Cost-effectiveness analysis of Lafiya Nigeria intervention

Editorial note

This document is intended to explain some of the modeling decisions we made in assessing the cost-effectiveness of Lafiya Nigeria’s intervention, along with some considerations for scaling and potential model extensions. We additionally provide explanation of uncertainty in some of the model parameters, which we use to inform our uncertainty analysis.

Due to the limited time allotted for this model-building exercise, we do not provide a comprehensive report describing the model and its findings. The initial research was conducted over a brief period (less than three weeks) in May and June 2024. This publication, and the model to which it pertains, were updated over a period of one week in June 2025 to incorporate more appropriate input data, refine outcome estimates, and update programming and cost assumptions in light of recent changes. This publication therefore supersedes the previous version.

We have tried to flag major sources of uncertainty in the report and are open to revising our views as more information becomes available. We are grateful for the invaluable input of our interviewees, and for the support and data that Klau Chmielowska and Céline Kamsteeg of Lafiya Nigeria provided throughout the project.

Summary and key takeaways

• We model the near-term cost-effectiveness (2025-2026) of Lafiya Nigeria’s intervention, in which they upskill trained female health workers to provide family planning counseling and distribute injectable contraceptives (Sayana Press, or DMPA-SC) to women in remote parts of rural northern Nigeria. This update supersedes earlier work modeling the intervention’s cost-effectiveness over 2024-2025.
• We estimate that the cost-effectiveness of the intervention is $23 per unintended pregnancy averted at the beginning of 2025, falling to $12 per unintended pregnancy by the end of 2026 when approximately calibrating the model to the current and planned rate of intervention scaling.
  • We provide additional cost-effectiveness estimates, including cost per DALY averted (maternal, child, total), death averted (maternal, child, total), couple-year protection, additional contraceptive user, unsafe abortion averted, and $1 increase in women’s income, which can be found here.
  • We also provide a rough estimate of income effects: increasing the income of women serviced by Lafiya Nigeria by $1 cost $0.42 at the beginning of 2025, declining to $0.35 by the end of 2026. Reproducing GiveWell’s approach to understanding these benefits in terms of multiples of cash transfers, we find that the intervention’s impact on income increases from approximately 35x cash transfers at the beginning of 2025 to 42x cash transfers by the end of 2026.
• The most substantive update to the model aims to account for the rise in self-injection among Lafiya Nigeria’s client base in 2025, and the resulting distribution of an increased number of doses for self-injection at home. We have additionally refined some of our maternal health outcome estimates to improve accuracy following conversations with practitioners in the field, and have adjusted parameter inputs based on changes to Lafiya Nigeria’s operations and costs (most notably the cost per dose, which increased from $0.27 to $0.85), and to better reflect their updated M&E data.
• We conduct uncertainty analysis using a Monte Carlo simulation to gain insight into the effect of uncertainties from 16 inputs/factors on our cost-effectiveness estimates (Figure 1).
  • Looking at the effect of uncertainty on cost per unintended pregnancy averted, in Q1 2025 the 90% highest density interval (HDI) spans from $21-$25 per unintended pregnancy averted (mean cost-effectiveness is ~$23). This uncertainty narrows over time, with a 90% HDI spanning $11-$13 in Q4 2026 (mean cost-effectiveness is ~$12).
  • Uncertainties affect our cost-effectiveness estimation for other modeled outcomes similarly, except for the intervention’s impacts on income in multiples of cash transfers: increasing cost-effectiveness over time causes the absolute magnitude of this outcome to increase over time (from ~33x in Q1 2025 to 39x in Q4 2026) and the uncertainty bounds widen slightly over time (14x-53x to 18x-67x). The 90% HDIs exceed a 10x multiplier for cash transfers in all quarters we assessed.
  • Overall, our uncertainty analysis maintains the impression of cost-effectiveness from the spreadsheet version of the model, and indicates continuing increases in cost-effectiveness over time.

Figure 1: Uncertainty analysis summary statistics for multiple outcomes. Points represent the mean of the Monte Carlo simulation (as do numbers on the outside of the ring), with error bars representing the 90% HDI. Numbers on the inside of the ring represent the year and quarter for each estimate.

Model structure, inputs, and assumptions

We model the near-term cost-effectiveness (=2025-2026) of Lafiya Nigeria’s intervention, in which they upskill trained female health workers to provide family planning counseling and distribute injectable contraceptives (Sayana Press, or DMPA-SC) to women in remote parts of rural northern Nigeria.

This document is a supplement to our cost-effectiveness model. The notes below are intended to explain some of the modeling decisions we made, along with some considerations for scaling and potential model extensions that we or others might make given more time and/or data. Due to the limited time allotted for this model-building exercise, we do not provide a comprehensive report describing the model and its findings, but instead provided a live walkthrough of the model to Lafiya Nigeria (LN) during the project timeframe.

Dynamic model with quarterly inputs

• Dynamic:
  • A dynamic model allows model users to account for expected changes in the landscape and for intervention scaling, and to better calibrate the model to data that is or becomes available.
  • While the model does not currently amortize costs (and therefore upfront investments with benefits lasting multiple quarters, such as training Lafiya Sisters, only “count” in the quarter in which they occur) or account for lags in the realization of benefits, the dynamic structure allows for model extensions that may account for lagged costs and benefits.
• Quarterly:
  • Coverage from the Sayana Press contraceptive method lasts for 13 weeks, so that women are encouraged to receive an injection every three months (i.e., quarterly).
  • A quarterly model allows for precise calibration of the model against LN’s monthly data.^[1]
• Model through 2026:
  • LN plans to operate solely in rural (or at most peri-urban) Nigeria for the foreseeable future, though they may expand to additional geographies.
  • LN has increasingly operated in collaboration with the government, though the base model does not account for possible changes to impact or costs resulting from integration with the Nigerian Ministry of Health (i.e., LN’s longer term plan for scaling the intervention).

Supply-side issues

The previous version of the model utilized a cost per dose of $0.27, with the possibility of an increase in year 2 due to supply-side constraints. LN has recently begun to procure Sayana Press for $0.85 per dose from Pfizer and this price per dose is used in all years and quarters in the current model.^[2]

Self-injection

In our previous publication, we had suggested a potential model extension to account for self-injection. In May 2024, an anonymous expert mentioned that increasing self-injection rates may be quite promising for scaling impact, and that rates of self-injection in Nigeria were 40%. They said that women who receive self-injections have higher continuation rates than do women who receive injections from a health worker, where continuation is defined as continuing to use Sayana Press in subsequent periods following the first injection. In Malawi, Senegal, and Uganda, self-injection led to 73%, 80%, and 81% continuation at the 12-month mark, whereas clinical injection led to 45%, 70%, and 65% continuation, respectively. In other words, self-injection led to an increase in continued use of Sayana Press through the first year of 28, 10, and 16 percentage points in Malawi, Senegal, and Uganda, respectively. Their recollection was that this data comes from three RCTs in which women were randomized into receiving health worker administration or self-injection, and the expert shared the following information in follow-up communications (see Table 1).

Table 1: Rates of 12-month continuous use, by method of injection

Self-injection rates (proportion of DMPA-SC visits for self-injection) in LN states in Q4 2023:

• Jigawa: 12%
• Kebbi: 16%
• Sokoto: 16%
• National: 40%

LN has successfully encouraged increases in self-injection rates among its client base since we created our first model. Additionally, in mid-2024, the Nigerian government approved distribution of three doses of Sayana Press for women to take home and administer themselves. Given the implications of these updates for reach, we have now incorporated self-injection into the model by introducing several additional parameters that differentiate between clients seeking provider-administered injections and those who self-inject. Women who prefer to self-inject may return for visits with Lafiya Sisters just once per year to receive four doses, leading to a decrease in the burden on Lafiya Sisters that allow for visits with more clients. The model also allows for differential discontinuation rates between the two types of clients, in line with the expert insights outlined above.

Unmodeled benefits

Additional contraceptive access

LN refers about 10% of clients to other sources of contraception following counseling on informed choice, though it is unclear how many clients who do not adopt Sayana Press subsequently act upon these referrals. Nonetheless it seems likely that women who are better informed about family planning’s benefits and about the options available to them are more likely to adopt from other sources if Sayana Press is not their preferred method. We do not currently capture this additional impact in the model.

Non-health benefits

• Several sources suggest that wider access to family planning information and resources in sub-Saharan Africa would have benefits beyond child and maternal health, and we broadly think such arguments carry weight (i.e., we think several of these benefits are feasible or even likely, but we have not critically investigated the evidence base). Some such benefits could include:
  • Education benefits^[3]
  • Household wealth benefits^[4]
  • Female empowerment and societal contributions^[5]
  • Economic growth and improved government spending^[6]
  • Stronger national institutions^[7]
  • Climate change resilience and food security^[8]
• In June 2024, we spent less than one day researching and attempting to incorporate one such benefit—household wealth—into our cost-effectiveness model (we have not conducted another related literature search since).
  • We started by spending about two hours looking into the available literature and models to try to identify existing conventions for modeling these benefits, including scanning through Founders Pledge’s BOTEC for Family Empowerment Media as well as MSI’s modeling (Weinberger et al., 2023). We did not quickly find any sources that attempt to model the cost-effectiveness of family planning interventions on household wealth, though we did not conduct an exhaustive search.
  • From a quick scan of the literature on Google Scholar, we identified Canning and Schultz (2012) as a source containing an estimated treatment effect of a program similar to LN’s.
    • The authors conduct a comparative analysis of villages sharing similar characteristics in Bangladesh (1977-1996) that did (treatment group) or did not (control group) receive a program wherein community health workers offered reproductive health services to married women of reproductive age, including offering contraceptives.
    • They found women’s monthly earnings were 40% higher in treated villages than in control villages in 1996 (controlling for age and schooling) and had 25% “more physical assets per adult in their household.”^[9]
      • The authors mention a similar program in Ghana (“The Navrongo Project”) starting in 1993, but note that long-term follow-up studies have not been completed; we did not have time to search deeply for any subsequent studies that identify effects on household income, though a quick search did not suggest one exists.^[10]
    • While we make significant adjustments to the treatment effect using internal and external validity discounts in the model, we would have liked to interview David Canning to get a deeper understanding of the state of the literature on this subject with the aim of identifying an estimate in which we would feel more confident.

Assumptions

• We make several assumptions in the model (see column titled “Type” to identify them). We provide explanations for these assumptions directly in the model (see column titled “Notes”), so we use this section to mention assumptions that we are making implicitly by omitting certain elements from the model.
• ​​We assume that non-adopters of Sayana Press also do not adopt an alternative contraceptive method following a visit with a Lafiya Sister.^[11]
  • Klau Chmielowska mentioned in our interview that women who do not want to use Sayana Press can be referred to health clinics and private providers, depending on availability and proximity. Approximately 11% of women receive such a referral.
• We do not include a discount rate in the model, under the assumption that the most relevant values to LN pertain to the current quarter (we model to the end of 2026); if LN plans to calculate the net present value of the program’s benefits into the future, it would need to consider an appropriate discount rate and model accordingly.
• To calculate the maternal and child health outcomes, we largely replicated MSI’s approach—which is widely used in the family planning space—and incorporated several inputs from MSI’s collated sources.
  • However, we judged that MSI’s approach to calculating the percent of unintended pregnancies that end in miscarriage, which relied on a ratio of miscarriages per abortion calculated from a 1992 paper, would be improved if replaced with a more recent direct estimate for Nigeria from the Guttmacher Institute (Guttmacher Institute, 2015).
  • We could not find an equivalent updated input for percent of unintended pregnancies that end in stillbirth, so we instead calculated this input using MSI’s input for ‘stillbirth per 1000 live births,’ which is not specific to unintended pregnancies. We therefore implicitly assume that this ratio is the same in intended and unintended pregnancies. We find it plausible that this ratio may differ between these groups, so we are somewhat uncertain in this assumption.
  • We did not calculate unintended pregnancies averted in the same way that MSI does, which is outside of the scope of our model (see Weinberger et al., 2023, p. 27-30). Instead, we estimate the number of women receiving contraception from LN this quarter (both existing clients and new adopters) that would counterfactually not have been using contraception, and we multiply this figure by MSI’s global default pregnancy rate, which we convert from an annual (44%) to a quarterly figure (13.5%).
    • This conversion is not simply one-fourth of the annual rate, because a woman who becomes pregnant is less likely to have unintended pregnancies in subsequent quarters (e.g., if their pregnancy went to term, they would not get pregnant again for nine months, or three subsequent quarters, though the significant abortion rate for unintended pregnancies along with the percent of pregnancies ending in miscarriages suggest that several women could still become pregnant again much sooner).
    • Thus, converting from annual to quarterly requires solving for QuarterlyRate in the equation (.
  • In adapting the MSI model, we re-scaled the MMR and cross-checked the relevant term magnitudes. Our ability to investigate this model was limited by the time allocated to this project, so we have been unable to fully vet all of the underlying assumptions and formulas. As a result, a small risk of specification error persists, but sensitivity checks suggest any realistic discrepancy is unlikely to affect the report’s bottom-line conclusions.

Challenges to scaling (not modeled)

An anonymous expert suggested there may be some challenges to scaling, including stockouts and reliance on upskilling, which means reliance on a supply of health workers being available that have been trained by others. We also imagine that additional factors—such as increased coordination—could complicate intervention implementation, which we do not capture in our modeling.

Potential future model extensions

There are several potential additions/extensions to the model that LN could consider in the future. We mention some of these below:

• Incorporate additional benefits (e.g., autonomy, education, subjective well-being, environment, broader economic benefits)
• Allow for different assumptions or impacts based on women’s age (e.g., adolescent vs. adult), marital status, ethnicity, or other demographic
• Amortize costs to spread upfront investment costs over time, and consider productivity over time (e.g., lagging the benefits of Lafiya Sister trainings by one quarter, or assuming some rate of productivity increase based on experience)
• Improve data collection and modeling around discontinuation
  • On the final day of the initial project in 2024, we received a response from Maryam Kaoje suggesting that “About 50% of the women discontinue the method for wanting to get pregnant again and 40% discontinue the method because they don’t want to continue to use any method of contraception and 10% want to test another method” (interview with Maryam Kaoje)
  • The model currently assumes women either discontinue while still in need (which would include women who prefer to seek out another method), or discontinue due to the desire to get pregnant
  • We do not include women who simply no longer want contraception, and Kaoje suggests this may be a sizable fraction (40%) of the women who discontinue
  • An update to the model could attempt to reduce uncertainty around discontinuation and its implications for impact
• Given the appropriate data, incorporate nuance around differential increases in protection between ‘new adopters’ and women who switch from a different contraceptive method (however, it is difficult to conceptualize the change in protection in terms of CYP and there is a lack of data on women who were using at the time of the Lafiya Sister visit)
• Going forward, use costs in Naira where possible, so that updates to the exchange rate apply throughout the model; currently several costs are hard-coded in USD (e.g., cost per worker trained)
• Adjust the model for a longer-acting DMPA-SC contraceptive, once available
  • Switching to a longer-acting version of Sayana Press will increase cost-effectiveness all else equal, since it would require fewer doses and increase the productivity of Lafiya Sisters in reaching more women
  • We hope the model can be helpful in informing the benefits of switching, as they trade off against costs and availability
  • We are unclear whether there is a cost difference between the short vs. long-acting versions and, as LN noted, whether there might be knock-on effects to other parts of the model (e.g., rate of discontinuation)

Uncertainty analysis

To generate estimates of uncertainty for cost-effectiveness calculations, we conducted a Monte Carlo simulation using the programming language R. Monte Carlo simulation refers to using random sampling as a means of including uncertainty in a calculation. Several inputs to the cost-effectiveness model were deemed to be sufficiently uncertain that they would be best included in the model not as single numerical values, but as a distribution of possible values that reflect the range of possibilities for that input.

The calculations in the uncertainty analysis are equivalent to the calculations in the spreadsheet, except that in each case where the input was deemed to be uncertain, we draw a random sample from the corresponding distribution. For this analysis, we drew 50,000 samples for each uncertain input. The final cost estimates therefore reflect the distribution of cost-related outcomes when taking into account the uncertainty we attributed to each of the different inputs. We then plot this overall distribution and provide summary statistics (e.g., the mean, median, and modal values, as well as the interval for the 90% most likely values or ‘highest density interval’ and 90% central values or ‘equal tailed interval’).

In the next section, we describe our uncertainties and the distributions that were used to reflect them in the model. We then provide the results of the Monte Carlo simulations.

Input uncertainty

We and/or LN are currently uncertain about a number of inputs in the model, and we selected several of these for inclusion in our uncertainty analysis based on the extent of our uncertainty as well as information provided by LN. Based upon literature review and discussions with the LN team, and with an eye to keeping the values within relatively conservative ranges, we suggested some upper and lower bounds on these uncertain parameters along with a shape of the distribution of their most likely values. We then used these intuitive suggestions to select corresponding distributions of values (primarily using truncated normal, PERT, and uniform distributions). Below we describe these distributions as well as the considerations that led us to select them.

• Rate of injectable discontinuation while in need of protection
  • In our interview with an anonymous expert, they mentioned that they think a Nigeria-wide estimate is likely to underestimate this input for rural areas. While we use the Nigeria estimate in the model, we think this value is a lower bound and that the real value is plausibly quite a lot higher for the women that LN is reaching.
  • Suggested uncertainty bound: [34.7, 90]
  • Suggested uncertainty distribution: left-skewed (peaking at 70)
  • Resulting distribution: PERT(min = 34.7, mode = 70, max = 90, shape = 4)
• Reduction in rate of discontinuation while in need due to LN distribution model
  • Discontinuation is likely to be due in some part to lack of access or lack of information, both of which LN’s distribution model aims to address. We therefore assume that LN’s model reduces the discontinuation rate by some proportion, which is captured in this input.
  • An anonymous expert thinks such an adjustment would be “modest” but non-zero since “most women discontinue for reasons outside of access,” including spousal opposition, side effects, and wanting to get pregnant (the last of which we capture separately in the model).
  • The expert mentioned that higher rates of self-injection tend to improve continuation, which suggests access is still an important consideration.
  • Suggested uncertainty bound: [5, 35]
  • Suggested uncertainty distribution: normal ~ (20, 6.5)
  • Resulting distribution: Truncated normal(min = 5, max = 35, mean = 20, sd = 6.5)
• Percent of previous contraception users with current unmet demand
  • According to LN’s surveys in 2024, only 31.4% of respondents are new to contraception, though the question asked in LN’s surveys pertains to use of contraception any time prior to the survey (i.e., not necessarily at the time of the visit with the Lafiya Sister).
  • We are uncertain about the extent to which women who had previously used contraception in their lifetime were using contraception prior to switching to Sayana Press. This input in the model allows for uncertainty analysis until a data-supported approach becomes available.
  • An anonymous expert’s best guess is that the number of current contraception users among women who had previously used any contraceptive method in their lifetime would be “really low”
    • They sent a paper (Adedini et al., 2023) analyzing Nigeria Demographic and Health Survey (DHS) data that suggests that among low- and middle-income countries (LMICs), the modern contraceptive prevalence rate in Nigeria is relatively very low at 12% (2% in Sokoto State in the North West), and the discontinuation rate is relatively high at 41%.
    • The discontinuation rate for all contraceptive methods in Nigeria is highest in the North West at 50.1% (see Adedini et al., 2023, p. ix).
    • Finally, 90.0% of women in rural Nigeria were not currently using contraception when they were surveyed (National Population Commission and ICF, 2019, p. 141, Table 7.4).
  • An anonymous expert said that DHS Nigeria should have within-subject data on whether a woman has used contraception before and whether they are using it now, which can inform this estimate. They also mentioned that DHS 2023 is complete, but not yet released, so we would suggest taking a look when the new data is released to see if it can help to reduce uncertainty in relation to this model input.
    • We did not have time to seek access to within-subject data to assess differences in the rate of non-use for women who have and have not ever used contraception, but given the expert’s intuition about current contraception usage being “really low” even among women who have previously used contraception, we assume the two rates are the same or roughly similar in the model.
  • Suggested uncertainty bound: [70, 100]
  • Suggested uncertainty distribution: left-skewed (peaking at 90)
  • Resulting distribution: PERT(min = 70, mode = 90, max = 100, shape = 4)
• Percent of abortions that are unsafe
  • We are uncertain about the percentage of abortions that are unsafe in (rural) Nigeria. We have seen estimates for Western Africa as high as 85% that are from over a decade ago, and more recent global numbers that are much lower (e.g., 45% according to WHO, 2024).
  • While PMA2020 focuses on states outside of the areas where LN is operating (according to an anonymous expert), we use their estimate of 60% given it falls between the two aforementioned estimates, and that it is an estimate specifically in Nigeria. However, we think it is plausible that the true value is between 50% and 80%.
  • Suggested uncertainty bound: [50, 80]
  • Suggested uncertainty distribution: right-skewed (peaking at 60)
  • Resulting distribution: PERT(min = 50, mode = 60, max = 80, shape = 4)
• Percent of unintended pregnancies that end in stillbirth
  • We are also uncertain about whether the percentage of pregnancies that end in stillbirth is the same for intended versus unintended pregnancies, as we currently assume in the model. Our best guess is that stillbirth rates for unintended pregnancies would be equal to or higher than the average for all pregnancies (e.g., due to women with demand for contraceptives being more likely to be at risk of health complications due to recent births, or due to other known maternal health issues that might lead them to use contraception).
  • We also have seen an estimate for Nigeria from UNICEF (2020) that is almost double the estimate in MSI’s default data (i.e., 4.3% versus 2.25%), though we use MSI’s default data in our model.
  • Suggested uncertainty bound: [2.25, 6.3]
  • Suggested uncertainty distribution: normal ~ (4.3, 0.66)
  • Resulting distribution: Truncated normal(min = 2.25, max = 6.3, mean = 4.3, sd = .66)
• Estimated average monthly earnings for women in rural Nigeria (Naira)
  • An estimate of mean income earned by rural Nigerian women in a study in Kwara State was 15,344.65 Naira per month (Falola et al., 2020).^[12] Asset ownership among women in Kwara State is between that in Sokoto State (slightly higher) and those in Jigawa and Kebbi States (lower), and women’s asset ownership in all of these states is quite low (between 0-15%; National Population Commission and ICF, 2019, p. 394).
  • Estimated median rural Nigerian household’s consumption from 2018-2019, with inflation adjustments to 2020, was 71,810 Naira (Anker Research Network and Global Living Wage Coalition, 2020, p. 5, Figure 1)
  • While we use a point estimate for monthly income of 15,000 Naira, we think it is possible this value could be as low as 0 and as high as 20,000 Naira.
  • Suggested uncertainty bound: [0, 20,000]
  • Suggested distribution: left-skewed, peaking at 15,000
  • Resulting distribution: PERT(min = 0, mode = 15000, max = 20000, shape = 4)
• Internal validity discount
  • We introduce an internal validity discount for the treatment effect in the study we identified (Canning & Schultz, 2012) on the effects of family planning exposure and access on income.
  • While the non-RCT setting introduces the possibility of bias in the study’s estimates (e.g., selection bias), a very brief scan through the paper suggests the analysis is at least somewhat econometrically sophisticated, and treatment and control areas were similar based on important observable information: “Before the programme began, according to a 1974 census, all 141 villages had similar surviving fertility (ie, child-to-woman ratios), average schooling, and housing characteristics” (Canning & Schultz, 2012, p. 2).
  • Suggested uncertainty bound: [0, 90]
  • Suggested distribution: right skewed, peaking at 20
  • Resulting distribution: PERT(min = 0, mode = 20, max = 90, shape = 4)
• External validity discount
  • Canning and Schultz (2012) has several dissimilarities with the context of interest. It takes place in Matlab, Bangladesh (i.e., a different continent and culture), and evaluates the results of a decades-old policy.
    • “In the district of Matlab, Bangladesh, outreach family planning programmes were set up in 71 of 141 villages from 1977 to 1996” (Canning & Schultz, 2012, p. 2).
  • At the same time, similar to northern Nigeria, the context in the study is also a rural area of a LMIC characterized by limited female labor force participation. Additionally, the intervention seems somewhat similar to that of LN, with health workers visiting women of childbearing age and offering family planning services, including access to contraception.
    • “Community health workers were trained to visit the homes of all married women of childbearing age in the outreach programme villages every 2 weeks to offer them various contraceptives and child and maternal health services and supplies, with some additional services provided after 1982” (Canning & Schultz, 2012, p. 2).
  • Therefore, some ways in which the generalizability seems strong is in the rural LMIC context and in the limited economic opportunity for women; moreover, the program does less targeting of women who seem to be the most likely candidates for uptake, indicating that a treatment effect among the women Lafiya Sisters reach could be even stronger (i.e., due to higher likelihood of adoption).
  • However, there are some ways in which the intervention is quite different: the intervention is (i) available only to married women (who may be more or less likely to be educated, or employed), (ii) provides more regular access to family planning services, and (iii) offers more contraceptive methods, which are immediately available to recipients (though our understanding is that injectables are by far the preferred contraceptive method in rural Nigeria, and a LN partnership with MSI means women in targeted areas can fairly easily access other contraceptives as well).
  • We are not confident that this study is the most generalizable study to our context of interest, but we did not conduct an exhaustive literature search due to time constraints.
  • Suggested uncertainty bound: [10, 70]
  • Suggested distribution: uniformly distributed
  • Resulting distribution: Uniform(min = 10, max = 70)
• Cost per health worker trained
  • While the cost of training per Lafiya Sister is currently $120, LN expects to bring this cost down to $107 due to provision of more localized training as the program scales, primarily reducing transportation costs.
  • Suggested uncertainty bound: [105, 120]
  • Suggested distribution: exponential,^[13] peaking at 120
  • Resulting distribution: PERT(min = 105, mode = 120, max = 120, shape = 3)
  • Note: The resulting distribution here is intentionally conservative, with only ~5% of values being $112 or less, and a median value of ~$117. Very few samples from the distribution reach $107.
• Number of Lafiya Sisters a PO [program officer] manages
  • Suggested uncertainty bound: [120, 150]
  • Suggested distribution: exponential distribution, peaking at 120
  • Resulting distribution: PERT(min = 120, mode = 120, max = 150, shape = 3)
• Contingency
  • Céline Kamsteeg also mentioned that the organization has tended to underspend relative to its budget, but maintains a contingency in the event of unforeseen changes.
  • We include this contingency, adding 10% of all delineated costs in the total cost. However, we think it is reasonable to assume that cost-effectiveness is actually higher than modeled if the organization does not spend this contingency.
  • Suggested uncertainty bound: [0, 10]
  • Suggested distribution: uniformly distributed
  • Resulting distribution: Uniform(min = 0, max = 10)
• Exchange rate
  • Note that we considered an uncertain input for the USD/Naira exchange rate, though we encountered a challenge in that it might also require modeling how possible devaluation of the Naira could have knock-on effects on wages and prices. Without including such impacts, adding the possibility of exchange rate changes might unreasonably bias estimates towards higher cost effectiveness due to greater dollar purchasing power. These considerations were beyond the scope of this report, but we highlight that changes in the exchange rate or general valuation of the Naira remain a source of uncertainty.
• Average number of monthly visits
  • Suggested uncertainty bound: [120, 150]
  • Suggested distribution: exponential distribution peaking at 120
  • Resulting distribution: PERT(min = 120, mode = 150, max = 150, shape = 3)
• Quarterly continuation rate
  • Suggested uncertainty bound: [60, 90]
  • Suggested distribution: Evenly distributed around 75
  • Resulting distribution: PERT(min = 60, mode = 75, max = 90, shape = 3)
• Percent who retrieve doses after 1 year
  • Suggested uncertainty bound: [60, 90]
  • Suggested distribution: Evenly distributed around 75
  • Resulting distribution: PERT(min = 60, mode = 75, max = 90, shape = 3)
• Percent with intention to self inject
  • Suggested uncertainty bound: [25, 50]
  • Suggested distribution: ramping up from 25 to 50
  • Resulting distribution: PERT(min = 25, mode = 25, max = 50)

Plotted distributions for each of the uncertain inputs are in the appendix.

Effects of uncertainty on our cost-effectiveness estimates

Price per dose of Sayana Press used in this updated model is $0.85 throughout, whereas a previous version of the model used a lower price per dose in year 1 ($0.27) and the possibility of a dose increase to $0.85 in year 2.

The mean cost-effectiveness figure is approximately $23 per unintended pregnancy averted in Q1 2025, with a 90% highest density interval (HDI)^[14] ranging from approximately $21-$25. The HDIs become increasingly narrow through time as cost-effectiveness improves^[15]: for instance, by Q4 2026, the intervention averts an unintended pregnancy for approximately $12 with a 90% HDI from $11 to $14. Our uncertainties affect our cost-effectiveness estimation for other modeled outcomes similarly, with the exception of the intervention’s impacts on income in multiples of cash, which are shown in Figure 2 below.

Figure 2: Uncertainty analysis for cost-effectiveness of impacts on women’s income, in multiples of “cash”

As noted, there is significant uncertainty in this estimate in the model, with quite wide HDIs that get wider over time, as shown in Table 2 below and Figure 2 above. For instance, mean cost-effectiveness estimates grow from 33.2x with a 90% HDI of 14.1x-53.0x in Q1 2025, to 39.1x with a 90% HDI of 18.1x-64.6x in Q4 2026. This widening uncertainty bound doesn’t seem to reflect some increase in uncertainty over time that is qualitatively different from how our uncertainty changes with respect to other outcomes – the bounds increase at least in part just because of the increasing absolute magnitude of this estimate as cost-effectiveness increases, whereas the other estimates are shrinking with greater cost-effectiveness over time.

The impact of uncertainty will only be decision-relevant if the uncertainty causes us to substantially diverge from our understanding of the impact based on the static/spreadsheet version of the model. For instance, let’s say there is a donor who (i) trusts our modeling, (ii) cares about improving income in LMICs, (iii) shares GiveWell’s ‘bar’ of 10x cash transfers, and (iv) will donate funds to LN if the lower bound of the 90% HDI is above this bar. The model spreadsheet suggests this intervention exceeds the 10x bar through all periods in the model, and increases through time, so the donor’s decision is to allocate funds to LN. Through all periods, the 90% bound around the uncertainty, not only the mean, also exceeds this 10x multiple.

Overall, our uncertainty analysis increases our confidence in the findings of the model, since defining and explicitly introducing uncertainty into our estimation approach does not introduce significant or particularly meaningful variation in our estimates. The distribution of potential cost-effectiveness figures for all outcomes from our Monte Carlo simulations—including the mean, mode, and median cost-effectiveness along with 90% uncertainty intervals based on these simulations—for both ‘cost per dose’ scenarios are available in the appendix.

Table 2: Summary of cost-effectiveness in Q1 2025 and Q4 2026

To maintain model credibility through time, we suggest that LN create quarterly versions of the model to resolve uncertainties and update inputs that resolve differently through time than the model suggests. Tracking updates to these changes is key to updating the uncertainty analysis, so we would suggest documenting any such changes clearly should LN wish to engage in further uncertainty analysis going forward. We are open to sharing the R code for the uncertainty analysis or to discuss working with our team to make updates if that is the case.

Appendix

Distributions of uncertain inputs

Figure A1: Distribution of uncertain inputs

Figure A2: Distribution of uncertain inputs

Figure A3: Distribution of uncertain inputs

Distribution of cost-effectiveness from Monte Carlo simulations

Figure A4: Cost per DALY averted

Figure A5: Cost per maternal DALY averted

Figure A6: Cost per child DALY averted

Figure A7: Cost per death averted

Figure A8: Cost per maternal death averted

Figure A9: Cost per child death averted

Figure A10: Cost per additional CYP

Figure A11: Cost per additional contraceptive user

Figure A12: Cost per unintended pregnancy averted

Figure A13: Cost per unsafe abortion averted

Figure A14: Cost per $1 increase in clients’ income

Figure A15: Cost-effectiveness estimate of intervention impacts on income, in multiples of “cash” (i.e., GiveDirectly’s unconditional cash transfers)

1. LN monthly data comes from their own data dashboard that initially contained data through April 2024 and subsequently contained data through May 2025. Some of our inputs may remain unchanged from the original model, despite possible changes reflected in the data. We primarily followed guidance from LN regarding which changes were the most consequential to update to reflect their program in 2025. If you would like access to these data, please contact info@lafiyanigeria.org. ↑
2. It is unclear whether supply-side issues will continue to affect costs. If these constraints relax or worsen in the future, this can be modeled indirectly by adjusting the model inputs that quantify Lafiya Sisters’ productivity (e.g., average number of client visits per month). ↑
3. “Since World War II, no country has gone from developing status to developed status without first reducing its population growth rate. Smaller family sizes enable couples to save a higher percentage of their income and invest some of it in education and infrastructure, leading to increased productivity of the economy, greater employment, and higher incomes” (Thurston, 2021). ↑
4. “The FPMCH program is unique because of the data available to measure improvements in community and household well-being. Over time, families in the program area were more likely than the comparison group to have higher incomes, increased home value, greater savings and assets, higher educational achievement, and improved access to water. . . . Family planning enables women to be healthier and have more equal opportunities to pursue an education, a career, and financial security. With fewer children to support, families can accumulate greater assets and invest more in their children’s health and well-being. The relationship between smaller families and greater wealth highlights the benefit of sustained investments in family planning and maternal and child health programs as an important poverty reduction strategy” (PRB, 2010). ↑
5. “… couples who have large numbers of children are more likely to experience financial struggles and might have to make some difficult choices about which children to send to school and support financially. Too often, girls are the ones who go without in these cases, because their education is considered less important. . . Investing in female education is a win not only for girls, but for families, society, countries, and the world as a whole. Girls who receive a full education are more likely to have lower fertility rates and are better equipped to make informed choices for themselves, take care of their families, and contribute to society in multiple ways” (Thurston, 2021). ↑
6. “Family planning helps countries free up resources to make the infrastructure investments needed to produce high-quality productive jobs, while reducing the number of future workers entering the job market. . . Reducing fertility gives countries the breathing room to invest in education and workforce development—human capital—and in the technology infrastructure to equip them to better meet the changing demands of the 21st century” (PRB, 2016).“But family planning does more than save lives; it also saves money. For every dollar invested in reproductive health services, $2.20 is saved in pregnancy-related health-care costs. Moreover, the longer a woman waits to have children, the longer she can participate in the paid labor force, thereby boosting the economic health and prosperity of poor communities” (Kanem, 2018). ↑
7. “Reduced fertility translates into more stable population growth rates, eased pressures on the job market, fewer unemployed youth, and as a consequence, an environment more conducive to cultivating strong democracies. Shifts in age structure from a youthful population to a more mature one helps lay the foundation for social and political stability—a cornerstone of robust national institutions” (PRB, 2016). ↑
8. “Population growth and climate change are currently the two greatest threats to food security in the Sahel region of Africa. The population of the countries that make up the Sahel is projected to nearly double by 2050, from 506 million to 912 million. Paired with the expected rise in temperature and increased frequency of extreme climatic events, these numbers could quickly overwhelm relief efforts. Strengthening human capital and economic stability are critical to prevent catastrophic suffering. This article recommends two evidence-based approaches that expand women’s autonomy and support their income-earning potential while building resilience to climate change. The first recommendation would be greater investments in adolescent girls’ education and autonomy, including efforts to delay marriage and childbearing. The second calls for an improvement in the availability and quality of reproductive health services, with a special focus on voluntary family planning. These interventions can increase incomes, reproductive autonomy and gender equity which build community resilience and adaptability to climate change” (Passano et al., 2023). ↑
9. “Women in villages with an outreach programme reported monthly earnings in 1996 that were 40% higher than were earnings in comparison villages, holding constant for age and schooling. Women of childbearing age in the outreach programme area also seemed to be healthier and more productive if they were part of the paid labour force than were those in the comparison area that did paid work. This advantage remained after correction for characteristics of women with paid jobs. Married women in programme villages reported 25% more physical assets per adult in their household than did those in control areas, and the composition of household assets in programme villages had shifted away from livestock, which depends on the availability of child labour, towards housing and financial assets, consumer durables, and jewellery” (Canning and Schultz, 2012, p. 3). ↑
10. However, there is at least one study on the long-term fertility impacts (see Phillips et al., 2012), which we did not review. ↑
11. In 2024, Céline Kamsteeg said that: “[Women] may decide [to use a different contraceptive method], if available (it frequently isn’t). We don’t capture what alternative method they may be using after a referral, because due to stockouts many items aren’t available, so even if they were to prefer another method, it’s chance whether or not they will actually receive it.” ↑
12. Mukaila et al. (2022) suggest a higher number (N22,561) in Kwara State, though we use the lower estimate given the relatively high poverty levels in northern Nigerian states. ↑
13. The term ‘exponential’ is used loosely here—not referring to the formal ‘exponential distribution’ but rather to a distribution with a ‘ramping’ up or down slope. ↑
14. Note we report the HDI since it represents the 90% most likely values. It is generally quite similar to the equal-tailed interval (ETI)—i.e., the set of values in the middle 90% of the distribution of simulated outcomes—in our analysis. ↑
15. It may seem counterintuitive that these HDIs narrow over time, especially as—if we were to look at the cost alone, or unintended pregnancies averted alone—the HDIs for their values get wider (i.e., increasing uncertainty) over time. However, over time the total program costs and program impact become increasingly highly correlated with one another: In Q1 and Q2 2025, the uncertainty is more reflective of the pure random uncertainty of the initial inputs, and the costs are relatively more fixed. In contrast, in later quarters, the increased coupling between cost and impact means that higher costs are more likely than in earlier quarters to be offset by greater impacts. Further, the program is generally more cost effective at scale, as some initially high costs (e.g., executive pay) do not increase as the program scales up, and the scale is generally greater at later time points. Consequently, in later quarters it tends to become more certain that the cost effectiveness values converge on values indicating more cost-effectiveness. ↑