Csae2014 293 road affects to production

Roads and Farming: The Effect ofInfrastructure Improvement on Agricultural

Input Use, Farm Productivity and MarketParticipation in Kenya.

Philemon KIPRONO ∗

and

Tomoya MATSUMOTO †

Revised 28th January 2014

Paper Presented at CSAE Conference 2014: Economic Development inAfrica, The University of Oxford.

23rd-25th March 2014

Abstract

Using longitudinal data together with geo-referenced roads data, we imple-ment a difference-in-difference(D-I-D) estimation to assess the impact of thechange in road access from 2004 to 2012 on the change in technology adoption, fer-tilizer intensification, farm productivity and market participation by smallholderfarmers in Kenya in Sub-Saharan Africa (SSA). We construct time distance vari-ables by motor-vehicle to the nearest market and to the nearest big town asmeasurements of road access. Our results show that land allocated to hybridmaize, inorganic fertilizer intensification, maize yield and milk market partici-pation increase more in areas with better road access improvement. It suggeststhat, since road access improvement occurs more in poorer road access areas inthe initial period, the recent infrastructure investment contributes to the produc-tivity enhancement especially in remote areas.

Keywords: Sub-Saharan Africa, Kenya, Road Access, Farm Input, Productivity,Market ParticipationJEL classification: Q12, Q13, O18

∗PhD Candidate, National Graduate Institute for Policy Studies,Tokyo, Japan. k [email protected]†Assistant Professor, Department of Economics, National Graduate Institute for Policy Studies,Tokyo, Japan. k

[email protected]‡ We want to thank Keijiro Otsuka, Sonobe Tetsushi, Aya Suzuki, and John Strauss for their excellent com-

ments and suggestions. This work was supported by the MEXT(Ministry of Education, Culture, Sports, Science, andTechnology) Global Center of Excellency Grant and JSPS KAKENHI Grant Number 25101002.

1

mailto:[email protected],[email protected]



1. Introduction

Alleviating poverty through roads infrastructure development has been one of the goalsin many developing nations agendas. Roads are thought to be the catalyzer in theprocess of economic development (Rostow, 1962) and are a particular important factorto the growth of rural areas (de Vera Garcia, 1984). In the recent past, there has beenhuge investments1 into improvement of roads in developing nations by major donors,yet very few studies have been carried out to ascertain the impact.

Improvement of roads in developing nations, especially rural-roads, has receivedgrowing funding and evaluation interest(Estache, 2010). The World Bank (The WorldBank, 2007) and leading donor communities have labeled improvement of roads as aninstrument of poverty alleviation in developing nations. As SSA nations depend onagriculture as the backbone of their economy(Gollin & Rogerson, 2014), improvementof roads will have far reaching effects on agricultural productivity. The African De-velopment Bank (ADB) estimates that 34% of rural Africa [where 80% of the citizensdwell] can access roads compared to 90% in the rest of the world (African DevelopmentBank, 2010). In SSA2 the total road network is estimated at 204 km per 1,000 km2

of land area, of which only about 25% is paved, compared to the world average of 944km per 1000 km2 of land (African Development Bank, 2010). The majority of roadnetworks is poorly developed or in a dilapidated state. Studies have shown stagnatedinvestment in agricultural sector(Otsuka & Yamano, 2005) and coupled with poor ac-cess to markets, these nations are prone to extreme poverty and are epic centers ofhunger today.

Studies on the impacts of road infrastructure in Asia and Latin America haveshown positive impacts on a number of outcomes: reduction in poverty levels (Gib-son & Rozelle, 2003; Jalan & Ravallion, 1998; van de Walle & Mu, 2011), alterationof land use (Jacoby, 2000), increased household income and consumption (Escobal &Ponce, 2003; Jalan & Ravallion, 2002; Khandker et al., 2006), investments in healthand education(Lokshin & Yemtsov, 2003; Stifel & Minten, 2008); crop intensification

1In Africa, the African Development Bank(ADB) invested more than USD 1 billion in 2007 alone. The World Bankapproved more than USD 9 billion in 2010 towards infrastructure development. In Kenya, about 20 percent of GDP($4billion) is dedicated for infrastructure development (Briceño-Garmendia & Shkaratan, 2011)

2United Nations (United Nations, 2008) categorises SSA countries as least developed.

2

and other production decisions (Khandker et al., 2006; van de Walle , 2009), migration(Fafchamps & Shilpi, 2009), reduction in transport costs (Jacoby & Minten, 2009) andexcitation of market activity (Mu & van de Walle, 2011). Other related literature re-porting benefits from transport infrastructure include firm expansion and gross domes-tic product(GDP) growth in China(Banerjee et al., 2012) and growth in interregionaland international trade in India(Donaldson, 2010). In Africa, this line of literature isthin. A few studies include Dercon et al. (2007) assessing improved road accessibilityon access to agricultural extension in Ethiopia who finds poverty reduced by sevenpercentage point with improvement or roads. Dorosh et al. (2012) using cross-countryregression in SSA found a substantial increase in agricultural production. Kingombe &di Falco (2012) on impacts of rural road improvements on farm productivity and cropchoices in Zambia found inconclusive results. Shiferaw et al. (2013) found a positiveimpact on manufacturing sector in Ethiopia. Gachassin (2013) found migration to havedecreased in Tanzania as a result of improved roads.

Estimating the impact of infrastructure improvement in SSA, especially roads, ischallenging due to data limitations and methodological constraints. Studies have doc-umented that benefits due to roads improvement are indirect and often dependenton interactions with other contextual factors such as physical infrastructure and thegeographical, community and household characteristics (Fan & Zhang, 2004; van deWalle , 2009). The fact that the roads are not randomly placed and people don’t ran-domly select where to settle, leads to the problem of endogeneity3. Using traditionalmethodologies such as instrumental variables(IV) approach or propensity score match-ing ex-post to assess impacts of roads improvement may yield unreliable results sinceobservable characteristics may be contaminated by the effects of the project (van deWalle , 2009). Furthermore, techniques approximating random trials are unsuitableand almost impossible to implement4. Use of panel data can mitigate these problems,but collecting large panel data is expensive undertaking and the timeframe for suchevaluations may be too short to capture the long-term benefits of intervention (Mu &

3 Jalan & Ravallion (1998) have cited endogeneity bias where the targeted poor areas typically lack infrastructureand other initial endowments, which could bias the results.

4And if resource allocation is politically motivated(Burgess et al., 2013), then if word go round that government israndomizing on road projects, then incumbent politicians may face backlash from their electorate, who expect theirelected leaders to amass resources for them. Hence randomizing interventions like road improvement are almost impos-sible.

3

van de Walle, 2011). Thus correct modeling is required for the precise outcomes 5.

In this paper we estimate the impact of improving road accessibility from 2004 to2012 on the change in technology adoption, fertilizer intensification, maize productivityand market participation by smallholder6 farmers in Kenya. In particular, we focus onthe impact of rural-road7 infrastructure improvement in 15 districts in Kenya. We usegeo-referenced panel data from Research on Poverty, Environment and AgriculturalTechnology (RePEAT), merged with Roads Network data. This dataset is uniquefor the following reasons: First, because of the geo-referenced data, we are able tomerge two very independent datasets collected almost at the same time to providea balanced panel. Second, our new panel data enable us to use modern techniquessuch as difference-in-difference(D-I-D), which makes it possible to control both thebiases due to omitted variables as a result of non-random placement of roads as wellas reverse causality, providing a substantial degree of explanatory power. To the bestof our knowledge, this is the first paper to empirically document the impacts of roadrehabilitation on technology adoption, agricultural input use, farm productivity, andmarket participation in Kenya.

Our study is closely related to a study by Dorosh et al. (2012). The point ofdeparture in our study is the use of new panel data covering the period before andafter the improvement of road infrastructure in Kenya.

Our results show increased land allocation to hybrid maize seeds, fertilizer inten-sification, manure application, yield of maize as well as market participation for bothmilk and maize in areas experiencing better road access. These are our explained vari-ables for technology adoption, input use intensification, farm productivity and marketparticipation respectively.

This paper is organized as follows: Section 2 gives an overview of Kenya’s roadinfrastructure. Section 3 present our data description as well as our estimation strategy.Section 4 combines our econometric estimates, result discussion and robustness check

5Ravallion (1999) noted that returns to infrastructure may seem enormous but when simultaneity bias and spuriouscorrelation have been econometrically controlled for, the returns are much reduced.

6By smallholder, we refer to an individual who manages a family farm holding with one or several small dispersedparcels of land or farms(≈ 20 acres in total)). Collier & Dercon (2009) notes that poverty is concentrated in rural areasamong smallholders, hence any policies targeting lower poverty levels must start with smallholders.

7Rural-roads: Small local roads and tracks in rural areas that have low or no motorized traffic volumes linkingvillages (van de Walle , 2009)

4

while section 5 concludes.

2. Background of Road Infrastructure in Kenya

The Development Blueprint of Kenya, Vision2030, identifies road infrastructure asone of the key components of communication and development under the blueprint’seconomic pillar8 (Republic of Kenya, 2008). Typically, Kenya’s road network is madeup unpaved roads such as tracks, trails, footpaths, and natural, gravel and earth roadsthat link rural villages and towns and often connect to secondary roads, which allowproducts and factor markets as well as other social services to reach the designatedvillages. The other type are engineered roads (paved) that connect small towns andurban centers for public transportation, and were traditionally used during colonialperiod to connect administrative headquarters. , which in optimal conditions enablesthe village dwellers to access urban centers in shorter travel timesRoad vehicles are thedominant mode of transport in Kenya. A total of 160,886 km of road network in thecountry , of which paved roads account for just 7% while unpaved roads account for93% (Ministry of Roads, 2011). The road conditions vary all over the country fromexcellent tarmac (asphalt) roads to very poor earth(dirt) roads.

2.1 Rural Access Roads

The Rural Access Roads Programme (RARP) was initiated in 1974 with the objective ofproviding more effective rural transport services and improving the standard of livingof the rural population. The hallmarks of the programme were the use of labour-intensive techniques and grassroots participation in the selection of roads from thethen unclassified road network, comprising mainly short links between farms and theclassified road network. Provision was made for 4 m wide gravel surfaces, followingexisting tracks, usually not more than 5-10 km in length, with traffic volumes unlikelyto exceed 10 vehicles per day following improvement. By 1985, when the Minor RoadsProgramme (MRP) began as a successor to RARP, about 8,000 km of farm-to-market

8Developed in 2007, this document aims at guiding the Kenyan Government in creating a globally competitive andprosperous nation with a high quality of life by 2030, hence transforming Kenya into a newly industrialized, middle-income country with a high quality of life in a clean and secure environment by year 2030. The vision is anchored onthree key pillars; economic, social and political governance.

5

roads had been constructed in 15 districts in six provinces using labour-based methods.Roads were usually numbered separately in each district, starting from R19 . Theaverage R link measure about about 7.3 km; but there exist road of lengths greaterthan this. Today, it is managed by the Kenya Rural Roads Authority (KuRRA). Theseclassifications are still used.

2.2 The Kenya Roads Act of 2007

Before the reforms that took place in the Kenyan government after the 2002 gen-eral elections, the roads sub-sector was marred by outright corruption and a lack ofaccountability in terms of road construction and rehabilitation. Uncertainties, dupli-cation of roles, and inconsistency in the road asset management were rampant andadversely affected road construction and rehabilitation, which resulted in dilapidatedroads in the country. Because of a lack of coordinating mechanism, ministries concur-rently exercised road management responsibilities through some of their departmentsand agencies. Furthermore, some authors have shown that historically the allocationof resources for the construction of roads have been hampered by political inclinationand patronage where roads have been traditionally constructed in areas perceived tobe friendly to government or where powerful government official come from(Burgesset al., 2013). Thus, there was a need for the new government to urgently address thisshortcoming. As a result of serious consultations, Sessional Paper No. 5 of 2006 waswritten and approved by Parliament on 19th October 2006. It spelled out policies tobe pursued by the Government in the medium term for sustained economic growthand provided the legal and institutional framework for the management of roads (Min-istry of Roads, 2011). The reforms under Sessional Paper No 5 of 2006 realized the fourbasic building blocks necessary for effective roads management: ownership, clarified re-sponsibility, stable financing, and commercialized management. Further amendmentsof Sessional Paper No. 5 of 2006 culminated in the drafting of the Kenya Roads Billof 2007. The bill provides for the establishment of various road-governing agencies,namely, the Kenya National Highways Authority, the Kenya Urban Roads Authority,and the Kenya Rural Roads Authority, to oversee various categories of roads (Republicof Kenya, 2007). In 2007, the Kenya Roads Act came into force and the three road

9R≡Roads

6

authorities with responsibility of clearly defined mandates on the management of re-spective sub-networks were created, thus ushering in a new dawn in road management.

3. Data and estimation framework

In this section, we describe the source of our data as well as the estimation strategy.

3.1 Data

The data used in this study are obtained from two main sources: (a) Research onPoverty, Environment, and Agricultural Technology (RePEAT) project and (b) theMinistry of Roads (Kenya). Supplementary data on rainfall were obtained from theMeteorological Agency in Kenya (MET).

3.1.1 RePEAT data

RePEAT data is a geo-referenced household and community survey conducted by Foun-dation of Advanced Studies on International Development (FASID) and its collabora-tors in three East African countries (Kenya, Uganda, and Ethiopia). Four follow-upsurveys have been conducted in Kenya since the initial one in 2004 namely in 2007,2009, 2012 and 2013. The 2013 survey collected information pertaining soil composi-tion, as well as ascertaining the condition of the roads. Initially, the randomly sampledhouseholds were selected from 2,966 households that were previously interviewed bythe International Livestock Research Institute (ILRI) in 1998 and 200010 and cov-ered 15 districts drawn from the Rift Valley, Central, Nyanza, Western and Easternprovinces.11

The survey contains detailed information on various household and communitycharacteristics including demographics, education attainment, household income andexpenditures, farm input use, land tenure system and land use. We use the data fromthe 2004 and 2012 surveys to match with the road data. The original data targeted 899.

10See Staal & ILRI (2001); Waithaka et al. (2002)11Full set of questionnaires and other information about RePeat can be obtained from http://www3.grips.ac.jp/

~globalcoe/e/index.html

7

http://www3.grips.ac.jp/~globalcoe/e/index.html

http://www3.grips.ac.jp/~globalcoe/e/index.html

In this study, we made adjustments to the data as follows: first we dropped householdsthat were replaced (82), no-contact (24), refused to be interviewed (7), head and spousesick (4) and no eligible member to respond (3) and missing data (57) in 2004 and 2012resulting in an attrition rate of ≈20% 12 and is distributed as shown in Table 1. Thus apanel of 715 records is used in the analysis. In addition to household data, communitylevel data were also collected by means of focused groups. These groups comprised ofvillage leaders headed by the Chief or the Sub-chief and are more knowledgeable onthe community matters 13.

3.1.2 Road Network Data

This second set of data is a detailed geo-referenced road network data obtained fromthe Ministry of Roads, Kenya, which was updated during the same period when theRePeat Survey was being undertaken.

The Ministry of Roads engaged a consultant to undertake a Road Inventory andCondition Survey (RICS) for the Classified Roads using Geographical Positioning Sys-tems (GPS) in 2001. This project was funded by the World Bank. The results ofthe study amalgamated to a database for classified roads with GIS information. Thismapping excluded unclassified rural and urban roads and hence made it difficult foreffective maintenance and development planning. Thus, in 2006, the Kenya RoadsBoard, with funding from the Nordic Development Fund under the Northern CorridorTransport Improvement Project, commissioned as second consultant to undertake aroad inventory and condition survey for the hitherto unclassified road network. Thesurveys were undertaken on a district by district basis between May 2007 and April2009. It captured detailed physical road and bridge inventory data (e.g., surface type,road width, culverts) and visual condition surveys (classified as excellent, good, fair,poor, and very poor). Because of the nature of RICS data, the data were compati-ble and were successfully merged to form the complete geo-database. Although thenumber of districts has since increased to over 250, the maps were drawn for the 71original districts in line with Terms of Reference (ToR). The complete GIS database

12Then we investigate whether there is any systematic attrition in our data. The F-ratio test indicates no evidenceof systematic attrition.

13Sub-chiefs are the representative of the central government, responsible for the lowest form of the central government,the sub-location. The chief represent the government at location level (herein referred to as community).

8

established at the Kenya Roads Board is being updated regularly to capture the currentdevelopments at the grassroots level. This provides a comprehensive road inventoryand condition data which can be used for planning and implementing maintenance anddevelopment programmes by road authorities (Ministry of Roads, 2011).

3.1.3 Algorithm for generating time distances

During the 2004 and 2012 surveys, smallholder farmers were asked on the condition ofthe road leading to the nearest market, as well as the average time to the nearest mar-ket. To construct our time distance variables14, we use the geo-referenced road networkdata. We employ an algorithm that takes into account road quality, slope, biophysicalcharacteristics of the land to minimize the size of bias due to road endogeneity.

First, we digitized the maps of road network of 200415 and superimpose them on2012 geo-referenced data using Desktop ESRI ARCGIS R© 10 software to measure thedistances. We also used historical Google R© maps to counter check on the physicalstatus of the roads as reported by smallholder farmers. We then assigned each sectionof the road in the network a speed according to the classification shown in Table A.1.These speeds were largely drawn from official speed limits in Kenya and where speedlimits were unavailable, speeds were assigned to mimic the actual speeds of the roads.Using the ESRI ARCGIS R© 10 software, we generated a network dataset and run aroute solver to pick the shortest time distance to the nearest market using householdsas the origin and markets as facilities. We choose qualitative time variables as opposedto dummy indicators because they are more informative: they reflect the type andquality of the infrastructure.

Using household identification numbers and community identification numbers, thedataset were merged with the RePEAT dataset.

14Travel times are by a vehicle, and not on foot.15We could not get ready database for 2004 road network, thus we digitized the maps to generate the 2004 road

network database.

9

3.2 Descriptive statistics

Table 1 shows the distribution of the sampled units used in this study. Summarystatistics are presented in Table 2. Households are aging with the household sizedeclining by 14% over the span of the survey. This is expected as children grew up andmigrate16 to urban centers to look for jobs and greater numbers of opportunities whileothers get married and start new households. In Kenya, household members are thekey supplier of farm labor, and the declining household size may negatively affect farmproductivity; especially where farm labour supply is scarce and farm mechanizationis unaffordable. A second notable point is the 5% increase in households headed byfemales. Female headed households are vulnerable to many vagaries and hence thisvariable may affect negatively our explained variables.

Numerous brands of inorganic fertilizer were available in the market in 2012 com-pared to 2004 as a result of fertilizer market liberalization (Freeman & Kaguongo, 2003).Therefore, in order to compare smallholder farmers using different types of fertilizer,we converted the fertilizers into primary nutrients measured in terms of nitrogen (N),phosphorus (P2O5), and potassium (K2O5) contained per kilogram of fertilizer (here-after, NPK). In the Table 2, although the percentage of households using inorganicfertilizer did increase significantly at 10%, the intensity of its usage increased by 64%on average. The amount of organic fertilizer usage also increased ,but with a higher de-gree of usage variation. This increase has been experienced in zones that have adopteddairy farming (zero grazing), as noted by Otsuka & Yamano (2005). Although totalland accessible (ha) increased by almost 5%, our data indicate that of the householdsinterviewed, 21% reported to have returned part of land they occupy to their ownerswhile approximately 4% transferred part of their land to their siblings.

Land set aside for planting hybrid maize increased by 35% on average over the studyperiod. Maize yield increased significantly by 57%. There seems a positive correlationbetween the use of HYV and organic fertilizer and the maize yield. But the relationshipmay not be causal.

Travel time by vehicles to a nearby active market improved by 27% while traveltime to a nearby big town improved by 18%. This is as a result of road infrastructure

16Household member migration has increased from 0.42 persons per household in 2004 to 0.61 persons in 2012

10

improvement initiated by the government after the implementation of the Roads Actof 2007.

As for market participation, proportion of milk marketed increased by 3% whilethe ratio of maize sold to harvest did not change significantly.

On mobile phone network coverage, all communities in our study were covered by amobile phone network in the year 2012, which was an improvement from 61% in 2004.This has led to an 81% jump in household mobile phone ownership.

[Insert Table 1 about Here]


3.3 Conceptual framework and empirical strategy

We investigate factors nudging smallholder farmers to expand land use for plantinghybrid maize and intensify the use of inorganic fertilizer and organic manure.

Smallholder farmers’ willingness to adopt productivity-enhancing technology is basedon the state of infrastructure and market conditions with which they are faced. Thusby improving rural connectivity through the rehabilitation of roads, it is expected thatthe propensity to adopt productivity-enhancing technology and the intensification offertilizer are bound to increase. This effect is assumed to be realized through lowertransportation costs of goods and services that raise smallholder farmers’ net outputprices as well as lower production cost due to cheaper and more accessible farm inputssuch as fertilizers, seeds and pesticides. Therefore, we hypothesize that:

Hypothesis 3.1 (Roads connectivity, hybrid maize land expansion, fertilizer inten-sification and farm productivity). Improving road quality increases the propensity toincrease land for hybrid maize and the intensification of fertilizer use and farm pro-ductivity in households living in remote areas because it is more profitable.

The second hypothesis that we wish to test is that by improving roads, smallholderfarmers naturally take advantage of market accessibility to market their farm produce.They will either sell their produce in the market or at the farm gate. For the perishable

11

farm produce like vegetables, fruits and milk, market accessibility is the key to fetchinghigh prices because the commodities are in demand while still fresh. In this hypothesis,the underlying idea is that, with improved road networks, many smallholder farmerswould engage in production of fresh farm produce like milk. Although we expectroads improvement not to influence market participation for non-perishable produce,we argue that the improvement of roads to a nearby big town may stimulate marketactivity, especially for dry cereals, because markets not reached hitherto may now beaccessible. Therefore, we hypothesize that:

Hypothesis 3.2 (Roads connectivity and market participation). Improving roads con-nectivity enhances market participation for perishable farm produce by smallholderfarmers living in remote areas.

Although the descriptive evidence suggests a positive impact of road improvementon the number of household outcomes, identifying factors that affect each of the out-come and assessing their importance requires econometric analysis. The goal is toexamine the impact of change in road access on various farm outcomes. We utilize thenew panel data and take advantage of the improvement of roads to obtain unbiasedestimates of our model.

We employ a D-I-D model to identify the impact of improving roads infrastructureon a number of household outcomes. Our base model is:

Yijt = β0 + βH .Hijt + βC .Cjt + βRM .RMijt + βRTRTjt

+ βinteract.(RMijt ×RTjt) + βyear×region + ζi + εijt(1)

where i indexes households, j indexes communities and t indexes year of survey. Yijtdenotes the variable of interest; proportion of land allocated for hybrid maize, fertilizerintensity, farm productivity or market participation of the ith household living in jthcommunity at time t. Hijt is a vector of household specific regressors namely: age of thehousehold head and its squared term, log of household size, a gender dummy equal to 1if the household head is female, number of years of education of the household head andits square term, household land size and its squared term and log of per capita assetvalue, number of men, women and girls in the household and wether the household

12

subscribe to a community self help group. Cjt control for observable community levelcharacteristics which include log of population density. RMijt is log of centered time17

distance in minutes to the nearby market from a household while RTjt is the centeredlog of time distance in minutes from community reference point to the nearby bigtown18. RMijt ×RTjt is the interaction term between time to the nearest market andtime to the nearest big town. This term will capture the additive(synergistic) effect ofimproving both types of roads. The synergy is important to analyze as it show(if any)the extra benefit from improving road to nearby big town, for an averaged improvedroad to nearby big market. ζi is unobserved household fixed effects. εijt is the householdidiosyncratic errors. δyear×region dummies were also included because we were concernedabout the dependent variables being correlated with any region-specific fixed effect.We also controlled for planting season where applicable. We use the heteroskedasticrobust standard errors clustered at the community level. βRM , βRT and βinteract areour coefficients of interest.

To analyze the factors influencing smallholder farmers to participate in the mar-keting of farm produce, we use a modified version of Equation 1. First, we define adummy variable (Sijt) for market participation as Sijt = 1 if a household sold part ofits produce (in this case either milk or dry maize). We then estimate the followinglinear probability model (LPM) with the household FE model and the controls as inEquation 1.

Sijt = α0 + αH .Hijt + αC .Cjt + αRM .RMijt + αRTRTjt

+ αinteract.(RMijt ×RTjt) + αyear×region + ∅i + ωijt(2)

where the control variables are as described above. Although our explained variableis a binary, we prefer LPM technique in the analysis because we do not need to makearbitrary assumptions like in logit, which entails interpretation of marginal effects whichare often difficult to infer. In this model, αRM , αRT and αinteract are our coefficients ofinterest.

For the proportion of milk and maize marketed, we use Equation 1.17Measurement of time distances is discussed in Section 318A big town here refers to market centers with a population over 100,000.

13

4. Econometric Results

In this section we present the estimates of our D-I-D regression with controls. First,we estimate Equation 1 and present the results in Tables 3 to 6. Then we estimateEquation 2 for market participation and present the results in Tables 7 to 10.

4.1 Effect on hybrid maize land expansion

Equation 1 is used to estimate the determinants inducing smallholder farmers to allo-cate more land to hybrid maize. Our variables of interest are the log of time distanceto nearby market and the log of time distance to the nearest big city; these capturesinfluence of the market through prices as well as inputs accessibility. Both of thesevariables are expected to influence adoption propensity positively. Therefore we ex-pect negative coefficient because of the reduction in travel times. In order to analyzeimpact of roads improvement in terms of remoteness of location of smallholder farmers,we classify our data into 4 quartiles using the 2004 physical distance to Nairobi [thecapital city of Kenya] namely from 0-84 Km, 85-165 Km , 166-305 Km and over 306Km.

Table 3 presents the results of the impact of road improvement on the propensityto allocate more land to hybrid maize in all the regions.


Column 1 of Table 3 presents results excluding the time distance to the nearbybig town. The sign of the coefficient is as expected and significant. Similarly, column2 presents results excluding the time distance to the nearby market. The coefficientis of expected sign and significant at 5% significance level. When we interact bothtime distances in column 3, the coefficient of the time distance to the nearby bigmarket reduces slightly and still remains significant at 5% significance level. All thespecifications are controlled for observable household and village characteristics, aswell as unobservable [fixed] household effects. Column 4-7 presents the results of oursub-samples. Within 84 Km from Nairobi, the coefficients of time to market as well astime to nearby big town remain significant but increased in magnitude. In column 5,the coefficient of roads to the nearby big market become is significant at 5%.

14

A clear pattern emerge in Table 3. The coefficient the time to big market remainssignificant within 166 Km from Nairobi. Similarly, the interaction term is positiveand significant in column 6. This suggests a synergistic effect between improvement ofroads to nearby markets and to big towns: in remote areas where roads to the nearbymarkets were improved, allocation of land for planting hybrid maize were higher inareas experiencing improvement in the roads to the nearby big towns. There seem tobe no effect of particular road improvements in the regions over 166 km from Nairobi.

A significant elasticity of -0.604 in column 5 implies that a 1% improvement inroads to the nearby big town will induce smallholder farmers to allocate more landto hybrid maize by almost 0.6%. Although the magnitude looks small, this is quitesubstantial in the sense that in Kenya, the planting of hybrid maize is not such a newtechnology (Matsumoto & Yamano, 2009).

4.2 Effect on intensification of inorganic fertilizer(NPKequivalent)

In this section, we estimate the impacts of road improvement on the intensity of fer-tilizer application19 using the same set of controls as in Equation 1. Here, we run ourestimation at plot level with fixed-effects at the household level. The dependent vari-able is the log of NPK equivalent(kg/ha) used in planting maize20. We expect fertilizerintensification on maize fields to increase because access to input market is improved.We measure the intensification rates by the amount of NPK equivalent fertilizer appliedto one hectare of maize field(kg/ha). The results are presented in Table 4.


Although column 1-3 of Table 4 show no significant coefficients suggesting no impactof road improvement, once we restrict our analysis to sub-samples there is evidenceof impact of road improvement on fertilizer intensification. First, column 5 shows a

19The household decision to expand land on hybrid maize is independent of fertilizer usage, but fertilizer usage isalmost conditional on the use of hybrid seeds because of their high responsiveness to fertilizer application Otsuka &Yamano (2005).

20In this analysis, we don’t distinguish between households that practice intercropping (planting maize and beanstogether) and those that practice pure stand cropping system. Less than 10% of the household had pure stand maizefields.

15

negative and significant elasticity implying that reduction of travel time to big towndid nudge smallholder farmers to increase inorganic fertilizer intensification. Between85 km and 305 km, the improvement of roads to big town has an impact on organicfertilizer intensification.

Matsumoto & Yamano (2009) in their study on factors behind the low applica-tion rates of external fertilizers in Kenya and Uganda found that Kenyan smallholderfarmers have achieved the optimal nitrogen application level. To further increase theapplication rates, they suggested a market-based approach that influences fertilizerprices [price reduction] as well as the maize output price [price increase]. Column 5 ofTable 4 reflects this scenario. A highly significant coefficient of improved roads to thebig town suggest a reduction in the transaction cost of inputs as well as increase inoutput prices making inorganic fertilizer investment profitable; thus smallholder farm-ers living in areas with improved roads are likely to adopt the application of fertilizeron their maize fields. However, even though our results suggest longer travel times tothe big towns discourages the use of fertilizer, we are cautious while interpreting themagnitude of the coefficient which is large.

4.3 Effects on the intensification of organic Fertilizer

The success of the green revolution in Asia have caused elicitation in the use of manurein SSA (Otsuka & Yamano, 2005). Many smallholder farmers practise mixed farming,both keeping livestock and planting crops. Some have adopted zero grazing and asa result, they are able to collect large amount of manure which they use to fertilizetheir farms. Because it releases nutrients to the soil slowly, the use of manure improvesorganic matter in the soil which have long-term benefits when used in farms.

To investigate the factors leading to intensification of manure and to check if therewas any substitution between inorganic fertilizer and organic manure, we regressedlog of manure intensification(kg/ha) on the same set of regressors as in the previoussection. The results are presented in Table 5.


16

Column 1 of Table 5 shows a significant coefficient at 5% for travel time to thenearby market. When we interacted the two time distances, the coefficient of traveltime to the nearby market increases. This aspect changes once we restrict our analysisto sub-sample. Column 7 shows a weakly significant coefficient of the travel timeto nearby market. The regression estimates however showed no evidence of manureintensification in areas within 305 km from Nairobi. This is due to the fact that per-capita land in these areas is still sufficient hence farmers practice free range system ofanimal husbandry.

4.3.1 Effects on farm Productivity

In this section, we estimate the determinants of maize productivity21. In Kenya, maizeis the staple food crop and is cultivated by almost all homesteads. Majority of thesehouseholds grow maize for subsistence.

To investigate the impact of road improvement on maize yield, we again regress logof yield of maize(kg/ha) on the same set of regressors as in Table 3. Table 6 presentsthe results of the regression.


Column 2 and 3 show coefficients(large) that are significant at one percent forthe time distance to the big town. This implies that improvement of roads to thenearby big town is important for maize yield, though indirectly. The effect may bethrough price reduction of farm inputs like fertilizer as well as their availability. Whenwe restrict our analysis to sub-samples, there is no evidence that roads improvementimpact positively on maize productivity up to 84 km from Nairobi. Beyond 85 Km,there is evidence that improving roads to nearby big town improves maize productivityin remote areas(column 5 and 7), though the coefficients are large. This shows that thefar flanked areas can now supply their maize to markets away from their traditionalmarkets within their locality.

21Maize is produced in favorable environment where production is expected to be profitable(specifically revenue mustbe greater than production costs). Mathematically crop production is given as y∗ijt = β′Xijt + εijt. We observemaize yield in household i(yijt = y∗ijt if y∗ijt > 0) only when it is profitable, otherwise zero production is observed(yijt = 0 if y∗ijt < 0).

17

4.4 Effect on market participation

In this section, we analyze the impact of roads improvement of market participationfor milk(which is perishable) and maize(non-perishable). We choose to analyze thesetwo farm outputs because over 80% of the surveyed households are engaging in theirproduction. Our conjecture was that the estimated coefficients of travel time wouldbe larger for milk than dry maize. Furthermore, we expected that the impacts of roadimprovement would be more pronounced in remote areas. First, we estimated factorsaffecting participation in the milk market and the ratio of sales to production for milk,after which we used the same estimation strategy for dry maize.

Table 2 shows about a 37% increment in the amount of milk sold per month by milkproducing households while the proportion of milk sold increased by about 8% duringthe same period. We also examined the effect of improving roads on the share22 ofmilk produced that was actually marketed. Equation 2 is used to analyze the impactof road improvement on market for both milk and maize.



Table 7 present the determinants of participating in the milk market while Table 8present the determinants of proportion milk actually sold. Although the coefficient oftime to the nearby market is of the right sign, non is significant. But the coefficient oftime to the big town is significant even when interacted with the time to nearby market.This implies that a 1% improvement of roads to nearby big town will nudge smallholderfarmers to participate in milk market by about 0.34 percentage points. Although weargued in subsection 4.3 that smallholder farmers living near Nairobi practice zerograzing, column 4 of Table 7 shows no evidence of improved milk market participation23. This is because before implementation of the Roads Act of 2007, the roads fromthe regions near Nairobi were already in good state. The impact of improving roads to

22Average proportion of liters of milk sold over the milk produced per month.23The modes of milk marketing included the private ventures include:Hawkers who increased by 9%, Private cool-

ers(8%), and other Private other co-operatives(5%) and Kenya Co-operative Creameries(KCC)(5%). Hawkers traversevast areas, moving from household to household looking for potential milk sellers. Improvement of infrastructure is ofgreat benefit to them.

18

the nearby big town is experienced in the regions between 166 to 305 km from Nairobi.The significant coefficients in column 5 and 6 implies that improvement of roads tonearby big towns enables milk produced from remote areas be transported to externalmarkets not accessed before.

On proportion of milk sold, column 2 and 3 of Table 8 show similar results to Table 7.Improvement of roads to the nearby big town increases the proportion of milk marketed.When we restrict our analysis to sub-samples, column 6 show a significant coefficientfor the roads to the nearby big town implying that a one percentage improvement ofroad to the nearby big town increases the proportion of milk sold by 0.4% given anaveraged improved road to the nearby market.

Next, we present the results of participating in selling maize. Table 9 presents theimpact of roads improvement on the probability of a small holder farmer participatingin the maize market.Except for column 5, the coefficient of road to the nearby bigmarket is consistently positive and significant. This implies that improvement of roadsto the big town lead to fewer people participating in maize market. This is contraryto our expectation that improvement of roads would nudge farmers to participate inmaize market.

Similarly, Table 10 presents the impact of roads improvement on the proportion ofactual maize marketed. Columns 2-4 shows positive and significant coefficients. Thisshows that the improvement of roads did not improve the amount of maize marketed.This is not very surprising in that, most of the farmers grow maize for subsistence henceif no surplus is available for sale, then even if markets are available, people would stillnot sell their maize.

4.5 Robustness check

In this section, we investigate on the robustness of our results. Because we have manyzeros in our data, running a fixed effect model may bias our results. Thus we run tobitfixed effect model as suggested by Honoré (1992). Results are presented in Table 11


19

From Table 11, the results are similar to the ones we presented in the main docu-ment, with some coefficients slightly becoming significant. This is because tobit modeltakes into account both true zero and generated zeros while estimating the coefficientsof the regression

Next, we interacted mobile phone network coverage together with our variablesof interest. Our conjecture is that both the improvement of roads and the mobilephone network expansion are synergistic thus providing a better platform for technologyadoption, fertilizer intensification as well as market participation to thrive. ?? presentthe results of the regression.


The coefficient of the interacted term for the time distance to the nearby big townbecomes insignificant for the proportion of land under hybrid maize, maize yield andproportion of maize sold. The absorption of the significance by interaction on thesignificant coefficients of ?? may be attributed to interdependence between roads im-provement and network expansion. In Kenya, mobile network expansion first occurredin areas with good infrastructure. Our results are robust to different specification.

5. Summary and policy implications

Dilapidated road infrastructure has resulted in slow agricultural development in Africabecause farm inputs become scarce hence become expensive and markets for farmproduce become inaccessible. Coupled with poor climatic conditions, the agriculturalyields are becoming poorer, further exacerbating the already widespread poverty condi-tions on the continent. Therefore, enhancing smallholder farmers’ capability to expandfarm productivity as well as market participation in rural Africa through road rehabil-itation is considered an important undertaking towards alleviating poverty. Kenya hasrecently experienced an expansive road rehabilitation programme that has seen manydilapidated rural roads brought to maintainable standards. As far as farm inputs areconcerned, the prices are expected to fall while those of farm outputs are expected torise, thus giving smallholder farmers more income, especially in the remote areas.

20

Using longitudinal survey data on smallholder households in Kenya and correspond-ing GIS road maps, we have estimated the impact of the change in road access from2004 to 2012 on the change of technology adoption, fertilizer intensification, maize pro-ductivity and market participation. Our results have provided empirical evidence onthe impact of roads improvement in Kenya. These results shows that the use of maizehybrid seeds, chemical fertilizers, maize productivity and milk market participationincrease more in areas with better road access improvement. There was no evidenceto support improvement of roads infrastructure could lead to increase in maize marketparticipation. However, overall the results show that even though there was a widespread improvement of roads, the impact was experienced more in areas with poorerroad access in the initial period. Therefore the recent infrastructure investment hascontributed to productivity enhancement, especially in remote areas.

Overall, our study shades light on the importance of improving infrastructure, es-pecially in the remote areas, where impact can have far reaching benefits. Because ourstudy was based in rural areas, we postulate that the results present here are lowerbound. More exploration is needed to ascertain the importance of our results to enableus to generalize to other settings such as urban areas. These findings imply that roadsimprovement is a key factor towards alleviating poverty in the country. It thereforewill guide policy drafters in Kenya to channel more funds to this type of infrastructureto realize the full potential of exploiting resources in the rural areas, which in mostcases are underutilized.

Finally, there could me more benefits to the improvement of road infrastructurein rural areas, for example, in land use alteration and human capital investment. Wetherefore propose further research on the importance of improving road infrastructureother than agricultural intensification and market participation.

21

Table 1: Distribution of sampled units, RePEAT (2004-2012)

2004 2012

Region Dist Sub HH Sub HH

Central Kiambu 12 87 12 87Kirinyaga 6 43 6 43Maragua 3 23 3 23Murang’a 8 62 8 62Nyandarua 7 26 7 26

Eastern Machakos 7 65 7 65Nyanza Kisii 7 61 7 61

Nyamira 4 29 4 29Rachuonyo 7 58 7 58

Rift Valley Nakuru 16 128 16 128Nandi 5 43 5 43Narok 3 16 3 16

Western Bungoma 4 23 4 23Kakamega 4 24 4 24Vihiga 4 27 4 27

Total 97 715 97 715

Dist ≡ Districts; Sub≡Sublocations; HH≡Households

22

Table

2:Sum

mary

stat

istic

sand

ttes

tfo

req

uali

ty

of

mea

ns

of

varia

bles

:K

enya

ReP

EAT

2004

and

2012

Year=

2004

Year=

2012

Variables

Obs

Mean

SdObs

Mean

SdMeanD

iff

Hou

seho

ldlevel

Averageeducationof

householdhead

(years)

715

6.61

4.51

715

6.76

4.42

0.151

Genderof

householdhead

(=1forfemale)

715

0.21

0.41

715

0.28

0.45

0.076***

Averageageof

householdhead

inyears

715

55.74

13.68

715

61.88

14.09

6.144***

Meanad

ulteducation(years)

715

7.93

2.72

715

7.59

2.92

-0.345**

Averageho

useholdsize

715

6.44

2.90

715

5.49

2.87

-0.950***

Averagenu

mbe

rof

depe

ndan

tsin

aho

usehold

715

2.67

2.09

715

1.97

1.63

-0.695***

Averagenu

mbe

rof

migrantspe

rho

usehold

715

0.42

0.81

715

0.61

1.09

0.190***

Percentageof

households

belong

ingto

aselfhelp

grou

pin

2012

715

0.00

0.00

715

0.46

0.50

0.462***

Percentageof

households

owning

atleaston

emob

ileph

one

715

0.12

0.33

715

0.93

0.25

0.811***

Percentageof

households

selling

milk

715

0.50

0.50

715

0.60

0.49

0.099***

Averagemilk

sold

mon

thly(liters)

715

71.80

130.83

715

98.30

176.11

26.500***

Propo

rtionof

milk

sold

715

0.28

0.31

715

0.36

0.34

0.084***

Averageho

useholdassetvalue(×10

4)

715

6.10

15.53

715

11.65

48.53

5.558***

Averageland

holding(H

a)715

1.68

2.74

715

1.76

2.55

0.080

Percentageof

households

usingfertilizer

715

0.68

0.40

715

0.73

0.41

0.041*

Percentageof

households

usingman

ure

715

0.49

0.41

715

0.52

0.43

0.027

Percentageof

households

selling

maize

715

0.23

0.36

715

0.26

0.38

0.028

Propo

rtionof

maize

sold

715

0.12

0.25

715

0.10

0.18

-0.013

Averagetimeto

nearby

market(m

inutes)

715

22.83

19.03

715

16.66

11.81

-6.167***

Parcellevel

Propo

rtionof

land

forplan

ting

Hyb

ridmaize

(Ha)

1069

0.52

0.46

990

0.70

0.39

0.178***

Plotlevel

Averageinorganicfertilizeruse(Kg/Ha,

NPK

equivalent)

1325

49.61

88.58

1072

81.95

87.23

32.343***

Averageman

ureuse(Kg/Ha)

1325

1752.07

4378.06

1072

6281.46

56371.54

4529.384***

Averagemaize

yield(Kg/Ha)

1325

1397.80

2556.97

1072

2193.14

1945.14

795.333***

Com

mun

itylevel

Percentageof

mob

ilene

tworkcoverage

970.62

0.49

971.00

0.00

0.381***

Averagetimeto

nearby

bigtown(m

inutes)

9795.27

46.39

9778.16

37.17

-17.110***

Averagecommun

itypo

pulation

density

97671.38

881.29

971006.49

1300.05

335.112**

Coefficientof

variationof

rainfall

970.37

0.19

970.32

0.09

-0.051**

Sign

ificancelevel:***p<0.01,*

*p<0.05,*p<

0.1

23

Table 3: D-I-D estimates of the impact of road improvements on expansion of land for hybrid maize: Kenya,2004-2012

Dependent Variable: Proportion of land under hybrid maize

(1) (2) (3) (4) (5) (6) (7)VARIABLES 0 - 84 85 - 165 166 - 305 Over 306

(a) Ln(Time to market) -0.140** -0.086* -0.273* -0.068 -0.122 0.015(0.036) (0.094) (0.055) (0.464) (0.132) (0.882)

(b) Ln(Time to big town) -0.297** -0.293** -1.048** -0.604** 0.097 -0.057(0.019) (0.018) (0.013) (0.020) (0.627) (0.710)

(c) (a)×(b) 0.187*** -0.198* 0.074 0.306*** 0.201(0.007) (0.094) (0.402) (0.001) (0.109)

Constant 0.168 0.199 0.252 1.029 0.537 0.251 -0.697(0.546) (0.478) (0.372) (0.104) (0.290) (0.612) (0.156)

Observations 3,087 3,087 3,087 855 708 777 747R-squared 0.121 0.117 0.130 0.259 0.211 0.205 0.088Number of hhid 715 715 715 173 181 203 158HH FE Yes Yes Yes Yes Yes Yes YesYear × Agrozone Yes Yes Yes Yes Yes Yes Yes

Notes:1). Reporting pval in parentheses2). Significance level: *** p<0.01, ** p<0.05, * p<0.13). Time distance in minutes by motor vehicle; Robust standard errors clustered at community level; All equations include year× agrozone; Controls include household and village level regressors as described in data section.

24

Table 4: D-I-D estimates of the impact of road improvements on inorganic fertilizer intensification(NPKequivalent, kg/ha): Kenya, 2004-2012

Dependent Variable: ln(NPK equivalent inorganic fertilizer(kg/ha))

(1) (2) (3) (4) (5) (6) (7)VARIABLES 0 - 84 85 - 165 166 - 305 Over 306

(a) Ln(Time to market) -0.200 -0.139 0.478 -0.537 0.170 -0.060(0.253) (0.471) (0.373) (0.169) (0.680) (0.789)

(b) Ln(Time to big town) -0.254 -0.231 1.604 -3.847** -1.297* 1.512(0.689) (0.710) (0.187) (0.021) (0.052) (0.170)

(c) (a)×(b) 0.228 0.526 -0.872* 0.688 0.974***(0.395) (0.290) (0.053) (0.146) (0.009)

Constant 1.550* 1.563* 1.630* 0.489 -1.520 5.611*** 1.235(0.081) (0.086) (0.071) (0.753) (0.601) (0.007) (0.432)


Notes:1). Reporting pval in parentheses2). Significance level: *** p<0.01, ** p<0.05, * p<0.13). Time distance in minutes by motor vehicle; Robust standard errors clustered at community level; All equations includeyear × agrozone; Controls include household and village level regressors as described in data section.

25

Table 5: D-I-D estimates of the impact of road improvements on organic fertilizer intensification(kg/ha):Kenya, 2004-2012

Dependent Variable: ln(Manure intensification (kg/ha))

(1) (2) (3) (4) (5) (6) (7)VARIABLES 0 - 84 85 - 165 166 - 305 Over 306

(a) Ln(Time to market) -0.397** -0.469** -0.741 -0.519 -0.489 -0.460*(0.022) (0.017) (0.273) (0.272) (0.201) (0.074)

(b) Ln(Time to big town) -0.726 -0.461 -1.234 -1.112 -0.389 -1.202(0.240) (0.459) (0.472) (0.465) (0.710) (0.268)

(c) (a)×(b) -0.414 -0.621 -0.566 -1.035*** 0.948*(0.123) (0.332) (0.397) (0.009) (0.079)

Constant 5.906*** 5.982*** 5.948*** 8.958*** 4.817** 5.291* 8.550***(0.000) (0.000) (0.000) (0.001) (0.048) (0.098) (0.001)



26

Table 6: D-I-D estimates of the impact of road improvements on maize yield(kg/ha), Kenya (2004-2012)

Dependent Variable: ln(Maize yield (kg/ha))

(1) (2) (3) (4) (5) (6) (7)VARIABLES 0 - 84 85 - 165 166 - 305 Over 306

(a) Ln(Time to market) -0.086 0.057 0.360 0.607 0.005 -0.531**(0.671) (0.794) (0.687) (0.277) (0.975) (0.032)

(b) Ln(Time to big town) -1.490*** -1.567*** 2.550 -3.045** -0.258 -2.050**(0.006) (0.005) (0.160) (0.011) (0.557) (0.021)

(c) (a)×(b) 0.336 -0.019 0.122 -0.093 0.717(0.340) (0.978) (0.856) (0.566) (0.128)

Constant 5.156*** 5.469*** 5.536*** 3.636* 3.233 7.025*** 4.334***(0.000) (0.000) (0.000) (0.095) (0.358) (0.000) (0.008)



27

Table 7: D-I-D estimates of the impact of road improvements on participation in milk market: Kenya,2004-2012

Dependent Variable: Milk seller ==1 if sold milk

(1) (2) (3) (4) (5) (6) (7)VARIABLES 0 - 84 85 - 165 166 - 305 Over 306

(a) Ln(Time to market) -0.008 0.007 -0.035 -0.097 0.051 -0.056(0.893) (0.919) (0.826) (0.429) (0.606) (0.482)

(b) Ln(Time to big town) -0.342** -0.342** 0.054 -0.200 -0.575** 0.409(0.035) (0.040) (0.949) (0.527) (0.048) (0.205)

(c) (a)×(b) -0.012 -0.016 -0.210 0.089 0.249*(0.856) (0.940) (0.214) (0.280) (0.071)

Constant -0.084 -0.014 -0.017 -0.719 -0.015 0.691 0.706(0.797) (0.964) (0.955) (0.203) (0.980) (0.355) (0.378)



28

Table 8: D-I-D estimates of the impact of road improvements on proportion of milksold: Kenya, 2004-2012

Dependent Variable: Proportion of milk sold

(1) (2) (3) (4) (5) (6) (7)VARIABLES 0 - 84 85 - 165 166 - 305 Over 306

(a) Ln(Time to market) -0.044 -0.037 -0.048 -0.128 -0.038 -0.002(0.199) (0.307) (0.674) (0.156) (0.432) (0.950)

(b) Ln(Time to big town) -0.326*** -0.300*** -0.254 -0.049 -0.497** 0.106(0.002) (0.004) (0.614) (0.705) (0.030) (0.496)

(c) (a)×(b) -0.040 -0.054 -0.096 -0.048 0.094(0.342) (0.751) (0.339) (0.305) (0.133)

Constant -0.139 -0.080 -0.083 -0.044 -0.197 0.109 0.397(0.507) (0.685) (0.677) (0.927) (0.583) (0.779) (0.220)



29

Table 9: D-I-D estimates of the impact of road improvements on participation in maize market: Kenya, 2004-2012

Dependent Variable: Maize seller==1

(1) (2) (3) (4) (5) (6) (7)VARIABLES 0 - 84 85 - 165 166 - 305 Over 306

(a) Ln(Time to market) 0.011 -0.004 0.099 -0.055 -0.115 -0.049(0.862) (0.956) (0.691) (0.750) (0.345) (0.692)

(b) Ln(Time to big town) 0.658*** 0.641*** 1.888*** -0.192 0.879** 0.756**(0.000) (0.000) (0.008) (0.664) (0.023) (0.024)

(c) (a)×(b) 0.079 0.023 0.059 0.004 0.413***(0.312) (0.894) (0.740) (0.963) (0.005)

Constant 0.570 0.435 0.451 0.971* -0.944 0.108 0.963*(0.100) (0.180) (0.162) (0.088) (0.140) (0.825) (0.072)



30

Table 10: D-I-D estimates of the impact of road improvements on proportion of maize sold: Kenya, 2004-2012

Dependent Variable: Proportion of maize sold

(1) (2) (3) (4) (5) (6) (7)VARIABLES 0 - 84 85 - 165 166 - 305 Over 306

(a) Ln(Time to market) -0.012 -0.008 0.100 -0.028 -0.049 -0.055(0.765) (0.848) (0.408) (0.790) (0.466) (0.480)

(b) Ln(Time to big town) 0.243** 0.230** 0.698** -0.164 0.376 0.345(0.016) (0.017) (0.028) (0.763) (0.119) (0.113)

(c) (a)×(b) 0.070 0.050 0.293 0.034 0.019(0.155) (0.625) (0.144) (0.642) (0.816)

Constant 0.374 0.320 0.335 0.506* -0.595 0.133 0.560(0.162) (0.216) (0.190) (0.066) (0.490) (0.733) (0.176)



31

Table 11: Tobit FE (Honoré, 1992) of the impact of road improvements on various outcomes of interest

Dependent Variable: See table notes

(1) (2) (3) (4) (5) (6)VARIABLES landhyvprop lnqfertnpk lnqmanure lnyieldmaize milkpropsold maizeratio

(a) Ln(Time to market) -0.080 -0.131 -0.469** 0.058 -0.072 -0.054(0.063) (0.486) (0.017) (0.728) (0.236) (0.736)

(b) Ln(Time to big town) -0.372** -0.296 -0.461 -1.567*** -0.440*** 0.469*(0.151) (0.572) (0.446) (0.000) (0.000) (0.071)

(c) (a)×(b) 0.240*** 0.243 -0.414 0.344 -0.066 0.189(0.071) (0.317) (0.118) (0.176) (0.292) (0.124)

Observations 3,087 3,256 3,256 3,256 1,430 1,430HH FE Yes Yes Yes Yes Yes YesYear × Agrozone Yes Yes Yes Yes Yes Yes

Notes:Standard errors in parentheses2). Significance level: *** p<0.01, ** p<0.05, * p<0.13). Time distance in minutes by motor vehicle; Robust standard errors clustered at community level; All equations include year ×agrozone; Controls include household and village level regressors as described in data section.4). Column (1)-Proportion of land under hybrid maize, Column(2)-ln(NPK equivalent inorganic fertilizer(kg/ha)), Column(3)-ln(Manure intensification (kg/ha)), Column(4)- ln(Maize yield (kg/ha)) Column (5)-Proportion of milk sold, Column (6)- Proportionof maize sold.

32

Table 12: D-I-D estimates of the impact of road improvements when interacted with community mobile network availability

Dependent Variable: See table notes

(1) (2) (3) (4) (5) (6)VARIABLES landhyvprop lnqfertnpk lnqmanure lnyieldmaize milkpropsold maizeratio

Time to market× network availability -0.039 -0.171 -0.155 -0.054 -0.003 -0.043*(0.380) (0.152) (0.242) (0.744) (0.918) (0.076)

Time to big town× network availability -0.148** -0.151 -0.021 -0.888*** -0.020 0.055(0.016) (0.552) (0.950) (0.002) (0.727) (0.283)

Time interaction× network availability 0.179** -0.053 0.207 0.116 -0.010 0.051(0.013) (0.801) (0.439) (0.675) (0.835) (0.285)

Constant 0.230 1.640* 5.912*** 5.557*** -0.137 0.361(0.406) (0.063) (0.000) (0.000) (0.499) (0.184)

Observations 3,087 3,256 3,256 3,256 1,430 1,430R-squared 0.130 0.086 0.036 0.233 0.074 0.063Number of hhid 715 715 715 715 715 715HH FE Yes Yes Yes Yes Yes YesYear × Agrozone Yes Yes Yes Yes Yes Yes

Notes:1). Reporting pval in parentheses2). Significance level: *** p<0.01, ** p<0.05, * p<0.13). Time distance in minutes by motor vehicle; Robust standard errors clustered at community level; All equations include year × agrozone; Controlsinclude household and village level regressors as described in data section.4). Column (1)-Proportion of land under hybrid maize, Column(2)-ln(NPK equivalent inorganic fertilizer(kg/ha)), Column(3)-ln(Manure intensifi-cation (kg/ha)), Column(4)- ln(Maize yield (kg/ha)) Column (5)-Proportion of milk sold, Column (6)- Proportion of maize sold.

33

Sour

ce:

Aut

hors

depi

ctio

n.

Fig

ure

1:M

ap

of

the

study

area

,R

ePEAT

2004

-201

2.

34

Fig

ure

2:V

isib

leim

proved

road

inK

arac

huonyo

Dis

tric

t.

Left

phot

osh

ows

the

road

in20

05whi

leth

eri

ghtph

oto

show

sth

ero

adin

2009

.M

apco

urte

syof

Goo

gle

R ©M

aps(0◦20′ 29.66

”S34◦ 48′41.26”

E)

35

References

African Development Bank. (2010, Sept). Infrastructure deficit and opportunities inAfrica (Economic Brief No. 1). Chief Economist Complex: The African DevelopmentBank Group.

Banerjee, A., Duflo, E., & Qian, N. (2012). On the road: access to transportationinfrastructure and economic growth in China [NBER Working Papers] (Tech. Rep.No. 17897). National Bureau of Economic Research, Inc.

Briceño-Garmendia, C. M., & Shkaratan, M. (2011). Kenya’s infrastructure: A con-tinental perspective. The World Bank, Africa Region: Sustainable DevelopmentDepartment .

Burgess, R., Jedwab, R., Miguel, E., Morjaria, A., & Padró i Miquel, G. (2013,September). The value of democracy: Evidence from road building in Kenya (CEPRDiscussion Papers No. 9627). C.E.P.R. Discussion Papers. Retrieved from http://ideas.repec.org/p/cpr/ceprdp/9627.html

Collier, P., & Dercon, S. (2009). African agriculture in 50 years: smallholders in arapidly changing world? In How to feed the world in 2050. proceedings of a technicalmeeting of experts, rome, italy, 24-26 june 2009. (pp. 1–13). Food and AgricultureOrganization of the United Nations (FAO).

de Vera Garcia, E. (1984). The Impact of Rural Roads. Philippine Journal of Devel-opment , XI (1).

Dercon, S., Gilligan, D. O., Hoddinott, J., & Woldehanna, T. (2007). The impactof roads and agricultural extension on consumption growth and poverty in fifteenEthiopian villages [CSAE Working Paper Series] (Tech. Rep. Nos. 2007–01). Centrefor the Study of African Economies, University of Oxford.

Donaldson, D. (2010, October). Railroads of the Raj: Estimating the impact of trans-portation infrastructure (NBER Working Papers No. 16487). National Bureau ofEconomic Research, Inc. Retrieved from http://ideas.repec.org/p/nbr/nberwo/16487.html

Dorosh, P., Wang, H. G., You, L., & Schmidt, E. (2012). Road connectivity, population,and crop production in Sub-Saharan Africa. Agricultural Economics , 43 (1), 89–103.

Escobal, J., & Ponce, C. (2003). The benefits of rural roads. Enhancing income oppor-tunities for the rural Poor [Documentos de Trabajo] (Tech. Rep.). Grupo de Análisispara el Desarrollo (GRADE): Updated 2009.

36

http://ideas.repec.org/p/cpr/ceprdp/9627.html

http://ideas.repec.org/p/cpr/ceprdp/9627.html

http://ideas.repec.org/p/nbr/nberwo/16487.html

http://ideas.repec.org/p/nbr/nberwo/16487.html

Estache, A. (2010). A survey of impact evaluations of infrastructure projects, pro-grams and policies [Working Papers ECARES] (Tech. Rep. No. 2010_005). ULB–Universite Libre de Bruxelles. Retrieved from http://ideas.repec.org/p/eca/wpaper/2010_005.html

Fafchamps, M., & Shilpi, F. (2009). Determinants of the choice of migration desti-nation (CSAE Working Paper Series No. 2009-09). Centre for the Study of AfricanEconomies, University of Oxford. Retrieved from http://ideas.repec.org/p/csa/wpaper/2009-09.html

Fan, S., & Zhang, X. (2004). Infrastructure and regional economic development inrural China. China Economic Review , 15 (2), 203–214.

Freeman, H. A., & Kaguongo, W. (2003). Fertilizer market liberalization and privateretail trade in Kenya. Food Policy , 28 (5-6), 505–518.

Gachassin, M. C. (2013). Should i stay or should i go? the role of roads in migrationdecisions. Journal of African Economies , 22 (5), 796-826.

Gibson, J., & Rozelle, S. (2003). Poverty and access to roads in Papua New Guinea.Economic Development and Cultural Change, 52 (1), 159–185.

Gollin, D., & Rogerson, R. (2014). Productivity, transport costs andsubsistence agriculture. Journal of Development Economics , 107 (0), 38 -48. Retrieved from http://www.sciencedirect.com/science/article/pii/S0304387813001508 doi: http://dx.doi.org/10.1016/j.jdeveco.2013.10.007

Honoré, B. E. (1992). Trimmed LAD and least squares estimation of truncated andcensored regression models with fixed effects. Econometrica, 60 (3), 533–65.

Jacoby, H. G. (2000). Access to markets and the benefits of rural roads. EconomicJournal , 110 (465), 713–737.

Jacoby, H. G., & Minten, B. (2009). On measuring the benefits of lower transportcosts. Journal of Development Economics , 89 (1), 28-38.

Jalan, J., & Ravallion, M. (1998). Are there dynamic gains from a poor-area develop-ment program? Journal of Public Economics , 67 (1), 65–85.

Jalan, J., & Ravallion, M. (2002). Geographic poverty traps? A micro model ofconsumption growth in rural China. Journal of Applied Econometrics , 17 , 329–346.

37

http://ideas.repec.org/p/eca/wpaper/2010_005.html

http://ideas.repec.org/p/eca/wpaper/2010_005.html

http://ideas.repec.org/p/csa/wpaper/2009-09.html

http://ideas.repec.org/p/csa/wpaper/2009-09.html

http://www.sciencedirect.com/science/article/pii/S0304387813001508

http://www.sciencedirect.com/science/article/pii/S0304387813001508

Khandker, S. R., Bakht, Z., & Koolwal, G. B. (2006). The poverty impact of ruralroads : Evidence from Bangladesh [Policy Research Working Paper Series] (Tech.Rep.). The World Bank.

Kingombe, C. K., & di Falco, S. (2012). The impact of a feeder road project oncash crop production in Zambia’s Eastern province between 1997 and 2002 [IHEIDWorking Papers] (Tech. Rep. Nos. 04–2012). Economics Section, The GraduateInstitute of International Studies.

Lokshin, M., & Yemtsov, R. (2003). Evaluating the impact of infrastructure rehabilita-tion projects on household welfare in rural Georgia [Policy Research Working PaperSeries] (Tech. Rep. No. 3155). The World Bank.

Matsumoto, T., & Yamano, T. (2009, December). Soil fertility, fertilizer, and the maizegreen revolution in East Africa [Policy Research Working Paper Series] (Tech. Rep.No. 5158). The World Bank.

Ministry of Roads. (2011). Roads investment programme 2010-2024 (1st ed., Vol. 1).Kenya Roads Board, Government of Kenya.

Mu, R., & van de Walle, D. (2011). Rural roads and local market development inVietnam. The Journal of Development Studies , 47 (5), 709–734.

Oosterhaven, J., & Knaap, T. (2003). Spatial economic impacts of Transport infras-tructure investments. In A. Pearman, P. Mackie, & J. Nellthorp (Eds.), TransportProjects, Programmes and Policies: Evaluation, Needs and Capabilities (pp. 87–105).Burlington, VT: Ashgate Publishing, Ltd.

Otsuka, K., & Yamano, T. (2005). The possibility of a green revolution in Sub-Saharan Africa: Evidence from Kenya. The Electronic Journal of Agricultural andDevelopment Economics , 2 (1), 7–19.

Ravallion, M. (1999, July). The mystery of the vanishing benefits : Ms. Speedy Analyst’sintroduction to evaluation [Policy Research Working Paper Series] (Tech. Rep. No.2153). The World Bank.

Republic of Kenya. (2007). Kenya Roads Act, 2007. The Kenya Gazzete.

Republic of Kenya. (2008). Kenya Vision 2030. (Ministry of planning and nationaldevelopment, Ed.). Nairobi: Government Printer.

Rostow, W. (1962). The process of economic growth. 2nd Ed. Oxfrod University Press.

38

Shiferaw, A., Söderbom, M., Siba, E., & Alemu, G. (2013). Road infrastructureand enterprise dynamics in Ethiopia (Working Papers). Department of Economics,College of William and Mary.

Staal, S. K., & ILRI. (2001). Dairy systems characterization of the greater Nairobimilk shed. Smallholder Dairy Project (SDP (R&D)), MALDM, KARI, ILRI.

Stifel, D., & Minten, B. (2008). Isolation and agricultural productivity. AgriculturalEconomics , 39 (1), 1–15.

The World Bank. (2007). A decade of action in transport : An evaluation of WorldBank assistance to the transport sector, 1995-2005 (No. 6695). The World Bank.

United Nations. (2008). Handbook on the least developed country category: Inclusion,graduation and special support measures. The United Nations Publishing Section,New York.

van de Walle , D. (2009). Impact evaluation of rural road projects. The Journal ofDevelopment Effectiveness , 1 (1), 15–36.

van de Walle, D., & Mu, R. (2011). Rural roads and local market development inVietnam. The Journal of Development Studies , 47 (5), 709–734.

Waithaka, M., Nyangaga, J., Staal, S., Wokabi, D., A.W.and Njubi, Muriuki, K.,Njoroge, L., & Wanjohi, P. (2002). Characterization of dairy systems in the WesternKenya Region.

39

A. Appendix

Table A.1: Speeds assigned to road sections (Km/h)

Road condition

Excellent Good Fair Poor Very poor Under construction

Surface typePremix 80 70 50 35 30 20Surface Dressing 80 70 50 35 30 20Brick 80 70 50 35 30 20Concrete 80 70 50 35 30 20Set Stone 80 70 50 35 30 20Concrete Block 80 70 50 35 30 20Gravel 50 50 35 20 5 5Sand 50 50 35 20 5 5Earth 40 30 25 10 5 5Natural 40 30 20 5 5 5Track 5 5 4 3 3 3Unknown 5 5 4 3 3 3Source: Traffic Act, Kenya and from authors projections

40

Table A.2: Channels under which rural-roads stir economic develop-ment

Expected Effects

Channel Temporary Permanent

Direct Via markets: Construction Exploitation and timesaving

External effects: Environmental Environmental, safetyetc.

Indirect Via demand: Backward expenditure Backward expenditureVia supply: Crowding-out Productivity and loca-

tion etc.External effects: Indirect emissions Indirect emissions etc.

Adapted from Oosterhaven & Knaap (2003)

41

Csae2014 293 road affects to production

Engineering

Transcript of Csae2014 293 road affects to production