Change Font: A A A A Contact Us What's New FAQs Subscribe home
HomePublicationsSpatial Effect in the Efficient Access of Rural DevelopmentTheoretical Framework

Theoretical Framework

The dynamics in a typical rural community are an irony between simplicity in rural life and the complexity of the economic system that is operating. The literature offers diverse theories and perspectives in trying to explain the rural economy. There seems to be a cycle over the years among these theories, postulated, reinvented, reformulated, refuted in some cases, and emerging again in recent literature. Lewis (1984) postulated that in the rural economy, growth is triggered by the initiation of trade. Farmers are producing not just for consumption but also for the demand in other communities. This is a valid assumption once productivity had surpassed the threshold for local needs. Otherwise, if the production level is still below the threshold, marginalization and subsequent exposure to vulnerability will dominate the rural production with growth hardly manifesting if not remaining impossible. Intensive intervention will be needed to push them initially to cross the threshold for growth. Growth will naturally push economic activities towards diversity at the community level and possibly (but not necessarily) specialization at the household level. In a growing rural economy, households cannot be competitive if they refuse to specialize. Given the limited technologies available to them (agriculture and non-agriculture), specialization will help maximize production in the light of economies of scale. As examples, working within a specific industry for microenterprise development (non-agriculture), raising specific crops requiring special farming systems (and technology) for agriculture, or even specialization of services offered in a diversifying economic environment, will continue to raise households¡¯ competitive advantage in that area. Specialization will stimulate efficiency in rural production and possibly curtail certain factors of production (in the hope of attaining efficiency). Among the factors of production, labor is easily substituted through the choice of appropriate technology, resulting in displacement of many rural workers. This phenomenon was observed in the rural Philippines, which has continuously been experiencing rural-urban migration for the past three decades or so. A sizeable proportion of labor migration spills over to other countries. In the desire for market efficiency, specialization can actually lead towards inequality because of the unequal utility values placed on different production activities. As Lewis (1984) points out, market efficiency is not the solution towards equilibrium in an agrarian economy; the concept rather equates social cost with the real gains from trade to serve as an engine of growth. The solution proposed then is empowerment of rural communities. Empowerment can include, but is not limited to, the provision of infrastructure and capacity building. The framework that this study is based upon revolves around the complementation of infrastructure and capacity building in forging a path towards rural development.

The initial role of the government is neither regulation nor governance but empowerment of local communities, similar to the paradigm proposed by the World Bank in poverty alleviation. Empowerment is defined in this paradigm as ¡°the expansion of assets and capabilities of poor people to participate in, negotiate with, influence, control, and hold accountable institutions that affect their lives¡± (Narayan, 2002). Focusing on empowerment in the framework, market efficiencies can be gradually attained since this will help in narrowing the information asymmetries among the stakeholders (the suppliers, the traders, the market/retailers, and the producers/farmers). The empowered stakeholders would like to gain access to pertinent information before they take specific decisions. Rural roads, other rural infrastructure, and capacity building activities will enable all the stakeholders to access relevant information of the supply-demand chains for rural/agricultural goods and services. The stakeholders can use such information in the efficient allocation of factors of production.

In the process, the government needs to facilitate the dynamics where the stakeholders interact towards attainment of efficiency. For certain interventions like credit, direct provision of say seed capital may be provided by the government or can be taken from some other forms of development assistance. This is also true for other infrastructure where the initial construction will need money that is beyond the capability of the stakeholders. It is important though to consider that rural infrastructure does not follow similar protocol as in mainstream public economics, where cost and maintenance have to be secured from the beneficiaries through the process of taxation. Many of the rural beneficiaries in developing countries fall short of the cut-off for taxable income brackets. However, direct provision should not be continuously done; the government and donors will have to veer away from direct provision and focus on facilitation to stimulate a participatory environment leading towards sustainability. It is important for the stakeholders to establish ownership. Hence, encouraging them to contribute (in cash or in kind) for maintenance to safeguard the sustainability plan should be part of the design of the intervention. The notion of user¡¯s fees is difficult to inculcate among the stakeholders especially because they have limited income and livelihood opportunities. A good advocacy strategy though will help rural stakeholders to eventually accept the concept of user¡¯s fees.

Models will be developed to explain the dynamics of the rural economy. The models will consider a household that would like to maximize its welfare function and will take into consideration spatial distribution. The spatial dimension will rationalize site-specific packaging of bundles of intervention. A stochastic frontier model, basically a production frontier, will also be developed with spatial dimensions. Note that the spatial dimension is justified in terms of soil fertility and diversity of economic activities determined by topography, among others. This model will help explain how inequality among rural households can be traced to how efficient/inefficient they are in accessing the factors of production available to them.

The data that will be used in the empirical investigation will be discussed and presented along with the empirical modeling strategies.


A rural road will be defined as an access route from the main road network to the rural communities and/or production areas. It is intended to provide an access path for individuals residing in rural communities and passage for light public vehicles carrying people and/or produce. Such roads allow transportation cost to be reduced because vehicles carrying farm loads are cheaper than the human carriers that are still used where there is no such road in many rural areas of the Philippines.

Farm roads are often constructed as dirt pavement, or are topped with gravel, with asphalt, or very seldom, with concrete (see Figure 4.1). Usually, only people and light vehicles pass through, but during harvest season, the local government or some community organization upgrades it so that haulers can reach as close as possible to the production areas. The roads in the main road network, called national roads in the Philippines, are usually constructed with concrete materials and are wider, thus accommodating heavy-duty haulers that will collect the produce and bring it to the main distribution depot (government or privately owned).

Figure 3.1 Typical Rural Road in the Philippines
Figure 3.1 Typical Rural Road in the Philippines

The path of rural development from the improvement of accessibility in the rural communities will start from the known direct impact of rural roads. Roads are intended to mitigate an area¡¯s state of isolation that otherwise hinders the initiation of various facets of development. Improved access roads among the rural households will lead to increased accessibility and movement because of lower transportation cost, increasing economic activities. The literature documents a wide range of percentages of reduction in transportation cost as a result of establishing new rural roads or improving existing ones. Regardless of the amount of inputs invested, rural roads are expected to contribute to lowering transportation cost.

Improvement in road networks starts up a feedback system of input procurement and marketing of produce. Producers are expected to pay less for the inputs of production because of the improvement in accessibility, so they become more capable of procuring more inputs. The different suppliers of inputs will lose monopoly and be forced to become competitive since the farmers will now have alternative sources. Marketing will also not be limited among a few traders, resulting in a negotiable pricing system since transportation cost reduction will open the ceiling of price negotiations. This is of course based on the assumption that commodity financing (usually associated with price ceilings for goods and not so fair to the farmers) is no longer practiced or that there is a sustainable credit facility in place. Knowledge of marketing avenues and demand for various commodities (to be facilitated by the government) will encourage farmers to diversify crops, and later on, to specialize in high value crops only viable in the production area (efficiency). Thus, increased production and increased gross value coupled with lower input cost will benefit the farmers in terms of increased earnings.

Improved accessibility will also facilitate provision of basic social services like education and health. Even if such services are not brought right into the community, it will be easier for the households to access those from the town centers or in another community. Social services should result in enhancement of human capital and along with other capacity building interventions, should contribute to empowering the rural community.

Rural roads will also generate multiplier effects. Foremost, they serve as catalysts for greater public investment into infrastructure and capacity building. Given that an improved access road will facilitate the construction of a health center (and visits of health professionals), a warehouse for agricultural commodities, and even the conducting of training and other capacity building activities. Provision of other physical infrastructure will be feasible because materials can be easily transported. Then for those manned by personnel from outside the community, or for capacity building where resource persons come from outside, traveling into the community will be viable now, reducing the lost time normally spent traveling to the site.

Because of the improved mobility of the households, they will be exposed to outside communities and may observe prototype development that will serve as a stimulus for their desire to realize similar development in their locality. It will then foster a good motivating factor for them to participate in the process of identification of strategies that can lead towards development. This is the start of community building that will later on evolve into a sustainability backbone.

With the growing demand for infrastructure, demand for support services will also increase, requiring more participation on the part of the household in planning and in sourcing for infrastructure and support services. This will encourage the local government to contribute as well, so sustainability will become clearer. All of this will lead to increased production. Because of the growing demand for infrastructure, there is now a viable input sourcing at reasonable cost (due to reduction in transportation cost). Better post-production handling will result in lower post-production losses, yielding a good profit margin for the farmers.

For the non-agricultural household, the direct impact of roads will be in terms of facilitating the emergence of new investments and new enterprises. Eventually, more diverse choices of livelihood will become available to them, an important manifestation of rural development.

The complementation between increased production among farming households and the non-farming households engaged in microenterprise development are early leads towards rural development. In rural areas where employment opportunities should extend beyond the traditional agriculture basis, the empowered households¡ªa stronger community that participates in intervention programs¡ªwill benefit not only the individual households, but the entire community, leading towards sustainability.


A client satisfaction survey was commissioned by the World Bank in 2005 (NEDAWB- ASEM, 2005) to develop a perception-based survey that will facilitate the verification of the effect of the outputs of the rural sector agencies (Department of Agriculture, Department of Agrarian Reform, and Department of Environment and Natural Resources) on rural development in the Philippines. A rural development and living condition scale (see Appendices 1 and 2) was developed and pilot-tested several times (see NEDA-WB-ASEM, 2005 and NEDA-WB, 2003). It was concluded that the scale can approximate the constructs of rural development. The survey was implemented in purposively selected barangays (villages) where households were then randomly selected. In the purposive selection of the barangays, prototype interventions of the departments were considered, along with an appropriate control group (no known intervention from the government in recent years). For the government interventions, the strata were defined in terms of whether the project is locally funded or with foreign funding for each of the three major departments working within the rural sector (agriculture, agrarian reform, and environment and natural resources). The delineation between local and foreign funding serves as a proximate indicator of the intensity of resources used in implementing the project, where resources from local sources are usually lesser than those coming from foreign sources. The barangays in the control group were also allocated according to expected income level (low, medium, high income), by topography (upland, coastal areas), and to include the KALAHI-CIDSS sites (a government project using an integrated strategy of facilitating rather than direct provisions, and a participatory approach rather than imposition of appropriate interventions). More than 6,000 households were included in the database. Only rural barangays were included.

The Family Income and Expenditures Survey (FIES), conducted every three years by the Philippine National Statistics Office (PNSO), will also provide data analyzed in this paper. It is a probability sample of about 20,000 households with rural-urban areas of the provinces as domains (until 2000). In 2003, the domain was raised to the regions. In return, more detailed information was collected. The units of analysis are also the households, but in contrast to the information from the Client Satisfaction Survey, longterm outcomes are collected. Transportation cost is used as a proxy indicator of road system improvement.


In a model with several variables including a good number that are dichotomous (dummy) variables, estimation using least squares may be affected because the designmatrix can become ill-conditioned. Estimates may yield reverse signs, so sensitivity analysis on each independent variable may not be feasible. Forecasting/prediction though may still be viable even when the least squares method is used in the presence of ill-conditioning in the design matrix.

To resolve the potential problem caused by ill-conditioning in the design matrix, the backfitting algorithm can be used in the estimation. The algorithm assumes that the postulated model is additive, a generalization of the linear regression model. The model is expressed as a sum of basic functions that can be linear, non-linear, or nonparametric. The additive model is given by

empty. The function ƒ can be of the empty, ε are independent of the x¡¯s, Ε(ε) and var (ε;) = σ2. The backfitting algorithm described by Hastie and Tibshirani (1990) enables additive model-fitting using any regression-type estimation mechanism, given by:

(i) Initialize: empty

(ii) Cycle: j = 1,2,...,r

(iii) Continue (ii) until the individual functions do not change where Sj denotes a smoothing of the response y against the predictor xj.

Smoothing may reduce to ordinary least square for simple regressions (one-at-atime) if the functions are linear.


Stochastic frontier analysis (SFA) will be used in analyzing efficiency of household production both from farm and non-farm sources. The model will be used in explaining inequality among rural households. It is postulated that inequality among rural households will depend on how efficient they are in utilizing infrastructure facilities towards increasing their income and other benefits in general. This is also affected by the combination of infrastructure and other interventions available and is needed in their production activities. Bundles yield more effect than simply adding the individual effect of each intervention.

It is further assumed that efficiency is also affected by spatial dependence in production/income-generation because of soil fertility that is site-specific, diversity of economic activities influenced by topography, homogeneity of agents of transportation, the source and availability of inputs, and markets in adjacent communities.

Technical efficiency will be computed for farming and non-farming activities of the household. The production function will consider income and the rural development index as the dependent variable.

3.4.1 Specification and Estimation of Production Frontier (Model 1)

Consider a cross-sectional production frontier model empty or empty is the actual production and is the empty theoretical production function. xi is a vector of production inputs needed to produce yi while vi is a random error. Note that the distribution of vi and the form of the function ƒ will dictate an efficient estimation procedure for the parameters. Assuming that the theoretical production function is correct, the ratio between actual and theoretical production level yields a reasonable account of technical efficiency (TE).

The function ƒ should satisfy the following conditions provided by Kumbhakar and Lovell (2000) summarized in Section 2 above. Let TEt = exp(-ut), then the production stochastic frontier model becomes empty yielding two error components vi and ui . The negative sign for ui will ensure that TE=1. TE=1 implies efficiency, while TE<1 indicates a shortfall (inefficiency) in a stochastic environment characterized by exp(vi), varying across households. The variable ui will be linked to some factors that are postulated to influence production efficiency of rural households. Reifschneider and Stevenson (1991) empty and empty . We will imbed a spatial autoregression model or SAR (Pace and Barry, 1997) with a general linear mixed model. Thus, the postulated technical efficiency model is empty where empty empty is a spatial parameter, empty , the spatial weight matrix where empty . Two households will be considered spatially related if they belong to the same barangay/village. wi is a vector of fixed factors, zi is a vector of random factors, and empty is pure error. If the observations are arranged so that households coming from the same barangay are next to each other, then the matrix D is block diagonal. The joint distribution of empty and empty is assumed to be normal with mean empty where ∑ and Γ are not necessarily diagonal. We are assuming a general dependence structure among the elements of z and ε , but independence of elements of z from elements of ε is imposed.

Thus, the production frontier equations can be summarized into empty or
where empty. The function ƒ may take the Cobb-Douglas form or a more general exponential or a non-linear function. Since dummy variables will be used in addition to factors of production that are zero for some households, an exponential function ƒ will be used. The location of an exponential function can be adjusted so that the six properties above are satisfied.

Estimation will be done using a modified backfitting algorithm (Landagan and Barrios, 2007), taking advantage of the additive nature of (3.1) and (3.2). The estimation algorithm follows:

1. Depending on the link function ƒ , ignore ui in (3.1) and estimate β using maximum likelihood estimation (MLE) or least squares estimation (LSE).
2. Compute the residuals from (3.1), empty . This now contains information on δ φ.
3. Estimate φ and from (3.2), setting aside the spatial effect, using the initial estimates of technical efficiency ( em>ui ). A maximum likelihood estimator for a mixed model can be used.
4. Compute residual technical efficiency empty . This contains information on δ.
5. Estimate δ from empty , which is a regression through the origin with a single covariate ( Du ).
6. Use the estimates derived from (5) in revising the estimates of technical efficiency from (3.2).
7. Estimate the overall constant term of (3.1) using a non-negative filter (e.g., logistic function), and deduct this from the revised estimate from (6).

The algorithm is expected to converge after (7), (see Landagan and Barrios, 2007 for details).

3.4.2 Specification and Estimation of Production Frontier (Model 2)

In the second model, spatial dependence is postulated on the production function instead of appearing in the efficiency equation.


A similar argument on spatial dependency can be made whether it is in the production function or in the technical efficiency equation. Estimation can be done using a similar algorithm to that in Model 1 above.

The advantage of simultaneous estimation of parameters through maximum likelihood estimation using a distribution of non-negative values for v (e.g., half-normal, logistic) is that it always produces estimates of technical efficiency ¡Ü 1. An alternative is to filter u using a function in a non-negative domain, similar to (7) in the algorithm above. Thus, instead of fitting a linear regression of the first residual from (3.3) above, filtering is done, e.g.,


A similar algorithm can be used with (3.4) above, in lieu of the linear regression of u on the determinants of efficiency. A no constant specification of (3.4) would ensure that u will always be positive, so that the estimate of technical efficiency will be between 0 and 1.

3.4.3 Specification of Variables

The response variables are total income and the rural development index (standardized so that values will range from 0 to 100). The total income coincides with farm income if the household derives all income from farming, non-farm income if it earns income from non-farm sources, and the aggregate of farm and non-farm income if it derives income from both sources.

The survey design imposes constraints in the choice of inputs of production (farming) among the households. Some proximate indicators were considered in lieu of real production inputs so that the production function becomes comprehensive. This will provide a rationale to the estimates of technical efficiency. The following inputs of production will be considered: area cultivated, access to irrigation, access and utilization of credit (as proximate indicators of procurement of farm inputs or capital availability for non-farm activities, a requirement for the development of small scale industries), whether single or multiple crops are planted (proximate indicator of farming system), health indicator of household members (as proximate indicator of human capital), number of household members with work (non-farm), and tenure of work. Two dummy variables will also be included: empty and empty. If the household derived incomefrom both farming and non-farming sources, then S1=S2=1. The interaction between S1 and farming inputs, and S2 with non-farming inputs, will be included to ensure that causation between output and production inputs are appropriate.

For the efficiency equation, the determinants are classified as fixed or random effects. Fixed effect determinants will register similar results regardless of the household being observed. On the other hand, random effect determinants are those whose effects are governed by a sampling distribution, i.e., one household may react differently from another household. The fixed-effect factors are household demographic characteristics (including dependency ratio), land tenure, female-male headed household, education of household members, and the spatial effect. The weight matrix for the spatial effect will be computed for the barangay (village) and will not differentiate households within the same barangay. The spatial indicator will account for socio-geographic characteristics that will affect production and income, soil fertility, and other site-specific unknown agronomic factors. For non-farming activities, the spatial effect will explain the kind of economic activities viable in the area and other site-specific unknown economic and cultural conditions.

Among the random factors to be included are availability of needed infrastructure or other intervention activities, bundles of such, whether the bundles include roads, membership in organization (as measure of participation), and whether they commit to contribute for maintenance. Since these factors are measured in terms of perception among the households, it is expected that the dichotomous responses will yield varying effects among the households.

Download this Discussion Paper [ PDF 333.1KB| 51 pages ].

[previous chapter] [next chapter]

Post a Comment

We welcome your feedback on this publication. Post a comment. ADBI is not obliged to acknowledge or publish comments and may abridge or edit them before web posting.


There are [0] comment(s) for this entry. Post a comment.

    The views expressed in this paper are the views of the authors and do not necessarily reflect the views or policies of the Asian Development Bank Institute (ADBI), the Asian Development Bank (ADB), its Board of Directors, or the governments they represent. ADBI does not guarantee the accuracy of the data included in this paper and accepts no responsibility for any consequences of their use. Terminology used may not necessarily be consistent with ADB official terms.

    Back to Top 
    © 2015 Asian Development Bank Institute.