- Split View
-
Views
-
Cite
Cite
Alison B Comfort, Paul J Krezanoski, The effect of price on demand for and use of bednets: evidence from a randomized experiment in Madagascar, Health Policy and Planning, Volume 32, Issue 2, March 2017, Pages 178–193, https://doi.org/10.1093/heapol/czw108
- Share Icon Share
There is an on-going debate about whether health products, such as insecticide-treated bednets (ITNs) for protection against malaria, should be distributed for free or at a positive price to maximize ownership and use. One argument in favour of free distribution is related to positive externalities. Like vaccines, individual use of ITNs provides a community-wide protective effect against malaria even for non-users. In addition, price may act as a barrier to ownership particularly among those most at-risk who are frequently poor. Alternatively, charging a positive price may reduce donor dependence, more efficiently allocate nets to those most at risk of malaria, and encourage use through a hypothesized sunk cost effect, where individuals are more likely to use goods they pay for. Using a randomized experiment in Madagascar, we evaluate the impact of price on demand for and use of ITNs. We find that price negatively affects both demand and use of ITNs. When price increases by $0.55, demand falls by 23.1% points (CI 19.6–26.6; P < 0.01) and effective coverage falls by 23.1% points (CI 19.6–26.6; P < 0.01). We fail to find evidence of a screening effect for prices greater than zero, but households eligible for free ITNs are more likely to use them if they have more self-reported fevers in the household at baseline. We also fail to find evidence of a sunk cost effect, meaning that households are not more likely to use nets that they pay for. Our results suggest that: (1) only partially subsidizing ITNs significantly limits ownership and (2) distributing ITNs for free or at a small nominal price will maximize demand and effective coverage. Alternative sources of financing should be identified to completely (or almost completely) subsidize the cost of ITNs in order to maximize coverage of ITNs among poor populations at risk of malaria.
• We evaluate the effect of price on demand for and use of anti-malarial bednets using a randomized controlled intervention.
• Consistent with a downward sloping demand curve, demand falls as price increases.
• There is no evidence of a screening effect at prices greater than zero: households with more self-reported fevers at baseline are not more likely to demand or use a bednet.
• Households eligible for free bednets are more likely to use them if they have more self-reported fevers in the household at baseline.
• Even partial subsidization of health products, such as bednets, is insufficient to overcome barriers to demand and use; effective coverage will be maximized when bednets are distributed for free or close to free.
Introduction
There is an on-going debate about whether preventive health products, such as insecticide-treated bednets (ITNs), should be distributed for free or sold at a positive price. One of the main policy arguments in favour of free distribution is related to the positive externalities generated from the use of ITNs since individual use of ITNs provides a community-wide protective effect against malaria through an impact on vector control [World Health Organization (WHO) 2006a]. In addition, price may be a barrier to ownership of ITNs, particularly among the poor populations most at risk of malaria. On the other hand, charging a positive price, typically partially subsidized, has been used by social marketing programmes to motivate private distributors to ensure the long-term availability of nets and to reduce donor dependence [Lengeler and deSavigny 2007; Population Services International (PSI) 2015]. In addition, a positive price could theoretically generate a screening effect by efficiently allocating ITNs to individuals who value them the most (Ashraf et al. 2010). Whereas, from a policy perspective, bednets are most valuable to individuals most at risk of malaria, there is conflicting evidence as to whether households that are more knowleadgeable about their malaria risk are more likely to demand bednets (Nganda et al. 2004; García-Basteiro et al. 2011; Krezanoski et al. 2014 vs Agyepong and Manderson 1999; Sangaré et al. 2012). A positive price could also create a sunk cost effect, based on a hypothesized psychological mechanism, where the act of paying for a good may increase the likelihood of using it in comparison to receiving it for free (Thaler 1980).
A growing body of experimental evidence investigates the effect of price on demand and use of ITNs among target populations. Cohen and Dupas (2010) provided the first results based on a randomized-controlled trial among pregnant women seeking antenatal care in rural Kenya. Their results showed that an increase in price significantly reduces both demand and use of ITNs and they failed to find evidence for a screening or sunk cost effect (Cohen and Dupas 2010). Another randomized evaluation among micro-lending clients in India found that individuals eligible for free ITNs had the highest ITN ownership and, contrary to expectations, usage was highest when ITNs were free (Tarozzi et al. 2014). Only one study has been conducted among a general population and confirmed, in rural Kenya, that demand for ITNs is price elastic and net usage is not higher among individuals who paid more for the net (Dupas 2009, 2014). Our study provides experimental evidence among a general population at risk of malaria in Madagascar on the effect of price on demand and use of ITNs. Focusing on the general population is particularly relevant given the WHO goal of universal coverage of ITNs for all individuals at risk of malaria (WHO 2013a). In addition, the general population may have different valuation of their health, knowledge about malaria, and willingness to purchase and use ITNs than targeted groups.
Background
World-wide, ∼3.2 billion individuals are at risk of malaria infection. In 2015, there were an estimated 438 000 deaths due to malaria world-wide, most of which occurred among children under 5 years of age in sub-Saharan Africa (WHO 2015a). The international community has consistently identified malaria control as one of its highest priorities for improving development, setting ambitious targets for reducing malaria incidence and mortality 90% by 2030 in the Sustainable Development Goals (WHO 2015b). Use of ITNs has been an important contributor towards the 60% decline in malaria-related mortality from 2000 to 2015 (WHO and UNICEF 2015) and accounting for an estimated 68% of the cases averted (Bhatt et al. 2015).
The WHO currently recommends universal coverage with ITNs for all individuals at risk of malaria at a ratio of 1 net per 1.8 persons (WHO 2013a). The recommended approach includes mass distribution campaigns of free ITNs and routine continuous distribution of free ITNs through antenatal and immunization clinics. Ultimately, mass distribution campaigns would give way to continuous distributions channels through health facilities, with mass campaigns used as adjuncts to sustain coverage. Non-free ITNs through the private sector are envisioned to play a supplemental role in ensuring universal coverage (WHO 2013a).
Almost all African countries with on-going malaria transmission have adopted policies to support mass distribution of free ITNs and estimated coverage has improved from 2% in 2000 to 55% in 2015 (WHO 2015a). Despite WHO recommendations for universal coverage, many countries in policy or practice continue to target primarily pregnant women and children under five years of age (WHO 2013b, 2015a). There continue to be insufficient funds to meet the WHO goal of universal coverage with ITNs, with an estimated requirement of 300 million additional ITNs per year for distribution (WHO 2015a). Currently, access to ITNs in many households is dependent on private sector providers selling partially subsidized nets. In addition, a number of African countries have adopted policies to directly sell ITNs, including some countries with national-level programmes (Willey et al. 2012; WHO 2013b). Given funding limitations, charging a positive price for ITNs may continue to be an attractive practical option to bridge the funding gap between current funding limitations and the goal of universal coverage with ITNs.
In general, the studies examining the effect of price on demand and use of ITNs suffer from selection bias because the populations targeted for social marketing interventions may be different from those targeted by free mass distribution campaigns. Multiple studies have confirmed that price may be a significant barrier to net ownership, particularly in poor, rural populations (Snow et al. 1999; Guyatt et al. 2002; Maxwell et al. 2006). Studies have also shown that free distribution can eliminate financial barriers to net use (Curtis et al. 2003; Teklehaimanot et al. 2007; WHO 2007). However, studies assessing social marketing and voucher programmes have shown that individuals are willing to pay some nominal fee for ITNs (Schellenberg et al. 2001; Guyatt et al. 2002). Recent experimental evidence has demonstrated that free nets are not more likely to be wasted and that price will not allocate nets to individuals who need them the most (Hoffmann 2009; Cohen and Dupas, 2010; Tarozzi et al. 2014). However, only one other study focuses on the effect of price in the context of a general population at risk of malaria (Dupas 2009, 2014).
Malaria in Madagascar
Malaria is transmitted primarily by Plasmodium falciparum in Madagascar. The entire population of Madagascar is at risk of malaria (WHO 2013b). We implemented the experiment in the central highlands, in the commune of Ambalavao in the Haute Matsiatra region, where the risk of malaria is seasonal, particularly high during the rainy season in December–April (INSTAT and ORC Macro 2010; INSTAT et al. 2013). The experiment was conducted from November 2004 to April 2005. At this time, malaria was the second leading cause of death for children under 5 years of age and the third leading cause of death overall (WHO 2006b). Bednet ownership was low, with 39% of households in Madagascar owning a bednet (whether treated or not1), and 36% of children under 5 years and 35% of pregnant women sleeping under a bednet (INSTAT and ORC Macro 2005).
In 2004, Madagascar adopted a policy to distribute ITNs free of charge, targeting pregnant women and young children. In practice, free ITNs were not available in the study location until after this study was completed. At the time of the experiment, bednets in Ambalavao were only available through social marketing programmes providing ITNs at partially subsidized prices through community health workers and selected retailers.2 Scale-up of malaria control activities in Madagascar began in late 2006, with ITNs and indoor residual spraying (IRS) (INSTAT and ORC Macro 2010). Only in 2009 did the Malagasy government adopt a policy of free ITNs for the general population (WHO 2013b). In summary, this study took place in an at-risk malarial region, during a time of limited access to and use of ITNs and before wide-spread scale-up of malaria control efforts with IRS and ITNs.
Methods
We used a randomized-controlled design to evaluate the effect of price on demand for and use of ITNs. We randomly varied, at the village level, the price at which households could purchase an ITN. The treatment categories included a price of zero (free) up to the social marketing price available at that time in the market (∼$2.20); the different possible price categories represented increments of $0.55 (equivalent to 1000 Ariary).3 These prices represent subsidies equal to 100, 75, 50, 25% and no subsidy, relative to the social marketing price of ITNs, which represents a subsidized price compared with the retail price. As a benchmark, 61% of the population in Madagascar was living on < $1 per day in 2006 (WHO 2006b). The social marketing price of an ITN is equivalent to 2 days of individual consumption, and the price difference in our randomization categories is equal to half a day of individual consumption in Madagascar.
The experimental sample consisted of twelve rural villages randomly selected from among all villages within a five kilometer radius of the center of the semi-urban town of Ambalavao (population ∼12 000). Randomization was conducted by lottery with all the village leaders (Table 1).4
Net price . | Subsidy level (relative to social marketing market price) . | Randomized villages . | Number of households per treatment assignment (experimental sample) . | Number of households per treatment assignment (analytic sample) . |
---|---|---|---|---|
Free ITN | 100% | Village 1 | 27 households | 27 households |
Village 2 | (0% attrition) | |||
$0.55 | 75% | Village 3 | 102 households | 64 households |
(1000 Ariary) | Village 4 | (37% attrition) | ||
$1.10 | 50% | Village 5 | 37 households | 36 households |
(2000 Ariary) | Village 6 | (3% attrition) | ||
$1.65 | 25% | Village 7 | 58 households | 43 households |
(3000 Ariary) | Village 8 | (26% attrition) | ||
$2.20 | No subsidy | Village 9 | 118 households | 92 households |
(4000 Ariary = social marketing price) | Village 10 | (22% attrition) | ||
Village 11 | ||||
Village 12 | ||||
342 | 262 |
Net price . | Subsidy level (relative to social marketing market price) . | Randomized villages . | Number of households per treatment assignment (experimental sample) . | Number of households per treatment assignment (analytic sample) . |
---|---|---|---|---|
Free ITN | 100% | Village 1 | 27 households | 27 households |
Village 2 | (0% attrition) | |||
$0.55 | 75% | Village 3 | 102 households | 64 households |
(1000 Ariary) | Village 4 | (37% attrition) | ||
$1.10 | 50% | Village 5 | 37 households | 36 households |
(2000 Ariary) | Village 6 | (3% attrition) | ||
$1.65 | 25% | Village 7 | 58 households | 43 households |
(3000 Ariary) | Village 8 | (26% attrition) | ||
$2.20 | No subsidy | Village 9 | 118 households | 92 households |
(4000 Ariary = social marketing price) | Village 10 | (22% attrition) | ||
Village 11 | ||||
Village 12 | ||||
342 | 262 |
Note: The analytic sample represents households for whom follow-up data from April is not missing.
Net price . | Subsidy level (relative to social marketing market price) . | Randomized villages . | Number of households per treatment assignment (experimental sample) . | Number of households per treatment assignment (analytic sample) . |
---|---|---|---|---|
Free ITN | 100% | Village 1 | 27 households | 27 households |
Village 2 | (0% attrition) | |||
$0.55 | 75% | Village 3 | 102 households | 64 households |
(1000 Ariary) | Village 4 | (37% attrition) | ||
$1.10 | 50% | Village 5 | 37 households | 36 households |
(2000 Ariary) | Village 6 | (3% attrition) | ||
$1.65 | 25% | Village 7 | 58 households | 43 households |
(3000 Ariary) | Village 8 | (26% attrition) | ||
$2.20 | No subsidy | Village 9 | 118 households | 92 households |
(4000 Ariary = social marketing price) | Village 10 | (22% attrition) | ||
Village 11 | ||||
Village 12 | ||||
342 | 262 |
Net price . | Subsidy level (relative to social marketing market price) . | Randomized villages . | Number of households per treatment assignment (experimental sample) . | Number of households per treatment assignment (analytic sample) . |
---|---|---|---|---|
Free ITN | 100% | Village 1 | 27 households | 27 households |
Village 2 | (0% attrition) | |||
$0.55 | 75% | Village 3 | 102 households | 64 households |
(1000 Ariary) | Village 4 | (37% attrition) | ||
$1.10 | 50% | Village 5 | 37 households | 36 households |
(2000 Ariary) | Village 6 | (3% attrition) | ||
$1.65 | 25% | Village 7 | 58 households | 43 households |
(3000 Ariary) | Village 8 | (26% attrition) | ||
$2.20 | No subsidy | Village 9 | 118 households | 92 households |
(4000 Ariary = social marketing price) | Village 10 | (22% attrition) | ||
Village 11 | ||||
Village 12 | ||||
342 | 262 |
Note: The analytic sample represents households for whom follow-up data from April is not missing.
Once a village was selected into a price category, each household was given a voucher to purchase an ITN at the treatment price from a designated local vendor located in Ambalavao.5 This distributor was selected as a convenient, centralized location since villagers typically travel to Ambalavao weekly for market day. Each household was eligible to purchase one ITN at the treatment price; the vendor maintained a list of the households and their treatment prices to prevent fraudulent purchases.
Ethical clearance for this study was provided by local officials in the town of Ambalavao and the Médicin Inspecteur of Ambalavao. Additionally, the village chiefs provided approval for the study to take place in their village. Data collection was conducted only among households that gave verbal consent. Whether or not the household provided verbal consent, each household received one voucher according to the price level designated for their village. This study was exempt from review by the Harvard University Committee on the Use of Human Subjects in Research (F18866-101).
Data
Prior to randomization, we conducted a census of all households by village (n = 348 households). Six households did not participate in the experiment. The 12 villages (with a remaining sample of 342 households) were randomized to different price categories. A household survey was conducted at baseline in November, prior to the intervention, and on a monthly basis from December through April. The baseline data included demographic and socio-economic indicators, ownership and use of ITNs, history of self-reported fevers in the household and other malaria prevention activities. Measurement of ITN use at baseline and follow-up was based on observation by the data collector of whether the household had an ITN hanging over a bed at the time of each survey. Among the experimental sample (342 households), 262 households completed both baseline and post-intervention surveys by endline in April (which we use for our outcome measures). This represents 23% sample attrition. While there is no attrition in the group randomized to free ITNs, loss to follow-up is 37% in the $0.55 category, 3% in the $1.10 category, 26% in the $1.65 category and 22% in the social marketing price ($2.20) category.
Statistical analysis
The variation in the treatment occurs at the village level, thus the small sample size restricts the number of village-level characteristics that can be included. To capture village-level characteristics in one measure, we create a pre-intervention index variable (Demand_indexij), which combines families of variables into one measure. In our case, the index is intended to reduce the number of cluster-level covariates in the regression. In other studies that have developed similar indices, such as Kling et al. (2007) and Bannerjee et al. (2015), the indices are used to reduce the number of outcome variables due to concerns with multiple comparisons. Our index captures the effect of village-level characteristics at baseline on inherent household demand for ITNs, excluding the effect of price; equal weight is given to all village-level covariates to generate the index (Appendix Table A3). This pre-intervention index is generated by separately regressing demand for ITNs on the village-level characteristics; similar regressions are run for use of ITNs and use of ITNs conditional on ownership as the dependent variables. The component characteristics include village-level factors potentially associated with malaria risk (measured as the percentage of the village population with a history of self-reported fevers during the month prior to the intervention), data collector fixed effects, average village distance to water source, number of households per village, and age and years of education of the village chief. We use the predicted value from the regression (e.g. predicted demand) and normalize it using the mean and SD of that variable. This predicted value represents the probability of ITN demand or use as a function of village-level factors (i.e. a higher z-score for this variable shows a higher proportion of ITN demand or use by village because of different village-level factors) and it is included as a regressor in the full regression.7
The main results show tests of statistical significance using the small-sample t-distribution with 13 degrees of freedom, given the small number of clusters (12 villages). We estimate cluster robust SEs, clustered at the village level.
We conduct a number of robustness checks to validate our findings. First, we test other specifications that do not assume a constant effect of price on the outcomes. We test the effect of free vs positive prices, as well as dummy variables by price category, similar to Kohler and Thornton (2012). Second, we use a probit model instead of a linear probability model. Third, we include village-level dummy variables instead of the village-level index. Fourth, we compare our results with the Wild bootstrap-t statistic estimations, following other experiments relying on randomization with fewer clusters (Burde and Linden 2013), since cluster robust SEs with a small number of clusters can still result in over-rejection of the null hypothesis (Cameron et al. 2008).9 Last, we include probability weights, generated using baseline characteristics of experimental sample (see Appendix section “Generating probability weights to adjust for non-response”), to correct for any non-random sample attrition (Wooldridge 2002).
Results
Baseline characteristics of the analytic sample
Seventy seven percent of the household heads in the analytic sample are male (Table 2). Heads of household are 43-years old on average and have four years of schooling. On average, there are five individuals per household. Almost all households live in houses with dirt floors, thatched roofs and mud walls. None of the households have electricity, nor do they own a television or a telephone, but 8 in 10 own a radio. The household members tend to be farmers and own animals for consumption and income generation.
. | . | Free . | $0.55 . | $1.10 . | $1.65 . | $2.20 . | Average . | F-stat . | Obs. . |
---|---|---|---|---|---|---|---|---|---|
Demographic indicators | |||||||||
Household head | Age | 39.52 | 45.23 | 42.81 | 41.09 | 43.53 | 43.03 | 0.83 | 262 |
Male (%) | 81% | 75% | 78% | 86% | 74% | 77% | 0.73 | 262 | |
Years of schooling | 3.81 | 4.33 | 2.92 | 3.33 | 4.54 | 3.99 | 2.88** | 262 | |
Married (%) | 89% | 75% | 78% | 79% | 73% | 77% | 0.81 | 262 | |
Spouse | Age | 33.63 | 38.08 | 36.54 | 35.71 | 33.91 | 35.54 | 0.84 | 201 |
Female (%) | 100% | 100% | 100% | 100% | 100% | 100% | . | 201 | |
Years of schooling | 3.71 | 4.52 | 4.04 | 3.47 | 4.90 | 4.30 | 2.05* | 201 | |
Household size | 5.56 | 5.08 | 5.56 | 5.12 | 5.51 | 5.35 | 0.49 | 262 | |
Socio-economic indicators | |||||||||
Assets | Radio (%) | 70% | 78% | 83% | 78% | 84% | 80% | 0.76 | 257 |
Television (%) | 0% | 0% | 0% | 0% | 0% | 0% | . | 257 | |
Telephone (%) | 0% | 0% | 0% | 0% | 0% | 0% | . | 257 | |
Electricity (%) | 0% | 0% | 3% | 0% | 0% | 0% | 1.55 | 257 | |
Bicycle (%) | 30% | 22% | 19% | 17% | 31% | 25% | 1.15 | 257 | |
Motorcycle (%) | 0% | 0% | 0% | 0% | 2% | 1% | 0.95 | 257 | |
Car (%) | 0% | 0% | 0% | 0% | 0% | 0% | . | 257 | |
Cattle cart (%) | 0% | 22% | 8% | 5% | 16% | 13% | 3.16** | 257 | |
Livestock | Pigs (%) | 7% | 3% | 8% | 10% | 8% | 7% | 0.53 | 257 |
Ducks (%) | 0% | 11% | 0% | 2% | 11% | 7% | 2.66** | 262 | |
Chicken (%) | 56% | 56% | 25% | 59% | 50% | 50% | 3.03** | 262 | |
Cattle (%) | 0.37 | 0.84 | 0.58 | 0.44 | 0.71 | 0.65 | 0.94 | 262 | |
Housing | Beds | 1.57 | 1.73 | 1.69 | 1.48 | 1.62 | 1.63 | 0.89 | 262 |
Rooms | 1.31 | 1.58 | 1.44 | 1.44 | 1.48 | 1.47 | 0.85 | 262 | |
Floor | Dirt (%) | 96% | 98% | 100% | 100% | 98% | 98% | 0.58 | 253 |
Roof | Thatch (%) | 96% | 100% | 94% | 100% | 98% | 97% | 3.35** | 262 |
Walls | Mud/Wood (%) | 100% | 100% | 97% | 100% | 97% | 98% | 1.07 | 262 |
Water source and access | |||||||||
Access to water | Number of water trips per day | 2.56 | 2.20 | 2.36 | 2.46 | 2.20 | 2.30 | 1.35 | 257 |
Distance from water source (mins) | 12.96 | 7.25 | 7.14 | 5.98 | 11.73 | 9.18 | 16.08*** | 257 | |
Reporting that water source is far (%) | 30% | 0% | 44% | 0% | 27% | 19% | 13.96*** | 262 | |
Water source | Pump (%) | 63% | 33% | 47% | 70% | 49% | 50% | 3.42*** | 262 |
Pond (%) | 0% | 67% | 53% | 30% | 30% | 39% | 262 | ||
Stream (%) | 37% | 0% | 0% | 0% | 21% | 11% | 262 | ||
Self-reported fevers and potential malaria risk | |||||||||
HH member goes out at night (%) | 45% | 12% | 14% | 19% | 31% | 24% | 11.78*** | 262 | |
Self-reported fever in last month (%) | 8% | 19% | 9% | 18% | 8% | 12% | 4.18*** | 262 | |
Consult doctor for fever in last month (%) | 6% | 13% | 2% | 11% | 6% | 8% | 3.62*** | 262 | |
Purchase medication for fever in last month (%) | 7% | 17% | 3% | 12% | 7% | 10% | 5.23*** | 262 | |
Malaria prevention | Own mosquito net (%) | 4% | 7% | 0% | 2% | 3% | 3% | 1.25 | 262 |
Use mosquito net (%) | 0% | 2% | 0% | 0% | 1% | 1% | 0.44 | 262 |
. | . | Free . | $0.55 . | $1.10 . | $1.65 . | $2.20 . | Average . | F-stat . | Obs. . |
---|---|---|---|---|---|---|---|---|---|
Demographic indicators | |||||||||
Household head | Age | 39.52 | 45.23 | 42.81 | 41.09 | 43.53 | 43.03 | 0.83 | 262 |
Male (%) | 81% | 75% | 78% | 86% | 74% | 77% | 0.73 | 262 | |
Years of schooling | 3.81 | 4.33 | 2.92 | 3.33 | 4.54 | 3.99 | 2.88** | 262 | |
Married (%) | 89% | 75% | 78% | 79% | 73% | 77% | 0.81 | 262 | |
Spouse | Age | 33.63 | 38.08 | 36.54 | 35.71 | 33.91 | 35.54 | 0.84 | 201 |
Female (%) | 100% | 100% | 100% | 100% | 100% | 100% | . | 201 | |
Years of schooling | 3.71 | 4.52 | 4.04 | 3.47 | 4.90 | 4.30 | 2.05* | 201 | |
Household size | 5.56 | 5.08 | 5.56 | 5.12 | 5.51 | 5.35 | 0.49 | 262 | |
Socio-economic indicators | |||||||||
Assets | Radio (%) | 70% | 78% | 83% | 78% | 84% | 80% | 0.76 | 257 |
Television (%) | 0% | 0% | 0% | 0% | 0% | 0% | . | 257 | |
Telephone (%) | 0% | 0% | 0% | 0% | 0% | 0% | . | 257 | |
Electricity (%) | 0% | 0% | 3% | 0% | 0% | 0% | 1.55 | 257 | |
Bicycle (%) | 30% | 22% | 19% | 17% | 31% | 25% | 1.15 | 257 | |
Motorcycle (%) | 0% | 0% | 0% | 0% | 2% | 1% | 0.95 | 257 | |
Car (%) | 0% | 0% | 0% | 0% | 0% | 0% | . | 257 | |
Cattle cart (%) | 0% | 22% | 8% | 5% | 16% | 13% | 3.16** | 257 | |
Livestock | Pigs (%) | 7% | 3% | 8% | 10% | 8% | 7% | 0.53 | 257 |
Ducks (%) | 0% | 11% | 0% | 2% | 11% | 7% | 2.66** | 262 | |
Chicken (%) | 56% | 56% | 25% | 59% | 50% | 50% | 3.03** | 262 | |
Cattle (%) | 0.37 | 0.84 | 0.58 | 0.44 | 0.71 | 0.65 | 0.94 | 262 | |
Housing | Beds | 1.57 | 1.73 | 1.69 | 1.48 | 1.62 | 1.63 | 0.89 | 262 |
Rooms | 1.31 | 1.58 | 1.44 | 1.44 | 1.48 | 1.47 | 0.85 | 262 | |
Floor | Dirt (%) | 96% | 98% | 100% | 100% | 98% | 98% | 0.58 | 253 |
Roof | Thatch (%) | 96% | 100% | 94% | 100% | 98% | 97% | 3.35** | 262 |
Walls | Mud/Wood (%) | 100% | 100% | 97% | 100% | 97% | 98% | 1.07 | 262 |
Water source and access | |||||||||
Access to water | Number of water trips per day | 2.56 | 2.20 | 2.36 | 2.46 | 2.20 | 2.30 | 1.35 | 257 |
Distance from water source (mins) | 12.96 | 7.25 | 7.14 | 5.98 | 11.73 | 9.18 | 16.08*** | 257 | |
Reporting that water source is far (%) | 30% | 0% | 44% | 0% | 27% | 19% | 13.96*** | 262 | |
Water source | Pump (%) | 63% | 33% | 47% | 70% | 49% | 50% | 3.42*** | 262 |
Pond (%) | 0% | 67% | 53% | 30% | 30% | 39% | 262 | ||
Stream (%) | 37% | 0% | 0% | 0% | 21% | 11% | 262 | ||
Self-reported fevers and potential malaria risk | |||||||||
HH member goes out at night (%) | 45% | 12% | 14% | 19% | 31% | 24% | 11.78*** | 262 | |
Self-reported fever in last month (%) | 8% | 19% | 9% | 18% | 8% | 12% | 4.18*** | 262 | |
Consult doctor for fever in last month (%) | 6% | 13% | 2% | 11% | 6% | 8% | 3.62*** | 262 | |
Purchase medication for fever in last month (%) | 7% | 17% | 3% | 12% | 7% | 10% | 5.23*** | 262 | |
Malaria prevention | Own mosquito net (%) | 4% | 7% | 0% | 2% | 3% | 3% | 1.25 | 262 |
Use mosquito net (%) | 0% | 2% | 0% | 0% | 1% | 1% | 0.44 | 262 |
Note: ***P < 0.01, **P < 0.05, *P < 0.1.
. | . | Free . | $0.55 . | $1.10 . | $1.65 . | $2.20 . | Average . | F-stat . | Obs. . |
---|---|---|---|---|---|---|---|---|---|
Demographic indicators | |||||||||
Household head | Age | 39.52 | 45.23 | 42.81 | 41.09 | 43.53 | 43.03 | 0.83 | 262 |
Male (%) | 81% | 75% | 78% | 86% | 74% | 77% | 0.73 | 262 | |
Years of schooling | 3.81 | 4.33 | 2.92 | 3.33 | 4.54 | 3.99 | 2.88** | 262 | |
Married (%) | 89% | 75% | 78% | 79% | 73% | 77% | 0.81 | 262 | |
Spouse | Age | 33.63 | 38.08 | 36.54 | 35.71 | 33.91 | 35.54 | 0.84 | 201 |
Female (%) | 100% | 100% | 100% | 100% | 100% | 100% | . | 201 | |
Years of schooling | 3.71 | 4.52 | 4.04 | 3.47 | 4.90 | 4.30 | 2.05* | 201 | |
Household size | 5.56 | 5.08 | 5.56 | 5.12 | 5.51 | 5.35 | 0.49 | 262 | |
Socio-economic indicators | |||||||||
Assets | Radio (%) | 70% | 78% | 83% | 78% | 84% | 80% | 0.76 | 257 |
Television (%) | 0% | 0% | 0% | 0% | 0% | 0% | . | 257 | |
Telephone (%) | 0% | 0% | 0% | 0% | 0% | 0% | . | 257 | |
Electricity (%) | 0% | 0% | 3% | 0% | 0% | 0% | 1.55 | 257 | |
Bicycle (%) | 30% | 22% | 19% | 17% | 31% | 25% | 1.15 | 257 | |
Motorcycle (%) | 0% | 0% | 0% | 0% | 2% | 1% | 0.95 | 257 | |
Car (%) | 0% | 0% | 0% | 0% | 0% | 0% | . | 257 | |
Cattle cart (%) | 0% | 22% | 8% | 5% | 16% | 13% | 3.16** | 257 | |
Livestock | Pigs (%) | 7% | 3% | 8% | 10% | 8% | 7% | 0.53 | 257 |
Ducks (%) | 0% | 11% | 0% | 2% | 11% | 7% | 2.66** | 262 | |
Chicken (%) | 56% | 56% | 25% | 59% | 50% | 50% | 3.03** | 262 | |
Cattle (%) | 0.37 | 0.84 | 0.58 | 0.44 | 0.71 | 0.65 | 0.94 | 262 | |
Housing | Beds | 1.57 | 1.73 | 1.69 | 1.48 | 1.62 | 1.63 | 0.89 | 262 |
Rooms | 1.31 | 1.58 | 1.44 | 1.44 | 1.48 | 1.47 | 0.85 | 262 | |
Floor | Dirt (%) | 96% | 98% | 100% | 100% | 98% | 98% | 0.58 | 253 |
Roof | Thatch (%) | 96% | 100% | 94% | 100% | 98% | 97% | 3.35** | 262 |
Walls | Mud/Wood (%) | 100% | 100% | 97% | 100% | 97% | 98% | 1.07 | 262 |
Water source and access | |||||||||
Access to water | Number of water trips per day | 2.56 | 2.20 | 2.36 | 2.46 | 2.20 | 2.30 | 1.35 | 257 |
Distance from water source (mins) | 12.96 | 7.25 | 7.14 | 5.98 | 11.73 | 9.18 | 16.08*** | 257 | |
Reporting that water source is far (%) | 30% | 0% | 44% | 0% | 27% | 19% | 13.96*** | 262 | |
Water source | Pump (%) | 63% | 33% | 47% | 70% | 49% | 50% | 3.42*** | 262 |
Pond (%) | 0% | 67% | 53% | 30% | 30% | 39% | 262 | ||
Stream (%) | 37% | 0% | 0% | 0% | 21% | 11% | 262 | ||
Self-reported fevers and potential malaria risk | |||||||||
HH member goes out at night (%) | 45% | 12% | 14% | 19% | 31% | 24% | 11.78*** | 262 | |
Self-reported fever in last month (%) | 8% | 19% | 9% | 18% | 8% | 12% | 4.18*** | 262 | |
Consult doctor for fever in last month (%) | 6% | 13% | 2% | 11% | 6% | 8% | 3.62*** | 262 | |
Purchase medication for fever in last month (%) | 7% | 17% | 3% | 12% | 7% | 10% | 5.23*** | 262 | |
Malaria prevention | Own mosquito net (%) | 4% | 7% | 0% | 2% | 3% | 3% | 1.25 | 262 |
Use mosquito net (%) | 0% | 2% | 0% | 0% | 1% | 1% | 0.44 | 262 |
. | . | Free . | $0.55 . | $1.10 . | $1.65 . | $2.20 . | Average . | F-stat . | Obs. . |
---|---|---|---|---|---|---|---|---|---|
Demographic indicators | |||||||||
Household head | Age | 39.52 | 45.23 | 42.81 | 41.09 | 43.53 | 43.03 | 0.83 | 262 |
Male (%) | 81% | 75% | 78% | 86% | 74% | 77% | 0.73 | 262 | |
Years of schooling | 3.81 | 4.33 | 2.92 | 3.33 | 4.54 | 3.99 | 2.88** | 262 | |
Married (%) | 89% | 75% | 78% | 79% | 73% | 77% | 0.81 | 262 | |
Spouse | Age | 33.63 | 38.08 | 36.54 | 35.71 | 33.91 | 35.54 | 0.84 | 201 |
Female (%) | 100% | 100% | 100% | 100% | 100% | 100% | . | 201 | |
Years of schooling | 3.71 | 4.52 | 4.04 | 3.47 | 4.90 | 4.30 | 2.05* | 201 | |
Household size | 5.56 | 5.08 | 5.56 | 5.12 | 5.51 | 5.35 | 0.49 | 262 | |
Socio-economic indicators | |||||||||
Assets | Radio (%) | 70% | 78% | 83% | 78% | 84% | 80% | 0.76 | 257 |
Television (%) | 0% | 0% | 0% | 0% | 0% | 0% | . | 257 | |
Telephone (%) | 0% | 0% | 0% | 0% | 0% | 0% | . | 257 | |
Electricity (%) | 0% | 0% | 3% | 0% | 0% | 0% | 1.55 | 257 | |
Bicycle (%) | 30% | 22% | 19% | 17% | 31% | 25% | 1.15 | 257 | |
Motorcycle (%) | 0% | 0% | 0% | 0% | 2% | 1% | 0.95 | 257 | |
Car (%) | 0% | 0% | 0% | 0% | 0% | 0% | . | 257 | |
Cattle cart (%) | 0% | 22% | 8% | 5% | 16% | 13% | 3.16** | 257 | |
Livestock | Pigs (%) | 7% | 3% | 8% | 10% | 8% | 7% | 0.53 | 257 |
Ducks (%) | 0% | 11% | 0% | 2% | 11% | 7% | 2.66** | 262 | |
Chicken (%) | 56% | 56% | 25% | 59% | 50% | 50% | 3.03** | 262 | |
Cattle (%) | 0.37 | 0.84 | 0.58 | 0.44 | 0.71 | 0.65 | 0.94 | 262 | |
Housing | Beds | 1.57 | 1.73 | 1.69 | 1.48 | 1.62 | 1.63 | 0.89 | 262 |
Rooms | 1.31 | 1.58 | 1.44 | 1.44 | 1.48 | 1.47 | 0.85 | 262 | |
Floor | Dirt (%) | 96% | 98% | 100% | 100% | 98% | 98% | 0.58 | 253 |
Roof | Thatch (%) | 96% | 100% | 94% | 100% | 98% | 97% | 3.35** | 262 |
Walls | Mud/Wood (%) | 100% | 100% | 97% | 100% | 97% | 98% | 1.07 | 262 |
Water source and access | |||||||||
Access to water | Number of water trips per day | 2.56 | 2.20 | 2.36 | 2.46 | 2.20 | 2.30 | 1.35 | 257 |
Distance from water source (mins) | 12.96 | 7.25 | 7.14 | 5.98 | 11.73 | 9.18 | 16.08*** | 257 | |
Reporting that water source is far (%) | 30% | 0% | 44% | 0% | 27% | 19% | 13.96*** | 262 | |
Water source | Pump (%) | 63% | 33% | 47% | 70% | 49% | 50% | 3.42*** | 262 |
Pond (%) | 0% | 67% | 53% | 30% | 30% | 39% | 262 | ||
Stream (%) | 37% | 0% | 0% | 0% | 21% | 11% | 262 | ||
Self-reported fevers and potential malaria risk | |||||||||
HH member goes out at night (%) | 45% | 12% | 14% | 19% | 31% | 24% | 11.78*** | 262 | |
Self-reported fever in last month (%) | 8% | 19% | 9% | 18% | 8% | 12% | 4.18*** | 262 | |
Consult doctor for fever in last month (%) | 6% | 13% | 2% | 11% | 6% | 8% | 3.62*** | 262 | |
Purchase medication for fever in last month (%) | 7% | 17% | 3% | 12% | 7% | 10% | 5.23*** | 262 | |
Malaria prevention | Own mosquito net (%) | 4% | 7% | 0% | 2% | 3% | 3% | 1.25 | 262 |
Use mosquito net (%) | 0% | 2% | 0% | 0% | 1% | 1% | 0.44 | 262 |
Note: ***P < 0.01, **P < 0.05, *P < 0.1.
At baseline, 12% of households report having had at least one household member ill with a fever during the last month. In the absence of blood smears, we use history of self-reported fevers as a proxy for malaria.10 Only 3% of households owned a bednet at baseline, and 1% reported using a bednet.
Although we use random assignment to balance differences in observable and unobservable characteristics in expectation, there are differences in certain observable characteristics across treatments, likely due to the relatively small sample of villages.11 We include these baseline characteristics as covariates in the analysis to increase the precision of the coefficient of interest and reduce potential bias from these differences. We also present baseline characteristics for the sample lost to follow-up (Table 3), consistent with recommendations for randomized evaluations (Dumville et al. 2006). There are some differences between the analytic sample and the sample lost to follow-up: the analytic sample has lower years of educations for spouses and a lower probability of owning certain assets, but a higher probability of owning a radio. The analytic sample is slightly more likely to go out at night.
. | . | Free . | $0.55 . | $1.10 . | $1.65 . | $2.20 . | Average . | F-stat . | Obs . | |
---|---|---|---|---|---|---|---|---|---|---|
Demographic indicators | ||||||||||
Household head | Age | 41.03 | 50.00 | 45.47 | 43.11 | 42.65 | 0.3 | 80 | ||
Male (%) | 89% | 0% | 73% | 68% | 78% | 2.91** | 79 | |||
Years of schooling | 3.92 | 4.00 | 4.00 | 4.18 | 4.02 | 0.07 | 80 | |||
Married (%) | 87% | 0% | 67% | 65% | 75% | 2.69* | 80 | |||
Spouse | Age | 33.88 | . | 41.70 | 35.88 | 35.75 | 1.17 | 60 | ||
Female (%) | 100% | . | 100% | 100% | 100% | . | 60 | |||
Years of schooling | 5.03 | . | 3.80 | 5.06 | 4.83 | 0.83 | 60 | |||
Household size | 5.13 | 4.00 | 5.20 | 5.31 | 5.19 | 0.1 | 80 | |||
Socio-economic indicators | ||||||||||
Assets | Radio (%) | 70% | 100% | 93% | 68% | 75% | 1.32 | 68 | ||
Television (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Telephone (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Electricity (%) | 3% | 0% | 0% | 0% | 1% | 0.41 | 68 | |||
Bicycle (%) | 30% | 0% | 33% | 41% | 34% | 0.38 | 68 | |||
Motorcycle (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Car (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Cattle cart (%) | 20% | 0% | 7% | 36% | 22% | 1.71 | 68 | |||
Livestock | Pigs (%) | 7% | 0% | 13% | 0% | 6% | 0.98 | 68 | ||
Ducks (%) | 2% | 0% | 0% | 9% | 4% | 1.62 | 80 | |||
Chicken (%) | 53% | 100% | 60% | 62% | 58% | 0.5 | 80 | |||
Cattle (%) | 0.79 | 0.00 | 0.49 | 0.99 | 0.79 | 0.62 | 80 | |||
Housing | Beds | 1.69 | 1.00 | 1.73 | 1.52 | 1.63 | 0.5 | 80 | ||
Rooms | 1.49 | 1.00 | 1.51 | 1.52 | 1.50 | 0.2 | 80 | |||
Floor | Dirt (%) | 100% | 100% | 100% | 100% | 100% | . | 68 | ||
Roof | Thatch (%) | 100% | 100% | 100% | 100% | 100% | . | 80 | ||
Walls | Mud/Wood (%) | 100% | 100% | 100% | 100% | 100% | . | 80 | ||
Water source and access | ||||||||||
Access to water | Number of water trips per day | 2.57 | 2.00 | 2.60 | 2.36 | 2.50 | 0.53 | 68 | ||
Distance from water source (mins) | 7.33 | 6.00 | 5.47 | 16.36 | 9.82 | 13.28*** | 68 | |||
Reporting that water source is far (%) | 0% | 0% | 0% | 35% | 11% | 9.05*** | 80 | |||
Water source | Pump (%) | 21% | 0% | 60% | 54% | 39% | 4.03** | 80 | ||
Pond (%) | 79% | 100% | 40% | 46% | 61% | 80 | ||||
Stream (%) | 0% | 0% | 0% | 0% | 0% | 80 | ||||
Self-reported fevers and potential malaria risk | ||||||||||
HH member goes out at night (%) | 21% | 0% | 22% | 23% | 21% | 0.22 | 80 | |||
Self-reported fever in last month (%) | 13% | 0% | 26% | 5% | 13% | 3.3** | 80 | |||
Consult doctor for fever in last month (%) | 11% | 0% | 23% | 5% | 11% | 3.26** | 80 | |||
Purchase medication for fever in last month (%) | 12% | 0% | 24% | 5% | 12% | 3.00** | 80 | |||
Malaria prevention | Own mosquito net (%) | 5% | 0% | 0% | 1% | 3% | 0.78 | 80 | ||
Use mosquito net (%) | 0% | 0% | 0% | 1% | 0% | 0.84 | 80 |
. | . | Free . | $0.55 . | $1.10 . | $1.65 . | $2.20 . | Average . | F-stat . | Obs . | |
---|---|---|---|---|---|---|---|---|---|---|
Demographic indicators | ||||||||||
Household head | Age | 41.03 | 50.00 | 45.47 | 43.11 | 42.65 | 0.3 | 80 | ||
Male (%) | 89% | 0% | 73% | 68% | 78% | 2.91** | 79 | |||
Years of schooling | 3.92 | 4.00 | 4.00 | 4.18 | 4.02 | 0.07 | 80 | |||
Married (%) | 87% | 0% | 67% | 65% | 75% | 2.69* | 80 | |||
Spouse | Age | 33.88 | . | 41.70 | 35.88 | 35.75 | 1.17 | 60 | ||
Female (%) | 100% | . | 100% | 100% | 100% | . | 60 | |||
Years of schooling | 5.03 | . | 3.80 | 5.06 | 4.83 | 0.83 | 60 | |||
Household size | 5.13 | 4.00 | 5.20 | 5.31 | 5.19 | 0.1 | 80 | |||
Socio-economic indicators | ||||||||||
Assets | Radio (%) | 70% | 100% | 93% | 68% | 75% | 1.32 | 68 | ||
Television (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Telephone (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Electricity (%) | 3% | 0% | 0% | 0% | 1% | 0.41 | 68 | |||
Bicycle (%) | 30% | 0% | 33% | 41% | 34% | 0.38 | 68 | |||
Motorcycle (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Car (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Cattle cart (%) | 20% | 0% | 7% | 36% | 22% | 1.71 | 68 | |||
Livestock | Pigs (%) | 7% | 0% | 13% | 0% | 6% | 0.98 | 68 | ||
Ducks (%) | 2% | 0% | 0% | 9% | 4% | 1.62 | 80 | |||
Chicken (%) | 53% | 100% | 60% | 62% | 58% | 0.5 | 80 | |||
Cattle (%) | 0.79 | 0.00 | 0.49 | 0.99 | 0.79 | 0.62 | 80 | |||
Housing | Beds | 1.69 | 1.00 | 1.73 | 1.52 | 1.63 | 0.5 | 80 | ||
Rooms | 1.49 | 1.00 | 1.51 | 1.52 | 1.50 | 0.2 | 80 | |||
Floor | Dirt (%) | 100% | 100% | 100% | 100% | 100% | . | 68 | ||
Roof | Thatch (%) | 100% | 100% | 100% | 100% | 100% | . | 80 | ||
Walls | Mud/Wood (%) | 100% | 100% | 100% | 100% | 100% | . | 80 | ||
Water source and access | ||||||||||
Access to water | Number of water trips per day | 2.57 | 2.00 | 2.60 | 2.36 | 2.50 | 0.53 | 68 | ||
Distance from water source (mins) | 7.33 | 6.00 | 5.47 | 16.36 | 9.82 | 13.28*** | 68 | |||
Reporting that water source is far (%) | 0% | 0% | 0% | 35% | 11% | 9.05*** | 80 | |||
Water source | Pump (%) | 21% | 0% | 60% | 54% | 39% | 4.03** | 80 | ||
Pond (%) | 79% | 100% | 40% | 46% | 61% | 80 | ||||
Stream (%) | 0% | 0% | 0% | 0% | 0% | 80 | ||||
Self-reported fevers and potential malaria risk | ||||||||||
HH member goes out at night (%) | 21% | 0% | 22% | 23% | 21% | 0.22 | 80 | |||
Self-reported fever in last month (%) | 13% | 0% | 26% | 5% | 13% | 3.3** | 80 | |||
Consult doctor for fever in last month (%) | 11% | 0% | 23% | 5% | 11% | 3.26** | 80 | |||
Purchase medication for fever in last month (%) | 12% | 0% | 24% | 5% | 12% | 3.00** | 80 | |||
Malaria prevention | Own mosquito net (%) | 5% | 0% | 0% | 1% | 3% | 0.78 | 80 | ||
Use mosquito net (%) | 0% | 0% | 0% | 1% | 0% | 0.84 | 80 |
Note: ***P < 0.01, **P < 0.05, *P < 0.1.
. | . | Free . | $0.55 . | $1.10 . | $1.65 . | $2.20 . | Average . | F-stat . | Obs . | |
---|---|---|---|---|---|---|---|---|---|---|
Demographic indicators | ||||||||||
Household head | Age | 41.03 | 50.00 | 45.47 | 43.11 | 42.65 | 0.3 | 80 | ||
Male (%) | 89% | 0% | 73% | 68% | 78% | 2.91** | 79 | |||
Years of schooling | 3.92 | 4.00 | 4.00 | 4.18 | 4.02 | 0.07 | 80 | |||
Married (%) | 87% | 0% | 67% | 65% | 75% | 2.69* | 80 | |||
Spouse | Age | 33.88 | . | 41.70 | 35.88 | 35.75 | 1.17 | 60 | ||
Female (%) | 100% | . | 100% | 100% | 100% | . | 60 | |||
Years of schooling | 5.03 | . | 3.80 | 5.06 | 4.83 | 0.83 | 60 | |||
Household size | 5.13 | 4.00 | 5.20 | 5.31 | 5.19 | 0.1 | 80 | |||
Socio-economic indicators | ||||||||||
Assets | Radio (%) | 70% | 100% | 93% | 68% | 75% | 1.32 | 68 | ||
Television (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Telephone (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Electricity (%) | 3% | 0% | 0% | 0% | 1% | 0.41 | 68 | |||
Bicycle (%) | 30% | 0% | 33% | 41% | 34% | 0.38 | 68 | |||
Motorcycle (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Car (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Cattle cart (%) | 20% | 0% | 7% | 36% | 22% | 1.71 | 68 | |||
Livestock | Pigs (%) | 7% | 0% | 13% | 0% | 6% | 0.98 | 68 | ||
Ducks (%) | 2% | 0% | 0% | 9% | 4% | 1.62 | 80 | |||
Chicken (%) | 53% | 100% | 60% | 62% | 58% | 0.5 | 80 | |||
Cattle (%) | 0.79 | 0.00 | 0.49 | 0.99 | 0.79 | 0.62 | 80 | |||
Housing | Beds | 1.69 | 1.00 | 1.73 | 1.52 | 1.63 | 0.5 | 80 | ||
Rooms | 1.49 | 1.00 | 1.51 | 1.52 | 1.50 | 0.2 | 80 | |||
Floor | Dirt (%) | 100% | 100% | 100% | 100% | 100% | . | 68 | ||
Roof | Thatch (%) | 100% | 100% | 100% | 100% | 100% | . | 80 | ||
Walls | Mud/Wood (%) | 100% | 100% | 100% | 100% | 100% | . | 80 | ||
Water source and access | ||||||||||
Access to water | Number of water trips per day | 2.57 | 2.00 | 2.60 | 2.36 | 2.50 | 0.53 | 68 | ||
Distance from water source (mins) | 7.33 | 6.00 | 5.47 | 16.36 | 9.82 | 13.28*** | 68 | |||
Reporting that water source is far (%) | 0% | 0% | 0% | 35% | 11% | 9.05*** | 80 | |||
Water source | Pump (%) | 21% | 0% | 60% | 54% | 39% | 4.03** | 80 | ||
Pond (%) | 79% | 100% | 40% | 46% | 61% | 80 | ||||
Stream (%) | 0% | 0% | 0% | 0% | 0% | 80 | ||||
Self-reported fevers and potential malaria risk | ||||||||||
HH member goes out at night (%) | 21% | 0% | 22% | 23% | 21% | 0.22 | 80 | |||
Self-reported fever in last month (%) | 13% | 0% | 26% | 5% | 13% | 3.3** | 80 | |||
Consult doctor for fever in last month (%) | 11% | 0% | 23% | 5% | 11% | 3.26** | 80 | |||
Purchase medication for fever in last month (%) | 12% | 0% | 24% | 5% | 12% | 3.00** | 80 | |||
Malaria prevention | Own mosquito net (%) | 5% | 0% | 0% | 1% | 3% | 0.78 | 80 | ||
Use mosquito net (%) | 0% | 0% | 0% | 1% | 0% | 0.84 | 80 |
. | . | Free . | $0.55 . | $1.10 . | $1.65 . | $2.20 . | Average . | F-stat . | Obs . | |
---|---|---|---|---|---|---|---|---|---|---|
Demographic indicators | ||||||||||
Household head | Age | 41.03 | 50.00 | 45.47 | 43.11 | 42.65 | 0.3 | 80 | ||
Male (%) | 89% | 0% | 73% | 68% | 78% | 2.91** | 79 | |||
Years of schooling | 3.92 | 4.00 | 4.00 | 4.18 | 4.02 | 0.07 | 80 | |||
Married (%) | 87% | 0% | 67% | 65% | 75% | 2.69* | 80 | |||
Spouse | Age | 33.88 | . | 41.70 | 35.88 | 35.75 | 1.17 | 60 | ||
Female (%) | 100% | . | 100% | 100% | 100% | . | 60 | |||
Years of schooling | 5.03 | . | 3.80 | 5.06 | 4.83 | 0.83 | 60 | |||
Household size | 5.13 | 4.00 | 5.20 | 5.31 | 5.19 | 0.1 | 80 | |||
Socio-economic indicators | ||||||||||
Assets | Radio (%) | 70% | 100% | 93% | 68% | 75% | 1.32 | 68 | ||
Television (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Telephone (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Electricity (%) | 3% | 0% | 0% | 0% | 1% | 0.41 | 68 | |||
Bicycle (%) | 30% | 0% | 33% | 41% | 34% | 0.38 | 68 | |||
Motorcycle (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Car (%) | 0% | 0% | 0% | 0% | 0% | . | 68 | |||
Cattle cart (%) | 20% | 0% | 7% | 36% | 22% | 1.71 | 68 | |||
Livestock | Pigs (%) | 7% | 0% | 13% | 0% | 6% | 0.98 | 68 | ||
Ducks (%) | 2% | 0% | 0% | 9% | 4% | 1.62 | 80 | |||
Chicken (%) | 53% | 100% | 60% | 62% | 58% | 0.5 | 80 | |||
Cattle (%) | 0.79 | 0.00 | 0.49 | 0.99 | 0.79 | 0.62 | 80 | |||
Housing | Beds | 1.69 | 1.00 | 1.73 | 1.52 | 1.63 | 0.5 | 80 | ||
Rooms | 1.49 | 1.00 | 1.51 | 1.52 | 1.50 | 0.2 | 80 | |||
Floor | Dirt (%) | 100% | 100% | 100% | 100% | 100% | . | 68 | ||
Roof | Thatch (%) | 100% | 100% | 100% | 100% | 100% | . | 80 | ||
Walls | Mud/Wood (%) | 100% | 100% | 100% | 100% | 100% | . | 80 | ||
Water source and access | ||||||||||
Access to water | Number of water trips per day | 2.57 | 2.00 | 2.60 | 2.36 | 2.50 | 0.53 | 68 | ||
Distance from water source (mins) | 7.33 | 6.00 | 5.47 | 16.36 | 9.82 | 13.28*** | 68 | |||
Reporting that water source is far (%) | 0% | 0% | 0% | 35% | 11% | 9.05*** | 80 | |||
Water source | Pump (%) | 21% | 0% | 60% | 54% | 39% | 4.03** | 80 | ||
Pond (%) | 79% | 100% | 40% | 46% | 61% | 80 | ||||
Stream (%) | 0% | 0% | 0% | 0% | 0% | 80 | ||||
Self-reported fevers and potential malaria risk | ||||||||||
HH member goes out at night (%) | 21% | 0% | 22% | 23% | 21% | 0.22 | 80 | |||
Self-reported fever in last month (%) | 13% | 0% | 26% | 5% | 13% | 3.3** | 80 | |||
Consult doctor for fever in last month (%) | 11% | 0% | 23% | 5% | 11% | 3.26** | 80 | |||
Purchase medication for fever in last month (%) | 12% | 0% | 24% | 5% | 12% | 3.00** | 80 | |||
Malaria prevention | Own mosquito net (%) | 5% | 0% | 0% | 1% | 3% | 0.78 | 80 | ||
Use mosquito net (%) | 0% | 0% | 0% | 1% | 0% | 0.84 | 80 |
Note: ***P < 0.01, **P < 0.05, *P < 0.1.
Unadjusted outcome data
The unadjusted results show that 39% of households own an ITN and 29% are using an ITN post-intervention (Table 4). Conditional on owning an ITN, 73% are using an ITN, meaning 27% have unused ITNs.
Net price . | HH own . | HH use . | HH use (cond’l own) . | Unused nets (cond’l own) . |
---|---|---|---|---|
Free | 100% | 33% | 33% | 67% |
$0.55 | 72% | 67% | 91% | 9% |
$1.10 | 47% | 33% | 71% | 29% |
$1.65 | 21% | 21% | 100% | 0% |
$2.20 | 2% | 2% | 100% | 0% |
F-stat | 64.46 | 28.25 | 12.02 | 12.02 |
Prof > F | 0.00 | 0.00 | 0.00 | 0.00 |
Obs | 262 | 262 | 101 | 101 |
Average | 39% | 29% | 73% | 27% |
Net price . | HH own . | HH use . | HH use (cond’l own) . | Unused nets (cond’l own) . |
---|---|---|---|---|
Free | 100% | 33% | 33% | 67% |
$0.55 | 72% | 67% | 91% | 9% |
$1.10 | 47% | 33% | 71% | 29% |
$1.65 | 21% | 21% | 100% | 0% |
$2.20 | 2% | 2% | 100% | 0% |
F-stat | 64.46 | 28.25 | 12.02 | 12.02 |
Prof > F | 0.00 | 0.00 | 0.00 | 0.00 |
Obs | 262 | 262 | 101 | 101 |
Average | 39% | 29% | 73% | 27% |
Notes: Ownership, use and conditional use measure whether HH owned or used ITN by April based on follow-up data.
Net price . | HH own . | HH use . | HH use (cond’l own) . | Unused nets (cond’l own) . |
---|---|---|---|---|
Free | 100% | 33% | 33% | 67% |
$0.55 | 72% | 67% | 91% | 9% |
$1.10 | 47% | 33% | 71% | 29% |
$1.65 | 21% | 21% | 100% | 0% |
$2.20 | 2% | 2% | 100% | 0% |
F-stat | 64.46 | 28.25 | 12.02 | 12.02 |
Prof > F | 0.00 | 0.00 | 0.00 | 0.00 |
Obs | 262 | 262 | 101 | 101 |
Average | 39% | 29% | 73% | 27% |
Net price . | HH own . | HH use . | HH use (cond’l own) . | Unused nets (cond’l own) . |
---|---|---|---|---|
Free | 100% | 33% | 33% | 67% |
$0.55 | 72% | 67% | 91% | 9% |
$1.10 | 47% | 33% | 71% | 29% |
$1.65 | 21% | 21% | 100% | 0% |
$2.20 | 2% | 2% | 100% | 0% |
F-stat | 64.46 | 28.25 | 12.02 | 12.02 |
Prof > F | 0.00 | 0.00 | 0.00 | 0.00 |
Obs | 262 | 262 | 101 | 101 |
Average | 39% | 29% | 73% | 27% |
Notes: Ownership, use and conditional use measure whether HH owned or used ITN by April based on follow-up data.
Regression results
First, we find that price negatively affects demand for ITNs, consistent with a downward sloping demand curve (Table 5). When price increases by $0.55, demand for ITNs decreases by 24.0% points (CI 19.1–28.9; P < 0.01) (column 1). The estimated effect is consistent (23.1% point decrease), when household- and village-level baseline characteristics are included. When price increases by $0.55 from free, demand falls by 28% relative to the regression-adjusted 83% ownership in the free category. Demand is relatively inelastic at low price levels (−0.39 at $0.55) but demand becomes more elastic at higher price levels (−1.25 at $1.10 −5.02 at $1.65). We estimate a price elasticity of −0.80 at mean price and mean ownership of ITNs.
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Own ITN . | Own ITN . |
Net price | −0.240*** | −0.231*** | −0.226*** |
(0.022) | (0.016) | (0.015) | |
[−0.289 to −0.191] | [−0.266 to −0.196] | [−0.260 to −0.192] | |
Net price × fever in HH | −0.060 | ||
(0.059) | |||
[−0.189 – 0.069] | |||
Fever in HH | 0.177 | ||
(0.149) | |||
[−0.151 to 0.506] | |||
Constant | 0.965*** | 0.831*** | 0.809*** |
(0.082) | (0.123) | (0.118) | |
[0.786–1.145] | [0.561–1.101] | [0.549–1.070] | |
Households controls | X | X | |
Village level index | X | X | |
Observations | 262 | 262 | 262 |
R-squared | 0.499 | 0.583 | 0.584 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Own ITN . | Own ITN . |
Net price | −0.240*** | −0.231*** | −0.226*** |
(0.022) | (0.016) | (0.015) | |
[−0.289 to −0.191] | [−0.266 to −0.196] | [−0.260 to −0.192] | |
Net price × fever in HH | −0.060 | ||
(0.059) | |||
[−0.189 – 0.069] | |||
Fever in HH | 0.177 | ||
(0.149) | |||
[−0.151 to 0.506] | |||
Constant | 0.965*** | 0.831*** | 0.809*** |
(0.082) | (0.123) | (0.118) | |
[0.786–1.145] | [0.561–1.101] | [0.549–1.070] | |
Households controls | X | X | |
Village level index | X | X | |
Observations | 262 | 262 | 262 |
R-squared | 0.499 | 0.583 | 0.584 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01,**P < 0.05, *P < 0.1.
The regressions use a linear probability model. Each column represents a separate regression with the dependent variable indicated in the column heading. The independent variable of interest, defined as net price, represents the price category starting from zero, up to $2.20. Net price increases in increments of $0.55. Fever represents the percentage of household members that self-reported having had a fever during the previous month at baseline. This indicator is a proxy for malaria risk and the interaction is meant to isolate potential screening effects. Regressions (where marked) include baseline household level and village level characteristics (index), as well as a control variable for the education campaign and an interaction with net price. Baseline household level variables include: ITN ownership and ITN use at baseline, age and education level of household head, household size, education of spouse, ownership of cattle, chicken, ducks, and cattle cart, housing material, roof material, number of bedrooms, number of beds, whether water source is considered far, distance to water source, type of water source own, whether go out at night, whether use mosquito coil, whether sought care for fever in last month, and whether purchased medicine for fever.
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Own ITN . | Own ITN . |
Net price | −0.240*** | −0.231*** | −0.226*** |
(0.022) | (0.016) | (0.015) | |
[−0.289 to −0.191] | [−0.266 to −0.196] | [−0.260 to −0.192] | |
Net price × fever in HH | −0.060 | ||
(0.059) | |||
[−0.189 – 0.069] | |||
Fever in HH | 0.177 | ||
(0.149) | |||
[−0.151 to 0.506] | |||
Constant | 0.965*** | 0.831*** | 0.809*** |
(0.082) | (0.123) | (0.118) | |
[0.786–1.145] | [0.561–1.101] | [0.549–1.070] | |
Households controls | X | X | |
Village level index | X | X | |
Observations | 262 | 262 | 262 |
R-squared | 0.499 | 0.583 | 0.584 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Own ITN . | Own ITN . |
Net price | −0.240*** | −0.231*** | −0.226*** |
(0.022) | (0.016) | (0.015) | |
[−0.289 to −0.191] | [−0.266 to −0.196] | [−0.260 to −0.192] | |
Net price × fever in HH | −0.060 | ||
(0.059) | |||
[−0.189 – 0.069] | |||
Fever in HH | 0.177 | ||
(0.149) | |||
[−0.151 to 0.506] | |||
Constant | 0.965*** | 0.831*** | 0.809*** |
(0.082) | (0.123) | (0.118) | |
[0.786–1.145] | [0.561–1.101] | [0.549–1.070] | |
Households controls | X | X | |
Village level index | X | X | |
Observations | 262 | 262 | 262 |
R-squared | 0.499 | 0.583 | 0.584 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01,**P < 0.05, *P < 0.1.
The regressions use a linear probability model. Each column represents a separate regression with the dependent variable indicated in the column heading. The independent variable of interest, defined as net price, represents the price category starting from zero, up to $2.20. Net price increases in increments of $0.55. Fever represents the percentage of household members that self-reported having had a fever during the previous month at baseline. This indicator is a proxy for malaria risk and the interaction is meant to isolate potential screening effects. Regressions (where marked) include baseline household level and village level characteristics (index), as well as a control variable for the education campaign and an interaction with net price. Baseline household level variables include: ITN ownership and ITN use at baseline, age and education level of household head, household size, education of spouse, ownership of cattle, chicken, ducks, and cattle cart, housing material, roof material, number of bedrooms, number of beds, whether water source is considered far, distance to water source, type of water source own, whether go out at night, whether use mosquito coil, whether sought care for fever in last month, and whether purchased medicine for fever.
In the third analysis, we interact price with history of self-reported fever. Within each price category, households with more self-reported fevers in the household at baseline are not more likely to demand an ITN. It is not possible to reject that there are no screening effects. The CI shows that we can reject a screening effect >6.9% points, meaning that households with a 1% more self-reported fevers in the household at baseline have < 7% point difference in demand for ITNs.
Second, we find that price has a negative impact on ITN coverage (Table 6). When price increases by $0.55, overall ITN coverage falls by 14.7% points (CI 1.5–27.9; P < 0.05) (column 1). When household- and village-level baseline variables are included, the effect size is 23.1% points (CI 19.6–26.6; P < 0.01). Relative to the regression-adjusted mean for effective coverage of ITNs among households eligible for free nets (90%), use falls by 26% when price increases by $0.55. Effective net coverage is highest when the price of ITNs is zero, compared with $0.55. We cannot conclude that net coverage is maximized when price is zero, since the intervention does not test small nominal price differences between 0 and $0.55. A sunk cost effect could exist at very small price increases above zero, but the intervention was not set up to detect such small differences. We find that households eligible for free ITNs with more self-reported fevers in the household at baseline are more likely to use their ITNs. There is no such effect among households eligible to pay a higher price (column 3).
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Use ITN . | Use ITN . | Use ITN . |
Net price | −0.147** | −0.231*** | −0.218*** |
(0.060) | (0.016) | (0.016) | |
[−0.279 to −0.015] | [−0.266 to −0.196] | [−0.253 to −0.183] | |
Net price × fever in HH | −0.102 | ||
(0.087) | |||
[−0.295 to 0.090] | |||
Fever in HH | 0.347** | ||
(0.144) | |||
[0.030–0.663] | |||
Constant | 0.642** | 0.904*** | 0.842*** |
(0.212) | (0.172) | (0.167) | |
[0.175–1.109] | [0.525–1.282] | [0.475–1.210] | |
Households controls | X | X | |
Village level index | X | X | |
Observations | 262 | 262 | 262 |
R-squared | 0.218 | 0.435 | 0.440 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Use ITN . | Use ITN . | Use ITN . |
Net price | −0.147** | −0.231*** | −0.218*** |
(0.060) | (0.016) | (0.016) | |
[−0.279 to −0.015] | [−0.266 to −0.196] | [−0.253 to −0.183] | |
Net price × fever in HH | −0.102 | ||
(0.087) | |||
[−0.295 to 0.090] | |||
Fever in HH | 0.347** | ||
(0.144) | |||
[0.030–0.663] | |||
Constant | 0.642** | 0.904*** | 0.842*** |
(0.212) | (0.172) | (0.167) | |
[0.175–1.109] | [0.525–1.282] | [0.475–1.210] | |
Households controls | X | X | |
Village level index | X | X | |
Observations | 262 | 262 | 262 |
R-squared | 0.218 | 0.435 | 0.440 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
The regressions use a linear probability model. Each column represents a separate regression with the dependent variable indicated in the column heading. The independent variable of interest, defined as net price, represents the price category starting from zero, up to $2.20. Net price increases in increments of $0.55. Fever represents the percentage of household members that self-reported having had a fever during the previous month at baseline. This indicator is a proxy for malaria risk and the interaction is meant to isolate potential screening effects. Regressions (where marked) include baseline household level and village level characteristics (see Table 5), as well as a control variable for the education campaign and an interaction with net price.
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Use ITN . | Use ITN . | Use ITN . |
Net price | −0.147** | −0.231*** | −0.218*** |
(0.060) | (0.016) | (0.016) | |
[−0.279 to −0.015] | [−0.266 to −0.196] | [−0.253 to −0.183] | |
Net price × fever in HH | −0.102 | ||
(0.087) | |||
[−0.295 to 0.090] | |||
Fever in HH | 0.347** | ||
(0.144) | |||
[0.030–0.663] | |||
Constant | 0.642** | 0.904*** | 0.842*** |
(0.212) | (0.172) | (0.167) | |
[0.175–1.109] | [0.525–1.282] | [0.475–1.210] | |
Households controls | X | X | |
Village level index | X | X | |
Observations | 262 | 262 | 262 |
R-squared | 0.218 | 0.435 | 0.440 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Use ITN . | Use ITN . | Use ITN . |
Net price | −0.147** | −0.231*** | −0.218*** |
(0.060) | (0.016) | (0.016) | |
[−0.279 to −0.015] | [−0.266 to −0.196] | [−0.253 to −0.183] | |
Net price × fever in HH | −0.102 | ||
(0.087) | |||
[−0.295 to 0.090] | |||
Fever in HH | 0.347** | ||
(0.144) | |||
[0.030–0.663] | |||
Constant | 0.642** | 0.904*** | 0.842*** |
(0.212) | (0.172) | (0.167) | |
[0.175–1.109] | [0.525–1.282] | [0.475–1.210] | |
Households controls | X | X | |
Village level index | X | X | |
Observations | 262 | 262 | 262 |
R-squared | 0.218 | 0.435 | 0.440 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
The regressions use a linear probability model. Each column represents a separate regression with the dependent variable indicated in the column heading. The independent variable of interest, defined as net price, represents the price category starting from zero, up to $2.20. Net price increases in increments of $0.55. Fever represents the percentage of household members that self-reported having had a fever during the previous month at baseline. This indicator is a proxy for malaria risk and the interaction is meant to isolate potential screening effects. Regressions (where marked) include baseline household level and village level characteristics (see Table 5), as well as a control variable for the education campaign and an interaction with net price.
Third, we do not find evidence that price affects the use of ITNs, conditional on ownership (Table 7). As noted, these results are non-experimental, since households self-select to purchase a net. We fail to reject a sunk cost of zero, meaning we do not find that households who paid more for an ITN are more likely to use it. We can reject a sunk cost effect of 23.8% points or larger (column 3), meaning that the difference in the conditional use of nets is <24 % points when price is increased by $0.55.12 Although we fail to reject a screening effect of zero, the large CI means that meaningful screening effects may still exist.
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Conditional use of ITN . | Conditional use of ITN . | Conditional use of ITN . |
Net price | 0.174 | −0.114 | −0.114 |
(0.101) | (0.159) | (0.158) | |
[−0.059 to 0.408] | [−0.480 to 0.251] | [−0.478 to 0.250] | |
Net price × fever in HH | −0.059 | ||
(0.129) | |||
[−0.355 to 0.238] | |||
Fever in HH | −0.087 | ||
(0.174) | |||
[−0.488 to 0.314] | |||
Constant | 0.534** | 0.986* | 0.967* |
(0.215) | (0.503) | (0.499) | |
[0.040–1.029] | [−0.174 to 2.147] | [−0.185 to 2.118] | |
Households controls | X | X | |
Village level index | X | X | |
Observations | 101 | 101 | 101 |
R-squared | 0.148 | 0.660 | 0.664 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Conditional use of ITN . | Conditional use of ITN . | Conditional use of ITN . |
Net price | 0.174 | −0.114 | −0.114 |
(0.101) | (0.159) | (0.158) | |
[−0.059 to 0.408] | [−0.480 to 0.251] | [−0.478 to 0.250] | |
Net price × fever in HH | −0.059 | ||
(0.129) | |||
[−0.355 to 0.238] | |||
Fever in HH | −0.087 | ||
(0.174) | |||
[−0.488 to 0.314] | |||
Constant | 0.534** | 0.986* | 0.967* |
(0.215) | (0.503) | (0.499) | |
[0.040–1.029] | [−0.174 to 2.147] | [−0.185 to 2.118] | |
Households controls | X | X | |
Village level index | X | X | |
Observations | 101 | 101 | 101 |
R-squared | 0.148 | 0.660 | 0.664 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
The regressions use a linear probability model. Each column represents a separate regression with the dependent variable indicated in the column heading. The independent variable of interest, defined as net price, represents the price category starting from zero, up to $2.20. Net price increases in increments of $0.55. Fever represents the percentage of household members that self-reported having had a fever during the previous month at baseline. This indicator is a proxy for malaria risk and the interaction is meant to isolate potential screening effects. Regressions (where marked) include baseline household level and village level characteristics (see Table 5), as well as a control variable for the education campaign and an interaction with net price.
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Conditional use of ITN . | Conditional use of ITN . | Conditional use of ITN . |
Net price | 0.174 | −0.114 | −0.114 |
(0.101) | (0.159) | (0.158) | |
[−0.059 to 0.408] | [−0.480 to 0.251] | [−0.478 to 0.250] | |
Net price × fever in HH | −0.059 | ||
(0.129) | |||
[−0.355 to 0.238] | |||
Fever in HH | −0.087 | ||
(0.174) | |||
[−0.488 to 0.314] | |||
Constant | 0.534** | 0.986* | 0.967* |
(0.215) | (0.503) | (0.499) | |
[0.040–1.029] | [−0.174 to 2.147] | [−0.185 to 2.118] | |
Households controls | X | X | |
Village level index | X | X | |
Observations | 101 | 101 | 101 |
R-squared | 0.148 | 0.660 | 0.664 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Conditional use of ITN . | Conditional use of ITN . | Conditional use of ITN . |
Net price | 0.174 | −0.114 | −0.114 |
(0.101) | (0.159) | (0.158) | |
[−0.059 to 0.408] | [−0.480 to 0.251] | [−0.478 to 0.250] | |
Net price × fever in HH | −0.059 | ||
(0.129) | |||
[−0.355 to 0.238] | |||
Fever in HH | −0.087 | ||
(0.174) | |||
[−0.488 to 0.314] | |||
Constant | 0.534** | 0.986* | 0.967* |
(0.215) | (0.503) | (0.499) | |
[0.040–1.029] | [−0.174 to 2.147] | [−0.185 to 2.118] | |
Households controls | X | X | |
Village level index | X | X | |
Observations | 101 | 101 | 101 |
R-squared | 0.148 | 0.660 | 0.664 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
The regressions use a linear probability model. Each column represents a separate regression with the dependent variable indicated in the column heading. The independent variable of interest, defined as net price, represents the price category starting from zero, up to $2.20. Net price increases in increments of $0.55. Fever represents the percentage of household members that self-reported having had a fever during the previous month at baseline. This indicator is a proxy for malaria risk and the interaction is meant to isolate potential screening effects. Regressions (where marked) include baseline household level and village level characteristics (see Table 5), as well as a control variable for the education campaign and an interaction with net price.
Overall, the evidence shows that the reduction in effective ITN coverage as a result of price is due to the negative effect of price on demand for ITNs. There is no countervailing positive effect from price on use of ITNs, conditional on ownership. Our results fail to reject a screening effect of zero both for demand and use of ITNs for prices greater than zero. However, we do find higher overall use of ITNs among households eligible for free bednets if they have more self-reported fevers in the household at baseline.
Robustness checks
We conducted various robustness checks. First, we test different functional forms since price may not have a constant effect on demand and use. We find that being eligible to pay a positive price relative to free ITNs significantly reduces demand but does not affect ITN coverage or use conditional on ownership. The CI for effective coverage is relatively large and meaningful effects cannot be ruled out (Appendix Table A4). Providing free ITNs relative to charging any price will maximize demand but not use. When we include dummy variables for each price category, we find consistent evidence of a linear relationship between price and demand, though we cannot reject that the effect on demand is similar for $1.10 and $1.65 (Appendix Table A5). There is suggestive evidence of a non-linear relationship between price and effective coverage, since we cannot reject that use is equal when price is free vs $0.55 or $1.10. Effective coverage is significantly higher when ITNs are free compared with the social marketing price and appears to be maximized at a price close to zero but not necessarily equal to zero. There is no evidence of a sunk cost effect.
Our original specification uses a linear probability model, consistent with Angrist and Pischke (2009). When we use a probit model instead, the main results for demand and effective coverage are consistent (Appendix Table A6). The model for conditional use can only be estimated when certain baseline variables are dropped because they perfectly predict success. When we replace the village-level index with village fixed effects, the findings are similar (Appendix Table A7).
We also include probability weights to weight observations in the analytic sample up to the observations in the experimental sample since sample attrition may affect the study’s internal validity. The main results are not meaningfully affected (Appendix Table A8), though we cannot rule out that unobservable factors, not included to generate the weights, may influence attrition.
Last, since there is concern that our tests of statistical significance may over-reject the null hypothesis because of the small number of clusters, we compare the P-values with those estimated using a Wild bootstrap-t. We find that the P-value using the Wild bootstrap-t continues to be <5% for the effect of price on demand (P = 0.002) and use (P = 0.012) and becomes larger for conditional use (P = 0.597). In summary, our main results hold and demonstrate a statistically significant effect of price on demand and use of ITNs not conditional on ownership.
Discussion
Our evaluation demonstrates that increasing the price of ITNs significantly reduces both demand and effective coverage. When ITNs are free (100% subsidized), ITN coverage reaches between 49 and 90%, depending on the specification. Our findings are consistent with the other experimental results demonstrating a negative effect of price on demand and use of ITNs (Dupas 2009, 2014; Cohen and Dupas 2010; Tarozzi et al. 2014). We find that, at a price close to zero, demand is relatively inelastic, reflecting a high willingness to purchase ITNs when price is close to zero. As price increases, the elasticity of demand increases. While Cohen and Dupas estimate a price elasticity of −0.37 at mean price and mean purchase probability, we estimate a larger elasticity (−0.80) at mean price and ITN ownership. The higher relative elasticity compared with Cohen and Dupas may reflect differences in the target population; this study focuses on the general population whose elasticity of demand for bednets is likely to be different than pregnant women seeking antenatal care, a prime target group for malaria prevention. We also fail to find evidence of a screening effect or a sunk cost effect at prices greater than zero, consistent with Cohen and Dupas (2010). However, we do find that, among households eligible for free ITNs, those with more self-reported fevers in the household at baseline have higher effective coverage. This finding may be consistent with a signaling effect from providing a preventive product for free, consistent with World Bank (2015) and Hoffman (2009). Households who perceive themselves as being more at risk of malaria may be interpreting free provision as a signal of encouragement to use the ITN. The absence of a positive correlation between a history of self-reported fever and demand and use for ITNs at a positive price is consistent with literature that calls into question whether households presumed to perceive themselves more at risk of malaria are necessarily more likely to value ITNs (Agyepong et al. 1999; Sangaré et al. 2012).
Our findings complement the evidence related to other preventive health products, including deworming medications (Kremer and Miguel 2007), water chlorination (Ashraf et al. 2010; Kremer et al. 2011) and hand washing soap (Spears 2014), which identifies a steep decrease in demand when price increases. Even partial subsidization of ITNs, through social marketing programmes, significantly limits ITN ownership and use. A price of zero or close to zero may represent a special threshold (Kremer and Glennerster 2011). For example, providing free ITNs may be conveying a social norm and alter intra-household allocation of ITNs to different individuals depending on whether the household receives the ITN for free or pays for it instead (Hoffman 2009; World Bank 2015). Alternatively, free ITNs may allow individuals to experience the value of these health products. Dupas (2014) shows that individuals who receive a free net are consequently more likely to pay a positive price in the future. Providing ITNs for free does not, however, guarantee particular levels of coverage. As a result, there may be a role for paying individuals to use ITNs, i.e. using ‘negative prices’ as shown in Madagascar (Krezanoski et al. 2010), if the goal is to achieve particular target rates of ITN use.
One of this study’s strengths is its implementation among a general population of poor households susceptible to malaria. Because the recent recommendation by the WHO focuses on universal coverage for all individuals at risk of malaria, it is important to understand demand and use of ITNs among the general population. Only Dupas (2014) provides similar evidence among the general population in Western Kenya. Other studies focus on pregnant women and micro-lending clients who may have a different valuation of their health and knowledge of malaria risk. Our findings inform policy related to pricing of ITNs in social marketing programmes since we provided access to ITNs in a way similar to these programmes.
One limitation is that this study is somewhat dated and access to ITNs and malaria prevalence in Madagascar has changed. The results of a similar experiment may be different now in the context of greater access to and experience with using ITNs, particularly as households already own an ITN or are seeking to provide coverage for other household members. Nonetheless, one of the challenges with malaria control is the need to sustain malaria control efforts and replace used ITNs that no longer provide adequate protection. There remains a coverage gap of 300 million ITNs according to the WHO (2015a) and households are making choices daily about obtaining and using ITNs depending on their demand for bednets at the prevailing price. This study’s findings provide relevant information about how households in malaria-endemic regions interact with this health technology, a cornerstone of malaria control strategies.
Another limitation of this study is the relatively small sample size. First, the small number of villages leads to greater imbalance in baseline characteristics than would otherwise occur with a larger sample in a randomized experiment. Ideally, a randomized experiment would have many more randomized units per treatment arm and this would reduce the likelihood of chance differences in baseline characteristics across treatment groups. The sample size was largely determined based on budget constraints. The sample size ended up being imbalanced across treatment arms (i.e. there were more villages randomized to the no subsidy group) because it was less costly to include additional villages in that group. This imbalance would reduce our statistical power but would not affect our main findings. Although we control for baseline differences, there may exist unobservable characteristics that may bias the results. Additionally, distance to the central distribution point may vary by village. Although we restricted the sample of eligible villages to those within a five kilometer radius of Ambalavao, small variations may still exist.
Sample attrition may also affect the internal validity of the results. One potential source of bias may be related to the number of households by price category. There is no/less attrition in the two price categories with fewer randomized households, compared with other categories with more households. Specifically, the data collectors reported that it was more difficult to survey all households in villages with more households. We control for the number of households by village in the indices, yet differences in village size may not completely explain the attrition. In addition, we also observe certain differences in baseline characteristics for the analytic sample compared with the sample lost to follow-up. Our findings are consistent when we include probability weights to weight the analytic sample up to the experimental sample. However, these weights may not completely capture unobservable factors that influence sample attrition. Certainly, a larger number of randomized units would have helped mitigate concerns regarding internal validity.
We are not able to reject potentially meaningful sunk cost and screening effects because of the small sample size. In addition, we are limited in our interpretation of the screening effects since we lack an objective measure of malaria risk. Self-reported fevers may be endogenous, meaning that these self-reports are likely to be related to households’ education level and this will also influence demand and use of ITNs. In this study, we were unable to collect blood smear samples. Not all fevers are due to malaria and are therefore an imperfect proxy of malaria risk. Nonetheless, the population’s perception of malaria risk may be strongly tied to fevers because this is the main symptom that health providers used to educate households at that time about their malaria risk. Indeed, at the time, the Malagasy term tazo (fever) and tazo moka (fever from mosquitoes/malaria) were used interchangeably in the villages. Even though our estimates are less precisely estimated, they are consistent with Cohen and Dupas (2010) in demonstrating the negative effect of price on both demand and use of ITNs among a general population at risk of malaria. By finding similar results to other studies but in a different country context and in the general population, our study further corroborates that price significantly limits demand for ITNs and effective coverage of ITNs.
Conclusion
The debate around the pricing of ITNs remains relevant as countries face funding gaps in achieving universal coverage of malaria control interventions and households at risk of malaria continue to lack access to ITNs. The optimal pricing of ITNs will continue to play an important role as countries transition away from free mass distribution campaigns towards more sustainable financing models. Our results demonstrate that even partial subsidization of ITNs significantly limits demand and effective coverage of ITNs. Rather than relying on partial subsidization of ITNs to maximize coverage of ITNs among poor populations at risk of malaria, alternative sources of financing should be identified to continue to finance free distribution.
Acknowledgements
We would like to thank the field staff in Madagascar (Association Fanilo and Association Avotra) and especially Ravo Harinirina for excellent field work. We would also like to thank David Cutler, Esther Duflo, Joseph Newhouse, Cynthia Kinnan, Peter O’Hanley, David Bangsberg, David Hamer, the Journal Authors Support Group at Abt Associates and colleagues for helpful comments on this study. Finally, we would like to thank the anonymous reviewers whose comments helped to improve this article.
Funding
Funding for this study came from the Justice, Welfare and Economics Dissertation Fellowship from the Weatherhead Center for International Affairs at Harvard University, the Daniel McGillis Development and Dissemination Grant Program at Abt Associates, and private donors.
Ethical approval
This study was exempt from review by the Harvard University Committee on the Use of Human Subjects in Research (F18866-101). Ethical clearance for this study was provided by local officials in the town of Ambalavao and the Médicin Inspecteur of Ambalavao. Additionally, village chiefs provided approval for the study to take place in their villages. Data collection was conducted only among households that gave verbal consent.
Conflicts of interest statement: None declared.
Notes
Originally, bednets were not treated with insecticide but provided only barrier protection for those sleeping under them. More recently, bednets have been impregnated with insecticides which not only provide a layer of barrier protection but also kill mosquitoes coming in contact with the netting. It is this mosquito killing capacity which enables ITNs to provide a community-wide protective effect and act as vector control agents.
At the time of the study, Population Services International (the social marketing programme working in Madagascar) was selling subsidized ITNs through private retailers. The social marketing price ($2.20) represents a partially subsidized price. The retail cost of ITNs at the time of the study was around $14 in 2004 (personal observation), but non-socially marketed ITNs were only available at retailers in the regional capital, some 40km away.
This is based on an exchange rate of 1USD = 1830 Ariary, as of January 2005.
A secondary intervention consisted in offering half of the villages a monthly health education campaign. The results are not reported in this study, because the primary interest of the evaluation was the effect of price. The results for this intervention show that the education campaign had no effect on demand for ITNs. For households that were eligible for free ITNs, the education campaign decreased their use of ITNs, whereas the education campaign increased use of ITNs among households eligible to pay a positive price for ITNs. The counterintuitive result related to the education campaign and the decreased use of ITNs among households eligible for free ITNs could be the result of village characteristics because only one village was randomized into this category. Table A1 reports the treatment assignment by price and education intervention. Table A2 shows the main analyses with coefficients reported for the education intervention and its interaction term with net price.
If heads of households were absent during the baseline data collection in November, they could receive the voucher in December and complete the baseline data collection then. After December, no new households were eligible to participate, to reduce the probability of households moving to become eligible for the experiment. The vouchers were printed on paper with household individual identification numbers inscribed corresponding to the rosters kept by the study team and ITN distributor.
The full set of baseline covariates that are included in the models are: ITN ownership and ITN use at baseline, age and education level of household head, household size, education of spouse, ownership of cattle, chicken, ducks, and cattle cart, housing material, roof material, number of bedrooms, number of beds, whether water source is considered far by subjective perception, estimated distance to water source, type of water source, whether household members go out at night, whether the household uses mosquito coil, whether individuals sought care for fever in last month, and whether the household purchased medicine for fever in the last month.
It can be interpreted as the effect on demand for ITNs from a 1 standard deviation increase in the probability of a household to demand an ITN as a function of village level characteristics, holding constant the price intervention and other individual level and household level characteristics.
Following Ashraf et al. (2010), Cohen and Dupas (2010) use a randomized two-stage pricing design to separate, experimentally, the sunk cost and screening effects. Among the pregnant women willing to purchase an ITN at the randomly offered price, they are then randomly offered a price reduction. This second randomized price reduction, representing the actual price ‘paid’, captures the sunk cost effect, as opposed to the screening effect, captured by the price ‘offered’.
Cameron et al. (2008) compare rejection rates, based on Monte Carlo simulations, using robust SEs, as well as different bootstrapping approaches, including pairs cluster bootstrapping, residual cluster bootstrapping, and wild cluster bootstrap. They find that wild cluster bootstrapping for standard errors has a rejection rate of 4% for clusters of 15, compared with a rejection rate of almost 10% for both cluster robust standard errors and pairs cluster bootstrap with the same number of clusters.
The study was implemented in 2004–2005 when malaria diagnostic technology, such as rapid diagnostic test kits, were not yet available in rural health centres in Madagascar, and access to microscopy was limited.
The observables that differ across price categories include years of schooling for the head of household and the spouse, ownership of cattle carts, ducks and chickens, roof material, distance to water source, percentage saying water source is far, type of water source, individuals going out at night, households with fevers in last month, and health seeking behaviour for fevers.
Cohen and Dupas (2010) also fail to reject a sunk cost effect of zero using an experimental approach. Their CI implies that the difference in conditional use will be <10% points if price is increased by $0.15. These results are assessing smaller price increases over a smaller range.
References
Appendix 1
Net price . | Price subsidy . | Education intervention . | No education intervention . |
---|---|---|---|
Free ITN | 100% | Village 1 | Village 2 |
(10 households) | (17 households) | ||
$0.55 | 75% | Village 3 | Village 4 |
(1000 Ariary) | (81 households) | (21 households) | |
$1.10 | 50% | Village 5 | Village 6 |
(2000 Ariary) | (17 households) | (20 households) | |
$1.65 | 25% | Village 7 | Village 8 |
(3000 Ariary) | (31 households) | (27 households) | |
$2.20 | 0 | Village 9 | Village 11 |
(4000 Ariary = social marketing price) | Village 10 (53 households) | Village 12 (65 households |
Net price . | Price subsidy . | Education intervention . | No education intervention . |
---|---|---|---|
Free ITN | 100% | Village 1 | Village 2 |
(10 households) | (17 households) | ||
$0.55 | 75% | Village 3 | Village 4 |
(1000 Ariary) | (81 households) | (21 households) | |
$1.10 | 50% | Village 5 | Village 6 |
(2000 Ariary) | (17 households) | (20 households) | |
$1.65 | 25% | Village 7 | Village 8 |
(3000 Ariary) | (31 households) | (27 households) | |
$2.20 | 0 | Village 9 | Village 11 |
(4000 Ariary = social marketing price) | Village 10 (53 households) | Village 12 (65 households |
Notes: This table shows treatment assignment by price and by education intervention. Villages were randomized to either receive a group health education intervention every month or not. The figure shows households in the experimental sample.
Net price . | Price subsidy . | Education intervention . | No education intervention . |
---|---|---|---|
Free ITN | 100% | Village 1 | Village 2 |
(10 households) | (17 households) | ||
$0.55 | 75% | Village 3 | Village 4 |
(1000 Ariary) | (81 households) | (21 households) | |
$1.10 | 50% | Village 5 | Village 6 |
(2000 Ariary) | (17 households) | (20 households) | |
$1.65 | 25% | Village 7 | Village 8 |
(3000 Ariary) | (31 households) | (27 households) | |
$2.20 | 0 | Village 9 | Village 11 |
(4000 Ariary = social marketing price) | Village 10 (53 households) | Village 12 (65 households |
Net price . | Price subsidy . | Education intervention . | No education intervention . |
---|---|---|---|
Free ITN | 100% | Village 1 | Village 2 |
(10 households) | (17 households) | ||
$0.55 | 75% | Village 3 | Village 4 |
(1000 Ariary) | (81 households) | (21 households) | |
$1.10 | 50% | Village 5 | Village 6 |
(2000 Ariary) | (17 households) | (20 households) | |
$1.65 | 25% | Village 7 | Village 8 |
(3000 Ariary) | (31 households) | (27 households) | |
$2.20 | 0 | Village 9 | Village 11 |
(4000 Ariary = social marketing price) | Village 10 (53 households) | Village 12 (65 households |
Notes: This table shows treatment assignment by price and by education intervention. Villages were randomized to either receive a group health education intervention every month or not. The figure shows households in the experimental sample.
. | (1) . | (2) . | (3) . |
---|---|---|---|
VARIABLES . | Own ITN . | Use ITN . | Use ITN (cond’l own) . |
Net price | −0.229*** | −0.229*** | 0.044 |
(0.018) | (0.016) | (0.075) | |
[−0.269 to −0.188] | [−0.265 to −0.192] | [−0.129 to 0.217] | |
Education intervention (dummy) | −0.104 | −0.761*** | −0.839*** |
(0.102) | (0.118) | (0.102) | |
[−0.328 to 0.121] | [−1.021 to −0.500] | [−1.074 to − 0.604] | |
Net price × education intervention | 0.048 | 0.209*** | 0.264** |
(0.028) | (0.037) | (0.079) | |
[−0.015 to 0.110] | [0.128–0.290] | [0.081–0.447] | |
Constant | 0.802*** | 0.925*** | 0.671** |
(0.135) | (0.163) | (0.225) | |
[0.505–1.099] | [0.566–1.284] | [0.152–1.191] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.581 | 0.435 | 0.656 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
VARIABLES . | Own ITN . | Use ITN . | Use ITN (cond’l own) . |
Net price | −0.229*** | −0.229*** | 0.044 |
(0.018) | (0.016) | (0.075) | |
[−0.269 to −0.188] | [−0.265 to −0.192] | [−0.129 to 0.217] | |
Education intervention (dummy) | −0.104 | −0.761*** | −0.839*** |
(0.102) | (0.118) | (0.102) | |
[−0.328 to 0.121] | [−1.021 to −0.500] | [−1.074 to − 0.604] | |
Net price × education intervention | 0.048 | 0.209*** | 0.264** |
(0.028) | (0.037) | (0.079) | |
[−0.015 to 0.110] | [0.128–0.290] | [0.081–0.447] | |
Constant | 0.802*** | 0.925*** | 0.671** |
(0.135) | (0.163) | (0.225) | |
[0.505–1.099] | [0.566–1.284] | [0.152–1.191] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.581 | 0.435 | 0.656 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
Notes: The regressions use a linear probability model. Each column represents a separate regression with the dependent variable indicated in the column heading. Net price represents the price category starting from zero, up to $2.20. Net price increases in increments of $0.55. Education intervention is a dummy variable which takes on a value of 1 if village was eligible for health education campaign. Net price X education intervention is an interaction term to assess where there is a differential effect of the education intervention by price level. Regressions (where marked) include baseline household level and village level characteristics (see Table 5).
. | (1) . | (2) . | (3) . |
---|---|---|---|
VARIABLES . | Own ITN . | Use ITN . | Use ITN (cond’l own) . |
Net price | −0.229*** | −0.229*** | 0.044 |
(0.018) | (0.016) | (0.075) | |
[−0.269 to −0.188] | [−0.265 to −0.192] | [−0.129 to 0.217] | |
Education intervention (dummy) | −0.104 | −0.761*** | −0.839*** |
(0.102) | (0.118) | (0.102) | |
[−0.328 to 0.121] | [−1.021 to −0.500] | [−1.074 to − 0.604] | |
Net price × education intervention | 0.048 | 0.209*** | 0.264** |
(0.028) | (0.037) | (0.079) | |
[−0.015 to 0.110] | [0.128–0.290] | [0.081–0.447] | |
Constant | 0.802*** | 0.925*** | 0.671** |
(0.135) | (0.163) | (0.225) | |
[0.505–1.099] | [0.566–1.284] | [0.152–1.191] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.581 | 0.435 | 0.656 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
VARIABLES . | Own ITN . | Use ITN . | Use ITN (cond’l own) . |
Net price | −0.229*** | −0.229*** | 0.044 |
(0.018) | (0.016) | (0.075) | |
[−0.269 to −0.188] | [−0.265 to −0.192] | [−0.129 to 0.217] | |
Education intervention (dummy) | −0.104 | −0.761*** | −0.839*** |
(0.102) | (0.118) | (0.102) | |
[−0.328 to 0.121] | [−1.021 to −0.500] | [−1.074 to − 0.604] | |
Net price × education intervention | 0.048 | 0.209*** | 0.264** |
(0.028) | (0.037) | (0.079) | |
[−0.015 to 0.110] | [0.128–0.290] | [0.081–0.447] | |
Constant | 0.802*** | 0.925*** | 0.671** |
(0.135) | (0.163) | (0.225) | |
[0.505–1.099] | [0.566–1.284] | [0.152–1.191] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.581 | 0.435 | 0.656 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
Notes: The regressions use a linear probability model. Each column represents a separate regression with the dependent variable indicated in the column heading. Net price represents the price category starting from zero, up to $2.20. Net price increases in increments of $0.55. Education intervention is a dummy variable which takes on a value of 1 if village was eligible for health education campaign. Net price X education intervention is an interaction term to assess where there is a differential effect of the education intervention by price level. Regressions (where marked) include baseline household level and village level characteristics (see Table 5).
. | −1 . | −2 . | −3 . |
---|---|---|---|
. | Own ITN . | Use ITN . | Use ITN (conditional on own) . |
Households in village with fever (%) | 0.307 (0.591) | 0.279 (0.637) | −0.070 (0.372) |
Distance to water source by village (min.) | −0.032** (0.013) | −0.013 (0.009) | −0.014 (0.020) |
Village main water source (dummy for pond) | −0.138 (0.079) | −0.094 (0.071) | 0.088 (0.122) |
Age (village chief) | 0.013** (0.006) | −0.007 (0.005) | −0.033*** (0.004) |
Years of schooling (village chief) | −0.050*** (0.013) | −0.039*** (0.010) | 0.040 (0.021) |
Health educator 1 | 0.049 (0.098) | 0.164 (0.128) | 0.431*** (0.117) |
Health educator 2 | −0.034 (0.069) | −0.031 (0.064) | 0.028 (0.057) |
Health educator 3 | 0.373* (0.197) | 0.213 (0.159) | 0.269 (0.165) |
Number of households per village | 0.006* (0.003) | 0.011*** (0.003) | 0.011*** (0.003) |
Constant | 0.093 (0.209) | 0.365** (0.161) | 1.506*** (0.174) |
Observations | 262 | 262 | 101 |
R-squared | 0.397 | 0.307 | 0.461 |
. | −1 . | −2 . | −3 . |
---|---|---|---|
. | Own ITN . | Use ITN . | Use ITN (conditional on own) . |
Households in village with fever (%) | 0.307 (0.591) | 0.279 (0.637) | −0.070 (0.372) |
Distance to water source by village (min.) | −0.032** (0.013) | −0.013 (0.009) | −0.014 (0.020) |
Village main water source (dummy for pond) | −0.138 (0.079) | −0.094 (0.071) | 0.088 (0.122) |
Age (village chief) | 0.013** (0.006) | −0.007 (0.005) | −0.033*** (0.004) |
Years of schooling (village chief) | −0.050*** (0.013) | −0.039*** (0.010) | 0.040 (0.021) |
Health educator 1 | 0.049 (0.098) | 0.164 (0.128) | 0.431*** (0.117) |
Health educator 2 | −0.034 (0.069) | −0.031 (0.064) | 0.028 (0.057) |
Health educator 3 | 0.373* (0.197) | 0.213 (0.159) | 0.269 (0.165) |
Number of households per village | 0.006* (0.003) | 0.011*** (0.003) | 0.011*** (0.003) |
Constant | 0.093 (0.209) | 0.365** (0.161) | 1.506*** (0.174) |
Observations | 262 | 262 | 101 |
R-squared | 0.397 | 0.307 | 0.461 |
Robust ses in parentheses.
P < 0.01, **P < 0.05, *P < 0.1.
Notes: These regressions are used to predict the propensity variables (propensity to own ITN, propensity to use ITN, and propensity to use ITN conditional on owning) which are normalized and used as regressors in the main specification to control for village level characteristics.
. | −1 . | −2 . | −3 . |
---|---|---|---|
. | Own ITN . | Use ITN . | Use ITN (conditional on own) . |
Households in village with fever (%) | 0.307 (0.591) | 0.279 (0.637) | −0.070 (0.372) |
Distance to water source by village (min.) | −0.032** (0.013) | −0.013 (0.009) | −0.014 (0.020) |
Village main water source (dummy for pond) | −0.138 (0.079) | −0.094 (0.071) | 0.088 (0.122) |
Age (village chief) | 0.013** (0.006) | −0.007 (0.005) | −0.033*** (0.004) |
Years of schooling (village chief) | −0.050*** (0.013) | −0.039*** (0.010) | 0.040 (0.021) |
Health educator 1 | 0.049 (0.098) | 0.164 (0.128) | 0.431*** (0.117) |
Health educator 2 | −0.034 (0.069) | −0.031 (0.064) | 0.028 (0.057) |
Health educator 3 | 0.373* (0.197) | 0.213 (0.159) | 0.269 (0.165) |
Number of households per village | 0.006* (0.003) | 0.011*** (0.003) | 0.011*** (0.003) |
Constant | 0.093 (0.209) | 0.365** (0.161) | 1.506*** (0.174) |
Observations | 262 | 262 | 101 |
R-squared | 0.397 | 0.307 | 0.461 |
. | −1 . | −2 . | −3 . |
---|---|---|---|
. | Own ITN . | Use ITN . | Use ITN (conditional on own) . |
Households in village with fever (%) | 0.307 (0.591) | 0.279 (0.637) | −0.070 (0.372) |
Distance to water source by village (min.) | −0.032** (0.013) | −0.013 (0.009) | −0.014 (0.020) |
Village main water source (dummy for pond) | −0.138 (0.079) | −0.094 (0.071) | 0.088 (0.122) |
Age (village chief) | 0.013** (0.006) | −0.007 (0.005) | −0.033*** (0.004) |
Years of schooling (village chief) | −0.050*** (0.013) | −0.039*** (0.010) | 0.040 (0.021) |
Health educator 1 | 0.049 (0.098) | 0.164 (0.128) | 0.431*** (0.117) |
Health educator 2 | −0.034 (0.069) | −0.031 (0.064) | 0.028 (0.057) |
Health educator 3 | 0.373* (0.197) | 0.213 (0.159) | 0.269 (0.165) |
Number of households per village | 0.006* (0.003) | 0.011*** (0.003) | 0.011*** (0.003) |
Constant | 0.093 (0.209) | 0.365** (0.161) | 1.506*** (0.174) |
Observations | 262 | 262 | 101 |
R-squared | 0.397 | 0.307 | 0.461 |
Robust ses in parentheses.
P < 0.01, **P < 0.05, *P < 0.1.
Notes: These regressions are used to predict the propensity variables (propensity to own ITN, propensity to use ITN, and propensity to use ITN conditional on owning) which are normalized and used as regressors in the main specification to control for village level characteristics.
. | (1) . | (2) . | (3) . |
---|---|---|---|
VARIABLES . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Positive price for ITNs | −0.700*** | −0.309 | 0.641 |
(0.161) | (0.290) | (0.411) | |
[−1.054 to −0.345] | [−0.946 to 0.329] | [−0.307 to 1.589] | |
Constant | 1.086*** | 0.802* | 0.063 |
(0.188) | (0.370) | (0.575) | |
[0.672–1.500] | [−0.012 to 1.615] | [−1.263 to 1.390] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.546 | 0.238 | 0.679 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
VARIABLES . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Positive price for ITNs | −0.700*** | −0.309 | 0.641 |
(0.161) | (0.290) | (0.411) | |
[−1.054 to −0.345] | [−0.946 to 0.329] | [−0.307 to 1.589] | |
Constant | 1.086*** | 0.802* | 0.063 |
(0.188) | (0.370) | (0.575) | |
[0.672–1.500] | [−0.012 to 1.615] | [−1.263 to 1.390] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.546 | 0.238 | 0.679 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
Notes: The regressions use a linear probability model. The independent variable “Positive price for ITNs” is a dummy variable that takes on a value of 1 if household is eligible to pay a price greater than 0. Dummy variable is 0 if household is eligible for free ITN. See Table 5 for a list of all baseline variables included in the specification.
. | (1) . | (2) . | (3) . |
---|---|---|---|
VARIABLES . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Positive price for ITNs | −0.700*** | −0.309 | 0.641 |
(0.161) | (0.290) | (0.411) | |
[−1.054 to −0.345] | [−0.946 to 0.329] | [−0.307 to 1.589] | |
Constant | 1.086*** | 0.802* | 0.063 |
(0.188) | (0.370) | (0.575) | |
[0.672–1.500] | [−0.012 to 1.615] | [−1.263 to 1.390] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.546 | 0.238 | 0.679 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
VARIABLES . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Positive price for ITNs | −0.700*** | −0.309 | 0.641 |
(0.161) | (0.290) | (0.411) | |
[−1.054 to −0.345] | [−0.946 to 0.329] | [−0.307 to 1.589] | |
Constant | 1.086*** | 0.802* | 0.063 |
(0.188) | (0.370) | (0.575) | |
[0.672–1.500] | [−0.012 to 1.615] | [−1.263 to 1.390] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.546 | 0.238 | 0.679 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
Notes: The regressions use a linear probability model. The independent variable “Positive price for ITNs” is a dummy variable that takes on a value of 1 if household is eligible to pay a price greater than 0. Dummy variable is 0 if household is eligible for free ITN. See Table 5 for a list of all baseline variables included in the specification.
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Price = $0.55 | −0.433*** | 0.168* | 0.097 |
(0.084) | (0.082) | (0.516) | |
[−0.619 to −0.248] | [−0.013 to 0.350] | [−1.093 to 1.287] | |
Price = $1.10 | −0.648*** | −0.092 | −0.127 |
(0.066) | (0.073) | (0.663) | |
[−0.794 to −0.502] | [−0.253 to 0.069] | [−1.655 to 1.401] | |
Price = $1.65 | −0.623*** | −0.287*** | −0.766 |
(0.102) | (0.092) | (0.866) | |
[–0.848 − –0.397] | [–0.489 − –0.084] | [–2.764 − 1.232] | |
Price = $2.20 | −0.947*** | −0.715*** | −1.453 |
(0.074) | (0.069) | (1.120) | |
[−1.110 to −0.784] | [−0.867 to −0.563] | [−4.035 to 1.130] | |
Constant | 0.875*** | 0.494*** | 1.267 |
(0.121) | (0.141) | (0.925) | |
[0.607–1.142] | [0.184–0.805] | [−0.865 to 3.400] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.597 | 0.465 | 0.686 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Price = $0.55 | −0.433*** | 0.168* | 0.097 |
(0.084) | (0.082) | (0.516) | |
[−0.619 to −0.248] | [−0.013 to 0.350] | [−1.093 to 1.287] | |
Price = $1.10 | −0.648*** | −0.092 | −0.127 |
(0.066) | (0.073) | (0.663) | |
[−0.794 to −0.502] | [−0.253 to 0.069] | [−1.655 to 1.401] | |
Price = $1.65 | −0.623*** | −0.287*** | −0.766 |
(0.102) | (0.092) | (0.866) | |
[–0.848 − –0.397] | [–0.489 − –0.084] | [–2.764 − 1.232] | |
Price = $2.20 | −0.947*** | −0.715*** | −1.453 |
(0.074) | (0.069) | (1.120) | |
[−1.110 to −0.784] | [−0.867 to −0.563] | [−4.035 to 1.130] | |
Constant | 0.875*** | 0.494*** | 1.267 |
(0.121) | (0.141) | (0.925) | |
[0.607–1.142] | [0.184–0.805] | [−0.865 to 3.400] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.597 | 0.465 | 0.686 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
Notes: The regressions use a linear probability model. Each independent variable represents a dummy variable which equals to 1 if the household was eligible for an ITN at that price. The omitted category is for price equal to zero. See Table 5 for a list of all baseline variables included in the specification.
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Price = $0.55 | −0.433*** | 0.168* | 0.097 |
(0.084) | (0.082) | (0.516) | |
[−0.619 to −0.248] | [−0.013 to 0.350] | [−1.093 to 1.287] | |
Price = $1.10 | −0.648*** | −0.092 | −0.127 |
(0.066) | (0.073) | (0.663) | |
[−0.794 to −0.502] | [−0.253 to 0.069] | [−1.655 to 1.401] | |
Price = $1.65 | −0.623*** | −0.287*** | −0.766 |
(0.102) | (0.092) | (0.866) | |
[–0.848 − –0.397] | [–0.489 − –0.084] | [–2.764 − 1.232] | |
Price = $2.20 | −0.947*** | −0.715*** | −1.453 |
(0.074) | (0.069) | (1.120) | |
[−1.110 to −0.784] | [−0.867 to −0.563] | [−4.035 to 1.130] | |
Constant | 0.875*** | 0.494*** | 1.267 |
(0.121) | (0.141) | (0.925) | |
[0.607–1.142] | [0.184–0.805] | [−0.865 to 3.400] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.597 | 0.465 | 0.686 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Price = $0.55 | −0.433*** | 0.168* | 0.097 |
(0.084) | (0.082) | (0.516) | |
[−0.619 to −0.248] | [−0.013 to 0.350] | [−1.093 to 1.287] | |
Price = $1.10 | −0.648*** | −0.092 | −0.127 |
(0.066) | (0.073) | (0.663) | |
[−0.794 to −0.502] | [−0.253 to 0.069] | [−1.655 to 1.401] | |
Price = $1.65 | −0.623*** | −0.287*** | −0.766 |
(0.102) | (0.092) | (0.866) | |
[–0.848 − –0.397] | [–0.489 − –0.084] | [–2.764 − 1.232] | |
Price = $2.20 | −0.947*** | −0.715*** | −1.453 |
(0.074) | (0.069) | (1.120) | |
[−1.110 to −0.784] | [−0.867 to −0.563] | [−4.035 to 1.130] | |
Constant | 0.875*** | 0.494*** | 1.267 |
(0.121) | (0.141) | (0.925) | |
[0.607–1.142] | [0.184–0.805] | [−0.865 to 3.400] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.597 | 0.465 | 0.686 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
Notes: The regressions use a linear probability model. Each independent variable represents a dummy variable which equals to 1 if the household was eligible for an ITN at that price. The omitted category is for price equal to zero. See Table 5 for a list of all baseline variables included in the specification.
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Net price | −0.184*** | −0.265*** | |
(0.0246) | (0.032) | ||
[−0.232 to −0.136] | [−0.327 to −0.203] | ||
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 257 | 253 | |
R-squared | 0.558 | 0.418 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Net price | −0.184*** | −0.265*** | |
(0.0246) | (0.032) | ||
[−0.232 to −0.136] | [−0.327 to −0.203] | ||
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 257 | 253 | |
R-squared | 0.558 | 0.418 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
Notes: The regressions use probit models and the reported coefficients represents marginal effects (dy/dx). The dependent variable Net price is defined as in Table 5. See Table 5 for a list of all baseline variables included in regressions 1 and 2. In regression 3, the model cannot be estimated when the full set of baseline characteristics are included.
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Net price | −0.184*** | −0.265*** | |
(0.0246) | (0.032) | ||
[−0.232 to −0.136] | [−0.327 to −0.203] | ||
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 257 | 253 | |
R-squared | 0.558 | 0.418 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Net price | −0.184*** | −0.265*** | |
(0.0246) | (0.032) | ||
[−0.232 to −0.136] | [−0.327 to −0.203] | ||
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 257 | 253 | |
R-squared | 0.558 | 0.418 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
Notes: The regressions use probit models and the reported coefficients represents marginal effects (dy/dx). The dependent variable Net price is defined as in Table 5. See Table 5 for a list of all baseline variables included in regressions 1 and 2. In regression 3, the model cannot be estimated when the full set of baseline characteristics are included.
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Net price | −0.215*** | −0.192*** | 0.051 |
(0.030) | (0.048) | (0.086) | |
[−0.282 to −0.149] | [−0.297 to −0.086] | [−0.147 to 0.250] | |
Constant | 0.554 | 0.642 | 0.432 |
(0.373) | (0.535) | (0.370) | |
[−0.267 to 1.376] | [−0.535 to 1.820] | [−0.421 to 1.285] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.610 | 0.474 | 0.685 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Net price | −0.215*** | −0.192*** | 0.051 |
(0.030) | (0.048) | (0.086) | |
[−0.282 to −0.149] | [−0.297 to −0.086] | [−0.147 to 0.250] | |
Constant | 0.554 | 0.642 | 0.432 |
(0.373) | (0.535) | (0.370) | |
[−0.267 to 1.376] | [−0.535 to 1.820] | [−0.421 to 1.285] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.610 | 0.474 | 0.685 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
Notes: The specification is the same as the one used in the main results, including control variables for household-level variables (see Table 5), but uses village fixed effects rather than the propensity scores to control for village-level characteristics.
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Net price | −0.215*** | −0.192*** | 0.051 |
(0.030) | (0.048) | (0.086) | |
[−0.282 to −0.149] | [−0.297 to −0.086] | [−0.147 to 0.250] | |
Constant | 0.554 | 0.642 | 0.432 |
(0.373) | (0.535) | (0.370) | |
[−0.267 to 1.376] | [−0.535 to 1.820] | [−0.421 to 1.285] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.610 | 0.474 | 0.685 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Net price | −0.215*** | −0.192*** | 0.051 |
(0.030) | (0.048) | (0.086) | |
[−0.282 to −0.149] | [−0.297 to −0.086] | [−0.147 to 0.250] | |
Constant | 0.554 | 0.642 | 0.432 |
(0.373) | (0.535) | (0.370) | |
[−0.267 to 1.376] | [−0.535 to 1.820] | [−0.421 to 1.285] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.610 | 0.474 | 0.685 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
Notes: The specification is the same as the one used in the main results, including control variables for household-level variables (see Table 5), but uses village fixed effects rather than the propensity scores to control for village-level characteristics.
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Net price | −0.221*** | −0.240*** | −0.169 |
(0.016) | (0.022) | (0.152) | |
[−0.257 to −0.186] | [−0.289 to −0.191] | [−0.518 to 0.181] | |
Constant | 0.836*** | 0.970*** | 1.219** |
(0.119) | (0.149) | (0.505) | |
[0.573–1.098] | [0.642–1.297] | [0.053–2.384] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.600 | 0.490 | 0.655 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Net price | −0.221*** | −0.240*** | −0.169 |
(0.016) | (0.022) | (0.152) | |
[−0.257 to −0.186] | [−0.289 to −0.191] | [−0.518 to 0.181] | |
Constant | 0.836*** | 0.970*** | 1.219** |
(0.119) | (0.149) | (0.505) | |
[0.573–1.098] | [0.642–1.297] | [0.053–2.384] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.600 | 0.490 | 0.655 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
Notes: The specification is the same as the one used in the main results, including control variables for household-level and village-level control variables (see Table 5). Probability weights are used in the regressions to weight the sample up to the experimental sample (see Appendix section “Generating probability weights to adjust for non-response” for details on probability weights).
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Net price | −0.221*** | −0.240*** | −0.169 |
(0.016) | (0.022) | (0.152) | |
[−0.257 to −0.186] | [−0.289 to −0.191] | [−0.518 to 0.181] | |
Constant | 0.836*** | 0.970*** | 1.219** |
(0.119) | (0.149) | (0.505) | |
[0.573–1.098] | [0.642–1.297] | [0.053–2.384] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.600 | 0.490 | 0.655 |
. | (1) . | (2) . | (3) . |
---|---|---|---|
Variables . | Own ITN . | Use ITN . | Use ITN cond'l own . |
Net price | −0.221*** | −0.240*** | −0.169 |
(0.016) | (0.022) | (0.152) | |
[−0.257 to −0.186] | [−0.289 to −0.191] | [−0.518 to 0.181] | |
Constant | 0.836*** | 0.970*** | 1.219** |
(0.119) | (0.149) | (0.505) | |
[0.573–1.098] | [0.642–1.297] | [0.053–2.384] | |
Households controls | X | X | X |
Village level index | X | X | X |
Observations | 262 | 262 | 101 |
R-squared | 0.600 | 0.490 | 0.655 |
Robust SEs in parentheses. 95% CIs in square brackets.
P < 0.01, **P < 0.05, *P < 0.1.
Notes: The specification is the same as the one used in the main results, including control variables for household-level and village-level control variables (see Table 5). Probability weights are used in the regressions to weight the sample up to the experimental sample (see Appendix section “Generating probability weights to adjust for non-response” for details on probability weights).
Author notes
These authors contributed equally to this work.