Lookback: Grant to Evidence Action’s Dispensers for Safe Water Program in Kenya, Uganda, and Malawi

Published: September 2025

Background

Evidence Action is an NGO that delivers low-cost health programs in Africa and Asia.1 In January 2022, GiveWell recommended a grant of up to $64.7 million to Evidence Action's Dispensers for Safe Water program. The grant length was between three and seven years, varying by location.2 At the time we made the grant, our best estimate was that it was around 7x GiveWell’s benchmark for comparing the impact of different programs per dollar spent.3

Dispensers for Safe Water installs, maintains, and promotes the use of chlorine dispensers in Kenya, Malawi, and Uganda.4 Dispensers are tanks of liquid chlorine, installed next to communal water points in rural areas. Community members turn a valve to dose chlorine into a bucket or jerrican, before filling it with water. The chlorine mixes with the water, killing pathogens and reducing the risk of waterborne disease.5

We made that grant because we thought:

• Chlorination provided by DSW was a cost-effective way to avert child mortality in these countries6
• The program was unlikely to be funded to the same size without GiveWell7
• Our funding could open up additional funding opportunities for safe water programs by (a) increasing our confidence that Evidence Action could expand its program, and so enabling further expansion to new countries, and (b) expanding DSW infrastructure that could be leveraged to deliver in-line chlorination (ILC), a separate chlorination program that we considered at the time for a different funding opportunity8
• Evidence Action would gather monitoring and evaluation (M&E) data that would improve our understanding of the program and the precision of our cost-effectiveness estimate9

The rest of this report provides our assessment of how the grant is going and next steps, based on what we’ve learned.

Would we have made this grant again, knowing what we know now

Knowing what we know now, we probably wouldn't have made this grant again at its full size. This is primarily due to our cost-effectiveness bar having increased since the original decision (from ~5x to ~8x our cost-effectiveness benchmark at the time of the grant to ~10x as of September 2025), which is in turn a reflection of GiveWell having less funding relative to available opportunities than was anticipated at the time we made this grant. We do not think this is a negative reflection on the program itself.

Our funding for the program in Kenya expired at the end of 2024. In August 2024, we decided not to renew this funding. The main reason for this is that our bar has gone up substantially since we made the grant, and Kenya was always below our bar even in the initial grant.10

If we were to make the grant today, we would consider making a smaller grant focused on Malawi and Uganda only, and possibly additional funding for expansion for scoping other countries where we think the program may be highly cost-effective. However, we’re unsure if a grant like this would exceed our current cost-effectiveness bar. We expect to learn more about this when we consider renewal of the Malawi and Uganda portions of the grant in 2025.

Part of the reasoning for the grant was also that it would test Evidence Action’s ability to expand DSW, which could enable further expansion to new countries. However, we now think a smaller grant could have achieved similar expansion objectives. We’re considering grants to expand DSW to Nigeria and Sierra Leone in late 2025, some of which may be small pilots.11 This suggests that we could have tested Evidence Action's expansion capabilities with less funding.

Our next steps:

• We plan to consider whether to renew funding for the program in Malawi and Uganda by the end of 2025. While we think that the grant discussed in this lookback is below our current cost-effectiveness thresholds, we expect our cost-effectiveness estimate for a possible renewal grant may be higher. This is because a renewal grant would not have to fund start-up costs in new locations (e.g., setting up offices, installing new dispensers). We have not yet updated our cost-effectiveness analysis for a possible renewal grant.
• We plan to consider grants to fund expansion of both DSW and ILC in Nigeria and Sierra Leone by the end of 2025. Evidence Action recently completed a survey of water point infrastructure in potential expansion locations in each country, which will inform this decision.

How did implementation go

Program implementation

Overall, our impression is that program implementation has gone well to date, although we have only investigated this question at a reasonably shallow level.

• Evidence Action installed ~12,000 new dispensers in both Malawi and Uganda (~24,000 total) as anticipated, and completed this expansion within budget.12 As of the end of 2023, it was operating slightly more dispensers than we projected when we made the grant (~53,000 dispensers in total compared to our expectation of ~51,000).13 Given the scale of this expansion in a relatively limited timeframe, we see this as a significant success.
• Data from Evidence Action’s routine monitoring and evaluation (M&E) suggests that the program has maintained similar chlorination rates to those we expected at the point we made the grant (56% overall in 2023 vs. 55% projected).14 We put less weight on this than other aspects of program implementation because we’ve seen data suggesting that Evidence Action’s M&E is overestimating the number of people reached in Kenya (more below), and Evidence Action concurs that this is likely an overestimate.

Operational challenges

Evidence Action also told us about some operational challenges they experienced over the course of this grant.15 We see these issues as in the range of normal operational challenges that would be expected when delivering a program of this scale:

• Uganda
  • Evidence Action identified fewer DSW-eligible water points than anticipated in target districts. As a result it extended its work into a few back-up districts to meet the targeted number of installations.
  • The Ebola virus outbreak in Uganda caused some disruption (e.g., because community members were hesitant to attend community engagement meetings).
• Malawi
  • Cyclone Freddy paused installation activities for about two weeks and resulted in flooding for some dispensers.
  • Fuel shortages and increased fuel prices slowed expansion activities because Evidence Action’s teams ran out of fuel.
• Kenya
  • New import requirements resulted in a four-month delay in receiving some dispenser parts (valves and padlocks). This had negative effects on dispenser functionality.16

Spending

The program has come in slightly under budget as of the end of 2023, with total costs incurred at 92% of projections in 2022 and 2023.17 Evidence Action attributed these savings to several factors including a weaker local currency in Malawi and Kenya resulting in lower dollar costs, reduced chlorine prices because of positive negotiations with vendors in Uganda, and lower-than-anticipated taxes on dispenser parts in Uganda due to identifying local suppliers.18

How cost-effective do we think this grant was?

Our current best estimate of the grant’s cost-effectiveness is ~5x our cost-effectiveness benchmark.19 This is down slightly from our initial best guess of ~7x and below our current funding bar of 10x.20 At the time of the original grant, our cost-effectiveness funding bar was 8x, and we were considering grants between 5x and 8x.

Country	2022 estimate21	Updated estimate22
Kenya	4x	4x
Uganda	6 - 7x	5x
Malawi	8 - 9x	5x
Overall	7x	5x

We currently estimate that this grant averted ~1,900 deaths over the course of 2022 and 2023, vs. ~2,300 deaths implied at the point we recommended the grant.23 We also estimate that the grant has averted or will avert ~7,600 deaths over the course of the whole grant period, vs our best estimate of ~9,000 deaths at the point we made the grant.24

The decrease in cost-effectiveness comes primarily from a lower estimate of the number of people with chlorinated water from dispensers, based on new evidence we’ve seen from Evidence Action and independent surveys (more).

Other updates to our CEA based on updated Evidence Action monitoring & evaluation and internal modeling changes we’ve made since the grant roughly canceled out. These are described in the footnote.25

We expect to more fully update our cost-effectiveness analysis when we make a decision about renewal for the Uganda and Malawi portions of the grant later in 2025. This includes reviewing updated data on the number of deaths due to diarrhea from the Institute for Health Metrics and Evaluation (IHME). We’ve found in some countries these data show surprising swings from different editions of IHME’s model and plan to investigate this further, or incorporate data from multiple sources, before making subsequent grants based on them.

Costs

The cost-effectiveness estimates that we reference above are “short-term” cost-effectiveness estimates that account for the costs and benefits over the course of the grant term itself. However, we expect that the cost-effectiveness would be higher over the longer term, as start-up costs (e.g. the costs of setting up new offices, dispenser installations, etc.) are spread out over a longer period. We estimate that using a “long-term” cost estimate where we account for both costs and benefits over an approximately 13-year period increases cost-effectiveness by around 10%.26 At the point we made the grant, we used the “long-term” cost estimate as the primary inputs for our decision-making.27

Our next steps: While we see both “short-term” and “long-term” estimates as informative, we are still discussing internally which to use as the “primary” input in our cost-effectiveness analysis.28 We plan to consider this for subsequent potential DSW expansion grants and understand if we’re being consistent about this across programs.

How many people did the program reach

We currently provisionally estimate that ~20% fewer people chlorinate water with dispensers than we believed at the point we made this grant. This is based on 2 new sources suggesting that our and Evidence Action’s previous estimates overestimated the number of people using dispensers in Kenya.

The first update comes from an independent study, the Kenya Study of Water Treatment and Child Survival (KSWTCS),29 whose focus area partly overlaps with Evidence Action’s program in western Kenya. In March 2024, the KSWTCS research team and Evidence Action shared an analysis suggesting that, in a subset of overlapping villages between 2019 and 2021, chlorination rates were lower than Evidence Action’s routine monitoring data had shown (see footnote for further details).30 Because the samples only partly overlap, and there were differences in data collection timing and methods, it was not possible based on this analysis alone to validate the discrepancies, or confirm the cause.

Evidence Action, in consultation with GiveWell and the KSWTCS research team, conducted a follow-up study in late 2024 in 69 villages in Kenya, with analysis completed in early 2025. This involved conducting two surveys in the same villages: (i) a replication of Evidence Action’s routine monitoring, conducted by Evidence Action monitoring staff, and (ii) a full census conducted by independent enumerators, where every household in the village was targeted for data collection.31 Doing both surveys in the same villages at the same time enables a more like-for-like comparison between different methods than KSWTCS.

At the time of writing (September 2025), Evidence Action’s central estimate, with substantial uncertainty in either direction, is that the census-based survey suggests ~40% fewer people in Kenya use dispensers than their routine monitoring previously suggested. They are currently in the process of investigating the reasons behind this overestimation.32

Our cost-effectiveness analysis provisionally uses Evidence Action’s estimate for the program in Kenya, and a rough assumption that the number of people using dispensers in Malawi and Uganda is -25% lower than previous estimates (i.e., two-thirds as large of an adjustment as Kenya). We would expect that the overestimation that Evidence Action’s survey discovered in Kenya is replicated to some extent in other program areas, which use the same approach to routine monitoring. But because none of the data we’ve seen to date demonstrates discrepancies with Evidence Action’s monitoring in Uganda or Malawi, we see the specific quantitative estimate we use as extremely rough, and it’s possible that the problem will not extend beyond Kenya.

To get a better understanding of the causes of the Kenya discrepancies, and understand how far they extend to Malawi and Uganda, GiveWell funded additional independent surveys in March 2025. These surveys will use a similar approach to Evidence Action’s Kenya survey, combining both a census-based survey and a replication of Evidence Action’s routine monitoring approach in the same villages at the same time. Unlike Evidence Action’s survey, both components will be conducted only by independent enumerators. We expect that this will enable us to (i) get a more precise and reliable estimate of the number of people using dispensers in Malawi and Uganda, and (ii) understand if Evidence Action’s routine monitoring approach, which we’d expect to be cheaper than a census-based approach, is able to accurately estimate the number of people reached by the program if conducted independently. We are collaborating with Evidence Action on the survey design and logistics.

Aside from these updates, the routine monitoring data we’ve received to date from Evidence Action is very similar to our projections at the point we made the grant. However, accounting for the lower program reach implied by these new datasets, our best guess estimates about the number of people using dispensers are ~20% lower than our projections.33

Indicator	2022 estimate for grant decision	Updated estimate (as of December 2023, Evidence Action routine data only)	Updated estimate (as of December 2023, with adjustment for new data sources)	Updated estimate as a % of 2022 estimate (Evidence Action routine data only)	Updated estimate as a % of 2022 estimate (with adjustment for new data sources)
Program costs for years 1 - 234	$30.9m	$28.5m	$28.5m	92%	92%
Number of dispensers installed	51,033	52,758	52,758	103%	103%
People drinking treated water from dispensers per year	4.7m	5.1m	3.7m	109%	79%
Children <5 drinking treated water from dispensers per year	630,000	674,000	491,000	107%%	78%

Our next steps:

• Review results from the new Malawi and Uganda surveys when they’re available (expected in September 2025). This will inform our decision about whether to renew funding in Malawi and Uganda in 2025, and expand the program to Nigeria and Sierra Leone, pending our and Evidence Action’s assessment of the opportunity in those countries. Evidence Action, GiveWell, and the independent research team may also discuss changes to Evidence Action’s monitoring approach depending on the results.
• Consider whether to commission more independent evaluations of monitoring information from grantees, to act as an independent check.

Major uncertainties we haven’t learned about

There were also some parts of the case for the grant that we haven’t collected more information on. The most significant of these are:

• The effect of water chlorination on under-5 mortality. Our cost-effectiveness analysis relies on a meta-analysis of randomized controlled trials of water quality interventions.35 This meta-analysis has a number of limitations, including a wide confidence interval, a significantly larger estimate of the reduction in mortality than would be expected from the effect on diarrheal disease alone, and limited information on the mechanisms that could avert mortality other than by averting diarrheal disease.36 We are currently in the process of scoping a large trial to better estimate the impact of chlorination on mortality, but we do not yet know whether such a trial is feasible. Our best guess is the program reduced under-5 all-cause mortality by ~6% across Kenya, Uganda, and Malawi.37 Our 25th - 75th percentile confidence interval for this value is 3%-10%,38 which implies a cost-effectiveness of ~4x-6x.39
• Baseline levels of mortality in program areas. This is an uncertainty across our grantmaking portfolio because we rely on modeled estimates of mortality whose methodology we don’t fully understand. Our best guess is baseline annual all-cause mortality for under-5s in program areas is ~1.2% per year on average across Kenya, Uganda, and Malawi.40 Our 25th - 75th percentile confidence interval for this value is 0.8% to 1.6%,41 which implies a cost-effectiveness of ~5x-6x.42 We’re currently interrogating the data we use to estimate the burden of disease across models and plan to begin putting weight on multiple sources going forward.

These are still major uncertainties that we think could substantially change our cost-effectiveness estimates. However, we don’t think it was feasible to collect this information as part of this grant.

Our next steps:

• By the end of 2025, update our approach for which burden of disease estimates we use and how much weight we put on other sources.
• Determine whether it’s feasible to conduct a large trial to learn more about the effect of water chlorination on mortality.

Did we set ourselves up to learn

We view the ability to learn from our grantmaking, to improve future decisions, as a crucial aspect of the grants we make. At the point we made the grant, we saw this grant as reasonably well designed to support this learning with Evidence Action collecting data on the following key indicators43 :

• Spending on the program.44
• Data to estimate the number of people served by dispensers:
  • Average number of dispensers maintained a year.
  • Average number of households using a water point where a dispenser is installed.
  • Average number of people and average number of under-five children per household among those using a water point with a dispenser.
• Data to estimate the impact of the program on chlorination rates, relative to the counterfactual:
  • Average chlorination rates among people using a water point with a dispenser.
  • Baseline water treatment rates (expansion areas45 only).
• Data to inform estimates of burden:
  • Baseline diarrhea rates (expansion areas46 only).

We now think the data on the number of people using dispensers, at least in Kenya, is overestimated. However, we have not yet come to firm conclusions on the reasons for this overestimation, and so do not yet have specific recommendations on how Evidence Action’s monitoring approach could be updated.

The following issues are less significant, but still potentially ways that we could have set the grant up better for learning:

• We lack data on chlorination rates in non-program areas, which limits our confidence in our estimates of the impact of the program on chlorination rates. Our best guess is that this isn’t a substantial limitation because we have not seen a general trend of increasing chlorination rates in other contexts, but it’s possible that we are missing something because we have not investigated this question systematically.
• Our understanding is that Evidence Action collects a large volume of data on other aspects of the program (e.g., water point functionality) that we have not reviewed because we’ve never asked for it. We may consider asking for more comprehensive program data in the future.

What about additional factors influencing our assessment of the impact of the grant?

Input from local stakeholders

We haven’t spoken to local stakeholders (e.g., local government officials or other NGOs) about how implementation has gone or how impactful Evidence Action’s activities were.

Our next step: We plan to do this as part of our decision about whether to renew funding for the program in Malawi and Uganda by the end of 2025.

Opening up room for more funding

One of the factors motivating the grant as a whole was opening up room for more funding (RFMF)47 for additional water programs.48 This had two components: (1) increasing our confidence that Evidence Action could expand its program, and so enabling further expansion to new countries, and (2) expanding DSW infrastructure that could be leveraged to deliver in-line chlorination (ILC), another chlorination program that we were considering funding separately at the time.49

On each of those:

• Evidence Action has successfully expanded DSW to new areas in Malawi and Uganda as anticipated (increasing the size of the program by ~3x to 4x in each country). We view this as a positive update on Evidence Action’s ability to scale up their program and a partial justification of the first part of our theory of change. However, we have two reservations:
  • It’s not clear that we needed to make such a large grant to do this. Some of the grants we’re planning to consider for Nigeria in Sierra Leone by the end of 2025 might be small pilots.50 We think it’s possible we could have gotten similar information on Evidence Action’s expansion capacity by expanding incrementally in this way.
  • Successful expansion in Malawi and Uganda doesn’t necessarily mean Evidence Action will be able to expand successfully elsewhere. We’ve previously heard that factors like water point infrastructure, chlorine costs, and chlorine acceptance vary significantly by country.51 These will have an impact on anywhere Evidence Action tries to expand further.
• We eventually decided to make a grant to support ILC in Malawi in 2022, but not in Kenya and Uganda.52 We think the ILC expansion has partially leveraged Evidence Action's DSW infrastructure funded through this grant (e.g., by sharing the costs of setting up new offices), although we are not sure of the precise mechanics.

Part of the case for making the Kenya-specific portion of the grant was that it hosted a number of functions (e.g., HR, finance, M&E) that support DSW, and therefore supported expansion, across all three countries.53 While we expect that funding the Kenya program helped enable expansion of the program in Uganda and Malawi, we have not investigated this assumption in detail.

Would other actors fund the program in GiveWell’s absence?

Part of the case for the grant was that we believed other funders would be unlikely to fill this full funding gap in GiveWell’s absence. We refer to the likelihood that other funders would fund a program in GiveWell’s absence as “funging” (from “fungibility”). We estimated a ~10% to ~20% probability that this would happen for this grant, varying by country.54 This was based on our understanding that other WASH funders typically prioritize improving access to water services over water quality, and that Evidence Action had struggled to fundraise for DSW in the years leading up to our grant.55

We haven’t deeply investigated whether this assumption was correct. Since we made the grant, we have had a number of conversations with sector experts that broadly corroborated our view that other funders tend to neglect water quality programs. We plan to look into this question in more detail when we decide whether to renew funding in Malawi and Uganda by the end of 2025.

As part of this grant Evidence Action and GiveWell agreed that ~$12.7m of program costs would not be covered by GiveWell, with Evidence Action seeking to fundraise them from other actors. Evidence Action has successfully met this target in the agreed time period,56 meaning the program has been fully funded during the grant period to date.

Our next steps: We plan to consider this question in more detail when we consider renewal of the Uganda and Malawi portions of the grant in 2025. One factor we’ll consider is how successful Evidence Action was at filling the funding gap for the Kenya program.

How calibrated were our forecasts?

Our forecasts57 for this grant were reasonably well calibrated (slightly over-optimistic). Overall, 7 of the 13 forecasts we made resolved positively (54%), compared to an average confidence level of 65%.58 Key themes in our forecasts were:

• We were slightly under-confident in the precision of our / Evidence Action’s monitoring data. Of 7 predictions we made about this (with an average confidence of 61%), 6 resolved positively (86%). Note that most of these predictions focus on aspects of Evidence Action’s monitoring other than chlorination rates, and so this estimate does not fully account for the monitoring issues described above.
• We were overconfident in our projections about the speed of GiveWell funding DSW expansion. We estimated a 70% chance that we would have made a grant in at least one new country with at least $30m of room for more funding by the end of 2022, and a 70% chance that for at least one new country with >$60m room for more funding by the end of 2023. As of September 2025, we have not made any new grants for DSW in new countries.

Summary of all our forecasts in the table below.

% confidence	Prediction	Resolution date (updated)	Did this happen?
75%	DSW has received government approval to operate in the selected areas	July 2022	Yes59
70%	Baseline coverage surveys and installations are concluded	March 2023	No. Too optimistic. All installations and surveys were completed by the end of June 2023.60
70%	Adoption rates in new areas are within 10% of adherence rates in current footprint in the country	May 202361	Yes,62 although we now think adherence rates (the % of people using dispensers) are overestimated, at least in Kenya.
60%	Based on baseline surveys and M&E program data in new areas, estimates for the following inputs will be within 20% of our current best guesses for…New dispensers in Uganda.	June 202363	Yes64
60%	… New dispensers in Malawi	June 202365	Yes66
60%	… Baseline water treatment rates for new areas in Uganda.	June 202367	No. Baseline water treatment rates were lower than anticipated.68
60%	… Baseline water treatment rates for new areas in Malawi.	June 202369	Yes70
60%	… People served per dispenser in new areas in Uganda.	July 202371	Yes72
60%	… People served per dispenser in new areas in Malawi.	July 202373	Yes74
60%	Our best guess cost-effectiveness across all countries (including Kenya) will be equal or above 6-8x cash.	October 2024	No. Too optimistic.75
70%	We have identified one or more countries where we expect DSW would be >8x for a total RFMF at scale of >$30m/year by December 2022.	December 2022	No. Too optimistic.76
70%	We have identified one or more countries where we expect DSW would be >8x for a total RFMF at scale of >$60m/year December 2023.	December 2023	No. Too optimistic.77
70%	We have recommended extending existing footprint in Malawi and Uganda for an additional year (~$10m), and funding at least three additional offices per each country (~$39m)	July 2023	No. Too optimistic.78

Sources

Document	Source
Development Innovation Lab, Comparison of chlorination rates from different data sources: Kenya Study on Water Treatment and Child Survival	Unpublished
Dispensers for Safe Water (DSW) Monitoring Review, Simultaneous Census and Monitoring Activity Design (August 2024)	Unpublished
Evidence Action, DSW Annual GiveWell report (2023) (redacted)	Source
Evidence Action, DSW Expansion Baseline Analysis_Feb2023_forGW	Source
Evidence Action, DSW Expansion Baseline Analysis_June2022_forGW	Source
Evidence Action, DSW Expansion Baseline Analysis_Nov2022_forGW	Source
Evidence Action, DSW Routine Monitoring Report_2023_GW	Source
Evidence Action, Evidence Action Doubles Safe Water Reach	Source (archive)
Evidence Action, External - DSW Expansion Operational Challenges	Source
GiveWell, Cost of Illness Averted Adjustment Write-up	Source
GiveWell, Dispensers for Safe Water grant size, costs and room for more funding [2024]	Source
GiveWell, Dispensers for Safe Water grant size, costs and room for more funding 2021	Source
GiveWell, Dispensers for Safe Water Organization basics, based on field notes, 2022	Source
GiveWell, Dispensers for Safe Water grant size, costs and RFMF_2021	Source
GiveWell, Dispensers for Safe Water program and M&E [2024]	Source
GiveWell, Evidence Action adherence adjustment calculations [2024]	Source
GiveWell, Evidence Action's Dispensers for Safe Water program – General Support (January 2022)	Source
GiveWell, Evidence Action's In-Line Chlorination Program — General Support (July 2022)	Source
GiveWell, GiveWell copy of DSW Routine Monitoring Analysis_2023_GW	Source
GiveWell, GiveWell's Cost-Effectiveness Analyses	Source
GiveWell, How We Produce Impact Estimates	Source
GiveWell, Room for more funding	Source
GiveWell, Tufts University — Expansion of Kenya Study of Water Treatment and Child Survival	Source
GiveWell, Water quality CEA (ILC and DSW)	Source
GiveWell, Water quality cost-effectiveness analysis [September 2025]	Source
Water quality cost-effectiveness analysis [September 2025] (baseline mortality sensitivity check, lower bound)	Source
Water quality cost-effectiveness analysis [September 2025] (baseline mortality sensitivity check, upper bound)	Source
Water quality cost-effectiveness analysis [September 2025] (effect size sensitivity check, lower bound)	Source
Water quality cost-effectiveness analysis [September 2025] (effect size sensitivity check, upper bound)	Source
GiveWell, Water Quality Interventions	Source
NOAA, Australia to Africa in 36 days: Freddy was the longest-lasting tropical cyclone in history	Source (archive)
WHO, Ebola Uganda, 2022	Source (archive)

• 1

  “Evidence Action scales low-cost health interventions that improve the wellbeing of hundreds of millions of people in Africa and Asia.” Evidence Action webpage.

• 2

  See this section of our 2022 grant page for more details on why the grant duration varies by location.

• 3

  GiveWell uses direct cash transfers as our benchmark for comparing different programs.

• 4

  For more on the program, see this section of our 2022 grant page.

• 5

  For more details, see Evidence Action’s overview of its dispensers program here.

• 6

  For more on the cost-effectiveness of the program, see this section of our 2022 grant page. Our intervention report on water quality interventions contains more detail on why we think chlorination programs are a cost-effective intervention in general.

• 7

  For more on the funding gap at the time, see this section of our 2022 grant page.

• 8

  For more on the potential room for more funding opened up by this grant, see this section of our 2022 grant page.

• 9

  For more on our ability to measure and evaluate impact from the grant, see this section of our 2022 grant page.

• 10

  For more on our reasoning for making the Kenya grant at the time even though it was below our cost-effectiveness bar, see this section of our 2022 grant page.

• 11

  We may also consider a “gated” approach, where we make a larger grant but each scale-up stage is conditional on certain criteria being met (e.g. chlorination adherence rates above a given threshold). We see this as functionally very similar to a grant where we pilot a program first, and then consider a subsequent scale-up.

• 12

  “From May 2022 through June 2023, we installed 12,161 dispensers across 11 districts in Uganda, and 12,222 dispensers across 7 districts in Malawi. This expansion was possible due to the dedicated efforts of more than 300 staff members, including full-time staff, temporary staff, and dispenser masons. Evidence Action’s dispensers now provide access to safe water to 10% of the total population of Uganda and 15% of the population of Malawi.” Evidence Action | Evidence Action Doubles Safe Water Reach

  For a comparison of Evidence Action’s spending vs expectations as of the end of 2023, see Evidence Action’s 2023 financial report.

• 13

  See this row in our analysis of Evidence Action’s monitoring data.

• 14

  See this row in our analysis of Evidence Action’s monitoring data.

• 15

  See a summary of the operational challenges Evidence Action has faced here. Note: this is a non-exhaustive list.

• 16

  We haven’t looked into this issue in detail, and note that the average number of dispensers installed in 2023 in Kenya remained higher than we anticipated when we made the grant. (~18,900 vs ~18,000)

• 17

  Source: DSW 2023 Annual Report to GiveWell (redacted) table 2. Total costs incurred in years 1 and 2 of the program (2022 and 2023) were $28,467,527 compared to $30,863,374 in projected costs = 92%.

• 18

  Source: DSW 2023 Annual Report to GiveWell (redacted) table 2 column H.

• 19

  For more on why we compare programs to direct cash transfers, see this post.

• 20

  Note that both the initial 7x estimate and our updated 5x estimate refer to cost-effectiveness over the course of the whole grant term, not the years 2022 to 2023 only. We use this method for comparability, since our main estimate we used for our grant decision in 2022 was for the grant as a whole, and we’d expect cost-effectiveness in the initial 2 years of the grant to be significantly lower. This is because program costs are frontloaded as Evidence Action opens new offices and installs dispensers.

• 21

  See this section of our previous water quality CEA for this grant.

• 22

  See this section of our updated water quality CEA for this grant, as of September 2025.

• 23

  We use 2023 as the cut off point for this estimate because this is the most recent program monitoring data we have reviewed in detail.
  Note that our 2022 estimate of deaths averted is based on our cost effectiveness analysis as of 2022, but we did not explicitly calculate deaths averted at the point we made the grant.

• 24

  This estimate is substantially higher than the 2022 to 2023 estimate because it accounts for deaths that we expect to be averted in the future over the course of the whole grant term, which was between 3 and 7 years long, depending on the specific location. More here.

• 25

  The following are a list of changes made to our water quality CEA since we made our original grant to DSW in 2022. Note: we have not prioritized determining how far each of the individual changes raised or lowered overall estimated cost-effectiveness as each individual change’s effect was small and the net effect was essentially to cancel each other out.

  • Incorporated updates from Evidence Action M&E (see the section below) into our estimates of chlorination rates achieved by the program (summarized here) and the cost per person served by the program (summarized here, calculations here). Note that in both cases we primarily use 2023 monitoring data for simplicity. When we investigate possible renewal grants for the program in Malawi and Uganda in 2025, we will consider using multi-year program data, which we’d expect to be less noisy.
    • Chlorination rates: We estimated that the proportion of households whose water tested positive for free chlorine residual was 45% in Kenya, 56% in Uganda and 60% in Malawi at the point we made the grant (see monitoring data here). We use 2023 estimates that are identical in Uganda and Malawi (56% / 60%). There was a small rise in Kenya (52%) (see monitoring data here, note, we are using the ‘Overall’ figures).
    • Cost per person served: We used updated Evidence Action estimates of the number of dispensers operating in 2023, the average number of households using each water point where a dispenser is installed, the average number of people per household, and Evidence Action’s spending on the program to date (which we estimate at 92% of projections) to estimate the cost per person served by the program.
  • Updated to using the “short-term” rather than “long-term” cost estimates as the primary cost input into our cost-effectiveness analysis. This change is discussed in more detail here. This change reduces our cost effectiveness estimate by approximately ~10%. Note that this is an internal GiveWell modeling change, not an update to the program itself.
  • Update to our internal mortality meta-analysis. We rely on an internal analysis of five studies included in a meta-analysis of water quality intervention to calculate the effect on mortality of chlorination interventions (see analysis and reasoning here. We dropped the passive control group from one of the 5 studies in the meta-analysis, Null et al. 2018 (see calculations here), because of a concern that it was introducing bias because it didn't have the baseline intervention (frequent home visits and interactions with people) that could itself have impacted mortality rate. This resulted in the following changes:
    • The main estimate of chlorination on under-5 mortality (pre-adjustments) fell from 14% to 12%.
    • The internal validity adjustment for under-5 mortality fell from -20% to -25%.
  • Updated estimate of counterfactual water treatment rates (Kenya 22% -> 10%, Uganda 17% -> 9%, Malawi 8% -> 9%), see 2022 estimates here and 2024 estimates here. This was driven by 3 changes: (1) using demographic and health survey (DHS) data only from the regions where Evidence Action was delivering the program, not each country as a whole, (2) a new source of data, a control group follow-up to Kremer et al. 2022, which we included in our estimates for Kenya, and (3) incorporated new data on baseline water treatment rates in program expansion areas in Uganda and Malawi, which Evidence Action collected as part of this grant (more below).
  • Increase in the plausibility cap for the reduction in under-5 and over-5 mortality (Kenya 5.6% -> 10.8%, Uganda 8.4% -> 11.2%, Malawi 10.9% -> 11.8%), see 2022 estimates here and updated estimates here. This was driven by (1) using an updated version of the Clasen et al meta-analysis of chlorination on diarrhea morbidity as our main input for the plausibility cap (RR 0.81 -> 0.74), and (2) updates to the external validity adjustment for the cap, based on other changes to our estimates of chlorination rates described above.
  • Updated internal validity adjustment for over-5 mortality: This was driven by (1) removing the Null et al. 2018 passive control group (described above), and (2) disaggregating the adjustment by country, rather than relying on Kenya only as a proxy (see 2022 version here vs 2024 version here).
  • Updated estimates of the increase in water treatment rates in the Kremer et al. meta analysis (47pp as of 2025 vs 52pp in 2022). This change accounts for the likelihood that some people switched from alternative forms of water treatment to chlorination as a result of the interventions in the Kremer et al. studies. We would expect this group not to benefit as much from chlorination. This changes slightly increases our cost effectiveness estimate, because it implies Evidence Action’s program reached more relative to the Kremer et al. studies than we originally thought.

• 26

  We estimate the overall cost-effectiveness of the grant would be 6.2x using the long-term cost estimate, and 5.6x using the short-term estimate (a difference of ~10%). Calculations here. Note that these calculations assume:

  • Evidence Action’s spending on the grant will be 92% of its expectations, based on lower spending than expected in years 1 and 2 of the program (more).
  • The “long-term” cost-effectiveness estimate accounts for costs and benefits over a ~13 year period (accounting for ~1.5 years of scale up, 10 years of maintenance, and a 1.25 year exit grant). The 10 years of maintenance included in this calculation is arbitrary, but broadly in line with other GiveWell cost-effectiveness analyses in similar scenarios.

• 27

  The overall short-term and long-term cost-effectiveness estimates are summarized here. In our grant page, we reported the long-term estimates as our primary estimates.

• 28

  In our initial grant decision we used the “long-term” estimate as the primary input into our cost-effectiveness analysis. More here.

• 29

  GiveWell partly funded this study. See this grant page for more information.

• 30

  This analysis found that under plausible assumptions, the number of people drinking chlorinated water in the KSWTCS dataset was only 33% as high as Evidence Action's in villages where both datasets overlap. Because of the difficulty comparing the datasets like-for-like, there is substantial uncertainty around this estimate in both directions. Development Innovation Lab, Comparison of chlorination rates from different data sources. March 2024, unpublished.

• 31

  Evidence Action, “Simultaneous Census and Monitoring Activity Design”, August 2024, unpublished.

• 32

  Evidence Action, “DSW Monitoring Review Results,” March 2025, unpublished.

• 33

  Note that this is a smaller discrepancy than would be implied by the ~-40% adjustment for Kenya and ~-25% adjustments for Malawi and Uganda cited above. This is because our 2022 cost effectiveness analysis already incorporated a skeptical 10% downward adjustment to account for possible overestimation in Evidence Action’s monitoring, e.g. because some of the monitoring data we relied on was old and there was some evidence that the number of people reached had been declining over time.

  Since our analysis now relies on more recent data and already incorporates a downward adjustment based on alternative data sources, we have removed this adjustment in our most recent analysis. This compresses the gap between our 2022 and 2025 estimates.

• 34

  Source: DSW 2023 Annual Report to GiveWell (redacted). Note: these figures use the project budget rather than the grant budget for these estimates, since we expected Evidence Action to contribute ~$12m to the program via fundraising. If we use the grant budget only, it looks like they’ve spent ~100% of the budget (I think implying that they used our funding ahead of funds from other sources, which makes sense since we think it was uncertain where their other funds would come from). We use the project budget here rather than the grant budget because we think this better captures the extent to which they’ve delivered vs expectations.

• 35

  See this section of our CEA for the methods and results of our meta-analysis, and this write-up for more details.

• 36

  See this section of our intervention report on water quality programs for more details.

• 37

  See this section of our cost-effectiveness analysis. This is a weighted average by country, weighted by the proportion of program spending in each country.

• 38

  Internal subjective GiveWell staff estimate.

• 39

  See our sensitivity analyses here (upper bound for program effect) and here (lower bound for program effect)..

• 40

  See this section of our cost-effectiveness analysis. This is a weighted average by country, weighted by the proportion of program spending in each country.

• 41

  Internal subjective GiveWell staff estimate.

• 42

  See our sensitivity analyses here (upper bound for under-5 mortality) and here (lower bound for under-5 mortality).

• 43

  These were summarized in our 2022 grant page here.

• 44

  See original budget here and costs to date (years 1 and 2) here.

• 45

  For the 2022 grant, expansion was primarily planned in Malawi and Uganda as DSW already had a substantial footprint in Kenya. See our 2022 grant page for more.

• 46

  For the 2022 grant, expansion was primarily planned in Malawi and Uganda as DSW already had a substantial footprint in Kenya. See our 2022 grant page for more.

• 47

  For more on how room for more funding affects our grantmaking, see this blog post.

• 48

  While we did not go into detail on our reasoning, we briefly mention the substantial room for more funding in our grant page at the time here.

• 49

  See our discussion of the potential for expansion and cost-sharing between DSW and in-line chlorination in this section of our 2022 grant page.

• 50

  We may also consider a “gated” approach, where we make a larger grant but each scale-up stage is conditional on certain criteria being met (e.g. chlorination adherence rates above a given threshold). We see this as functionally very similar to a grant where we pilot a program first, and then consider a subsequent scale-up.

• 51

  This is based on unpublished expert conversations.

• 52

  We declined to fund ILC in Kenya and Uganda because of lower cost-effectiveness. See our discussion of this decision here.

• 53

  See our discussion of the case for the Kenya grant at the time of grantmaking in this section of our 2022 grant page

• 54

  We estimate this as follows:

  • We roughly guess there’s a high (50%) chance of other funders covering the costs of the program’s current footprint in our absence. However, this only makes a small contribution to overall funging risk because (i) extension is <50% of the program budget, and (ii) the grant size already assumed that Evidence Action would fundraise enough to cover ~half of this.
  • We estimated a low (7.5%) chance that other funders would cover the costs of expansion. This was based on our understanding that other WASH funders tend to focus on improving people’s access to water than on water treatment, and on Evidence Action stopping further expansion of DSW in 2016 due to lack of funding.

• 55

  For more details, see this section of our 2022 grant page.

• 56

  This understanding is based on feedback from Evidence Action on a draft of this page, February 6th, 2025.

• 57

  At the time of grantmaking, GiveWell generally publishes probabilistic forecasts about grant outcomes to record the implicit predictions that inform our decisions, and to make it possible for us to look back on how well-calibrated and accurate those predictions were. For more on our use of probabilistic forecasts, see this blog post.

• 58

  This is the simple average of all the confidence levels in the table below.

• 59

  Confirmed by Evidence Action on an update call on May 3rd 2022 (unpublished).

• 60

  See discussion of the installations here. The conclusion of the surveys was confirmed to us in unpublished conversations with Evidence Action.

• 61

  The original resolution date for this forecast was March 2023. We used a delayed resolution date for this forecast to account for the survey data being later to arrive than we initially forecast.

• 62

  In May 2023, Evidence Action told us (unpublished conversations) adherence rates were 51% in Malawi expansion areas and 54% in Uganda expansion areas, compared to 56% (Uganda) and 60% (Malawi) when we made the grant (original calculations here).

• 63

  The original resolution date for this forecast was April 2023. We used a delayed resolution date for this forecast to account for the survey data being later to arrive than we initially forecast.

• 64

  When we made the grant, we projected 12,000 new dispensers would be installed per country in Uganda and Malawi. In July 2023, Evidence Action reported that it had installed 12,161 dispensers in Uganda as of June 2023.

• 65

  The original resolution date for this forecast was April 2023. We used a delayed resolution date for this forecast to account for the survey data being later to arrive than we initially forecast.

• 66

  When we made the grant, we projected 12,000 new dispensers would be installed per country in Uganda and Malawi. In July 2023, Evidence Action reported that it had installed 12,222 dispensers in Malawi as of June 2023.

• 67

  The original resolution date for this forecast was April 2023. We used a delayed resolution date for this forecast to account for the survey data being later to arrive than we initially forecast.

• 68

  We estimate baseline water treatment rates in Uganda were lower than we forecast (7.6% vs 17.2%). See calculations here and our original projection here.
  The original resolution date for this forecast was April 2023. We used a delayed resolution date for this forecast to account for the survey data being later to arrive than we initially forecast.

• 69

  The original resolution date for this forecast was April 2023. We used a delayed resolution date for this forecast to account for the survey data being later to arrive than we initially forecast.

• 70

  We estimate baseline water treatment rates in Malawi were 7.1% in expansion areas, compared to our 7.9% forecast. Note the original resolution date for this forecast was April 2023. We used a delayed resolution date for this forecast to account for the survey data being later to arrive than we initially forecast.

• 71

  The original resolution date for this forecast was April 2023. We used a delayed resolution date for this forecast to account for the survey data being later to arrive than we initially forecast.

• 72

  The first routine M&E we received for expansion areas indicated that the number of people served by each dispenser was 269 on average, compared to 261 at the time we made the grant, based on unpublished internal analysis. Note that this data only includes data from 3 expansion districts, out of 11 total expansion districts over the whole course of the grant. Note that “people served” refers to the number of people using a water point where dispensers are installed, not the number of people using the dispenser.

• 73

  The original resolution date for this forecast was April 2023. We used a delayed resolution date for this forecast to account for the survey data being later to arrive than we initially forecast.

• 74

  The first routine M&E we received for expansion areas indicated that the number of people served by each dispenser was 187 on average, compared to 181 at the time we made the grant, based on unpublished internal analysis. Note that this data only includes data from 3 expansion districts, out of 7 total expansion districts over the whole course of the grant. Note that “people served” refers to the number of people using a water point where dispensers are installed, not the number of people using the dispenser.

• 75

  As of September 2025, our best guess cost-effectiveness estimate for the grant as a whole is 5x (see calculations above). The original resolution date for this forecast was April 2023. We used a delayed resolution date for this

• 76

  This forecast refers to GiveWell making a grant for DSW in new countries. As of September 2025, we have not made any new grants for DSW expansion since we made this grant.

• 77

  This forecast refers to GiveWell making a grant for DSW in new countries. As of September 2025, we have not made any new grants for DSW expansion since we made this grant.

• 78

  As of September 2025, we have not made a decision about whether to renew our funding in Uganda / Malawi, with a renewal decision expected later in 2025.