1. Introduction
Poverty reduction programs in developing countries are typically not concerned with their environmental impacts. Conservation programs in these settings, however, often seek to achieve ‘win-wins’ for environment and development, believing these twin goals to be inextricably linked (Wunder, Reference Wunder2001; Angelsen and Atmadja, Reference Angelsen, Atmadja and Angelsen2008). Yet poverty-environment relationships in remote rural areas are complex and theoretically ambiguous (Reardon and Vosti, Reference Reardon and Vosti1995; Scherr, Reference Scherr2000; Wunder, Reference Wunder2001). For example, numerous studies indicate that the poor often cope with income and consumption shocks by increasing their use of natural resources (e.g., Pattanayak and Sills, Reference Pattanayak and Sills2001; Takasaki et al., Reference Takasaki, Barham and Coomes2004; McSweeney, Reference McSweeney2005; Debela et al., Reference Debela, Shively, Angelsen and Wik2012) and there is some limited evidence that positive income shocks can reduce such reliance (Fisher and Shively, Reference Fisher and Shively2005). But reducing poverty may not benefit biodiversity conservation in all cases – it likely depends on whether environmental products, such as wild foods, are inferior or normal goods in particular communities. Similarly, impacts on land use may also be variable: reducing poverty may increase households’ ability to expand their farms – thereby increasing pressure on ecosystems – or it may expand households’ ability to participate in markets for off-farm labor, thus decreasing such pressure.
Poverty-environment relationships should also differ depending on whether one takes a short-run or long-run view, though the directional change over time is not clear. On the one hand, the environmental Kuznets curve theory posits that societies exploit natural resources to escape poverty and at a certain inflection point of wealth then begin demanding environmental amenities that flow from conserved resources (Koop and Tole, Reference Koop and Tole1999). On the other hand, the environmental impacts of reducing poverty can have important feedback effects for poverty itself: Barbier (Reference Barbier2010) argues that the drive to raise incomes through agriculture and resource exploitation can lead to ‘poverty-environment traps’ if markets for land, off-farm labor and credit are incomplete. And if reducing poverty increases consumption of wild game, over-exploitation can lead to the collapse of populations that may be an important source of protein for the (future) rural poor.
This theoretical ambiguity may be clarified by more nuanced consideration of heterogeneous access to markets when examining poverty-environment relationships. Research by Ferraro et al. (Reference Ferraro, Hanauer and Sims2011) has shown that the ability of conservation programs to achieve poverty and environment ‘win-wins’ is moderated by market access, though Alix-Garcia et al. (Reference Alix-Garcia, McIntosh, Sims and Welch2013) also note that we are of course only able to observe the full environmental impacts of human behavior where poor transportation networks effectively localize them.
We explore these themes by examining the short-run impacts of a poverty alleviation program that extends unconditional cash transfers to households in rural Zambia. This study is the first to examine the environmental impacts of a cash transfer program in Africa. We investigate impacts on households’ consumption of fuelwood, charcoal and bushmeat, as well as their use of land for farming. These are the main pathways through which rural livelihoods affect natural resources in Zambia. The principal livelihood strategy for most households in rural Zambia is small-scale farming. Bushmeat is an important source of protein in rural areas. Rural households rely on fuelwood for energy and 82 per cent of urban households in Zambia use charcoal (Tembo et al., Reference Tembo, Mulenga and Sitko2015). Charcoal production is one of the main drivers of deforestation and degradation in Zambia (Mulenga et al., Reference Mulenga, Hadunka and Richardson2017). Mulenga et al. (Reference Mulenga, Richardson, Tembo and Mapemba2014) report that about 6 per cent of farming households in Zambia earn income from the sale of non-timber forest products and these sales account for 35 per cent of these households’ cash income. Seventy per cent of these households sell fuelwood.
We also examine an element of the causal chain hypothesized by Barbier (Reference Barbier2010) to be important for avoiding poverty-environment traps: program impacts on off-farm business enterprises. Particular attention is paid to investigating whether impacts vary according to market access. We hypothesize that distance to market significantly affects household decision-making regarding conversion of the transfer into consumption and production.
2. Cash transfers, transaction costs and impact heterogeneity: theory and evidence
Agricultural households in subsistence economies clearly live in a world characterized by multiple market imperfections. Lack of opportunities for wage labor and access to markets for cash and loans create liquidity and credit constraints, limiting purchase of productive agricultural inputs. High transaction costs associated with selling crops also hinder specialization and commercial activity – as does the high uncertainty regarding the purchase price of food, which further encourages self-sufficiency in food production rather than commercial agriculture. All of these points and the importance of considering non-separability when examining cash transfer programs in rural areas have been raised by Handa et al. (Reference Handa, Davis, Stampini and Winters2010) and Boone et al. (Reference Boone, Covarrubias, Davis and Winters2013).
But the focus in the literature thus far has been on considering agricultural households’ shadow value of time (e.g., Handa et al., Reference Handa, Davis, Stampini and Winters2010) and how cash transfers can help increase agricultural production by relaxing farmers’ liquidity and credit constraints (e.g., Boone et al., Reference Boone, Covarrubias, Davis and Winters2013). And while the concepts of non-separability and price bands created by transaction costs are generally accepted in the literature, examination of how variable transaction costs moderate households’ production and consumption decisions is not routine. And yet, as Lofgren and Robinson (Reference Lofgren and Robinson1999: 1) note:
The existence of such non-separability indicates the presence of market imperfections or failures that may have important policy implications. For example, depending on the nature of the market imperfections, there may be ‘threshold’ effects whereby policy changes have no effect on household behavior until the change is ‘large’ in some measure. In this environment, policy analysis assuming the existence of perfect markets may badly misstate the impact of policy changes on producer behavior and household welfare.
De Janvry and Sadoulet (Reference de Janvry, Sadoulet, de Janvry and Kanbur2003) also call for more examination of how transaction costs affect rural household behavior. Consideration of how distance to markets affects households’ production decisions is receiving increasing attention in the conservation planning and evaluation literature (e.g., Joppa and Pfaff, Reference Joppa and Pfaff2009; Ferraro et al., Reference Ferraro, Hanauer and Sims2011), but attention to this topic has received comparatively less attention in the development economics and cash transfer literatures. One exception is a study by Alix-Garcia et al. (Reference Alix-Garcia, McIntosh, Sims and Welch2013), which examines the ecological impacts of Mexico's Oportunidades program.
In addition to commenting on the importance of understanding market linkages when implementing poverty alleviation programs in rural areas, this study aims to contribute evidence on the heterogeneous impacts of cash transfer programs. Few studies have explored impact heterogeneity in the context of cash transfers and those that do focus on cases from Latin America and health and schooling outcomes. For example, Handa et al. (Reference Handa, Davis, Stampini and Winters2010) test for heterogeneous impacts of Progresa in Mexico and whether use of schooling and health services differs between agricultural and non-agricultural households. Dammert (Reference Dammert2009) examines heterogeneous impacts of Nicaragua's Red de Proteccion Social program and finds that impacts on schooling and child labor differ according to the age and gender of the child, gender of the household head and degree of poverty in the community. Galiani and McEwan (Reference Galiani and McEwan2013) look at variation in impacts of the Honduran Programa de Asignacion Familiar and find that the program's positive impacts on children's schooling and labor activity are much larger for the poorest of the poor.
In order to improve the design and success of public policies, we often need to understand more than just their average impacts and compare impacts for different sub-populations. For example, if the impacts of cash transfers are more muted in remote communities, then policymakers should consider how to simultaneously address both supply-side and demand-side factors driving poverty (or perhaps just focus on supply-side factors). As noted by Handa and Davis (Reference Handa and Davis2006), cash transfers implicitly assume that the poor consume less schooling and health services due to constraints on their demand – and not that issues of access and quality might pose constraints on the supply-side. Similarly, in remote rural areas, supply-side factors may also constrain basic consumption of food and non-food items as well as agricultural production. Rawlings and Rubio (Reference Rawlings and Rubio2005) and Handa and Davis (Reference Handa and Davis2006) argue that uncovering heterogeneous impacts and any supply-side constraints are some of the most pressing questions facing the next phase of cash transfer research.
3. The Zambian child grant programme
The Zambian Child Grant Programme is an unconditional cash transfer program being implemented by Zambia's Ministry of Community Development, Mother and Child Health. The program aims to reduce extreme poverty and the intergenerational transmission of poverty to children. The only eligibility criterion for the program is that households have a child under the age of five. Enrolled households receive 60 kwacha (about US$12) per month, which is estimated to be the cost of purchasing one meal per day for an average-sized household for a month. Payments are picked up on a bimonthly basis from a local paypoint manager.
Program implementation began in 2010, in three districts with the highest rates of child mortality and malnutrition in Zambia: Kalabo, Kaputa and Shong'ombo. These districts are extremely remote, situated more than two days car travel from the country's capital, Lusaka, at Zambia's frontiers with Angola and the Democratic Republic of Congo.
4. Conceptual framework
Following Singh et al. (Reference Singh, Squire and Strauss1986), Sills et al. (Reference Sills, Lele, Holmes, Pattanayak, Sills and Abt2003), and Barbier (Reference Barbier2010), we use household production theory to present a model adapted to the specific characteristics of the agricultural household in rural developing economies. In such models, time is the primary input to production and the household consumes most of its own production. From this household production framework, we derive equations for estimating household consumption and production. We assume households maximize utility, which depends on consumption of three goods: those purchased in the market (X), those produced at home (H) and leisure (L). Households consume these goods, conditioned on their preferences (Φ), subject to four constraints: an agricultural and environmental goods production function (the technological constraint), the time constraint, an earned income constraint and a cash constraint.
The objective function and constraints are defined as follows:
Own production of agricultural and environmental goods (A) is a function of time inputs (T A), purchased inputs (X A) and the quality of natural resources, including plot fertility (F A), which is exogenously fixed. Time (T) is composed of leisure time, time in production of agricultural and environmental goods (T A) and time spent in the market on wage labor (T M). Production of environmental goods includes fuelwood harvesting, bushmeat hunting and charcoal production. Earned income (E) is the multiple of any market time and the wage (w). The cash constraint dictates that expenditures on market goods and agricultural inputs must be less than or equal to the sum of the marketed surplus from home production [p A(A−H)], earned income and exogenous sources of income (C), such as the cash transfer.
The time, earned income and cash constraints can be combined into a full income constraint such that the Lagrangian can be written as:
The choice variables are consumption of market goods (X), consumption of own produced goods (H), leisure time (L), time in own production (T A), time in market T M, own production (A) and agricultural inputs (X A).
Solving this constrained optimization problem yields first-order conditions which reveal that households equate marginal costs with marginal benefits when making consumption and production decisions. The shadow values measuring how binding the technological (λ 1) and full income constraints (λ 2) are play key roles in the household's choice of optimal bundles. These shadow values are specific to each household and, like the six choice variables identified in the Lagrangian, are thus endogenous and a function of all exogenous variables in the system.
Where the shadow price for a consumption-production good (functions of the constraint's shadow value and the good's marginal utility) equals the market price for the good, household decisions can be viewed as separable (Lofgren and Robinson, Reference Lofgren and Robinson1999). The household first maximizes producer profits and then maximizes its consumption utility according to this income. In such cases, markets can be viewed as complete and prices and income are key determinants of household production and consumption in line with standard theory (Sills et al., Reference Sills, Lele, Holmes, Pattanayak, Sills and Abt2003). But where the shadow and market prices differ, household production and consumption decisions are said to be non-separable. As Lofgren and Robinson (Reference Lofgren and Robinson1999: 2) state, this non-separability exists ‘…whenever the household shadow price of at least one producer-consumer good is not given exogenously by the market but instead is determined endogenously by the interaction between household demand and supply’.
Non-separability arises whenever markets are incomplete (Sills et al., Reference Sills, Lele, Holmes, Pattanayak, Sills and Abt2003). Lofgren and Robinson (Reference Lofgren and Robinson1999) note that farm households in developing economies typically face these market imperfections due to the following circumstances: (1) the market purchased good is not a perfect substitute for the home produced good, (2) the household is not a price-taker, and (3) there are gaps between the sales price and purchase price of a good. Sills et al. (Reference Sills, Lele, Holmes, Pattanayak, Sills and Abt2003) describe how these ‘price bands’ for goods are caused by variable transaction costs facing households, which are influenced both by exogenous sources of market integration (e.g., distance) and endogenous sources, such as connections to traders. These variable transaction costs imply that markets for consumption and production goods may be complete for some but incomplete for others. These insights motivate our adoption of a conceptual framework that explicitly tests how constraints on household's market participation affect their consumption and production decisions. Specifically, we use distance to market as a proxy for transaction costs and thus market imperfections faced by households, and posit that the effect of the cash transfer will vary accordingly, as suggested by the theory.
5. Data and descriptive statistics
Zambia's Child Grant Programme is being rolled out in phases, enabling the program to first conduct a randomized impact evaluation of the pilot phase before scaling up. The evaluation employs two levels of randomization. Thirty communitiesFootnote 1 from each of three districts were randomly assigned to either treatment or control status. All eligible households within treatment communities were then enrolled in the program and, of these households, 28 were randomly selected to participate in the study. Baseline surveys were administered to 2,515 households living in 90 communities (45 control, 45 treatment) just before program implementation began in 2010. The second round of data was collected in 2012.
There were 221 households that migrated out of the study area after the collection of baseline data (table 1). Handa et al. (Reference Handa, Seidenfeld, Davis and Tembo2014) examine the effect this out-migration had on the sample and find no differential attrition between the control and treatment groups in terms of out-migration rates or their observable household characteristics. These authors also investigate whether out-migration caused those that remain in the sample to be, on average, different from the overall baseline sample. They find that the sample stays generally the same over time, in terms of observable household characteristics, but those who remained in the sample were less likely to have experienced a weather shock at baseline. Seventy-two per cent of the households that left the study lived in Kaputa, where Lake Mweru Wantipa – important for fishing and farming livelihoods – is drying up, causing mass migration out of the area.
Study sample sizes
Note: 221 households migrated out of the sample.
We collected in-depth information on households’ use of natural resources and establishment of non-farm business enterprises. In the case of fuel and food, we asked households the amounts of resources consumed within the previous 4 weeks (charcoal and firewood) or 2 weeks (bushmeat); specifically, the amounts purchased at market, received as gifts, and produced/collected themselves. We also asked households the price of these amounts (or what it would have been had it been purchased) and aggregated these values into one consumption value (expressed in kwacha) for each resource. We asked households about their agricultural production in the previous season (between October 2011 and September 2012), including the area of land used. In 2012, households also reported whether they owned a non-farm business. For consumption outcomes (fuelwood, bushmeat and farmland), we use per capita amounts to adjust for the household size. (Importantly, analysis of the Child Grant Programme's fertility impacts found that the cash transfer is not incentivizing households to have more children (Palermo et al., Reference Palermo, Handa, Peterman, Principe and Seidenfeld2016), although transfers could create incentives for family relatives to join and expand the household, as was seen in the South Africa pension program (Duflo, Reference Duflo2003).
Households in the sample are quite poor, with 92 per cent living below the poverty line of 93.37 kwacha per person per month and 90 per cent ranked as severely food insecure (table 2). On average, households live between 16 and 22 kilometers (km) from a food market, 32–36 km from a national road, and 69–72 km from a paved road. At baseline, roughly 90 per cent of households consumed firewood, 5 per cent charcoal, and only 2 per cent bushmeat. Households with bushmeat consumption tend to have higher food consumption and be less food insecure than the panel average. We also see that both charcoal and bushmeat consumers are more likely to live in Kaputa. Bushmeat consumption is driven by purchases (87 per cent) whereas firewood and charcoal consumption are driven by household's own production (98 per cent and 73 per cent, respectively).
Mean characteristics and tests for equivalence between control and treatment groups at 2010 baseline
Notes: All samples restricted to those who remain in the panel survey in 2012. Means and tests for significant difference are regression-adjusted to account for clustered randomized design. Monthly per capita food consumption and food security regressions include controls for recipient characteristics (age, education, marital status), household characteristics (wealth, household size, and demographic composition), district fixed effects, and a vector of baselines prices (maize/grain, rice, beans, fish, oil, sugar, salt, hand soap, liquid soap).
**Indicates significant differences between the treatment and control groups at the 95% level and * at the 90% level.
To investigate whether there are heterogeneous impacts of the cash transfer due to a household's distance to market, we first explored the data graphically using Lowess-smoothed plots to see if there appear to be differential effects of cash that vary according to market distance (see figures A1–A11 in the online appendix). These graphs were plotted using the market distance data contained in our survey, which is kilometers to the nearest food market, as reported by households. The Lowess-smoothed graphs clearly show that, for the treatment group, impacts bifurcate around the 10 km mark for all of our outcomes of interest. Intuitively, the 10 km point makes sense when one considers that humans walk, on average, 5 km an hour, and so a one day round-trip to a market more than 10 km away implies more than 4 h of walking in one day. Households farther from markets may make different consumption and production decisions because they have less access to market goods; their economic behavior may also differ because the quantity (e.g., fuelwood, bushmeat) or quality (e.g., soil fertility) of natural resources may be greater farther from markets.
Because the graphs show a clear bifurcation around the 10 km mark, this indicates that there are threshold effects which may not be picked up by measuring the marginal effects of each kilometer using a continuous measure of distance. Therefore, we use the 10 km cutoff to split the sample into two sub-groups and use this binary variable in our triple-difference models to test for heterogeneous impacts of cash. Because household reports of kilometers to food markets may suffer from respondent and enumerator error and also (as one reviewer noted) yield household-reported estimates that cluster around increments of 5 km, we use two additional objective measures of distance to market that are based on geospatial analysis. Using road network data from the Government of Zambia's National Spatial Data Infrastructure road map, we calculate each household's distance to both the nearest (1) national road (paved or unpaved) and (2) paved road.Footnote 2 Histograms of these three measures of distance show correspondence between the left-skewed distributions of the household-reported distance to food market measure and the distance to national road measure, with the distance to paved road measure having a more even distribution across the sample (see figures A12–A14 in the online appendix).
The 10 km food market distance cutoff splits the sample into 1,201 households living more than 10 km from a market and 1,097 living within 10 km from a market (tables A1 and A2, online appendix). For those living far from markets, average distances range from 20 to 39 km across the natural resource/control-treatment sub-groups. For the sub-groups living close to markets, average distances range from only 2–4 km. Comparing the two market distance sub-samples, we see that the recipient and demographic characteristics tend to be similar. However, somewhat surprisingly, households living closer to markets tend to be poorer, with higher food insecurity and lower food consumption – though they have higher wealth scores (i.e., more assets).
The vast majority of households in the sample – 89 per cent – farmed land in 2012 (see table A3 in the online appendix). Note that we lack baseline data on agricultural production and non-farm businesses and therefore report descriptive statistics for the 2012 data. Maize, cassava, and rice are the most common crops in the sample, followed by millet, groundpeas, and sorghum. These small-holders farmed, on average, less than one hectare (ha) each. The largest plots measure between 10 and 12 hectares.
Households were asked to name up to three non-farm businesses that they own (table A4, online appendix). Of the 885 households (39 per cent of the sample) that own a non-farm business, 73 (8 per cent) own more than one business. We examine whether or not these businesses are based on exploitation of natural resources, which in this sample includes fishing, charcoal production and haying. Thirteen per cent of households own a business based on natural resources while 27 per cent own other types of businesses. The most important non-farm businesses represented in the sample are fishing (12 per cent of households), home brewery (10 per cent) and petty trader (6 per cent). Only 2 per cent of households produce and sell charcoal.
6. Estimation strategy
Because non-separable models of household consumption and production decisions are functions of exogenous household preferences and characteristics, they can be estimated using a reduced form approach (de Janvry and Sadoulet, Reference de Janvry, Sadoulet, de Janvry and Kanbur2003). To estimate the average impacts of cash on natural resource use we run a series of difference-in-difference models, which compare the temporal change in the treatment group with the temporal change in the control group. This nets out the effect of any general time trend not associated with the cash transfer on natural resource use in the Kaputa, Kalabo and Shang'ombo districts.
The key assumptions of our difference-in-difference models are that: (1) natural resource use is balanced between the control and treatment groups at baseline, and (2) the control and treatment groups would experience the same general time trend with respect to natural resource use in the absence of the cash transfer program. We test that the first assumption holds and while the second assumption is fundamentally unknowable, our research design provides strong assurance that it holds as well. Because the cash transfer program was randomly assigned within and across three districts, treatment status should not be systematically correlated with observed or unobserved characteristics of participating households or communities that vary over time or are time-invariant.
We lack baseline data on land use and non-farm business enterprises and therefore run a series of first-difference models to examine the impact of cash on these outcomes. These models measure the difference between the control and treatment groups in 2012, and therefore assume baseline equivalence between these groups regarding land use and business enterprises.
To examine whether there are heterogeneous impacts of the cash transfer due to distance to market, we run a series of triple-difference estimates for the fuelwood and bushmeat models and a series of difference-in-difference models for the farming and non-farm business models. We consider all three measures of distance for each series.
We consider two measures of natural resource use: whether households used firewood, charcoal, bushmeat or farmland at all; and the amount of the given resource used amongst households consuming it at baseline. Examination of these two trends separately allows us to explore whether the cash transfer is having strong income effects that induce changes in households’ livelihood strategies (i.e., moving into or out of farming), dietary patterns, or encourage fuel-switching. It also provides a means of dealing with the high frequency of zeros (i.e., non-users) in the firewood, charcoal and bushmeat data when examining the transfer's impact on overall amounts used.
6.1 Testing assumptions of the impact estimates’ econometric models
We confirm that randomization yielded similar observable characteristics between treatment and control households by testing for their equivalence at baseline. We test for equivalence at baseline in terms of basic characteristics of the recipient and household and our key outcomes of interest (natural resource use) and report these results in table 2 and tables A1–A5 in the online appendix. We restrict our analysis to just the panel of households that remained in the survey for both rounds and cluster robust standard errors at the community-level (and do so for all subsequent models). Equivalence at baseline tests are run for all variations of the sample used in our average impact estimates.
We find charcoal, fuelwood and bushmeat use to be well balanced between the control and treatment groups at baseline (table A5). On average, amongst consumers, fuelwood consumption is roughly equivalent to 14 per cent the value of their food consumption. For charcoal it is about 12 per cent. Bushmeat accounts for roughly 12 per cent of bushmeat consumers’ monthly food budget. Similar frequencies and amounts of natural resource use at baseline across the sub-samples suggest remoteness is not associated with different fuelwood, charcoal or bushmeat consumption patterns in these regions of Zambia.
Control and treatment households are generally equivalent along their observable characteristics at baseline, although there are important differences amongst those with baseline charcoal consumption (n = 124) and baseline bushmeat consumption (n = 46) (table 2). For charcoal users, control households are farther from markets and a greater percentage lives below the poverty line while a higher percentage lives in Shang'ombo. Amongst bushmeat consumers, the treatment group has a significantly higher percentage living below the poverty line, as well as slightly different household size and demographic composition. However, when the 10 km food market distance cutoff is used to divide the full panel into two sub-samples, any significant differences between the control and treatment groups at baseline disappear (tables A1–A2, online appendix).
We also consider whether distance to market is picking up the effect of another key variable that might affect livelihood strategies and responses to the cash transfer. In these rural regions of Zambia, education may be such a variable and it could be correlated with distance to market. We test whether those living more than 10 km from markets have, on average, different levels of education than those living closer to markets. We examine whether the recipient ever attended school or not is correlated with each of our three measures of distance. While we find no correlation between distance to food market and education, we do find that those living more than 10 km from a paved road or any national road are 10–11 percentage points less likely to have ever attended school (table A6, online appendix). We conduct additional robustness checks to ensure that our market distance measures are not confounded with educational attainment and discuss these results in section 8.
6.2 Identification strategy for impact estimates
The difference-in-difference model we use to identify the average impacts of cash on natural resource use can be specified as follows:
where Y igt measures whether household i in district g in period t used the natural resource in question or the per capita amount used, Postigt is a dummy variable equal to 1 if the observation is in 2012, Cashig is a dummy variable equal to 1 if the household is in the treatment group, Xig represents a vector of household and recipient characteristics measured at baseline, Zg is a vector of baseline prices for food and other important consumption goods, Wg is a district fixed effect, and E igt is the error term. We include controls for baseline characteristics and prices and district fixed effects to increase the precision of our estimates. The coefficient of interest in this model is B 3, which captures the effect of being in a treatment community on natural resource use.
Because we lack baseline data on land use and non-farm businesses, we run a series of first difference models using the 2012 data to test for the impact of cash. This model is similar to equation (1) and is written as:
here, the treatment effect is captured by B 1.
Next, we test for impact heterogeneity by running a series of triple-difference models on our fuelwood and bushmeat outcomes and a series of difference-in-difference models on our land use and non-farm business outcomes. The difference-in-difference models follow the specification laid out in equation (1), although now the Post variable is replaced with each of our three distance measures and the heterogeneous treatment effect is given by the coefficient on the interaction of Distance and Cash. The triple-difference models are specified as:
where Distance ig is a dummy variable equal to 1 if the observation is more than 10 km from a food market/national road/paved road at baseline, and the coefficient on the triple interaction B 7 identifies a heterogeneous treatment effect, if any. The coefficient on the triple interaction can be interpreted as the effect of cash on natural resource use for those far from markets relative to those close to markets. The model works by estimating the time trend in consumption for those far from markets in the cash group and differencing out the time trend for those far from markets in the control group as well as the time trend for those close to markets in the cash group.Footnote 3
7. Results
We do not find evidence that the cash transfer program significantly affects consumption of firewood or bushmeat, on average.Footnote 4 Cash does, however, significantly increase the decision to use charcoal, on average, by 8 percentage points (table 3). However, these average impacts obscure the heterogeneous effects of distance to market. We find that cash increases the likelihood of consuming bushmeat by 7 percentage points for those living more than 10 km from a food market relative to those close to food markets, and increases the likelihood of firewood consumption by 15 percentage points for those far from roads relative to those close to roads (table 4). For charcoal, we find that cash increases its consumption by 716 per cent for those living close to paved roads, as compared to those living far from paved roads (table 4). Given that between 68 and 73 per cent of charcoal users (at baseline) live in Kaputa (table 2) and 25 of the 36 charcoal businesses are located there, we also investigate the impacts of cash on charcoal use in Kaputa district. We find that cash increases the likelihood of charcoal use in Kaputa by 24 percentage points (table A10, online appendix).
Average impacts of cash on fuelwood, bushmeat and land for farming
Notes: Sample restricted to those who remain in the panel survey in 2012. Robust standard errors are clustered at the community level to account for the clustered randomized design and included in parentheses below coefficients.
***Indicates significant differences at the 99% level and * at the 90% level. Coefficients for recipient, household, and community characteristics as well as baselines prices are presented in tables A7–A9 in the online appendix.
Heterogeneous impacts of cash on fuelwood and bushmeat, according to market distance: triple-difference models
Notes: Sample restricted to those who remain in the panel survey in 2012. Robust standard errors are clustered at the community level to account for the clustered randomized design and included in parentheses below coefficients.
***Indicates significant differences at the 99% level and * at the 90% level. Coefficients for recipient, household, and community characteristics as well as baselines prices are presented in tables A7–A9 in the online appendix.
We also find that impacts of cash on both the decision to farm and the total area of land used vary according to market access. While we do not detect an impact of cash on the decision to farm amongst the full panel (table 3), cash significantly increases the likelihood of farming by six percentage points amongst those living more than 10 km from food markets versus those living close (tables 5 and 6). In terms of land area used, cash increases the area farmed by 27 per cent on average (tables 3, 5, and 6), but we only see evidence of differential treatment effects with the paved road model, with those more than 10 km from a paved road increasing the size of the farms by 34 per cent more than those living close to paved roads in response to cash.
Heterogeneous impacts of cash on land use and non-farm businesses, according to market distance: difference-in-difference models
Notes: Sample restricted to those who remain in the panel survey in 2012. Robust standard errors are clustered at the community level to account for the clustered randomized design and included in parentheses below coefficients.
***Indicates significant differences at the 99% level, ** at the 95% level, and * at the 90% level. Coefficients for recipient, household and community characteristics as well as baselines prices are presented in tables A7–A9 in the online appendix.
Impacts of cash on farming: average impacts and heterogeneous impacts of distance to markets
Notes: Sample restricted to those who remain in the panel survey in 2012. Robust standard errors are clustered at the community level to account for the clustered randomized design and included in parentheses below coefficients.
***Indicates significant differences at the 99% level, ** at the 95% level, and * at the 90% level. Coefficients for a vector of baseline prices (maize/grain, rice, beans, fish, oil, sugar, salt, hand soap, liquid soap) and household age composition controls not shown. Kalabo District omitted.
Cash significantly increases the likelihood of owning a non-farm business by 17 percentage points, on average (table 7) and these impacts are most pronounced for those living far from a paved road (tables 5 and 7). For households living more than 10 km from a paved road, cash increases the likelihood of owning a non-farm business by 16 percentage points more than for those living within 10 km of a paved road.
Impacts of cash on non-farm business ownership and farming amongst business owners: average impacts and heterogeneous impacts of distance to markets
Notes: Sample restricted to those who remain in the panel survey in 2012. Robust standard errors are clustered at the community level to account for the clustered randomized design and included in parentheses below coefficients.
***Indicates significant differences at the 99% level, ** at the 95% level, and * at the 90% level. Coefficients for a vector of baseline prices (maize/grain, rice, beans, fish, oil, sugar, salt, hand soap, liquid soap) and household age composition controls not shown. Kalabo District omitted.
Because establishment of non-farm enterprises could represent a shift away from farming, we also investigate if cash has differential impacts on land use amongst those with a non-farm business. We find that, on average, cash increases the area farmed amongst business owners by 13 per cent (table 7). Again, we see differential impacts depending on proximity to paved roads: For non-farm business owners, living far from paved roads and receiving cash is associated with an increase in area farmed that is 36 per cent higher than the treatment effect for those living close to markets.
8. Robustness checks
As a reviewer noted, being far from a market not only means that producers and consumers face higher transaction costs, but also likely means that individuals have lower levels of education (due to either less access to schools or different norms regarding the utility of formal schooling). Lower levels of education could also possibly drive the behavioral outcomes we examine in this study and thus confound the impacts we attribute to market distance. Educational attainment may affect responses to the cash transfer, particularly for farming, as those with higher levels of education may be able to use the cash transfer to transition out of farming. More educated households may have more information about non-farming livelihood options or they may simply have more non-farming options as a function of their educational qualifications. They may also know more about, or be more likely to adopt, environmentally-friendly livelihood practices, such as alternative fuels or land-intensive farming. If cash is constraining their ability to pursue these other options, we may see differential impacts of the cash transfer on natural resources with respect to education. Indeed, in the environmental quality literature we see that more educated households have higher demand for clean water, and that extending information about environmentally-friendly farming practices and water purification can significantly change behavior and improve public health in developing countries (Somanathan, Reference Somanathan2010). And experimental studies show that educational levels significantly increase the likelihood of adopting such practices (see, for example, Jalan and Somanathan, Reference Jalan and Somanathan2008).
As noted in section 6.1, educational level is correlated with distance to a paved road as well as any road, with those more than 10 km from roads less likely to have attended school. Because it may be possible that lower levels of education are what causes those far from markets to make greater investments in their farms than those close to markets, we run two additional sets of models on all of our outcome variables to further investigate the impact of education and ensure that our market distance impact estimates do not suffer from omitted variable bias. We first run a set of models that investigates the heterogeneous impacts of cash with respect to educational attainment. These models follow the construction of our distance impact heterogeneity models, using triple-difference models for the outcomes measured at two points in time and difference-in-difference models for those measured only in the post-transfer period. The triple-difference specification of these models is:
where B 7 is the coefficient of interest and tells us whether there are differential impacts of the cash transfer based on the recipient's educational level. We then run a second set of models on all of our outcome measures that examines the heterogeneous impacts of cash and distance while also controlling for the interaction of cash and education. The specification of these models is similar to equation (4), but with our original distance interactions now included, and is written as:
The primary coefficient of interest in this model set is given by B 11, which tells us the heterogeneous impact of distance while controlling for the heterogeneous impact of education, which is given by B 8. The first set of robustness check models allows us to compare the heterogeneous impacts of education with our distance impact heterogeneity results presented in tables 4–7. The second set of robustness check models allows us to see if the results presented in tables 4–7 suffer from omitted variable bias by not controlling for cash-education interactions.
For our robustness checks we considered two measures of education: whether the transfer recipient ever attended school (a binary measure) as well as their highest grade completed (a continuous measure). While the former measure is the variable used as a control in our original impact estimates, highest grade completed might better capture how variation in educational attainment affects behavioral responses to the cash transfer. We ran all robustness checks using both measures of education and found the results to be similar, though we only detect educational impact heterogeneity in the first set of robustness checks (i.e., without distance controls) with the continuous measure of education. While the highest grade completed variable is missing for 18 households in our dataset, this is a small number and this continuous variable provides a more nuanced measure for the construct of education by allowing for marginal effects. It is thus our preferred measure of educational attainment for the robustness check models and we discuss these results below.
On average, transfer recipients in our sample have attended school through the 4th grade, although the years of schooling completed range from 0 to 17 (table A11, online appendix). Twenty-eight per cent have not ever attended school – the highest frequency response, followed by having completed seven years of school (i.e., finished primary school) at 16 per cent. While education does not appear to affect impacts of the cash transfer on consumption of fuelwood or bushmeat (table A12, online appendix), we find that each additional year of education is associated with the transfer decreasing the area farmed by 3 per cent amongst all farmers and by 5 per cent amongst non-farm business owners (table A13, online appendix). Given that these are marginal effects, they are fairly large impacts. In comparing these results to the distance impact heterogeneity results presented in tables 5–7, we see that while distance to paved roads strongly increases impacts of the cash transfer on farm expansion, education works in the opposite direction and in fact causes cash recipients to decrease the size of their farms. If those that live more than 10 km from paved roads have lower levels of education than those that live closer, it is possible that the distance variable is merely picking up the effect of education on behavioral responses to the cash transfer.
Our next set of robustness checks tests this possibility by again examining the heterogenous impacts of cash according to market/road distance, this time while controlling for cash-education impact heterogeneity (tables A14 and A15, online appendix). Here we see that our distance impact heterogeneity results (tables 4 and 5) are robust to the inclusion of cash-education interactions for all of our distance measures (distance to food market, any road or paved road), with only some slight changes to the size of the significant coefficients. However, we also detect impact heterogeneity for education's effect on farming, while controlling for the differential effects of distance. Each additional year of education is again associated with cash decreasing the area farmed, by between 2 and 5 per cent (table A15). Looking across all of the results in tables A12–A15, we see that the sign on education is consistently negative. Taken together, these robustness checks indicate that while cash transfers are causing those living far from paved roads to expand their farms and this effect is not just due to differences in educational levels, education itself is also an important factor affecting households’ farming response to the transfers. Whether close or far from markets, those with more education respond to the cash transfer by decreasing the size of their farms.
9. Discussion and conclusions
Our findings provide further evidence of the complexity of poverty-environment relationships. We find that, on average, cash increases the likelihood of using charcoal and owning a non-farm business as well as the land area used for farming. Even amongst households owning a non-farm business, cash, on average, increases the area farmed. However, these average impacts mask heterogeneity in resource use, moderated by households’ distance to market.
We find strong evidence that cash is dramatically increasing the consumption of charcoal amongst charcoal users living close to paved roads. Since charcoal is more common in urban areas and is a good that must be purchased, it is likely that the cash transfer is facilitating an energy transition from fuelwood to charcoal for those households with access to charcoal markets and a preference for charcoal. Far from roads, we see that cash significantly increases the likelihood of firewood consumption beyond what we see close to markets, indicating that those too poor to consume firewood are now able to do so. The effect may be more pronounced farther from markets because there is less forest degradation and thus greater availability of firewood and/or because those closer to markets not consuming any fuelwood at baseline choose to consume charcoal instead of firewood, although we do not find that cash significantly increases the likelihood of charcoal consumption closer to markets.
There is weaker evidence that cash increases the likelihood of bushmeat consumption far from food markets but not close to food markets. Interestingly we do not see a similar pattern of results for our other two measures of distance, indicating that being far from a food market is a more important determinant of bushmeat consumption than being far from roads.
On the decision to farm, cash only has an impact for those households living far from food markets, though the effect is weak and the magnitude of the impact small. The strongest impacts identified in this study are for the paved road models for farm size and non-farm business ownership. Full results for these models are presented in tables 6 and 7. We see that for households far from paved roads, cash increases the size of farms and the likelihood of owning a non-farm business by substantially more than what we see close to paved roads. We also see that non-farm business owners living close to markets do not use the cash to increase the area of their farms, while those living far from markets do.
Taken together, these results show that in this particular region of Zambia, the biggest impacts of a cash transfer program on natural resources – in the short run – come from small-holder farming. Our findings also demonstrate that proximity to paved roads is an important moderator of the transfer's effects. In more remote areas where households lack access to non-farm employment, they are more likely to invest the cash transfer in expanding their farms but also more likely to start non-farm businesses – and to invest profits from non-farm businesses in farm expansion rather than transition out of farming. We may see greater establishment of non-farm businesses farther from paved roads because there are simply more market opportunities in remote areas, whereas closer to paved roads markets for small businesses may be more saturated. The fact that cash transfers help diversify livelihood strategies but do not seem to facilitate a transition away from farming – but rather an expansion of farming – has important implications for land use and rural livelihoods. However it is important to stress that these are short-term impacts; over time it is still possible that the livelihood diversification that cash transfers are promoting could result in some transitioning out of farming.
Further research could investigate the specific types of land that are being used for this farming expansion. Households could be expanding their farms onto lands to which they hold customary rights but which had previously been left fallow, or transferring farmland amongst rightsholders in the village, or expanding farms into open-access woodland areas. These various pathways clearly have different implications for sustainable development.
We observe the strongest evidence of heterogeneous impacts for the paved road models. This is likely because access to a paved road, rather than any road or a small village food market, is what really matters for access to robust markets. And while remoteness is associated with farm expansion, we find that education works in the opposite direction, causing households to decrease the size of their farms in response to the cash transfer. Further research should investigate the causal pathways behind this impact and examine how cash transfers are enabling this livelihood transition. It may be that more educated households are using the transfer to diversify their livelihoods and spend more resources (time and money) on non-farm activities, either because they have more information about other options or because they have more options because they have more qualifications than less-educated households. For example, infusions of cash may allow more educated households to forgo farming time and income and instead travel for paid work, invest resources in their own non-farm businesses, or even pursue additional education. Or it may be that more educated households are still invested in farming, but are now able to use the infusions of cash to pursue more costly farming practices (requiring more or different inputs) that are more land-intensive rather than extensive (i.e., have a greater output per land area and thus use less land).
What are the implications of these findings for policy and programming targeting rural environment and development issues? First, cash transfers enable households to diversify their livelihood strategies, but this does not necessarily imply, at least in the short term, that they will reduce their land use for farming. Second, cash transfers may also cause households with access to charcoal markets to increase their consumption of charcoal. Given that charcoal is the number one driver of deforestation in Zambia (Day et al., Reference Day, Gumbo, Moombe, Wijaya and Sunderland2014) and many other regions of Africa, development programs should consider pairing cash transfers with alternative fuel programs and heightened attention to charcoal supply chains. Third, because households living far from markets will likely use cash to expand their farms, cash transfer programs may wish to consider increasing agricultural extension services in these areas to encourage land-intensive practices and safe use of fertilizers and pesticides. Finally, the role of education in reducing households’ land use pressure should not be overlooked.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1355770X19000305
Acknowledgements
This paper analyzes data collected for the impact evaluation of the Zambian Child Grant Programme, which is being implemented by the American Institutes for Research and the University of North Carolina at Chapel Hill under contract to UNICEF-Zambia. The evaluation is commissioned by the Government of Zambia's Ministry of Community Development, Mother and Child Health, with support from DFID, Irish Aid, and UNICEF-Zambia. Other organizations that are part of the evaluation team include FAO and Palm Associates. The members of the evaluation team, listed by affiliation and then alphabetically within affiliation are: American Institutes of Research (Juan Bonilla, Cassandra Jesse, Leah Prencipe, David Seidenfeld); FAO (Benjamin Davis, Josh Dewbre, Silvio Diadone, Mario Gonzalez-Flores); UNICEF-Zambia (Charlotte Harland Scott, Paul Quarles van Ufford); Government of Zambia (Vandras Luywa, Stanfield Michelo); DFID-Zambia (Kelley Toole); Palm Associates (Alefa Banda, Liseteli Ndiyoi, Gelson Tembo, Nathan Tembo); UNICEF Office of Research (Sudhanshu Handa); University of North Carolina at Chapel Hill (Sudhanshu Handa). We are grateful to all members of the Zambia Cash Transfer Evaluation Team. This paper has also benefitted from very helpful comments from Erin Sills, Pete Andrews, Dick Bilsborrow, and two anonymous reviewers. Luke Aitken provided excellent research assistance on geospatial analysis. UNICEF-Zambia plans to make the data used in this study publicly available after the overall evaluation of the programme is officially complete.
