Farm households' participation in rural non-farm employment in post-war Rwanda: Drivers and policy implications

Abstract

Despite the post-war government's unprecedented efforts to stimulate growth of the rural non-farm (RNF) sector in Rwanda, evidence suggests that participation in this sector remains low compared with other developing and transition economies. This study investigates the micro and meso-level factors defining farm household's capacity and incentives to participate in RNF employment in the post-war Rwanda. Based on the household's time allocation theory, this study employs household survey data collected in Gisagara District in a double-hurdle regression. The results reveal that female-headedness, labour availability, education, social networks, access to finance and rural towns increase the probability of participating in RNF activities, whereas for participating households, the time allocated to RNF activities tends to decrease with age, land productivity, distance to market and dispersed settlements. The article concludes with key implications for rural development policies such as basic education and umudugudu settlements.

Keywords:

• household model
• rural non-farm employment
• double-hurdle model
• Rwanda

1. Introduction

Rural economic transformation through livelihood incomes diversification remains a challenge for developing countries, particularly in Africa (Davis et al., 2010; Winters et al., 2010). The recent decades were marked by increasing emphasis by development researchers and policy-makers on the role of the rural non-farm (RNF) sector in spurring development at meso and micro levels. For resource-limited and agriculturally backwards regions, the thrusts in promoting RNF sectors are expected to generate multiple benefits, including absorption of the unemployed or underemployed labour force in agriculture and slowing down rural–urban migration (Clay et al., 1990; Elbers & Lanjouw, 2001; Gordon & Craig, 2001; Lanjouw & Lanjouw, 2001; Knight & Song, 2003; de Janvry et al., 2005; Zhu & Luo, 2006; Haggblade et al., 2010). At the micro level, particularly for resource-limited households, promotion of the RNF sector is a pathway to food security in the face of low farm productivity and shocks, and to serve as a risk management strategy or safety net in the face of dysfunctional credit and insurance markets (Reardon, 1997; Schneider & Gugerty, 2011).

This consideration underscores the relevance of RNF sector promotion for the development of countries with largely subsistence-oriented agricultural economies such as Rwanda. The country is characterised by galloping population growth (Wali et al., 2011) on a small arable land surface (Jayne et al., 2010). Currently, the country's primary agriculture employs around 80% of the population and contributes to the overall gross domestic product to a level of 36% (World Bank, 2011). This performance of the agricultural economy is realised in spite of low and declining farm size, coupled with issues of overreliance on highly variable rainfall, largely hilly terrains, and declining soil fertility caused by soil erosion and overexploitation of arable lands as indicated by previous studies (Clay, 1995; Byiringiro & Reardon, 1996; Clay et al., 1998), and hence the growing number of people engaged in mostly subsistence farming. These problems are exacerbated by several factors, including the tenure insecurity, the weak agricultural research base and extension system, the lack of agricultural finance services, the missing or failing agricultural markets, and the poor rural infrastructure, causing little technological progress (Reardon et al., 1995; Musahara & Huggins, 2005; Musahara, 2006; Jayne et al., 2010; Pritchard, 2013).

Analyses of labour productivity in rural Rwanda unveiled low and declining farm labour productivity, suggesting that, from a static microeconomic viewpoint, the poor have better opportunities in off-farm employment. Byiringiro & Reardon (1996) found that the marginal value product of farm labour among the poor households was below the off-farm labour market wage. They also showed that diversified households were less poor. Dabalen et al. (2004) further found that RNF employments generate higher returns compared with other off-farm activities, and farm households diversifying their income-generation strategy into the RNF sector are less likely to be poor and have higher consumption levels compared with specialised farm households. These findings imply that poor farmers should spend more of their labour time in RNF employments. This in turn, as argued by Clay et al. (1995), would increase incomes and facilitate investment in soil fertility and land improvement, and ultimately household income.

The deepening of poverty among farming communities and the low labour productivity in agriculture have led Rwandan policy-makers to design new policies and strategies aimed at reducing rural poverty. Under its Vision 2020 that sets out to transform the country from an agrarian to a knowledge-based middle-class economy (Republic of Rwanda, 2000), the country has adopted a two-pronged rural development strategy, focusing on both farm and non-farm sectors. The former focuses on increasing the performance of the agricultural sector through targeted interventions (Republic of Rwanda, 2009). On the other hand, an unprecedented attention has been given to the promotion of the RNF sector to ease the pressure on agriculture as a major employer. Policies such as the umudugudu (i.e. agglomerated) settlements (Isaksson, 2011), small-scale agro-based industry development (Republic of Rwanda, 2011), rural financial services development (Papias & Ganesan, 2010; Republic of Rwanda & DFID, 2011), and the labour-intensive local development programmes (Republic of Rwanda, 2008) underscore this strategic intent.

Theoretically, both market and policy-level incentives are expected to generate increased participation rates in the RNF sector among farm households. However, empirical evidence reveals a noteworthy discrepancy between the pre-war and post-war eras. When Clay et al. (1990) investigated the emergence of non-farm employment in Rwanda using the 1988 Nonfarm Strategy Survey, they found that close to 32% of Rwandan farm households engage in non-farm employment. However, studies conducted in the post-war era by Dabalen et al. (2004), Strode et al. (2007), and NSIR (2011), using the 2004, 2007, and 2010 Integrated Rwanda Household Living Conditions Survey (EICV), respectively, found that only around 20% of farm households are employed in the RNF sector, with a minimal rate of expansion. This rate of participation in RNF employment is among the lowest in developing and transition countries.

The authors have argued simplistically that this discrepancy could be due to the 1994 war and genocide that destroyed the social and economic base of the country. This assertion is simplistic and disregards the unprecedented efforts made by the post-war government of Rwanda to revitalise the rural economy through promotion of the RNF sector. It overlooks the potential influences that certain micro and meso-level factors exert on RNF employment participation, leaving a vacuum in the understanding of other confounding factors. This study aims at addressing this knowledge gap by primarily examining capacity factors conditioning farm households' participation in RNF employment in the post-war Rwanda at the current market and policy incentive levels.

The remainder of this article is organised into three sections. The subsequent section elicits the theoretical underpinning of the study, as well as the empirical model and data collection technique adopted. This is followed by a section discussing the results of the econometric estimation. Lastly, a concluding section outlines the major policy implications.

2. Methods

2.1 Theoretical framework

This study is based on household's non-farm labour supply theory, as developed by Huffman (1980) and Gebauer (1988). The basic concept of non-farm employment participation decisions is the trade-off between leisure (all non-work activities) and consumption goods for individuals in a farm household. Farm household members are assumed to maximise a unitary household utility function (U) (with consumption goods and leisure as arguments, and exogenous preferences structures, human capital and household's locational characteristics), subject to time, income, and farm production constraints, as follows:⁴

After simple mathematical⁵ simplifications of the constraints, the⁶ model becomes:

The corresponding⁷ Lagrangian function can be written as follows:

Solving the set of first-order conditions of Equation (6), an interior solution for the optimal allocation of time between leisure, farm and/or non-farm work is determined by:

The non-farm market-wage rate should in optimum be equal to the marginal rate of substitution of leisure and consumption. is the value of marginal productivity of labour on-farm. Optimal labour use in agriculture requires that this value is equal to the non-farm wage rate; that is, the opportunity cost of family labour. A farm household will choose to work more on farm if marginal productivity of labour on-farm is above the non-farm market wage rate (). If marginal productivity of labour in on-farm work is less than the wage rate, then the household will work more non-farm hours. This participation decision rule can be summarised as follows:

2.2 Empirical model

To estimate empirically the participation decision rule in Equation (8) as well as the intensity of participation thereof, this study adopts a double-hurdle econometric technique (Blundell et al., 1987; Matshe & Young, 2004; Moffatt, 2005; Ground & Koch, 2008; Paul et al., 2008). Under this econometric technique, a farmer has to cross two hurdles to become participant in RNF employment. A farmer may be a potential participant after crossing the “first hurdle”. Given that a farmer is a potential participant, the socio-economic scenario will lead to actual participation. This is termed the “second hurdle”. The double-hurdle model is a⁸ two-equation framework:

where:

The log-likelihood function for the double-hurdle model is:

The analysis of marginal effect helps to assess the impact of the exogenous variables on the dependent variables. To do so, the unconditional mean is decomposed into the effect on the probability of participating and the effect on the conditional level of participation and differentiating these components with respect to each explanatory variable. The unconditional mean can be written as:

The probability of participation and the work hours conditional on participation are:

and

For the discrete or categorical variables, the marginal effects are used to calculate percentage changes in the dependent variable when the variable shifts from zero to one, ceteris paribus.

In this study, the regression analysis of the decision to participate (Equation (12)) is estimated using probit regression and the second stage (Equation (13)) is calibrated using a truncated regression.

2.3 Data

The household-level data used in this study were collected in Gisagara District of the Southern Province (see Figure 1) from May to June in 2009. The purposive selection of Gisagara District was based on its diversely rich demographic, agro-climatic, and economic characteristics, which could be representative of other rural regions of the country. It is one of the districts making up the Southern Province. The district is classified as a rural district by the Rwanda National Institute of Statistics, based on population density and indicators of remoteness. Gisagara District is also one of Rwanda's smallest districts, with a surface area of 816.4 square kilometres (km²) and a population of 267,161 inhabitants. It is one of the districts with the highest population densities in Rwanda (more than 320 inhabitants per km²), although this density is not equally distributed across its administrative sectors, with some recording up to 550 inhabitants per km², reflecting the problem of acute land scarcity and environmental degradation. Around 54% of the Gisagara District population is female.

Figure 1: Map of Gisagara District

Source: District's official website (www.gisagaradistrict.gov.rw).

With an altitude ranging between 1600 and 1800 metres and a bi-modal rainfall regime, the district cuts across two agro-ecological zones: the Dorsale Granitique zone characterised by soil with sandy-clay loam texture and high gravel content; and the Plateau Central with similar soil texture but lower gravel content (Steiner, 1998). The relief is marked with slope inclinations ranging from 13 to 25%, and soil erosion defines the micro-scale soil suitability for key crops such as maize, sorghum, beans, potatoes, cassava, soy and banana (Steiner, 1998).

Although the majority of community members farm at a subsistence level, the district has some well-organised commercial farmers such as rice and coffee growers. Coffee and (to some extent) rice are the only cash crops grown in the district. Agricultural processing (coffee-washing stations and rice-processing factory) is the major agro-industry found in the district. Small-scale coltan and cassiterite mining activities are also found in the area. The tertiary sector is mainly comprised of retail commerce, transport services, construction, and artisanship.

Although the habitat of the district is predominantly dispersed settlements, the umudugudu (or conglomeration) settlements as well as rural towns are common in some sectors. The district possesses a dense road network, even though the majority is rough roads. Access to infrastructure such as electricity is limited to areas near roads, especially the highway and major rough roads. Only three out of the 13 sectors in Gisagara District have access to electricity. Commercial centres (shops, banks, businesses) are also located in areas served by country roads and the electricity.

Data were collected from a random cross-section of 241 farm households, generated using a multistage stratified sampling technique, where stratification was based on differences in area's non-farm labour market potentiality. From a review of the literature, the potentiality is portrayed by population density and proximity to rural and urban towns. In total, 498 adult members (i.e. in the official age of employment and not currently attending full-time studies) of the selected households were interviewed using a structured questionnaire. Details of the sample design can be found in Table 1.

Table 1: Sampling design

Stage	List used	Sampling method	Sample size
One	Administrative sectors(or simply sectors)	Purposive sampling	Six sectors shared equally into the following strata:
			High population density sectors (Save and Kibilizi)
			Average population density sectors (Musha and Mugombwa)
			Low population density sectors (Ndora and Gishubi)
Two	Administrative cells(or simply cells)	Purposive sampling	12 cells, i.e. two cells within each sector as follows:
			One situated within 3 km of the rural town
			One situated more than 3 km away from town
Three	Farm households	Systematic random sampling	241 farm households, i.e. 20 farm households within each cell

Source: Authors.

The dependent variable was generated using the typology shown in Figure 2. It is based on the definition of RNF employment proposed by Barrett et al. (2001); that is, all income-earning activities outside the primary agricultural sector, regardless of location or function. In most cases, a household has one distinct occupation that it considers primary, and to which more labour time is allocated, relative to other activities. Stylised questions depicting time allocation to various activities in a typical day/week (Bryant et al., 2004) were employed to measure household's labour time allocation.

Figure 2: Structural diagram for rural employment participation

Source: Authors’ typology.

A detailed description of the dependent variable in Table 2 reveals that although agriculture remains the major employer in Gisagara District, 18.5% of interviewed farm household members (25% of the households) engage in some form of non-farm employment. Low-skilled labour dominates among self-employment and wage employment. The Pearson chi-square statistic suggests that among Gisagara farmers the preference for secondary jobs tends to vary according to sex. Females prefer to undertake low-skilled non-farm activities as their secondary source of income, whereas men go for own farming as their secondary job. This shows that there is division of labour among surveyed household members.

Table 2: Sectoral distribution of individual activities by gender

	Main job		Secondary job		Total
Employment status	Female	Male	Female	Male
Agriculture – own farming	236	154	5	11	406
Agriculture – off-farm wage employment	17	6	6	4	33
Non-agricultural – self employment – low skilled	11	19	15	5	50
Non-agricultural – self employment – high skilled	2	4	3	5	14
Non-agricultural – wage employment – low skilled	7	11	4	3	25
Non-agricultural – wage employment – high skilled	2	5	3	1	11
Total	275	199	36	29	539
Pearson chi-square statistic	6.665		7.905
Significance	0.155		0.049

Source: Authors’ survey, 2009.

Table 3 presents the distribution of individuals within participating households by field and gender. It shows that when the members of farm households seek non-farm employment, it is most commonly in restaurant and accommodation services, retail commerce, and construction works. The remaining non-farm time is spent in textile and straw manufacturing, transport and education. Table 3 indicates that Gisagara women exhibit more preference for retail trade and manufacturing activities.

Table 3: Distribution of non-farm activities by sub-sector and gender

Non-farm subsector/field of work	As primary job		As secondary job		Total
	Female	Male	Female	Male
Mining and natural resources	0	0	0	1	1
Manufacturing – beverages	0	0	3	0	3
Manufacturing – food processing	0	2	2	0	4
Manufacturing – textiles	4	2	1	1	8
Manufacturing – shoes and leatherwear	1	0	0	0	1
Manufacturing – Straw	3	0	3	0	6
Manufacturing – wood	0	2	0	0	2
Manufacturing – other	0	0	0	1	1
Construction	0	13	0	2	15
Utilities	0	1	0	1	2
Commerce – wholesaling	1	3	0	0	4
Commerce – retailing	8	4	9	4	25
Services – restaurant and hotels	2	3	2	2	9
Services – transport	0	3	0	1	4
Services – Education	1	5	1	0	7
Services – administration	1	0	2	0	3
Services – domestic services	0	1	0	0	1
Services – cultural and religious(only if remunerated)	0	0	2	1	3
Services – others	1	0	0	0	1
Total	22	39	25	14	100

Source: Author's survey, 2009.

2.4 Independent variables and theoretical expectation

In the presence of fixed-cost market entry and unemployment, two groups of individuals will record zero hours of work (Blundell et al., 1987): a group consisting of those who do not want to work at their market wage; and another group consisting of those who want to work but remain unemployed. For the first group of farmers, the first hurdle reflects the proxies of fixed (time and money) costs of entry causing individuals not willing to work below a minimum number of work hours, commonly referred to as reservation hours (Cogan, 1981). The size of discontinuity depends on the individual's preference function, as defined by gender, education, labour availability and opportunity cost (Cogan, 1981).

Moreover, the first hurdle introduces local demand factors directly into the labour market model, reflecting the observable unemployment factors for individuals wishing to work (Blundell et al., 1987). Such demand variables affect P(Y_i > 0) directly through their effect on the probability of finding employment or indirectly through their effect on wage rates. The impact of these factors over time and regions provides useful information on the degree to which zero hours of work reflect unemployment as opposed to desired non-participation (Blundell et al., 1987).

In the second hurdle, observable and unobservable factors constrain the farmers' time allocation choice set. For example, if the two discrete hours' packages (part-time, full-time) are all that is available, then household labour supply decisions will reflect this and will depend on the actual discrete labour market outcomes rather than the underlying latent hours' variables (Blundell & Smith, 1994).

For this study, the variables indicating the micro-level and regional factors are presented in Table 4. These variables are drawn from the conceptual literature review by Barrett et al. (2001) and Reardon et al. (2007), and were selected based on the log-likelihood ratio test (Wooldridge, 2002). These variables are arranged in accordance with the scale level, from micro-level assets towards meso-level factors (Reardon et al., 2007).

Table 4: Description of independent variables used in the double-hurdle regression model

			Key references
Variable definition	Variable description	Indicator of …	Effect on the participation decision	Effect on RNF labour time allocation
Household-specific factors
Female-headedness	1 = if the head of the household is female; 0 = otherwise	Gender effect	Haggblade et al. (2010)	Bittman et al. (2003)
Mean age of an adult in the household	(Sum of age of adult members in the household) / total number of adults in the household	Lifecycle effect	Leinbach & Smith (1994), Bodie et al. (1992)	Leinbach & Smith (1994), Bodie et al. (1992)
Number of adult workers in the household	Number of household members above the official working age (16 years)	Labour availability	Rosenzweig & Wolpin (1985)	Rosenzweig & Wolpin (1985)
Mean years of education of an adult in the household	(Sum of years in formal educations system) / total number of adults in the household	Efficiency effect	Huffman (1980), Fafchamps & Quisumbing (1999)	Sadoulet et al. (1998), Rosenzweig (1980)
Farm-specific factors
Log of total landholding	Log of sum of plot sizes in hectares	Scale (dis)advantage	Fernandez-Cornejo et al. (2007)	Fernandez-Cornejo et al. (2007)
Log of total landholding squared	Log of sum of plot sizes in hectares squared	Marginal rate of return on the dependent variables
Plot's topographic position	1 = urubumba (mainly found at valley bottoms); 2 = mugugu (found at foot slopes and valley bottoms); 3 = umusenga (found at lower slope and foot slope); 4 = urunombe (found at lower slopes and foot slopes); 5 = urukurwe (found at upper slopes and lower slopes); 6 = ikigwagwa (found at middle slopes and upper slopes); 7 = igishonyi (found at upper slope concavities); 8 = urubuye (found at hill tops and middle slopes); and 9 = urusenyi (found at upper slopes)	Land productivity	Goodwin & Mishra (2004)	Goodwin & Mishra (2004)
Social capital factors
Density of group membership	Sum of household members’ memberships in local groups and organisations	Structural or explicit social capital through fixed transaction costs abatement	Traikova et al. (2007), Key et al. (2000), Ioannides & Loury (2004)	–
Mean level of generalised trust in a household	4 = strongly agree; 3 = agree; 2 = neither agree nor disagree; 1 = disagree; 0 = strongly disagree	Cognitive or implicit social capital through fixed transaction costs abatement	Traikova et al. (2007)	–
Conditioning variables
Local economic and demographic characteristics
Distance to the nearest microfinance organisation	Estimated kilometres between homestead and the nearest microfinance organisation	Access to finance	Khandker et al. (2013), Reardon et al. (1994)	–
Mean distance to nearest market and shopping centre	(Estimated kilometres between homestead and the nearest market + estimated kilometres between homestead and the nearest shopping centre) / 2	Market access	Jones (1984)	Jones (1984)
Access to electricity	1 = Electricity poles available in the area; 0 = otherwise	Access to infrastructure	Chowdhury (2010), Gibson & Olivia (2010), Wandschneider (2003)	Chowdhury (2010), Gibson & Olivia (2010), Wandschneider (2003)
Population density of the cell	Number of administrative cell inhabitants/km^2	Access to RNF growth centres (i.e. RNF wage)	Deichmann et al. (2009), Fafchamps & Shilpi (2003)	Deichmann et al. (2009), Fafchamps & Shilpi (2003)

Source: Authors.

Table 5 presents the key descriptive statistics and results of tests of differences in means between participants in RNF employment and non-participants. The table suggests that most participating households are headed by women. Participating households have more than two adults. Education levels among interviewed households also differ significantly between the two groups, suggesting that households with more schooled members are likely to participate in RNF employment.

Table 5: Descriptive statistics of variables used in the econometric model

		Mean		t test for equality of means
Variable definition	Unit ofmeasurement	Non-participants(n = 179)	Participants(n = 62)	t statistic	p value
Dependent variables
Participation decision (at least one individual participated in the non-farm sector)	0/1
Intensity of participation (an aggregate number of hours of non-farm work declared by each individual)	Count
Independent variables
Household-specific factors
Female-headedness	0/1	0.41	0.53	2.33	0.022
Mean age of an adult in the household	Year	38.62	36.85	–0.17	0.198
Number of adult workers in the household	Count	1.97	2.6	3.63	0.007
Mean years of education of an adult in the household	Year	2.51	4.99	6.01	0.000
Farm-specific factors
Log of total landholding	Hectare	0.28	0.42	1.35	0.134
Log of total landholding squared	Ha^2	– 4.13	– 2.55	2.06	0.034
Plot's topographic position	Categorical	3.21	3.87	1.84	0.102
Social capital factors
Density of group membership	Count	1.74	2.13	2	0.034
Mean level of generalised trust in a household	Count	3.24	2.912	–1.56	0.123
Conditioning variables
Local economic and demographic characteristics
Distance to the nearest microfinance organisation	km	5.34	3.39	–2.72	0.019
Mean distance to nearest market and shopping centre	km	5.91	3.73	–1.61	0.139
Access to electricity	0/1	0.02	0.15	1.09	0.244
Population density of the cell	Hab/km^2	369.62	540.6	7.89	0

Source: Authors’ survey, 2009.

Descriptive statistics also show significant differences in farm characteristics between participants in RNF employment and their counterparts. Participating households have significantly more non-land assets and lower agricultural income than their counterparts. Among institutional factors, there is significant difference in density of group membership between participating and non-participating households, suggesting that participants are those households with good access to social networks. There are also statistically significant differences between participants and non-participants with regard to access to microfinance institutions. For economic and demographic factors, there is a statistically significant difference with regard to population density.

3. Results and discussion

The results of the double-hurdle regression technique are presented in Table 6 and Table 7. The goodness of fit was tested using a log-likelihood ratio test (Wooldridge, 2002), and the test results indicated that the variables used in the chosen models give the best fit. The results of the examination of the correlation coefficients of the variables used in the quantitative models suggest that multicollinearity in not a serious problem. To curb potential heteroscedasticity, this study used the heteroscedasticity-robust standard errors for parameter estimates. To measure and correct the possible sample selection bias in this model, the probit model in the first stage was used to generate a sample selection correction term referred to as the inverse Mills ratio (IMR), and then used as an explanatory variable in the truncated model (Wooldridge 2002).

Table 6: Determinants of farm household participation in RNF employment: results estimated using probit regression

	Dependent variable: decision to participate in RNF employment
Independent variable	Coefficient	Marginal effect	p value
Constant	–5.311		0.003
Household-specific factors
Female-headedness	0.329	0.053	0.078
Mean age of an adult in the household	–0.018	–0.002	0.131
Number of adult workers in the household	0.849	0.132	0.000
Mean years of education of an adult household member	0.306	0.047	0.000
Farm-specific factors
Log of total landholding	–0.261	–0.028	0.746
Log of total landholding squared	0.106	–0.012	0.263
Plot's topographic position	0.064	0.009	0.375
Social capital factors
Density of group membership	0.205	0.032	0.089
Mean level of generalised trust in a household	–0.153	–0.013	0.454
Local economic and demographic characteristics (conditioning variables)
Distance to the nearest microfinance organisation	–0.023	–0.003	0.053
Mean distance to nearest market and shopping centre	–0.038	–0.006	0.224
Access to electricity	1.024	0.026	0.150
Population density of the cell	0.006	0.001	0.000

Notes: Number of observations = 241, Wald χ²(13) = 91.50, Prob > χ² = 0.001.

Source: Authors’ survey, 2009.

Table 7: Determinants of time allocation to RNF employment: results estimated using truncated regression

	Dependent variable: total labour hours spent in non-agricultural activities
Independent variable	Coefficient	p value
Constant	3 721.426	0.071
Household-specific factors
Female headedness	–352.280	0.244
Mean age of an adult in the household	–60.534	0.001
Number of adult workers in the household	198.475	0.462
Mean years of education of an adult household member	135.765	0.183
Farm-specific factors
Log of total landholding	–538.500	0.515
Log of total landholding squared	45.195	0.634
Plot's topographic position	–103.315	0.049
Local economic and demographic characteristics (Conditioning variables)
Mean distance to nearest market and shopping centre	–11.548	0.097
Access to electricity	410.324	0.112
Population density of the cell	1.105	0.000
Λ (Inverse Mill's ratio)	387.227	0.556

Notes: Number of observations = 62, Wald χ²(11) = 59.32, Prob > χ² = 0.015.

Source: Authors’ survey, 2009.

3.1 Determinants of the decision to participate in RNF employment

The results from the probit regression model (first hurdle) are presented in Table 6. Based on the marginal effects, household-specific factors turn out to be major drivers of RNF participation decisions through reduced reservation hours. These results show that female-headedness adds significantly 5.3% chance of engaging in RNF jobs, ceteris paribus, implying that farm households that become female-headed reduce their reservation hours. These results can be explained by the congruence of childcare responsibilities and other social expectations that limit women's mobility, forcing them into home-based non-farm works (Haggblade et al., 2010). Although these results corroborate empirical findings of studies in neighbouring countries such as Uganda (Canagarajah et al., 2001) and Kenya (Lay et al., 2008), the peculiarity of the Rwanda case is that the exclusion from land inheritance on females imposed by the pre-war constitution could have forced those who become head of households to seek ways of earning a livelihood outside agriculture.

The results also show the number of adult workers living in a farm household is a strong predictor of household participation in the RNF sector. One extra adult adds 13.2% probability of engaging in RNF activities, suggesting a sort of intra-household division of labour created particularly when the available manpower exceeds on-farm labour requirements. As Rosenzweig & Wolpin (1985) explain, family extension facilitates occupational diversification and thus serves to reduce income risk. This in turn reduces farm household's reservation hours in the RNF labour markets.

As expected, education also turns out to be a significant socio-economic capacity driver of the decision to take up RNF works. This suggests that households with educated people are likely to possess the skills needed to participate in the RNF sector either as self-employed entrepreneurs or as wage workers, such as the ability to manage a business, process-relevant information, adapt to changing demand patterns, and liaise with public and private service providers (Wandschneider, 2003). Although the response of the RNF labour supply to education levels in farm households is lower due to the fact that education and experience increases managerial efficiency in both farm and non-farm activities (Rosenzweig, 1980), Sadoulet et al. (1998) demonstrate that only unskilled labour would respond positively to shadow the farm wage rise. Therefore, education increases farmers' confidence and aspiration towards working off their farms (Reardon, 1997). This finding corroborates the evidence that the returns to education are higher in the off-farm labour market than on the farm (Fafchamps & Quisumbing, 1999).

With regard to social capital, the regression results indicates that, all other factors remaining constant, joining one more community organisation group increases significantly the probability of participating in RNF employment by 3.2%. This finding suggests that positive participation decisions are constrained by the congruence of fixed transaction costs, as explained by Traikova et al. (2007). Fixed transaction costs (mainly pertaining to imperfect information) force farm households to defer their market participation decisions (Key et al., 2000). Information networks therefore play a critical role in job search (e.g. through referrals) (Ioannides & Loury, 2004). This finding implies that farm households who have access and claims to a wider social networks are more likely to take advantage of their informational supports to reduce the extent of transaction costs they face in the RNF labour markets (Fafchamps & Minten, 1998; Zhang & Li, 2003; Traikova et al., 2007).

On local economic and demographic characteristics, the results show that the distance to the nearest microfinance organisation negatively predicts the decision to engage in RNF activities. These results suggest that longer distances to financial organisations constrain farmers' access to financial resources needed to draw labour out of the farms to become RNF entrepreneurs (Deininger et al., 2007; Papias & Ganesan, 2010). As Khandker et al. (2013) demonstrate, inadequate access to finance constitutes a disincentive to the RNF enterprise through reduced profitability.

The results also indicate that the population density significantly affects the likelihood of engaging in RNF activities, suggesting that, on one hand, high population density may put pressure on agricultural land available to a household, creating the impetus to seek employment in RNF sector; or, on the other hand, high population may create a derived demand for non-farm goods and services that increases the scope of jobs and business opportunities in the non-farm sector (Reardon et al., 1998). This finding corroborates the evidence that high returns to RNF activities are more concentrated around urban agglomerations (Deichmann et al., 2009). These meso-level factors therefore define the extent of unemployment among farm households willing to work in the RNF sector.

3.2 Determinants of the intensity of participation in RNF employment

For the truncated regression model, the self-selection bias is corrected for by generating the IMR for each household using the probit model. The IMR is then included as an explanatory variable in the subsequent truncated regression. Since both hurdles do not have equal vectors of explanatory variables due to the exclusion of three variables (group membership, trust, and distance to microfinance organisation) in the intensity model, the IMR cannot be correlated with the vector of explanatory variables in the truncated regression model (Wooldridge, 2002). The results are presented in Table 7. The coefficient of the IMR turns out to be statistically insignificant in the truncated model, indicating that self-selection was not an issue.

On household-specific characteristics, the regression results show that the mean age of a household member affects negatively the time allocated in the non-farm sector, suggesting that, given positive participation decision, time allocation choice sets reduces with the mean age of an adult in the household. As Leinbach & Smith's (1994) model (based on the Chayanovian model) demonstrates, younger households with higher consumer-to-labour ratio tend to participate more in off-farm works due to higher demand for family labour and labour shortage, and this trend falls as the household moves to later stages of the demographic lifecycle. This result supports the view that labour flexibility and risk-taking decrease over the lifecycle (Bodie et al. 1992).

Unexpectedly, the sign of the slope position turns out to be negative and significant, suggesting that the time allocated to RNF activities decreases with increasing scope of soil erosion uphill. A plausible explanation is that, although the erosion process is highly pronounced at the convex slopes uphill, these parts of the landscape generally remain relatively undeveloped and more productive than the concavities downhill often characterised by deeply weathered parent materials (Steiner, 1998). Therefore, the results imply that higher implicit farm wages uphill tend to lower RNF employment efforts (Goodwin & Mishra, 2004).

With regard to regional factors, the results indicate that the distance to the nearest market and shopping centres reduces the amount of household labour time allocated to the RNF activities. This result reflects a von Thünen-like model, suggesting that even though the demand for RNF activities could be more pronounced in remote areas, the effect of (proportional) transaction costs such as transport cost outweigh the positive RNF response to reduced land values in remote areas (Jones, 1984), creating underemployment in this sector.

Finally, the results show that population density positively influences the time allocated to the RNF activities. This finding suggests that conglomerated settlement increases the time allocation choice sets available to farm households willing to participate in RNF activities, thereby reducing the extent of underemployment in the RNF sector. This finding corroborates the empirical evidence showing that RNF employment declines quickly as one leaves urban areas and moves into the hinterland (Fafchamps & Shilpi, 2003).

4. Conclusion

From the analysis of the results reported in the previous section, one can conclude that the ability to take advantages or respond to the RNF employment opportunities brought about by the market and policy incentives in post-war Rwanda has been confounded by differential access to livelihood assets (human, social, natural, and financial assets) coupled with regional economic and demographic circumstances. The double-hurdle econometric estimation results suggest that female-headedness, labour availability, education and social networks reduce the RNF work reservation hours, thereby increasing the probability of positive participation decision. They also suggest that age and land productivity reduce farmers' time allocation choice sets. Furthermore, the positive influence of regional factors such as access to finance and population density on the participation decision implies a certain scope of underemployment among farm households willing to work in the RNF sector, whereas the effect of market access and population on the labour supply reflects the scope of underemployment.

These empirical results raise several issues pertaining to farm households' integration into the RNF sector, if poor households are to take full advantage of expanding economic opportunities offered by economic growth in post-war Rwanda. The first challenge revolves around human capital formation in rural areas of Rwanda. The finding that human capital conditions the ability to participate in RNF employment calls for increased investment of resources in village schools and other vocational and adult educations. The Government of Rwanda and its development partners need to allocate more efforts in raising education funds for the most vulnerable strata of Rwandan society such as women and the poor. Therefore, the results of this study strongly support the free primary education policy launched in 2004. Drawing parallels with the recommendations of Yabiku & Schlabach (2009), the present study recommends that this policy should not only emphasise enrolment, but also completion.

The second challenge lies within the balancing of sectoral development strategies. The finding that farm-specific characteristics do influence the farmers' income diversification strategy implies that policy-makers should recognise the complementarities of agriculture and non-agricultural activities in the development of rural economies and hence in sustaining livelihoods in rural areas of Rwanda. Focusing on both sectors (i.e. farm and non-farm) has the advantage of generating forward–backward production linkages and rural–urban linkages that can promote simultaneous development of the two sectors.

The third challenge relates to development of local institutions and organisations. The finding that participation in local organisation such as farmers' groups, microfinance, and so forth, increases farm household's likelihood of participation in RNF employment implies that local institutions and organisation have a significant say in rural poverty alleviation through access to non-farm employment resources and markets. This calls on the necessity to devise a social capital development strategy relevant to war aftermaths, a strategy that should focus more on bridging than on bonding ties to ensure that beneficial organisations thrive around them. Furthermore, the government will need to empower local organisation through direct actions (such as capacity-building, leadership training) and/or indirect actions (such as creating a democratic, open and corruption-free environment).

The fourth challenge resides in stimulating services and infrastructure in rural settlement schemes and rural areas in general. The findings of this study support the microfinance revolution underway in Rwanda which extends the financial services to historically underserved areas such as Umurenge (administrative sector) Savings and Credit Cooperatives. The finding that access to the market allows greater integration of farm households into the RNF sector strongly supports the local economic development approach based on collaborative efforts by the public sector, private entrepreneurs and the local communities in the identification and facilitation of local competitive edges for locally based, self-reliant markets.

Lastly, the finding that rural towns promote RNF employment opportunities and social networks determine the ability to take up such opportunities strongly supports the ongoing agglomeration (umudugudu) settlements policy in Rwanda. This strategy evolved from a post-genocide resettlement programme consisting of closely spaced dwellings meant for genocide survivors and returnees, which became compulsory relocation and villagisation policy, aiming at community welfare and social cohesion through social networking (Bruce, 2007).

Attention should therefore be given to these micro and meso factors during the design and implementation of the strategies aiming at rural economic transformation in post-war Rwanda.

Acknowledgement

The authors are grateful to the African Economic Research Consortium (AERC) for the award of the scholarship grant which made this study possible.

Notes

⁴Definitions: C, consumption goods; D, leisure; X_M, quantity of market goods; X_C, quantity of home-grown goods; T, total time available; H_F, on-farm work hours; H_N, non-farm work hours; L, leisure; E, exogenous preferences structures; H, human capital; G, household's locational characteristics; W_F, farm wage; W_N, non-farm wage; V, unearned income; P_Q, the anticipated price of farm outputs; P_Z, price of non-labour variable inputs; P_M, price of marketed goods; Z, quantity of non-labour variable inputs; W_F, on-farm wage rate; L, hired labour hours; L_D, labour demand; Q, farm output; K, farm specific characteristics.

⁵ and .

⁶ L_D = H_F + L.

⁷Assuming competitiveness in inputs and output markets, and were simplified into P.

⁸The diagonality of the covariance matrix shows that the two error terms are assumed to be independently distributed; is the binary choice variable in Equation (8); Z is a vector of capacity factors explaining the decision to participate in RNF employment, whereas α represents their respective influences; the second forumla of Equation (9) explains the factors affecting the extent of participation ( is the latent variable that reflects farmer's time allocation in RNF employment), X, and β being the vector of factor explaining the intensity of participation and their influences respectively; and the observed variable, Y_i, is determined as .