A benefit-cost ratio (BCR) compares the total present value of a program's benefits to its total costs. A BCR above 1.0x means benefits exceed costs; below 1.0x means costs exceed benefits.
How Banerjee et al. (2015) calculated it: Costs are the full program cost per household, inflated to the equivalent of year 3. Benefits are the sum of: (1) nondurable consumption gains in years 1–2, (2) the year 3 asset and consumption gain, and (3) the year 3 consumption gain assumed to persist in perpetuity — all discounted at a 5% social discount rate.
Bandiera et al. (2017) and Bedoya et al. (2019) use the same BCR framework. The perpetuity assumption is what makes this metric favorable: it converts a $251/year consumption gain (India, year 3) into a present value of ~$5,300. This is the key driver of BCRs above 1.0x for most countries.
Banerjee et al. (2015) BCR values from Table 4, Row 11. Bangladesh and Afghanistan BCRs from respective papers. Note: different PPP base years are used across studies.
LIF's ROI metric asks: for every $1 spent on a program, how much additional household welfare do participants gain over the following five years? LIF's investment threshold is a 5x ROI.
Each term is the annual household gain in that year, summed over the 5-year horizon, undiscounted. In practice, most studies report only two endlines — typically around years 1 and 3 — so the sum is approximated as (2 × Y1 + 3 × Y3), assuming years 1–2 track the first endline and years 3–5 track the second. This is a data-constrained shorthand, not the underlying method: where richer annual data exist, they should be used directly.
Two variants are shown: one using total household consumption gains (food + nonfood + durables) and one using household income gains. For single-endline studies, the observed endline gain is used for both Y1 and Y3.
A third column, LR ROI (latest data), sums all observed years of impact to the most recent long-run follow-up and divides by cost. Where multiple survey rounds exist, annual values are linearly interpolated between observed data points (e.g. India: Y2, Y3, Y7, Y10 income estimates). Where only a single terminal estimate is available, the formula is lrYears × terminal annual gain / cost.
All values in USD PPP, but PPP base years differ across studies: 2014 (Banerjee et al.), 2007 (Bandiera et al.), 2018 (Bedoya et al.), 2016 (Bossuroy/Niger; Rahman et al.), 2024 (Beam et al.). Absolute values (costs, consumption levels) are not directly comparable across these vintages, but ROI ratios are — within each study, costs and benefits use the same base year, so the ratio is a dimensionless multiple that is comparable across studies.
The two sets of numbers measure different things using different methods — this is expected, not a sign of error. Three factors drive the gap:
1. Metric type. LIF's ROI is a simple undiscounted sum of dollar gains relative to cost (e.g. 2.5x = households received $2.50 in consumption gains per $1 spent over 5 years). GiveWell's figure is a welfare-weighted, discounted multiple of a cash benchmark — it is not in dollar units at all. A GiveWell value of 4.6x means the program delivers 4.6× as much welfare benefit per dollar as GiveDirectly cash transfers to Kenya.
2. Time horizon and discounting. LIF uses a 5-year undiscounted horizon by default. GiveWell assumes benefits persist for 15 years, discounted at 4% per year, with a +30% bonus for spillovers and health. A longer horizon with no fadeout mechanically produces higher figures.
3. Cost base. GiveWell typically uses its own (often lower) nominal-USD cost estimates. LIF uses the PPP-adjusted costs from the papers. Since CE = welfare / cost, a lower cost denominator raises the result.
The GW-style × GW benchmark toggle (in the controls above) bridges these two views by applying GiveWell's methodology to LIF's own data — holding GW's framework constant while swapping in LIF's cost estimates. Any remaining gap between that column and the published GiveWell CE column reflects cost-base differences specifically.
GiveWell's figures come from their Graduation BOTEC and meta-analysis (April 2026). GiveWell does not assess Honduras, Peru, Niger (psychosocial & full arms), Philippines, Uganda RTV, or Zambia SWL (Financial Capital arm) — these show "—". For India, GiveWell uses the Banerjee, Duflo & Sharma (2021) 10-year follow-up; for Niger, the capital arm only.
The "GW-style × GW benchmark" toggle replaces the LR ROI Latest Data column with an alternative calculation that applies GiveWell's methodology to LIF's own data. It answers the question: if we evaluated LIF's programs using GiveWell's framework — rather than LIF's standard ROI metric — how would they rank, and how do those results compare to GiveWell's published figures?
Key parameters: A 15-year benefit horizon. A 4% annual discount rate. A 30% upward adjustment for positive spillovers onto non-recipients (+20%) and likely health improvements (+10%). Log welfare weighting using a moral weight of 1.44 per unit of ln(consumption). Results are expressed as multiples of GiveDirectly's cost-effectiveness (× GW benchmark benchmark ≈ 0.003355 welfare units per dollar).
How the year-by-year profile is constructed: LIF's best available consumption data is used for each year of the 15-year window. If a long-run follow-up profile exists (e.g., India's Y2/Y3/Y7/Y10 income profile), it drives the annual shape; values are linearly interpolated between survey rounds and held flat beyond the last observed year. If only a single long-run endpoint is available (e.g., Ethiopia's Year-7 estimate), that level is applied for all 15 years. If no long-run data exists, the short-run endline (cons_y3) is held constant — equivalent to GW's own no-fadeout assumption.
The % conversion uses GiveWell's row-13 ITT figure from their BOTEC spreadsheet as the consumption-gain percentage at the reference endline. Dollar impacts at other years are scaled proportionally.
Why results still differ from GW's published figures: This toggle uses LIF's PPP cost per household, divided by an assumed household size of 4.5. GiveWell uses its own (often lower) nominal-USD cost estimates. The remaining gap between this column and the GiveWell CE column reflects cost-estimation differences. Programs with no GiveWell ITT% (Honduras, Peru, Niger psychosocial/full, Philippines, Uganda RTV, Zambia FC) show "—".
| Program ↕ | BCR Published ∞ ↓ | LIF ROI Consumption ↕5-yr horizon | LIF ROI Income ↕5-yr horizon | LR ROI Latest data ↕cons. & inc. · long-run follow-up | GiveWell Cost-Effectiveness ↕× cash · no decay | Cost / HH ↕ | N (treatment) ↕ |
|---|
Methodology Notes
BCR Published BCR methodology
BCRs for the six Banerjee et al. (2015) countries are taken directly from Table 4, Row 11. Bangladesh (Bandiera et al. 2017): Table X. Afghanistan (Bedoya et al. 2019): Table 9. Niger (Bossuroy et al. 2022): main text (capital arm, consumption-only). Beam et al. (2025): Table 2, 100% persistence scenario. Bangladesh Hybrid (Rahman et al. 2021): Table 8 (ITT). All BCRs use a 5% social discount rate except Beam et al. and Rahman et al., whose BCRs are taken directly from the published papers.
All studies assume consumption gains persist in perpetuity from the final endline, which is the primary driver of BCRs exceeding 1.0x. Without this assumption, only India would pass a 5-year threshold.
BCR ratios are comparable across studies as dimensionless multiples (benefits divided by costs, both within the same currency base). What limits comparability is methodology, not currency vintage: studies differ in which benefit components are included (consumption only vs. income), how long-run persistence is modelled, and — for Uganda/RTV — time horizon and discount rate treatment. PPP base years differ across studies (2014, 2007, 2018, 2016, 2024), but this does not itself affect comparability of the ratios.
LIF ROI Methodology — consumption & income
Both columns sum annual household gains over a 5-year horizon and divide by cost, undiscounted: (Y1 + Y2 + Y3 + Y4 + Y5) / Cost. Because most studies report only two endlines, this is approximated in practice as (2×Y1 + 3×Y3) / Cost — assuming years 1–2 hold at the first endline and years 3–5 hold at the second. For single-endline programs, the observed endline is used for all five years.
Key notes by program group: Banerjee et al. countries use per-capita monthly consumption × HH size × 12 (Tables S5a-1/S5a-2). Bangladesh BRAC TUP (Bandiera et al.) uses year-by-year HH consumption (Table IX). Afghanistan uses per-capita monthly ITT × HH size × 12. Niger uses HH-level values from SI.27 Panel 3: Y1 = $186/yr, years 2–5 = $274/yr (Y1 multiplier = 1). Beam et al. (2025) uses annual total HH consumption ITT directly from Table 2. Bangladesh Hybrid (Rahman et al. 2021) uses BDT figures converted at 24.5 PPP factor (2016).
- Uganda/RTV (Mahmud & Riley) achieves the highest consumption ROI in the dataset (10.44x) and meets LIF's 5x income threshold (7.08x). Important caveat: this is a universal program — every household in the village receives it, not just the ultra-poor. Average gains include wealthier HHs, which inflates per-household impact relative to a targeted ultra-poor comparison. BCR (2.71x) uses a 3-year horizon, not perpetuity — not comparable to Banerjee/Bandiera BCRs.
- Bangladesh Hybrid (Rahman et al.) is adjusted for long-run dissipation: Ahmed et al. (2025) shows impacts fade to zero by year 7, so Y4–Y5 gains are set to zero (Y3 multiplier = 1 rather than 3). This reduces consumption ROI from 5.06x (standard formula) to 2.62x. Hover the LIF ROI cell to see the breakdown.
- Zambia SWL (Botea et al.): all values in nominal USD, not PPP. ROI ratios are internally consistent and comparable as dimensionless multiples. However, because the values are not PPP-adjusted, the absolute dollar figures are not on the same purchasing-power basis as other programs — so interpret the scale of gains in context. Income ROIs appear very high (19-51x) because the grant cost is low relative to nominal income gains; consumption gains are more modest (1.82–2.11x). BCR not formally reported.
- India remains highest on income among targeted programs with extractable PPP values (3.41x). Philippines group (Beam et al.) reaches 3.17x on consumption despite a short horizon.
- Honduras: negative consumption ROI (−0.32x) but positive income ROI (1.03x). Asset loss suppressed consumption while earnings rose.
- Niger (Bossuroy et al.): layered on government cash transfer; cost = $482 PPP 2016 (marginal add-on only). BCR = 11.26x (perpetuity, 5% discount; SI.27 Panel 3 Row 11) — updated for comparability with Banerjee/Bandiera BCRs. Paper text reports 0.80x under zero-persistence (Scenario A), which is not comparable. Consumption from SI.27 Panel 3: Y1 = $186/yr, years 2–5 = $274/yr (unadjusted HH-level). LIF ROI (5yr) = 2.66x.
- PPP base years differ across study groups (2007–2024). Absolute values (costs, consumption levels in dollars) are not directly comparable across these vintages. ROI ratios are comparable — within each study, costs and benefits use the same base year, yielding a dimensionless multiple.
LR ROI Long-run ROI column
The "Latest data" column shows ROI using all available follow-up data. Where multiple survey rounds exist, annual gains are linearly interpolated between observed data points and summed over the full follow-up period (sum Y1–YN ÷ cost). Where only a single terminal estimate is available, the legacy formula applies: lrYears × terminal annual gain / cost. Both consumption and income ROIs are shown where data are available.
Currently populated: India (Banerjee/Duflo/Sharma 2021, 10yr — full income profile Y2/Y3/Y7/Y10, consumption terminal year only) and Ethiopia (Barker et al. 2024, year-7 follow-up; AER:I 6(2)). Hover over the 📋 icon to see source notes and observed data points for each program.
PPP Currency Converter
Convert a nominal cost estimate (in USD or local currency) to the PPP dollar base year used by a specific dashboard entry. Select a program, enter the nominal cost, and choose the year of the estimate. The converted figure is ready to paste into the cost text box.
LIF ROI Explorer
Enter estimates from an emerging or unpublished study to compute LIF ROI in real time. For single-endline studies, enter the same value for Y1 and Y3. All values in USD PPP (note the PPP base year for comparability).
Government Leverage Calculator
When a government co-finances a programme, it covers part of the total cost — so donor funding goes farther. Enter the government contribution per household (G) to see how this reduces the donor's effective cost and improves their ROI. The opportunity-cost adjustment asks whether the government's money is well-spent relative to its next-best alternative. G can be entered in paper units (USD PPP) or converted from current nominal USD using an approximate PPP factor.
Data sources: Banerjee et al. (2015), Science 348(6236) · Bandiera et al. (2017), QJE 132(2) · Bedoya et al. (2019), NBER WP 25981 · Bedoya et al. (2023), World Bank WPS 10596 · Barker et al. (2024), AER: Insights 6(2) · Banerjee, Duflo & Sharma (2021), AER: Insights 3(4) · Bossuroy et al. (2022), Nature 605 · Beam, Brune, Das, Dercon, Goldberg et al. (2025), NBER WP 34309 · Rahman, Bhattacharjee & Das (2021), Review of Development Economics 25(4) · Mahmud & Riley (2025), "We're All in This Together" (working paper) · Botea, Brudevold-Newman, Goldstein, Low & Roberts (2023), NBER WP 31625. · LIF ROI methodology: livelihoodimpactfund.org/roi · Dashboard prepared by LIF Evidence (Ventures), April 2026.