-effective as cash transfers, with half of the benefit coming from bridges themselves and half coming from value of information. We view this as a rough check on where the value of this grant comes from and a way to make sure we've considered key drivers of value. We do not put a lot of stock in the specific cost-effectiveness estimate provided.
  • This is a unique opportunity to rigorously study B2P’s program, so enhancing its quality seems especially important. The research team has already made agreements with B2P and the Rwandan government to randomize roll-out,34 for example, and there is already a team in place. As a result, it feels like an opportune time to enhance the quality of the evaluation.
  • The findings of this study could be informative to policymakers and researchers, and our funding could improve how informative the study is. Our broad impression is that there are few, if any, RCTs of transportation infrastructure interventions like B2P’s trailbridges. We think it’s possible there could be broader learnings for researchers, and the findings may also inform infrastructure investments by other decision-makers. However, this is speculative, and we have not tried to quantify this benefit. This additional benefit is excluded from our value-of-information model.
  • We have positive preliminary impressions of B2P as an organization. B2P seems to have an above-average interest in research and evaluation (this Rwanda RCT is their second large-scale study, following the quasi-experiment in Nicaragua), and our impression is they collect detailed monitoring data on bridge utilization.35 It also seems like they’ve been effective in growing over the past several years and forming partnerships with governments. We have only looked into these qualitative elements at a shallow level, though.

Risks and reservations

  • We have several uncertainties about key parameters for the value of information from the grant. These include:
    • The probability we update our cost-effectiveness of B2P with or without this grant. We think that funding additional bridges will increase the chance we update our estimate of cost-effectiveness, based on the findings, relative to the case where the trial proceeds without these enhancements. However, our current estimates are subjective guesses, intended to capture roughly how likely we would be to positively update if the trial did not include additional bridges and how likely we would be to positively update if it did.36 These are rough and are not based, for example, on specifying prior and expected posterior distributions on key parameters (effect on income, number of households affected, duration of benefits).
    • The extent to which additional confidence in the results would cause us to recommend more funding to B2P. With a more informative trial, we may also feel more confident directing more funding to B2P, even given the same cost-effectiveness.37 This adjustment is also highly subjective.
    • The probability another funder would fund additional bridges in the RCT and how that would limit the cost-effectiveness of this grant. B2P told us there are other funders they have spoken to about this gap who may step in and fill the gap for bridge-building in Rwanda.38 We have tried to build in an adjustment for risk of other funders filling this gap into our value-of-information cost-effectiveness analysis, but our adjustment is highly speculative and stronger adjustments would imply lower value.39 (For the direct effect of the bridges themselves, we also included a funging adjustment, though this value is also speculative.40 )
    • The amount of funding we expect to recommend to B2P, conditional on believing B2P is above our cost-effectiveness bar after seeing the results. We estimate that B2P could absorb approximately $12 million per year in settings similar to Rwanda over the next several years, based on estimates B2P has shared for funding gaps.41 We have made rough adjustments to account for gaps being filled by other funders and some gaps not being cost-effective, but these are highly subjective.42 On the other hand, these estimates do not factor in the potential for the findings of the trial (and any enhancements to quality as a result of this grant) to cause other funders to shift more funding to trailbridges, which would increase the amount of funding that could be allocated.
    • The extent to which results from Rwanda will generalize to other settings. We also adjust room for more funding downward to account for findings from Rwanda not generalizing to other settings. It’s possible that results from Rwanda don’t extend to other countries in East Africa where B2P may build in the future (e.g., because Rwanda is especially prone to flooding,43 which increases the impact of bridges, or its high population density means an unusually large number of households per bridge,44 which improves cost-effectiveness). While we guess the trial will permit tests of how effects vary across sites, we’re uncertain about how much we’ll be able to use those results to learn more about effects in other settings.
  • Our cost-effectiveness estimates of the direct effect of bridges built by B2P is based on a shallow analysis. We have high uncertainty about the effect of trailbridges in Rwanda, how many households benefit from trailbridges, and how long benefits persist. While we expect the Rwanda RCT to provide updates to our estimates of these parameters, which is part of the value of this grant, it’s also possible we’ve missed critical factors that cause us to overestimate cost-effectiveness. In that case, the value of the grant would be limited, both because the direct effect of trailbridges and value of information would be lower (since it would be even more unlikely we update toward believing trailbridges are above our cost-effectiveness bar).
  • We’re unsure about bridge durability and maintenance, though our impression is that safety risk from bridge failure is low. B2P has told us that there have been two cases of bridge failures during B2P’s 20-year history—once in Haiti during a hurricane and once in Rwanda during extreme flooding. In both cases, no one was hurt.45 Bridges also require regular maintenance, and bridge maintenance is the responsibility of districts.46 We have incorporated probability of bridge breakdown into our estimates of durability of benefits,47 though we’re uncertain about the value of this parameter. We also include bridge maintenance paid by districts into our estimates of bridge cost.48
  • We have only done a shallow review of B2P’s program. We have engaged less with B2P than with our top charities or with organizations receiving larger grants because this is a relatively smaller grant. It’s possible that more thorough review would lead us to update our assessment of the program.

Plans for follow-up

We expect to have preliminary results in September 2022 (including data for 25 intervention sites), an updated report of results in August 2023,49 and a summary report of final results in August 2024.50

After receiving results, we plan to update our cost-effectiveness analysis. We plan to update the following parameters:

  • Effect of bridges on household income and consumption among those closest to bridges. Our current best guess is that bridges will lead to an effect of approximately 8% on household consumption for households in villages closest to bridge sites. In producing our current best guess, we put weights of 50% on a small-scale, pilot study in Rwanda, 20% weight on a quasi-experiment in Nicaragua, and 30% weight on two studies of road-building in Ethiopia and India. We anticipate we’ll put 90% or more weight on the results of the Rwanda RCT results, given that it’s the highest quality and most relevant to the setting we’re interested in.
  • Effect of bridges on household income and consumption among those further from bridges. Our current best guess is that bridges lead to an effect of approximately 3% on more distant villages. We plan to use specifications that test how effect varies with distance to update this parameter.
  • Number of households impacted by each bridge. We currently estimate there are 150 households close to bridges (and receiving the highest impact) and 350 households farther from bridge sites but still seeing a benefit. We plan to work with B2P and the research team to understand how many households are in both nearby and distant villages and combine these with estimates of how effects vary across distance to determine the number of households “impacted” by each bridge.
  • Changes in benefits over time. We currently assume the benefits of bridges endure for 15 years. Because the Rwanda RCT includes three years of follow-up surveys, it may be possible to understand whether effects persist, at least in the short- to medium-term. However, we still expect to have substantial uncertainty about long-term effects. In the future, it may also be possible to do longer-term follow-up.
  • Additional outcomes. We also expect to learn about other key parameters in our model from the study, including effects on health and education and effects on the “less affected” side of the bridge (i.e., those who were less isolated initially).51

    Internal forecasts

    We think there’s a 30% chance that we will think this is at least 6x as effective as cash transfers once we receive the evaluation results and make a grant of $30 million or more by the end of December 2025.

    Process

    • We published an intervention report on B2P in March 2020.
    • We updated that report in 2021, based on some initial pilot findings from Rwanda.
    • We had several conversations with B2P and the research team between December 2021 and May 2022.

    Sources

    Document Source
    Bridges to Prosperity, call with GiveWell, March 16, 2022 (unpublished) Unpublished
    Bridges to Prosperity, call with GiveWell, May 6, 2022 (unpublished) Unpublished
    Bridges to Prosperity, Remaining Research Bridge Budget Analysis, 2022 Source
    Bridges to Prosperity, Response for GiveWell, May 5, 2022 Source
    Christina Barstow, Bridges to Prosperity, response to questions from GiveWell, May 3, 2022 Source
    GiveWell blog, "Revisiting leverage," 2018 Source
    GiveWell, "Bridges to Prosperity" Source
    GiveWell, "Bridges to Prosperity," 2021 version Source
    GiveWell, "New Incentives' Coverage Assessments: Plans as of October 2021" Source
    GiveWell, "Recommendation to Open Philanthropy for Grants in November 2020," 2021 Source
    GiveWell, Bridges to Prosperity power calculations, 2022 Source
    GiveWell, Bridges to Prosperity value of information BOTEC (May 2022) Source
    GiveWell, Update to Bridges to Prosperity CEA for intervention report (August 2022) Source
    Macharia et al. 2022 Source (archive)
    University of Colorado Boulder, Bridges to Prosperity Revised Study Design, 2021 Source
    Wyatt Brooks, Arizona State University, email to GiveWell, May 6, 2022 (unpublished) Unpublished
    Wyatt Brooks, Arizona State University, response to questions from GiveWell, May 10, 2022 Source
    • 1

      See our intervention report on Bridges to Prosperity’s program.

    • 2
      • B2P told us that the drivers were (1) reduced government funding due to COVID-19 – they had to reallocate funds from bridges to other programs, reducing their contribution from 40% of the total cost to 30% of the total cost (from approximately $40,000 to $30,000 of the total cost of $100,000) and (2) reduced philanthropic giving, much of which is from corporate donors who give funding as part of an experience of traveling to bridge sites. Bridges to Prosperity, call with GiveWell, March 16, 2022 (unpublished)
      • “As an example, for FY21 we had originally projected to build 40 bridges. However, the Rwandan government cut funding to the bridge program in order to serve immediate COVID needs. Through our discussions with them we agreed to reduce their contribution from 40% to 30% and were able to build 18 bridges during FY21.” Christina Barstow, Bridges to Prosperity, response to questions from GiveWell, May 3, 2022.

    • 3

      See our intervention report on Bridges to Prosperity’s program.

    • 4

      See our cost-effectiveness analysis, row "x cash after leveraging and funging," column "Best Guess of Effect of Bridges to Prosperity in Rwanda."
      As of early 2022, our bar for directing funding is about 6x as cost-effective as GiveDirectly. For examples of the cost-effectiveness of our recommendations, see this page.

    • 5

      “We are using a stepped-wedge block randomized design (Hemming et al., 2015) for the scale-up study. The steps in the wedge design are yearly, with each site surveyed once a year.” University of Colorado Boulder, Bridges to Prosperity Revised Study Design, 2021, Pg 13.

    • 6

      "The study is anchored on a stepped-wedge randomized controlled trial (RCT) implemented in 147 sites: 97 phased-in intervention sites and 50 long-term control sites." Macharia et al. 2022, Pg 1.

    • 7

      “In total, we are following 147 sites over four years (July 2020 to June 2024), for a total sample of 588 site-surveys. Of these, 17 will be built in the first year, 35 in the second and 45 in the third. The remaining 50 sites will not receive a bridge during the study period. Data collection at these sites includes household-level surveys, sensor-based monitoring of bridge use and administrative data collection. Household-level surveys will be conducted at each site once a year for a total of four rounds (one baseline survey and three follow up surveys).” University of Colorado Boulder, Bridges to Prosperity Revised Study Design, 2021, Pg 13.

    • 8

      Macharia et al. 2022, Figure 3, Pg 6.

    • 9
      • “At each bridge site, six members of the research team completed the process of determining which three villages would be most impacted by the completion of the bridge and, therefore, would be the villages in which household surveying was conducted.” Macharia et al. 2022, Pg 5.
      • “We have established 30 distant villages that we are following, conducting the same household-level survey intended to investigate the extent of impacts by the bridges in the scale-up study. Specifically, at 20 intervention sites and 10 long-term control sites, ten additional household-level surveys were conducted in distant villages. A distant village was defined as one whose boundary is 2.5 km from the bridge site, not within 2.5 km from another completed or anticipated bridge site, and adjacent to a national or district road. Where multiple villages met that criteria at a single site, one was randomly selected. These surveys will support the development of a statistical model examining the impact of the trailbridges among communities at varying distances from the bridge sites.” Macharia et al. 2022, Pg 16.

    • 10

      “The main outcomes we will study are

      • Household income and farm productivity
      • Health outcomes
      • Educational attainment and attendance

      In pursuit of the mechanisms behind those changes we will study secondary outcomes:

      • Farm outcomes: Farm outputs (harvest size, crop choice), inputs (fertilizer expenditures, labor), and land use (in- and out-rentals, acreage planted)
      • Off-farm outcomes: earnings, the location of those earnings, wages
      • Connection to market: sales of crops in market, expenditure patterns on non-crop consumption in the market

      We will also study heterogeneity in these effects. The main interactions will include

      • Household-level: savings, wealth
      • Village-level: distance, indicator for whether the village is on the side of the bridge closer to market, market prices for main consumption goods.” Macharia et al. 2022, Pgs 5–6.

      Measures for income and consumption

      • “Total Consumed is equal to the sum of market spending in all consumption categories plus the market value of all harvested crops that were consumed within the household.
      • Total Income is equal to the sum of labor market spending, business net income, and the value of all sold crops produced by the household less agricultural intermediate spending.” Bridges to Prosperity, Response for GiveWell, May 5, 2022.

      “We hypothesize that the B2P trailbridges lead to positive effects on key indicators including increased labor income and access to health services, lower transportation costs (time and money), and higher agricultural productivity and income for people in the intervention villages.” Macharia et al. 2022, Pgs 4-5.

    • 11

      “These surveys will support the development of a statistical model examining the impact of the trailbridges among
      communities at varying distances from the bridge sites.” Macharia et al. 2022, Pg 6.

    • 12

      “Our study design enables us to measure short and mid-term effects based on the data collected and the outcomes of interest.” Macharia et al. 2022, Pg 4.

    • 13

      “Sub-study question 2: Is there a correlation between weather events, river flow/discharge and bridge use? Rural mobility can be affected by various biophysical and geographical factors. We hypothesize that there is increased use of the trailbridges during periods of high rainfall and river flows.” Macharia et al. 2022, Pg 5.

    • 14

      Question 2: What mechanisms and intermediate outcomes from these trailbridges lead to impacts? Macharia et al. 2022, Pg 4.

    • 15

      Macharia et al. 2022, Pg 1.

    • 16

      $5.32m - $1.93m = $3.39m. Bridges to Prosperity, Remaining Research Bridge Budget Analysis, 2022, "Summary" tab.

    • 17

      See Bridges to Prosperity, Remaining Research Bridge Budget Analysis, 2022, "Budget Breakdown" tab.

    • 18

    • 19

      See our cost-effectiveness analysis, "Best Guess of Effect of Bridges to Prosperity in Rwanda" column.

    • 20

      “‘Have you done checks of baseline balance to confirm that there are no issues with the randomization? This is something we’re especially interested in, given the note in the pre-analysis plan (attached) that delays in bridge construction could compromise randomization.’

      “Yes, the randomization order was not compromised by budget delays, which is the most critical thing for our identification strategy. This means that our treatments are orthogonal to observable and unobservable charac- teristics. Our balance checks look as we would expect with a large number of observations. In a few outcomes there are differences that are statistically significant (though the differences are not economically meaningful). The key is that, by design, treatment is orthogonal to characteristics. Following the literature, we will control for baseline characteristics in our estimating equations.” Wyatt Brooks, Arizona State University, response to questions from GiveWell, May 10, 2022.

      We have not seen the actual baseline balance regressions, however.

    • 21
      • Our current best guess is that bridges will lead to an 8% effect on household consumption for households in villages closest to bridge sites, and we guess that an effect of 12% or greater for those households will put this within range of cost-effectiveness
      • The minimum detectable effect with 97 bridges is a 12% increase in household income and 14% increase in household consumption. The researchers estimate these are upper bounds. Bridges to Prosperity, Response for GiveWell, May 5, 2022, see table. A 12% increase in household income is calculated by dividing the pooled effect of 97 bridges on income by the mean total income: 1967.67/16442.81=12%. A 14% increase in household consumption is calculated by dividing the pooled effect of 97 bridges on consumption by the mean total consumption: 1649.05/12011.84=14%.
      • “Since we think that further cleaning is likely to reduce the standard deviation of these measures, we interpret these MDEs as an upper bound.” Bridges to Prosperity, Response for GiveWell, May 5, 2022, see table.

    • 22

      “‘How concerned are you about contamination of control areas (i.e., people in control areas using bridges in nearby treatment areas)? Are there ways we can account for or measure the extent of contamination/spillovers? Based on Figure 1 in the pre-analysis plan (attached), it looks like several of the control villages are fairly close to treatment villages.’ This is possible, but unlikely due to B2P’s selection process. Their needs assessment explicitly accounts for alternative means of crossing, including their own bridges when completed, and bridges are not built in places that have alternative options. Moreover, given the extremely rugged terrain in Rwanda, linear distance is a poor indicator of travel time, so two locations that look close on the map may actually be very difficult to travel between. That said, we can check and account for this possibility since we collect data on travel times and frequently visited locations outside the village. For example, we expect that bridges will cause intensive margin changes in travel times to schools, banks, markets and clinics, and extensive margin changes in the locations that are visited. We can check to see if nearby villages in control locations experience changes in these outcomes because of construction of nearby bridges.” Wyatt Brooks, Arizona State University, response to questions from GiveWell, May 10, 2022.

    • 23

      “The answers to all of these questions is “Yes.” Our pre-analysis plan is in the publication process at this moment. Also, data and codes will be publicly available in a de-identified form at the conclusion of the project in fulfillment of commitments made to other donors, and according to the requirements of our affiliations with J-PAL. This will be available on a public server, such as through Yale’s library system, or the J-PAL dataverse.” Wyatt Brooks, Arizona State University, response to questions from GiveWell, May 10, 2022.

    • 24

      Calculations can be found here.

    • 25

      Calculations can be found here.

    • 26

      Wyatt Brooks, Arizona State University, email to GiveWell, May 6, 2022 (unpublished).

    • 27

      Calculations can be found here.
      “We report two MDEs. The first is the pooled treatment effect (pooling across years and ages of bridges). We separately report the MDE on the difference in outcomes for villages where the bridge was completed one year ago compared to those where it was completed three years ago in the final project year. This second comparison is informative about the ability of the research design to measure the dynamic effects of bridge construction, such as paying fixed costs, learning about new crop markets, or expansion of social networks in response to bridge construction. We note that MDEs are greatly impacted by additional bridges. This suggests that a research design with only 50 bridges has essentially no possibility of detecting dynamic effects.” Bridges to Prosperity, Response for GiveWell, May 5, 2022.

    • 28

      Calculations can be found here.

    • 29

      Calculations can be found here.

    • 30

      Calculations can be found here.

    • 31

      “Furthermore, we will identify cost effective, easily assessed measures that are highly correlated to the economic and health benefits of the intervention. These measures will include electronic counting of bridge crossings, correlated to village level impacts. These measures may then be used by a portfolio of interventions across multiple geographies without always requiring complex trials.” Macharia et al. 2022, Pg 10.

    • 32

      See this page.

    • 33

      Moreover, taking general equilibrium effects into account is crucial for getting a complete picture of how many people are affected by bridges. As discussed above, it is very interesting to think about how far bridge use extends beyond the closest village that is connected. Taking into account the addition of multiple further-away new villages to outside markets, general equilibrium effects on wages and prices in those markets start to impact the entire region. Again, this is crucial for providing a complete picture of how bridges impact all the people in the areas where they are built.

    • 34

      “B2P partnerships and operations staff in Rwanda worked with government partners to ensure understanding of the randomization process and adhere as closely as possible to the randomized construction order, some times at the expense of operational efficiency and government priorities." Macharia et al. 2022, Pg 3.

    • 35

      “Motion-activated cameras have been installed at eleven bridge sites. These cameras record short videos and still photographs of objects passing through the bridges at short intervals. The cameras are enabled with infrared sensors in order to also track bridge use at night.” Macharia et al. 2022, Pg 7.

    • 36

      See here.

    • 37

      See "Expected annual funding to B2P under this scenario (m)" row.

    • 38

      B2P has a 50% or higher chance of securing $1.85m from other funders. Bridges to Prosperity, call with GiveWell, May 6, 2022 (unpublished)
      If we assume these are 75% likely to materialize, then that is 34% of the funding gap (equals $1.85 * 0.75 / ($3.40 + $0.65). However, B2P also indicated that this funding would delay Rwanda bridge-building. This would delay the results and could potentially affect study design. As a result, we adjust this percentage downward and assume a roughly 30% chance these bridges would be built and achieve the same benefit without our funding.
      We are also not taking into account the counterfactual value of this funding (which would slightly improve cost-effectiveness).

    • 39

      See here.

    • 40

      See here. B2P has also told us that if GiveWell were to recommend funding and other funders stepped in, the other funders were likely to fund bridge-building in Uganda. If bridges in Uganda and Rwanda have the same cost-effectiveness, then this funging adjustment would cancel out.

    • 41

      See here.

    • 42

      B2P’s strategy seems to be to have philanthropic funders provide initial support then have governments and multilaterals pay for the larger expansions. We don’t know if this means that B2P or government partners would be against our funding larger scale-up (provided we thought this was really promising). We’ve tried to factor this into the RFMF estimates, but if these are off, that would change the value of information as well. (There’s also some “insurance” here: If other funders step in—and we think that our funding now makes that more likely, either because bridges we fund now provide proof of concept or because the RCT makes an even more compelling case—then there's less RFMF but maybe higher cost-effectiveness by causing multilaterals to fund B2P over something that's maybe less cost-effective. We haven’t modeled that out at all, though.)

    • 43

      “Floods and landslides are significant threats to household and community resilience in the country. Reported flood impacts include damages and losses to property, livelihoods and lives (Nsengiyumva and Habiyaremye, 2014; Mind’je et al., 2019). The most recent widespread flood events occurred between 2019 and 2020 when East Africa experienced one of the wettest short rains season [sic] (October-December) in recent decades (Wainwright et al., 2020). During this period, seasonal rainfall exceeded 200% of the historical mean for the same period.” Macharia et al. 2022, Pg 6.

    • 44

      “Rwanda is one of the most densely populated countries in Africa with approximately 500 people per square kilometer and the majority living in rural areas that rely heavily on non-motorized means of transport (Shirley
      et al., 2021).” Macharia et al. 2022, Pg 4.

    • 45

      Bridges to Prosperity, call with GiveWell, May 6, 2022 (unpublished)

    • 46

      Bridges to Prosperity, call with GiveWell, May 6, 2022 (unpublished)

    • 47

      See here.

    • 48

      See here.

    • 49

      “We’ll have the preliminary first pass of the analysis in August/September of this year (noting only 25 bridges will have moved to intervention by then), with a substantial update in August of next year after all the bridges are completed.” Christina Barstow, Bridges to Prosperity, response to questions from GiveWell, May 3, 2022.

    • 50

      “Final surveying will be complete in June 2024, with a summary report available by August 2024.” Christina Barstow, Bridges to Prosperity, response to questions from GiveWell, May 3, 2022.

    • 51

      Effects on health and education: We currently include a small upward adjustment for better access to healthcare, education, and government services. The RCT will include measures of effects on health outcomes (include symptoms, medical care received, vaccination records) and educational attainment and attendance.

      Effects of those on the “less affected” side of the bridge (i.e., those who were less isolated initially). We currently include a small downward adjustment for negative impacts on those in villages receiving more workers as a result of bridges (e.g., as a result of wage decreases). We may use effects on wages to estimate general equilibrium effects on those receiving migrants from previously isolated villages.