ADOPTION OF PRODUCTIVITY-ENHANCING INPUTS AND IMPROVED 

FARM PRACTICES IN CAMBODIA’S RICE PRODUCTION 

 
 
 
 

 
By 

Socheat Keo 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

A DISSERTATION 

Submitted to 

Michigan State University 

in partial fulfillment of the requirements 

for the degree of 

Community, Agriculture, Recreation and Resource Studies – Doctor of Philosophy 

2018 

 

 

 

 

 

 

ABSTRACT 

ADOPTION OF PRODUCTIVITY-ENHANCING INPUTS AND IMPROVED 

FARM PRACTICES IN CAMBODIA’S RICE PRODUCTION 

 
By 

Socheat Keo 

Agricultural intensification, particularly, the adoption of improved farm technologies has 

been considered by the government of Cambodia as a driving force for agricultural development, 

which will contribute to improving living standards, particularly in rural areas. Meanwhile, little 

research has empirically analyzed the factors that influence farmers’ decisions to adopt improved 

farm technologies. Hence, this dissertation aims at adding to this literature and contributing to 

Cambodia’s agricultural development policies through three empirical studies with a focus on the 

rice production – the dominant sector of Cambodia’s agriculture. 

The first paper of this dissertation investigates the impact of formal and semi-formal land 

titles on the adoption of chemical fertilizer and manure in Cambodia’s paddy production using 

pooled cross-sectional data of the Cambodia Socio-Economic Survey (CSES) from 2009 to 

2014. Propensity Score Matching (PSM) and regressions on the matched samples were used to 

estimate the effects for each type of land title, controlling for village heterogeneity. The 

empirical findings show that both formal and semi-formal land titling significantly increases the 

average adoption rates of chemical fertilizer and manure, but the impact of formal land titling on 

manure adoption is higher than that of semi-formal titling. However, the empirical evidence 

shows that land titling does not significantly increase fertilizer expenditure and productivity. In 

general, this study suggests that formal and semi-formal land titling are equally conducive to 

fertilizer use and productivity improvement.  

The second paper addresses two key issues: first, it examines whether farmers’ decisions 

to adopt improved rice varieties and chemical fertilizer are interrelated (interdependent); second, 

it analyzes the determinants of the improved farm technology adoption. The quantitative data is 

based on the HARVEST (Helping Address Rural Vulnerabilities and Ecosystem Stability) 

household panel survey (2012-2016) in four provinces of Cambodia, while the qualitative data 

was collected from 25 semi-structured interviews with some of the surveyed households. The 

study applies a bivariate probit model to test for the interdependence of technology adoption, and 

the correlated random effects (CRE) framework to detect the determinants of adoption. The 

results indicate that adoption of an improved rice variety and chemical fertilizer at the plot level 

are complementary. The empirical results further suggest that irrigation, social learning in the 

form of information from neighbors, age of household head, secondary education, TV ownership 

(as a means of accessing the media), and remittances are positively associated with the adoption 

of improved farm technologies.  

The third paper follows up on the second one by further examining the role of credit in 

the adoption of the interrelated inputs. Propensity Score Matching (PSM) and regressions on the 

matched samples were applied to examine the linkage between agricultural credit and the 

adoption of the interrelated inputs. This study relies on a cross-sectional survey from the Census 

of Agriculture of Cambodia (CAC), 2013. The results suggest that credit for agricultural 

activities increases the probability of adoption of high-yield rice variety, fertilizer, pesticide (or 

herbicide/fungicide) and the combination of the three types of modern input. The effect of credit 

on adoption of pesticides is the most robust, particularly when farm households contract loans 

from both formal and informal sources. Our empirical finding suggests that affordable credit for 

farm activities increases adoption of modern inputs in rice production. 

 

Copyright by  
SOCHEAT KEO 
2018 
 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ACKNOWLEDGEMENTS 

 

Doing this PhD, particularly writing this dissertation has been a long and demanding 

process. Therefore, I would like to take this opportunity to extend my gratitude to the people and 

institutions that have contributed to my work. 

First of all, my most profound and sincere thanks are extended to my advisor, Dr. John 

Kerr, for his tremendous support and continuing technical guidance. Without his contributions 

and guidance throughout my PhD program, I would not be able to achieve the goal in my studies 

in the United States. Additionally, I appreciate his kind understanding and his patience to 

supervise me. He is the most inspiring and helpful advisor in my academic life. 

My profuse thanks are also due to the other members of my PhD guidance committee: 

Dr. Robert Richardson, Dr. Mywish Maredia and Dr. Kimberly Chung. Their support and ideas 

have also helped me navigate my PhD. research process. In addition, I wish to thank my local 

mentor, Dr. Kimsun Tong, for his technical supports during my dissertation writing. 

Above all else, I am indebted to the U.S government for financially supporting my PhD 

program which is based upon work supported by the United States Agency for International 

Development, as part of the Feed the Future initiative, under the CGIAR Fund, award number 

BFS-G-11-00002, and the predecessor fund the Food Security and Crisis Mitigation II grant, 

award number EEM-G-00-04-00013. 

I am also grateful to the staff and the management team of the Borlaug Higher Education 

for Agricultural Research Development (BHEARD) program for their assistance. Additionally, 

my sincere thanks go to the faculty and staff of the Department of Community Sustainability, 

College of Agriculture of the Michigan State University for their support in facilitating my study. 

 

v 

Thanks go also to the management team of Cambodia Development Resource Institute (CDRI) 

for providing me with a decent office space to write my dissertation in Cambodia. 

Finally, my heartfelt thanks also go to my family, parents and friends for spiritual support 

and understanding, without which pursuing this PhD would be even more challenging. 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

vi 

TABLE OF CONTENTS 

 

LIST OF TABLES ......................................................................................................................... ix 

LIST OF FIGURES ....................................................................................................................... xi 

CHAPTER 1: INTRODUCTION ................................................................................................... 1 
REFERENCES ............................................................................................................................... 6 

CHAPTER 2: IMPACT OF LAND TITLING ON THE ADOPTION OF PRODUCTIVITY-  
                     ENHANCING INPUTS IN CAMBODIA’S RICE PRODUCTION ........................ 8 
2.1. Introduction .............................................................................................................................. 8 
2.2. Literature review .................................................................................................................... 11 
2.2.1. Administrative capacity of land institutions ............................................................... 14 
2.2.2. Socio-economic context .............................................................................................. 14 
2.3. An overview of the country context ....................................................................................... 15 
2.3.1. Land title in Cambodia ................................................................................................ 15 
2.3.1.1. Formal land registration (land title issued by the government) ........................... 16 
2.3.1.2. Semi-formal land registration (land title issued by the local authority) .............. 18 
2.3.2. Cambodia’s rice sector ................................................................................................ 19 
2.4. Conceptual framework and empirical methodology .............................................................. 21 
2.4.1. Conceptual framework ................................................................................................ 21 
2.4.2. Empirical methodology ............................................................................................... 23 
2.4.2.1. Steps in propensity score matching ...................................................................... 25 
2.4.2.2. Assumptions and limitations of PSM .................................................................. 27 
2.4.2.3. Regressions on the matched samples ................................................................... 28 
2.5. Data and descriptive statistics ................................................................................................ 29 
2.5.1. Background on the Cambodia Socio-Economic Survey (CSES) ............................... 29 
2.5.2. Definitions and summary of outcome variables and covariate ................................... 31 
2.6. Empirical results and discussions .......................................................................................... 33 
2.6.1. Getting the matched samples ...................................................................................... 33 
2.6.2. Regressions on the matched samples .......................................................................... 36 
2.7. Conclusion and policy implications ....................................................................................... 41 
APPENDICES .............................................................................................................................. 44 
Appendix A: Extra tables and figures for the empirical analysis ............................................. 45 
Appendix B: Choosing the matching technique or algorithm .................................................. 67 
REFERENCES ............................................................................................................................. 69 

CHAPTER 3: ADOPTION OF INTERRELATED FARM TECHNOLOGIES IN CAMBODIA:   
                     THE CASE OF IMPROVED RICE VARIETIES ................................................... 75 
3.1. Introduction ............................................................................................................................ 75 
3.2. Literature review .................................................................................................................... 77 
3.3. Conceptual framework ........................................................................................................... 79 
3.4. Empirical methodology .......................................................................................................... 82 

 

vii 

3.4.1. Multivariate probit model ........................................................................................... 82 
3.4.2. Correlated random effects (CRE) model .................................................................... 84 
3.4.3. Qualitative fieldwork .................................................................................................. 86 
3.5. Data and descriptive statistics ................................................................................................ 87 
3.6. Empirical results .................................................................................................................... 93 
3.7. Conclusions and policy implications ................................................................................... 101 
REFERENCES ........................................................................................................................... 104 

CHAPTER 4: THE LINKAGE BETWEEN AGRICULTURAL CREDIT AND ADOPTION OF   
                     INTERRELATED MODERN INPUTS IN CAMBODIA’S RICE         
                     PRODUCTION ...................................................................................................... 109 
4.1. Introduction .......................................................................................................................... 109 
4.2. Literature review .................................................................................................................. 112 
4.3. Country context .................................................................................................................... 113 
4.3.1. Cambodia’s agriculture ............................................................................................. 113 
4.3.2. Cambodia’s rural credit ............................................................................................. 115 
4.4. Conceptual framework and empirical methodology ............................................................ 117 
4.4.1. Conceptual framework .............................................................................................. 117 
4.4.1.1. Adoption of interrelated inputs .......................................................................... 117 
4.4.1.2. Use of credit in agricultural activities ................................................................ 118 
4.4.2. Empirical methodology ............................................................................................. 119 
4.4.2.1. Simultaneous adoption ....................................................................................... 119 
4.4.2.2. Propensity Score Matching (PSM) .................................................................... 120 
4.4.2.3. Regressions on the matched samples ................................................................. 122 
4.5. Data and descriptive statistics .............................................................................................. 123 
4.5.1. Background on Census of Agriculture of Cambodia (CAC) 2013 ........................... 123 
4.5.2. Definitions and summary of outcome variables and covariates ............................... 125 
4.6. Empirical results .................................................................................................................. 126 
4.6.1. Simultaneous adoption of the modern inputs ............................................................ 126 
4.6.2. Getting the matched samples .................................................................................... 127 
4.6.3. Regressions on the matched samples ........................................................................ 130 
4.7. Conclusions .......................................................................................................................... 136 
APPENDIX ................................................................................................................................. 139 
REFERENCES ........................................................................................................................... 164 

	

	
	
	
	
	
	

 

 

 

 

viii 

LIST OF TABLES 

 

Table 1: Number of plots, households and villages by year ......................................................... 31 

Table 2: Number of plots by type of land title and year ............................................................... 31 

Table 3: Summary of PSM quality indicators from 5 nearest-neighbor matching ....................... 34 

Table 4: Number of observations in common support region for each matching ........................ 36 

Table 5: Average outcome variables by group of plots (before matching) .................................. 37 

Table 6: Average outcome variables by group of households (the common support observations 

with propensity score weights) ............................................................................................. 37 

Table 7: Impacts of land titling ..................................................................................................... 39 

Table 8: Definitions of outcome variables and covariates ............................................................ 45 

Table 9: Descriptive statistics for treatment and control plots ..................................................... 47 

Table 10: Test of matching quality (land title from the government versus control group) ......... 51 

Table 11: Test of matching quality (land title from local authority versus control group) .......... 53 

Table 12: Test of matching quality (land title from the government versus local authority) ....... 55 

Table 13: Test of matching quality (general land title versus control group) ............................... 57 

Table 14: Impact of formal land title ............................................................................................ 59 

Table 15: Impact of semi-formal land title ................................................................................... 61 

Table 16: Formal land title versus semi-formal land title ............................................................. 63 

Table 17: Impact of general land title ........................................................................................... 65 

Table 18: Variable definitions and summary statistics by year .................................................... 92 

Table 19: Correlation coefficients for MVP regression equations ............................................... 94 

Table 20: Correlated random effects (CRE) estimation results of the determinants of adoption of 
agricultural technology ......................................................................................................... 95 

Table 21: Correlation coefficients for MVP regression equations ............................................. 126 

Table 22: Summary of PSM quality indicators from 5 nearest-neighbor matching ................... 129 

 

ix 

Table 23: Number of observations in common support region for each matching .................... 130 

Table 24: Average modern input adoption rates by group of households (before matching) .... 131 

Table 25: Average modern input adoption rates by group of households (the common support 

observations that have propensity score weights) ............................................................... 131 

Table 26: Impacts of credit for agriculture ................................................................................. 132 

Table 27: Definitions of outcome variables and covariates ........................................................ 140 

Table 28: Descriptive statistics for households with agricultural credit ..................................... 142 

Table 29: Descriptive statistics for households by different category of agricultural credit ...... 144 

Table 30: Test for matching quality (impact of credit from any sources) .................................. 146 

Table 31: Test for matching quality (impact of getting only formal credit) ............................... 148 

Table 32: Test for matching quality (impact of getting only informal credit) ............................ 150 

Table 33: Test for matching quality (impact of getting both formal and informal credit) ......... 152 

Table 34: Impact of credit for agriculture (from any sources) .................................................... 156 

Table 35: Impact of credit for agriculture (from only formal source) ........................................ 158 

Table 36: Impact of credit for agriculture (from only informal source) ..................................... 160 

Table 37: Impact of credit for agriculture (from both sources) .................................................. 162 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 

x 

LIST OF FIGURES 

 

Figure 1: Conceptual framework .................................................................................................. 22 

Figure 2: Distribution of propensity scores for matching between treatment group (land title from  

        the government) and control group ....................................................................................... 49 

Figure 3: Distribution of propensity scores for matching between treatment group (land title from 
the local authority) and control group ................................................................................... 49 

Figure 4: Distribution of propensity scores for matching between treatment group (land title from 
the government) and land title from the local authority ....................................................... 50 

Figure 5: Distribution of propensity scores for matching between treatment group ( general land 
title) and land title from the local authority .......................................................................... 50 

Figure 6: Distribution of propensity scores for matching between treatment group (HHs with any 
category of agricultural credit) and control group .............................................................. 154 

Figure 7: Distribution of propensity scores for matching between Treatment Group (HHs with 

only formal credit for agriculture) and control group ......................................................... 154 

Figure 8: Distribution of propensity scores for matching between Treatment Group (HHs with 

only informal credit for agriculture) and control group ...................................................... 155 

Figure 9: Distribution of propensity scores for matching between Treatment Group (HHs with 

both types of credit for agriculture) and control group ....................................................... 155 

 

 

xi 

CHAPTER 1: INTRODUCTION 

Though the share of agricultural GDP and labor force may have declined in low-income 

countries, the number of rural households that depend on agriculture as the main source of 

income remains significant (Timmer, 1988; Hazell, Poulton, Wiggins, & Dorward, 2010). In 

general, the sources of growth matter for poverty reduction, with growth in labor-intensive 

sectors such as agriculture having greater poverty-reducing effects than growth in capital-

intensive sectors (Christilansen & Devarajan, 2013; Ravallion, 2001). Further, the two common 

characteristics of agriculture in low-income countries are: (1) agriculture comprises a relatively 

larger share of the total economy; and (2) the productivity is low (Timmer, 1988). Hence, in 

order for a developing or low-income country to achieve a goal of self-sustained growth, there is 

a need for agriculture to shift from subsistence farming to a more productive and commercialized 

agricultural sector. To this end, improving productivity and production practices at the farm level 

is of particular importance for agrarian economies. An approach to this agenda has centered on 

promoting adoption of improved farm technology, which is expected to have a significant impact 

on livelihood improvement and poverty reduction through better income generation (Minten & 

Barrett, 2008; Ali & Abdulai, 2010). 

Cambodia’s agriculture is not an exception to the challenges that other low-income 

economies are facing. Some major constraints on Cambodia’s agricultural productivity include 

limited use of cultivation techniques, and appropriate use of inputs (RGC, 2013). Rice is the 

main crop of Cambodia’s agriculture because it covers 75 percent of the total cultivated area 

(FAO, 2014) and has a significant share (around one third) of total agricultural output (NIS, 

2011).  The government promotes the adoption of improved farm technologies (irrigation, 

fertilizer and improved seeds) which are considered one of the driving forces for agricultural 

 

 

1 

production, especially in the rice sector through productivity improvement (RGC, 2010)  In line 

with this, the empirical study by Yu & Fan (2011) also suggests that there is a great potential for 

Cambodia to enhance rice productivity by expanding the adoption of modern technology and 

modern inputs, and by improving rural infrastructure, i.e. irrigation and transportation. 

However, the research on the adoption of agricultural technology in Cambodia has not 

been extensive, though Cambodia’s agriculture is instrumental to rural livelihood improvement. 

Hence, motivated by the aforementioned gap, this dissertation aims to examine the issues of 

adoption of agricultural technologies among farmers in Cambodia by using econometric analysis 

in order to have better insights than previous literature could provide. More specifically, the 

dissertation takes the form of three empirical research papers intending to understand the 

household’s decision to adopt improved farm technologies, e.g. fertilizer and improved seed 

varieties. 

The first paper, “The Impact of Land Titling on the Adoption of Productivity-enhancing 

Inputs in Cambodia’s Rice Production,” estimates the impact of formal and semi-formal land 

titles on the adoption of chemical fertilizer, manure and productivity in Cambodia’s paddy 

production.  It uses Propensity Score Matching (PSM) and regressions on the matched samples 

to estimate the effects for each type of land title, controlling for village heterogeneity. The study 

is based on pooled cross-sectional data from a nationally representative household survey – the 

Cambodia Socio-Economic Survey (CSES) conducted in 2009, 2010, 2011, 2012, 2013 and 2014 

by Cambodia’s National Institute of Statistics (NIS). 

The empirical results indicate that both types of land title have a positive and significant 

impact on the likelihood of fertilizer and manure adoption. However, the impact of formal land 

titling on manure adoption is higher than that of semi-formal titling. Further, neither type of land 

 

 

2 

titling has a significant effect on fertilizer expenditure and yield. The fact that it does not detect 

any effect of land tenure status on yield is puzzling and may relate to the fact that rice is a short-

term crop and fertilizer expenditure is also a short-term investment, and land titling is more 

important for longer term investments. Regarding the policy implications, the lack of significant 

difference between formal and semi-formal title on fertilizer use implies that land users are 

unlikely to apply for the more expensive formal title except when a systematic program is in 

place to reduce its cost. The government may wish to seek to reduce the cost of land transfer and 

registration through the formal system more widely. This is because the formal land titling 

enables the government to have an up-to-date cadastral system, which is an essential legal basis 

for handling potential land conflicts. 

The second paper, “Adoption of Interrelated Farm Technologies in Cambodia: The Case 

of Improved Rice Varieties and Chemical Fertilizer,” examines whether farmers’ decisions to 

adopt improved rice varieties and chemical fertilizer are interrelated (interdependent), and 

analyze the factors that determine decisions to adopt those improved farm technologies. The 

study is based on plot-level panel data from the HARVEST household panel survey (2012-2016) 

in four provinces in Tonle Sap region – one of the four agro-ecological zones in Cambodia. Also, 

it uses qualitative data from 25 semi-structured interviews which were conducted with some of 

the surveyed households in order to assist with the interpretation and enrich the discussion of the 

empirical findings.  

The results suggest that adoption of an improved rice variety and chemical fertilizer at the 

plot level are complementary. Further, there is empirical evidence that the probability of 

adopting improved farm technologies is positively associated with various factors: irrigation, 

social learning in the form of information from neighbors, age of household head, secondary 

 

 

3 

education, TV ownership (as a means of accessing the media), and remittances. At the same 

time, our results show that the probability of adoption is negatively related with land size of a 

household. These findings provide a better understanding about the design and implementation 

of policy intervention to promote the adoption of improved farm technologies in Cambodia. 

The third paper, “The Linkage between Agricultural Credit and Adoption of Interrelated 

Modern Inputs in Cambodia’s Rice Production,” follows up on the second paper by further 

exploring the role of credit in the adoption of interrelated inputs (high-yield variety, chemical 

fertilizer, and pesticide, herbicide or fungicide). It applies Propensity Score Matching (PSM) and 

regressions on the matched samples to examine the linkage between agricultural credit and the 

adoption of the interrelated inputs. This study relies on a cross-sectional survey from Census of 

Agriculture of Cambodia (CAC) 2013, conducted by Cambodia’s National Institute of Statistics 

(NIS). 

The results indicate that credit for agricultural activities increases the probability of 

adoption of high-yield rice variety, fertilizer, pesticide (or herbicide/fungicide) and the 

combination of the three types of modern input. Additionally, we also find that the high-yield 

rice variety, fertilizer and pesticide or herbicide or fungicide are all complements. Nonetheless, 

the impact on high-yield rice variety adoption is not robust. The effect of credit on adoption of 

pesticide (or herbicide/fungicide) is the most robust, particularly when farm households contract 

loans from both formal and informal sources. Further, the linkage between credit and adoption of 

the combination of the three types of modern inputs is positive and statistically significant. Our 

empirical finding suggests that the affordable credit for farm activities increases adoption of 

modern inputs.  

 

 

4 

 

These three papers add to the literature on improved farm technology adoption in Cambodia, 

and  the  findings  serve  as  inputs  for  the  design  and  implementation  of  policy  intervention  to 

promote the adoption of agricultural technology in Cambodia. The key message from the first 

essay is that formal land titling does not play a more important role than semi-formal titling in 

fertilizer use and productivity of Cambodia’s rice production. Besides, the important implication 

from essay two is that in addition to irrigation and other significant factors, farmers’ access to 

information and their knowledge related to rice production technology also have a positive and 

significant linkage with farmer’s decision to adopt the complementary improved farm technologies 

in Cambodia’ rice farming. Finally, the main message of the third paper is that access to affordable 

credit for farm activities has a significant role in promoting adoption of agricultural technology. 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5 

REFERENCES 

	

6 

REFERENCES 

 

Ali, A., & Abdulai, A. (2010). The Adoption of Genetically Modified Cotton and Poverty 

Reduction in Pakistan. Journal of Agricultural Economics, 61(1), 175–192. 

Christiaensen, L., & Devarajan, S. (2013). Making the Most of Africa ’ s Growth. Current 

History, 112(754), 181–187. 

Hazell, P., Poulton, C., Wiggins, S., & Dorward, A. (2010). The Future of Small Farms: 

Trajectories and Policy Priorities. World Development, 38(10), 1349–1361. 

NIS -National Institute of Statistics. (2011). National Accounts of Cambodia 1993-2010. Phnom 

Penh: NIS. 

Ravallion, M. (2001). Growth, inequality and poverty: Looking beyond averages. World 

Development, 29(11), 1803–1815. 

RGC-Royal Government of Cambodia. (2010). Policy Document on Promotion of Paddy Rice 

Production and Export of Milled Rice. Phnom Penh. 

RGC-Royal Government of Cambodia. (2013). Rectangular Strategy-Phase III, 2014-2018. 

Phnom Penh. 

Timmer, P. (1988). The Agricultural Transformation. In Handbook of Development Economics 

(Vol. I, pp. 275–331). Amsterdam: North-Holland. 

Yu, B., & Fan, S. (2011). Rice Production Response in Cambodia. Agricultural Economics, 

42(3), 437–450. 

 
 

 

	

	

	

	

7 

        PRODUCTIVITY-ENHANCING INPUTS IN CAMBODIA’S      

CHAPTER 2: IMPACT OF LAND TITLING ON THE ADOPTION OF  
 
                     RICE PRODUCTION  
 

2.1. Introduction  

Land as a factor of agricultural production is one of the most important assets. In this 

regard, land tenure security has been of interest to policy makers and researchers because of its 

expected positive linkage with productivity-improving measures such as farm investments and 

greater use of variable inputs (Abdulai, Owusu, & Goetz, 2011; Feder, Onchan, Chanlamwong, 

& Hongladarom, 1988). The aforementioned linkage has been well documented in literature, 

much of which has shown positive impacts of land title on agricultural performance such as 

investment and adoption of improved technology. On the other hand, many studies have not 

shown this relationship. In other words, empirical evidence does not always suggest that land 

title has significant impact on agricultural production. For example, Besley (1995), Goldstein & 

Udry (2008), Deininger & Jin (2006), Minten & Barrett (2008),  Abdulai et al. (2011), Kassie, 

Jaleta, Shiferaw, Mmbando, & Mekuria (2013), Teklewold, Kassie, & Shiferaw (2013), Nguyen, 

Bauer, & Grote (2016) found a positive effect of land rights on land investment and adoption of 

improved farm technology, whereas others did not (see, for instance, Pender & Kerr, 1999; Place 

& Otsuka, 2002; Brasselle, Gaspart, & Platteau, 2002). In general, under certain circumstances 

land title can play a critial role in providing land security and serving as collateral for credit, but 

emprical evidence shows that the importance of this role varies depending on the specific 

conditions prevailing under any given sitution (Place, 2009).    

In Cambodia, after the collapse of genocidal regime in 1979, the government did not 

provide households with any legal documents to certify ownership of their cultivated land. They 

had only rights to residential land while cultivated land was owned and managed collectively. 

	

8 

However, since 1989 it has been possible to obtain land title through sporadic land registration, a 

process by which individual landowners choose to register for land titling.  They can get the 

certificate if they meet all conditions required by law, but it is quite costly in both money and 

time for rural landholders to apply for land title by themselves. Meanwhile, there was a 

significant increase in conflicts over land rights in the 1990s because of rising demand for arable 

land and an underdeveloped land registry system (Markussen, 2008; Thin, 2012; So, 2009). To 

address this problem and speed up land registration in the country, systematic land titling was 

introduced in 2002 under the government and a World Bank program (Thin, 2012; So, 2009). 

Apart from these approaches to obtain formal title, the routine practice for rural land transfer is 

that people go to the commune office to have their written agreement certified with a signature 

and stamp from a commune chief. This is an alternative, less formal approach to registration that 

is attractive when landowners are not able to afford formal land title (So, 2009). This semi-

formal system provides landowners with a land certificate issued by the local authority, but the 

name of the owner of the transferred land is not recorded in the cadastral registrar of the Ministry 

of Land Management, Urban Planning and Construction (MLMUPC). More details on land 

governance in Cambodia will be presented in section 2.3 of this chapter. 

There have been a number of empirical studies on the causal impact of land title on farm 

investment and productivity in Cambodia. For example, Markussen (2008) used an instrumental 

variable approach using cross-sectional plot-level data from the Cambodia Socio-Economic 

Survey 2004 and found that land title has a significant positive impact on crop productivity. 

Also, Tong (2010) conducted a follow-up study using the Cambodia Socio-Economic Survey 

(CSES) 2007 and the same empirical method, and his results were consistent with Markussen’s 

(2008) findings. Thin (2012) used CSES 2003-2004 to estimate the impact of land rights on land 

	

9 

investment, consumption and child health in the landowner’s household by employing propensity 

score matching and instrumental variable approaches. This study found that the impact of a land 

title was significant and positive for all of the outcomes studied. However, thus far, the impact of 

land tenure and titling on adoption of important variable inputs such as fertilizer and manure has 

yet to be documented in the literature on land tenure in Cambodia.  This is important because 

there is great potential for Cambodia to enhance rice productivity by expanding the use of 

fertilizer and irrigation (Yu & Fan, 2011).   

This study intends to add to the literature on land titling and adoption of fertilizers by 

documenting the impact of land title on the adoption of chemical fertilizer and manure in 

Cambodia. More specifically, the main objective of this essay is to estimate the average impact 

of the different types of land titling (title from the government and certificate from local 

authority) at the plot level. There are four main outcome variables in this study: yield, 

expenditure per hectare on chemical fertilizer, and dichotomous adoption variables (1 for 

adoption; 0 otherwise) for both chemical fertilizer and manure.  

I use propensity score matching (PSM) to minimize the selection bias of land titling, and 

the covariates for the PSM approach include plot and household characteristics. Then, I run 

regressions on the matched samples to estimate the impact of land titling by controlling for plot, 

household, and village characteristics (the latter through village fixed effects). The regressions 

are weighted by propensity score weights so as to make the treatment and comparison groups 

comparable, and standard errors are clustered at the village level to get robust standard errors. 

We use pooled cross-sectional data from a nationally representative household survey – the 

Cambodia Socio-Economic Survey (CSES) conducted in 2009, 2010, 2011, 2012, 2013 and 2014 

by Cambodia’s National Institute of Statistics (NIS).  

	

10 

The chapter is organized as follows. It begins with the background of the study (this 

section), and this is followed in section 2.2 by a literature review about the impact of land titling 

on agricultural production, especially on variable input use and adoption of improved farm 

technology. Section 2.3 presents the Cambodian context of land title and rice production. The 

conceptual framework and estimation strategy are shown in section 2.4. Section 2.5 describes the 

data source and descriptive statistics. Section 2.6 is the core section which focuses on the 

findings and a discussion of their implications. The conclusion and policy implications are 

presented in the last section. 

2.2. Literature review 

This section discusses the empirical evidence of the impacts of land tenure on adoption of 

improved technology and farm investment, particularly adoption of and expenditure on fertilizer 

and manure. Because empirical studies of the impacts of land title on agricultural investment and 

adoption of improved technology are inconclusive, our literature review examines both the 

studies that support the theory of positive impact of land title and the ones showing insignificant 

effect. Also, we review how local context such as land governance institutions and socio-

economic characteristics can affect the role of land titling in increasing farm investment. In the 

areas where land is abundant, customary land tenure system is prevailing and land renting is 

common (Holden & Otsuka, 2014).  As population keeps growing, however, land becomes 

scarce. As a result, land rights have to be well-defined and safeguarded by an effective state 

institution. 

One of the seminal early works on the impacts of land title on agricultural outcomes is 

the study by Feder, Onchan, Chanlamwong, & Hongladarom (1988). The study used reduced 

form regressions and found that land title had a positive and significant impact on input use and 

	

11 

access to credit in Thailand. Another prominent empirical study on the causal relationship 

between land title and farm investment was conducted by Besley (1995), which used data from 

Ghana and controlled for endogeneity of land tenure. He found that tenure security encouraged 

investment in one study area but not in another. Also, Goldstein & Udry (2008) found that 

households with good social and political networking had more secure land rights, which in turn 

motivated them to invest more in land fertility. 

 Further, Holden, Deininger, & Ghebru (2009) used three rounds of household- and plot-

level data to evaluate the impact of low-cost land registration and certification on maintenance of 

soil conservation structures and land productivity in the Tigray region of Ethiopia. The results 

suggest the significant and positive impact of land certification. In connection with this, Abdulai, 

Owusu, & Goetz (2011) empirically found that farmers with secure land rights are likely to 

embrace some types of investments such as soil improvement measures using mulch and manure. 

It is worth noting that manure has benefits that last more than one season by adding organic 

matter to the soil, but fertilizer’s benefits normally last just one season (Motavalli, Singh, & 

Anders, 1994). Kassie et al. (2013) and Teklewold, Kassie, & Shiferaw (2013) also found a 

positive association between land tenure and adoption of fertilizer and animal manure. However, 

the empirical results of both of these studies may suffer from selection bias because the land title 

is treated as a regressor in cross-sectional analysis without addressing endogeneity. 

Furthermore, there have been other empirical studies providing evidence of a positive 

linkage between land title and the use of productivity-enhancing inputs, particularly in Asia.  A 

study in Vietnam by Do & Iyer (2008)  found that the share of land area allocated to multi-year 

crops increased as a result of land titling. In northeast rural China, Jacoby, Li, & Rozelle (2002) 

found that tenure insecurity significantly reduced application of organic fertilizer. Banerjee, 

	

12 

Gertler, & Ghatak (2002) found that tenancy reform had a positive effect on agricultural 

productivity in the Indian State of West Bengal after a tenancy reform program was launched 

there in the late 1970s. Li, Rozelle, & Brandt (1998) found a positive linkage between the length 

of tenure and use of organic manure and phosphate fertilizer in maize production in rural 

northeast China. Additionally, Nguyen, Bauer, & Grote (2016) used panel household data (1993, 

1998 and 2006) and random effect regression and confirmed that land tenure security had a 

positive impact on manure use in the northern uplands of Vietnam.  

On the other hand, some studies did not find a significant impact of land tenure.  Pender 

& Kerr (1999) examined the effects land sales restrictions on credit, land investment and 

cultivation decisions in two villages in South India. Their empirical results suggest that land sale 

restrictions were not significantly associated with inefficiency in the two villages.  Place & 

Otsuka (2002) found that tenure had no impact on farm investment and crop productivity in 

Uganda, but the impact was estimated by ordinary least squares (OLS) regression, which did not 

enable them to deal with endogeneity of land tenure. The emprical study by Brasselle, Gaspart, 

& Platteau (2002) controlled for endogeneity and their findings suggest no causal relationship 

between land rights and agricultural investment in Burkina Faso. The insignificant impact of land 

tenure on farm investment and the use of yield-enhancing inputs has been commonly found in 

sub-Saharan Africa (Atwood, 1990; Place & Hazell, 1993; Platteau, 1996). In this regard, 

empirical evidence shows that the impact of formal land tenure varies depending on the specific 

conditions prevailing under any given sitution (Place, 2009). It suggests that two main factors, 

discussed below, determine the ability of land titling to perform its role in providing land 

security: 1) administrative capacity of land institutions, and 2) socio-economic conditions 

(Jacoby & Minten, 2007; Benjaminsen, Holden, Lund, & Sjaastad, 2009; Thin, 2012). 

 

	

13 

2.2.1. Administrative capacity of land institutions 

In order for land titling to provide tenure security, it has to be endorsed by an effective 

state system capable of enforcing land rights (Feder et al., 1988). To this end, successful land 

governance requires strong administrative capacity. First, the government body in charge of land 

title registration must be able to facilitate land registration with an affordable cost and a simple 

procedure. Second, the supporting infrastructure of land registry, particularly the information 

system, also matters because it enables the land institution to update the information of land 

transfers. That is, failure to have up-to-date records of land holding may cause a lack of legally 

recognized rights, thereby negatively affecting the successes of the land registration institution. 

Third, the judicial system is a key supporting institution, ensuring justice when there is a land 

conflict and protecting the vulnerable against potential land grabbing by elites.  

The study by Jansen & Roquas (1998) confirmed that one of the reasons for the failure of 

the Land Titling Project for Small Farmers in Honduras was the bureaucracy’s organizational 

incapacity. Further, Benjaminsen, Holden, Lund, & Sjaastad (2008) indicated that insufficient 

administrative preparedness is a main constraint to formalization of land rights in Niger. The 

empirical evidence from Jacoby & Minten’s (2007) study showed that land title has no 

significant impact on land investment in rural Madagascar, and the key institutional constraints 

include out-of-date records of land registration and costly and complex procedures, which also 

has been shown to be a major issue in some African countries (Platteau, 1996; Atwood, 1990).    

2.2.2. Socio-economic context  

Apart from administrative capacity of land registration institutions, a country’s socio-

economic context is also a key determinant of the importance of land titling. In particular, failure 

to determine whether or not informal institutions and practices can provide sufficient land 

	

14 

security may result in counterproductive outcomes of a land titling program. Fundamentally, land 

titling may not have any significant role in the countries where indigenous land rights are still 

prevalent (Atwood, 1990) or where there is a well-functioning government program offering 

tenure rights but without full titling (Pender & Kerr, 1999). As suggested by Feder et al. (1988), 

titling is more appropriate in countries where the potential return on land is increasing, and 

landowners need rules to secure their land tenure to minimize land disputes and land grabbing. 

Under some circumstances, if land title is undertaken without comprehensive knowledge of a 

country’s land issues it may even exacerbate the inequitable distribution of land in cases where 

local elites take advantage of land reform to grab more land. For instance, some land rights 

formalization (titling) attempts in Africa were not successful in providing tenure security, 

because the customary land ownership system was already well-functioning while the land 

reform law was in favor of the elites at the expense of the poor and vulnerable (Atwood, 1990; 

Brasselle et al., 2002; Benjaminsen et al., 2008). Hence, in order for land titling to reach its full 

potential, there is a need for a social adaptation and response in terms of culture, politics and 

other socio-economic characteristics, such as increased commercialization of land and 

population growth. 

2.3. An overview of the country context 

2.3.1. Land title in Cambodia 

In Cambodia, between 1979 and 1989 following the downfall of the genocidal Khmer 

Rouge regime, all land officially belonged to the government, and cultivated land was 

collectively managed (Boreak, 2000). However, the collectivization system was not successful 

for various reasons. First, farmers were not motivated to work because output was also 

distributed to the group members who did not contribute their labor to cultivation. Second, the 

	

15 

rules were not strictly enforced to deal with group members who were not willing to work. Third, 

farm inputs such as seeds, fertilizers and fuels were only offered to the model solidarity groups. 

Hence, with the failure of collectivization and a shift in the economic system from a centrally-

planned economy to free markets in 1989, the government reintroduced private property rights 

(Boreak, 2000).  

2.3.1.1. Formal land registration (land title issued by the government) 
	

In 1989, the government started introducing private land ownership rights. Following the 

adoption of the Land Law of 1992, landowners were increasingly encouraged to apply for formal 

ownership by submitting all relevant documents as proof of their occupation without any conflict 

over the plot. Under this law, land title can be achieved through sporadic land registration which 

is a process by which individual landowners choose to register for a land title, which they can get 

if they meet all conditions required by law. However, it is quite costly for rural landholders to 

apply for land title by themselves because oftentimes they need to pay unofficial fees to get their 

application approved (So, 2009; Un & So, 2011). Consequently, only a small portion of tenure 

applications were granted formal land titles (Markussen, 2008).  

According to Chan, Tep, & Sarthi (2001) and Thin (2012), sporadic land registration is 

conducted through a multi-step bureaucratic procedure. First, an application form is issued to the 

applicant and a schedule is set for land survey and demarcation, if there is no counter claim or 

conflict. Second, when the land survey and demarcation have been completed, the district 

cadastral office signs a form recording the land holder’s identity and land’s boundaries. Third, 

this form is shown to the public for 30 days. Fourth, the land is registered by the district cadastral 

staff into a land registration map. Fifth, the application and relevant supporting documents are 

forwarded to the provincial or city cadastral office where information of the land claim is 

	

16 

recorded into the land registry. In the final step, the application is sent to the General Department 

of Cadastre and Geography (GDCG) under the Ministry of Land Management, Urban Planning 

and Construction (MLMUPC) for issuance of land title.  

There was a significant increase in conflicts over land rights in the 1990s because of 

rising demand for the arable land and an underdeveloped land registry system (Markussen, 2008; 

Thin, 2012). To address this problem through speeding up land registration in the country, 

Systematic Land Registration (SLR) was introduced in 2002 (Thin, 2012). This project was 

mainly funded by the World Bank with the goal of completing land registration from 2002 to 

2017. SLR was founded on a 2001 amendment to the Land Law that allowed people to legally 

receive land title if they had already occupied a parcel free of conflict for no less than five years 

before the promulgation of the law. In general, SLR is targeted at the commune level and most 

parcels in a commune are registered at the same time by cadastral teams composed of 

registration specialists and land surveyors. After collecting information about landholders and 

conducting land surveys, the team proceeds with registering land into the cadastral system (So, 

2009). The annual report from the Ministry of Land Management, Urban Planning and 

Construction (MLMUPC) shows that by December 2014, around 3.8 million titles had been 

issued via sporadic and systematic land registration.  

As of now, landholders can still apply for title through sporadic land registration if the 

systematic land registration program is not available in their commune. Thin (2012) reveals that 

it takes 3 to 4 months for a land title to be issued under sporadic land registration and 5 to 7 

months under the SLR. This may be due to the fact that systematic land registration is conducted 

at the commune level, so the SLR teams need time to register numerous parcels into the cadastral 

system. It costs landholders $0.25 per 1,000 m2 for agricultural land in rural areas for issuance of 

	

17 

a land title under SLR, while the official fee under sporadic land registration is approximately 

$2.50 per 1,000 m2. In practice, the fee charged under sporadic land registration is higher than 

this, and it is not clearly defined by the size of the landholding (So, 2009).  

Furthermore, the formal land transfer system requires a complicated procedure and a high 

transfer fee. For example, after the transfer form is signed by the district governor, land holders 

have to pay a tax of 4 percent of the sale’s value at the provincial tax department (Thin, 2012; 

So, 2009). Subsequently, the transfer form and tax receipt are verified by the provincial 

Department of Land Management before being forwarded to the MLMUPC. This implies that 

that the cost of registration declines with the size of the parcel. Hence, if one owns a small and 

inexpensive parcel of agricultural land, one may have a low incentive to pursue formal land 

transfer as opposed to semi-formal. This is because of the high costs of formal land transfer and 

the complicated procedure. Nonetheless, there is no information on the exact costs of formal land 

transfer, because it is sometimes associated with corruption, which is a sensitive issue (So, 

2009). 

2.3.1.2. Semi-formal land registration (land title issued by the local authority) 
 

 

Apart from formal land registration, the routine practice for rural land transfer is that 

landholders who are exchanging rights to a piece of land go to the commune office to have their 

written agreement certified with a signature and stamp from a commune chief. This is an 

alternative, less formal approach to registration that is attractive when people are not able to 

afford formal land transfer (So, 2009). The official fee under sporadic land registration is 

approximately $2.50 per 1,000 m2, but information about the unofficial fees is not available. 

Also, we have no information about the exact cost of informal land registration, but it is 

significantly lower than for sporadic formal land registration. The semi-formal system provides 

	

18 

landowners with a land certificate issued by the local authority, but the owner’s name of the 

transferred land is not updated in the cadastral registrar of the MLMUPC – the only government 

institution responsible for land title registration. The most important reason for landholders to 

prefer this semi-formal practice to the formal land transfer system is that the latter is costly, as 

mentioned in the discussion of the formal process.  

The semi-formal system is convenient for rural and poor landholders, but it creates 

difficulties for the systematic land registration process. This is because under the semi-formal 

system the owner’s name of the transferred land is not recorded in the cadastral registrar of 

MLMUPC, thereby potentially resulting in an out-of-date cadastral system when the commune 

chiefs fail to regularly report the land transactions to the cadastral office. Then, the outdated 

cadastral system can negatively affect the land governance institution. In connection with this, 

Thin (2012) and So (2009) argue that systematic land registration was successful in building an 

initial record of landholding, but not in updating the cadastral registrar when land ownership is 

transferred. 

2.3.2. Cambodia’s rice sector 

Cambodia is an agrarian and low-income country (FAO, 2014) with a narrow base of 

economic growth because the economy relies on three main pillars: industry (garments), services 

(tourism and construction) and agriculture. Agriculture is the second biggest contributor to GDP 

(around 34 percent of GDP in 2013), following the service sector (ADB, 2014). Agriculture’s 

moderate average growth rate was around 4 percent over the period 1996-2013, compared to 

industry (12 percent) and service sectors (18 percent). Regarding the labor force, the share of 

employment in agriculture was 65 percent in 2012 (FAO, 2014; NIS, 2011). The majority of 

rural livelihoods depend on agriculture. Rice is the main crop (NIS, 2011), covering 3.7 million 

	

19 

hectares or 75 percent of the total cultivated area (FAO, 2014). Most Cambodian farmers are 

smallholders with each household owning less than two hectares of land (FAO, 2014). 

Cambodia’s agriculture has faced challenges such as vulnerable farming systems and low 

productivity because of an underdeveloped irrigation system, poor cultivation techniques, and 

inadequate use of high quality inputs including fertilizer and seeds. (RGC, 2013). The country’s 

average rice yield is lower than that of neighboring Vietnam and Laos (CDRI, 2013; Sok, Chap, 

& Chheang, 2011). In 2008, for instance, rice yield in Cambodia was 2.58 tons/ha while it was 

4.88 tons/ha in Vietnam, 3.53 tons/ha in Laos and 2.76 tons/ha in Thailand (Yu & Fan, 2011). 

Additionally, lack of government investment in agriculture and agricultural research is also a 

constraint to improving Cambodia’s agricultural growth. For example, the share of government 

expenditure on agriculture has ranged from 1.7 to 2.4 percent between 2002 and 2010 (Keo & 

Theng, 2013). 

The empirical study by Yu & Fan (2011) suggests that there is great potential for 

Cambodia to enhance rice productivity by expanding the adoption of modern inputs, particularly 

fertilizer. A descriptive and qualitative study of fertilizer markets in Cambodia by Theng, Khiev, 

& Phon (2014) shows that the fertilizer use rates remain low for several reasons including: (1) 

complicated import licensing procedures and regulations; (2) underfunding of research on 

fertilizer application; and (3) lack of clarity regarding roles and responsibilities between 

government institutions in charge of regulating fertilizer trade. 

Chemical fertilizer and organic fertilizers (manure and plant residues) are the two 

productivity-improving inputs that are commonly used in Cambodian agriculture, including rice 

production. The proportion of households reporting the use of chemical fertilizer is 70 percent 

(NIS, 2015b). Among the four agro-ecological zones of Cambodia, proportions of households 

	

20 

using chemical and organic fertilizers were highest in the Plains Zone, followed by the Tonle Sap 

Lake, Coastal, Plateau and Mountainous Zones.  

2.4. Conceptual framework and empirical methodology 

2.4.1. Conceptual framework 

This sub-section conceptualizes how land title affects farmers’ behavior of adopting and 

spending on productivity-enhancing inputs such as fertilizers, and manure. In general, there are 

two ways in which a land title provides landowners with incentives to make productivity-

enhancing investments on farmland. First, because it is backed by the land governance institution 

capable of enforcing the property rights, it acts as an assurance of ownership for one’s land, 

therefore incentivizing the owner to invest more in their farm (Besley, 1995). In this regard, the 

land title will safeguard the farm from any prospective grabbing from a third party before 

production and income from an investment in the farm can be realized (Fenske, 2011). Second, 

having secure property rights will facilitate farmers’ access to credit from both informal and 

formal financial institutions since land can be used as collateral.  As a result it enables farmers to 

scale up their investments (Besley, 1995) including the adoption of improved farm technology. 

The conceptual framework of our study is based on the above theoretical perspectives 

and a modification of Feder et al. (1988) and Nguyen, Bauer, & Grote (2016), which 

conceptualized causal linkages between land titling and investment in farming, particularly soil 

fertility improvement. As our outcome variables are fertilizer and manure use and productivity, 

we do not focus on the land market or land prices in our framework. As shown in Figure 1, land 

title from either the central government or a local government authority minimizes land 

ownership uncertainty, therefore enhancing incentives to invest and put forth management effort. 

Additionally, titled land can serve as collateral for credit, which enables farmers who have 

	

21 

liquidity constraints to contract loans for their input purchases. Greater use of inputs, particularly 

chemical fertilizer and manure, leads to improvements in soil fertility and productivity. 

 

 

Figure 1: Conceptual framework 

Land titling 

Greater management 

effort  

Larger investment 

 
 
 
 
 
 
 
 

Greater use of chemical fertilizer 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
Source: Adapted from Feder et al. (1988); Nguyen, Bauer, & Grote (2016) 

Greater use of manure 

Improvements in soil 

fertility and productivity  

This conceptual framework depends on various assumptions that hold to a certain extent 

in the context of Cambodian land titling. First, the framework is more appropriate in countries 

where arable land becomes scarcer with an increasing population, and landowners demand rules 

or institutions supportive of well-defined land rights, which is the case in Cambodia (Markussen, 

2008; RGC, 2014). This is because land titling does not necessarily have any critical role in 

providing land security and serving as collateral for credit in the countries where there is a well-

functioning customary land ownership system, or where there is a well-functioning government 

program offering tenure rights but without full titling (Pender & Kerr, 1999; Place, 2009). 

Second, this conceptual framework is applicable when land titling is secured by an effective 

state-enforced system. Even though there have been some limitations in land governance 

	

22 

institutions (as discussed in section 3 of this article), Cambodia has a legal framework and a 

government body in charge of land title registration, which enable landowners to legally own a 

given plot of land without relying on social customs. Hence, we can hypothesize that, other 

things equal, a titled plot in Cambodia has more secure tenure than a plot without any title. 

Figure 1 succinctly summarizes the linkages between land titling and improvement in soil 

fertility and productivity through great uses of chemical fertilizer and manure.  

It should be noted that the link from land titling and management effort will not be 

empirically tested in this study, because there is no data about that. The main reason for 

mentioning it in the framework is to explain the relationship between land titling and the 

outcomes. 

2.4.2. Empirical methodology 

In Cambodia, there is no evidence that receiving or providing land title is random, so its 

impact estimator is subject to selection bias or endogeneity. Fundamentally, selection bias refers 

to pre-existing differences between the treatment and comparison group that can influence the 

outcomes of the treatment. The two main types of selection bias are self-selection bias and 

program placement bias (Duflo, Glennerster, & Kremer, 2006). In the context of Cambodia land 

title, our estimation can be subject to both types of selection bias. First, landowners can choose 

to apply for land title issued by the government through sporadic land registration if there is no 

systematic land registration assigned to their village, or to get a land certificate from local 

authority if they are not able to afford the cost of formal land registration or transfer. In this 

situation, our estimated impact is confounded by self-selection which is a major problem in 

impact evaluation studies because people decide to participate in a program on a voluntary basis, 

which is not random (Duflo, 2004). For example, if households with better education, more 

	

23 

wealth and higher social status are more likely to get land title, attributing greater use of fertilizer 

and manure to land title is biased. Second, households may obtain land title through the 

systematic land registration whose placement criteria are not random. That is, for instance, if the 

program is targeted to poorer areas, the estimated impacts of the treatment would be 

underestimated because the treatment group is worse off than the comparison counterpart, 

regardless of land title. 

Our study uses cross-sectional data, so instrumental variables (IVs) and the Propensity 

Score Matching (PSM) are the potential methods to address the selection bias. The mode of land 

acquisition is available in our sample; it has been shown to be a valid IV under some conditions 

(Besley, 1995; Tong, 2010; Brasselle, Gaspart, & Platteau, 2002). However, the logic of IVs is 

not valid when some of the households in the sample received land title through the systematic 

land registration because the IV method is applicable when households self-selected to apply for 

land rights (sporadic land registration). That is, as argued by Besley (1995) and Markussen 

(2008), landowners are more likely to be motivated to apply for land title on plots acquired by 

purchase than plots acquired by forest clearance for free. On the other hand, with systematic land 

registration, the program targets certain villages, and it is possible for landowners to obtain a 

land title regardless of their motivation to apply for it. Thus, the IV method is not feasible 

because I have no way to distinguish plots that received sporadic versus systematic land 

registration, as this information is not available.   

Therefore the method for this study is PSM, which has been used to quantitatively assess 

the impacts of policy and program interventions using cross-sectional data (Rosenbaum & 

Rubin, 1983; Caliendo & Kopeinig, 2008; Kuwornu & Owus, 2012; Dillon, 2011; Davis et al., 

2012 ). There are also some impact evaluation studies of land titling using this method such as 

	

24 

Galiani & Schargrodsky (2004); Galiani & Schargrodsky (2010); Thin (2012).  Our large sample 

size is a considerable advantage for the matching process because there will be sufficient sample 

size of the comparison group, which enables the common support (matched sample) to have 

large enough statistical power to reduce the bias in impact estimation (Khandker, Koolwal, & 

Samad, 2010). The two treatments are defined as (1) plot with land certificate from the 

government, and (2) plot with certificate from the local authority, while the control plot is the 

one with no paper to certify ownership.  

 I use PSM to get the matched samples to make four sets of comparisons regarding the 

outcome variables: 1) control plots to plots with a land certificate from the government; 2) 

control plots to plots with a certificate from the local authority; 3) plots with a certificate from 

the central government to those with a certificate from the local authority; and 4) control plots to 

plots with any kind of land certificate1 – either from the central government or from the local 

authority. When there is more than one treatment group it is common to take this approach of 

comparing all possible comparison groups (see, for example, Dillon, 2011; Blimpo, 2014).  

2.4.2.1. Steps in propensity score matching 

In the PSM approach, the treatment and comparison groups with comparable propensity 

score – the estimated conditional probability of receiving the intervention given observed 

characteristics – are matched. The observations whose propensity scores are not comparable (not 

in common support) are dropped from the analysis. By using propensity score matching adapted 

from Guo & Fraser (2014) and Ravallion (2001), the analytical process of PSM is presented as 

follows.  

																																																								
1 Previous studies in Cambodia applied only the fourth comparison conducted in this study; that is, they estimated 
the impact of land title in general, regardless of its type. 

	

25 

First, I must estimate the probability of a plot receiving land title by using logit 

regression. A binary logit model can be specified as:  

IP
(

i

=

|1

X

)

=

i

1
e
X
-b

1

+

=

XP
(

i

)

     (equation 1) 

Where Xi is a vector of explanatory variables or covariates including attributes of plot, household, 

and village which are likely to affect the probability of receiving land titles. After running the logit 

model, I predict propensity score, P(Xi), for every sample of the treatment and comparison groups. 

Second, after estimating the propensity score, I match the treated and comparison observations 

based on propensity scores using the matching techniques – e.g. nearest neighbor (NN) or kernel 

estimators; the pros and cons of each matching technique will be outlined in the appendix. Third, 

I check the region of common support to avoid comparing incomparable observations which could 

result in evaluation bias.  

To check the quality of matching, one can compare the covariates (Xi) before and after 

matching.  The  mean  and  median  of  absolute  bias  are  expected  to  decrease  markedly  after 

matching. In addition, the standardized bias of each control variable (covariate) in the logistic 

regression  before  and  after  matching  is  also  used  to  figure  out  whether  there  are  systematic 

differences in the means of the control variables across both groups (Rosenbaum & Rubin, 1983). 

That is, after matching, no significant differences in the covariates between both groups should be 

found,  suggesting  that  the  observed  characteristics  between  the  two  matched  groups  are 

comparable. To this end, Caliendo & Kopeinig (2008) suggest that median and average of this 

standardized bias below 3 and 5 percent after matching are sufficient. In addition, the Pseudo-R2 

from the estimation of propensity score after matching should be lower than that before matching, 

and the P-values of likelihood ratio tests are non-significant after matching, indicating there are 

no systematic differences in the distribution of observable covariates between both groups. 

	

26 

2.4.2.2. Assumptions and limitations of PSM 

The first assumption of PSM is called the conditional independence assumption (CIA) or 

exogeneity  assumption.  It  states  that  given  observable  covariates  X  that  are  not  affected  by 

treatment, potential outcome Y is independent of the intervention. Rosenbaum & Rubin (1983) call 

this unconfoundedness; it implies that uptake of an intervention or program is based entirely on 

observable characteristics. Rosenbaum & Rubin (1985) suggested that “when appropriate common 

support  is  established,  the  conditional  independence  assumption  becomes  valid”  (Kuwornu  & 

Owus, 2012, p. 12). In this regard, one can support the conditional independence assumption by 

controlling for many observed characteristics that can possibly affect the uptake of an intervention 

or program (Khandker et al., 2010). 

The common support or overlap condition is the second assumption of PSM; it ensures 

that the treatment units have comparison counterparts nearby in the propensity score distribution. 

The implication of this assumption is that the effectiveness of PSM also depends on a large and 

comparable number of the treatment observations and the non-treatment observations so as to have 

a  substantial  region  of  common  support.  As  a  result,  there  is  a  sizable  overlap  in  propensity 

distribution across the treatment and control units, which supports the precision of the impact 

estimator.   

The main limitation of PSM is that the approach assumes that selection bias is based only 

on observed characteristics, so it does not address the unobserved factors influencing the likelihood 

of receiving treatment or program participation (Khandker, Koolwal, & Samad, 2010; Cerulli, 

2015). I acknowledge this disadvantage because there has not been any nationally representative 

household panel survey; the Cambodia Socio-Economic Survey (CSES) is our best choice because 

it is the only national survey that contains useful data on land titles. In general, PSM can help to 

	

27 

reduce  (not  eliminate)  selection  bias  caused  by  unobserved  confounding  factors  (Kuwornu  & 

Owus, 2012). 

Despite  this  limitation,  the  bias  in  PSM  estimates  can  be  low,  given  three  broad 

requirements as revealed by Heckman, Ichimura, & Todd (1997, 1998). First, the data on treatment 

and control groups should be collected from the same survey (the same questionnaire, the same 

interviewer  training  and  survey  period).  Second,  a  representative  sample  survey  of  eligible 

treatment  and  comparison  observations  can  also  significantly  improve  the  precision  of  the 

propensity score. Third, the larger the sample of the eligible comparison group the smoother the 

matching  process,  because  the  treatment  observations  will  keep  searching  for  the  comparison 

counterparts with the comparable value of the predicted propensity score. The sample for our study 

meets these three broad provisions because CSES is a nationally representative household survey. 

2.4.2.3. Regressions on the matched samples  

After matching, I run regressions on the matched samples to estimate the effects of land 

titling.  The  regressions  are  weighted  by  propensity  score  weights  from  PSM  to  minimize  the 

selection bias of land titling, and the standard error is clustered at village level to ensure that it is 

robust for making inference. Also, I control for village heterogeneity (or village fixed effects) in 

the regressions. The propensity score weights are defined as 1/#$%#&'()*+	(-%$& for treatment 
group and as 1/(1−#$%#&'()*+	(-%$&) for comparison group (Khandker et al., 2010). 

 Since our dependent variables are continuous and binary, I specify the empirical models 

based on this. To look at the effects of land titling on fertilizer expenditure and yield, I use Ordinary 

Least Square (OLS) which is expressed in the following matrix form: 

12=425+72    (equation 2) 

Where i is the observation running over 1 to n; Yi represent continuous outcome variables for 

	

28 

(either  fertilizer  expenditure  or  yield); 42  is  a  set  of  independent  variables  including  plot  and 
household characteristics, and village dummies; and 5 is vector of parameters associated with the 
independent variables; and  72 is an error term. 
model which can be derived from introducing a latent variable 12∗, which is defined as: 
In practice, one cannot observe 12∗, but instead one observe 12, which takes the values of 1 if 
12∗>0, and 12=0 if  12∗≤0.  

12∗=425+92    (equation 3) 

To examine the effects of land titling on the adoption of fertilizer and manure, I use probit 

2.5. Data and descriptive statistics 

2.5.1. Background on the Cambodia Socio-Economic Survey (CSES) 

In this study I use pooled cross-sectional data from a nationally representative household 

survey: the Cambodia Socio-Economic Survey (CSES), conducted in 2009, 2010, 2011, 2012, 

2013 and 2014 by the National Institute of Statistics (NIS) of Cambodia. The sampling design of 

the CSES is a three-stage sampling process beginning from villages, enumeration areas2 to 

households. These surveys are cross-sectional, so each sample did not cover the same households 

every year. The available information includes plot characteristics, crop production, household 

characteristics, consumption, migration, asset, credit and village characteristics. Sample size is 

much larger for the 2009 and 2014 surveys than the other years. On average, 18 to 20 households 

were randomly selected from each village in 2009 and 2014. In 2009, 12,000 households from 

720 villages were randomly selected, and in 2014 12,960 households from 1,008 villages were 

selected  (NIS, 2010; NIS, 2015a). On average, there were approximately 10 households 

randomly chosen from each village in 2010, 2011, 2012 and 2013 (NIS, 2013;  NIS, 2014). The 

																																																								
2 In the Cambodia Socio-Economic Survey (CSES), the enumeration area is a smaller unit of a village. 

	

29 

survey in 2010 included 3,592 households from 360 villages, which is the same as the sample 

size for 2011. In 2012 about 3,600 households were randomly interviewed from 384 villages, 

and the sample size for 2013 is 3,840 households from 384 villages. It is worth noting that 

sample size of CSES varies across the years because the NIS can financially afford a large 

survey only once every 5 years (e.g. 2009 and 2014).  

I constructed the sample for our empirical analysis with a focus on the needs of our study.  

Its purpose is to estimate the impact on agricultural input use of formal land title issued by the 

government and semi-formal land title from the local authority, so we kept plots in the sample 

with these land titles as well as control plots with no title of either kind. Because of our focus on 

land titling, we drop the approximately 5 percent of plots in the sample that were rented or 

sharecropped. Some plots with some potential measurement errors were dropped to enhance the 

internal validity of our data. For example, with regard to land title, only plots that had proper 

answers for land titles – “Yes” or “No” – were kept while those with answers such as “lost it” or 

“don’t know” were dropped. Additionally, for the types of land title, plots for which the 

respondent had no paper to show the interviewers were dropped.  Finally, we limited the plot-

level data to rice production, which is the most common crop in the sample.  This ensures that 

we have comparable data for all observations in the sample. After data management, we obtain 

18,210 plots (12,142 households in 1,616 villages3), which is for all years combined.  Table 1 

illustrates the distribution of sample size by year, including the number of plots, households and 

villages from each year in the sample for this study.  

 

 

																																																								
3 Among these villages, there are 1,138 unique villages. 

	

30 

Table 1: Number of plots, households and villages by year 
Households 

Plots 

Proportion 
(percent) 

Proportion 
(percent) 

Villages 

Proportion 
(percent) 

Number 
4,467 
1,024 
1,098 
1,034 
1,048 
3,471 
12,142 

36.79 
8.43 
9.04 
8.52 
8.63 
28.59 
100 

Number 

401 
174 
181 
181 
183 
496 
1,616 

24.81 
10.77 
11.2 
11.2 
11.32 
30.69 
100 

Year 

2009 
2010 
2011 
2012 
2013 
2014 
Total 

Number 
7,201 
1,601 
1,635 
1,507 
1,411 
4,855 
18,210 

39.54 
8.79 
8.98 
8.28 
7.75 
26.66 
100 

Source: Calculation based on CSES 2009 to 2014 

Table 2 shows that in the sample for empirical analysis, there are 3,773 plots with a land 

certificate from the government, 5,003 plots with a certificate from the local authority and a 

control group of 9,441 plots with no certificate of any kind.  The fact that there are more plots 

with semi-formal papers from the local authority than land with title issued by the government 

implies that a written agreement certified with a signature and stamp from a commune chief 

remains the most common practice for rural land transfer, as suggested by So (2009). 

Table 2: Number of plots by type of land title and year 

Type of land title 

2013 
2009 
325 
Land title from the government  1,286 
Paper from local authority 
363 
1,744 
No title 
4,171 
723 
Total 
7,201  1,601  1,635  1,507  1,411 
Source: Calculation based on CSES 2009-2014 

2,011 
344 
436 
855 

2010 
356 
485 
760 

2012 
325 
493 
689 

Total 
2014 
3,773 
1,137 
5,003 
1,482 
2,236 
9,434 
4,855  18,210 

2.5.2. Definitions and summary of outcome variables and covariate 

There are four outcome variables for our study. They are expenditure for chemical 

fertilizer per hectare, dichotomous adoption variables (1 for adoption; 0 otherwise) for chemical 

fertilizer and manure, and yield. Regarding the covariates, we control for 33 variables including 

characteristics of plots and households.  Definitions of all variables are reported in Table 8 of the 

appendix – the Table is long, so we have put it in the appendix.  If the plot was planted in two 

	

31 

seasons per year, we considered it as one observation, with one unique plot ID for empirical 

analysis. In such cases the continuous outcome variables (i.e. fertilizer expenditure and yield) 

were defined as the average of both periods. For dichotomous variables, for example, fertilizer 

adoption was defined as 1 if this input was either in dry season, wet season or both seasons. This 

captures whether or not fertilizer was used on that plot. This definition also applies to manure 

adoption. 

Table 9 of the appendix displays descriptive statistics for all variables, and we choose to 

describe only some main variables. The average expenditure for chemical fertilizer is 176 

thousand riels (around USD 45) per hectare. It is worth noting that plots with government land 

title had higher average expenditure for both chemical fertilizer and manure than the plots with 

semi-formal title and the control plots. The adoption rates of chemical fertilizer and manure were 

77 percent and 58 percent, respectively. Plots with government title had higher adoption rates 

than those with semi-formal title and the control plots, but the yield is similar across the groups 

(Table 9). The average plot size of land with government title (0.689 hectare) is smaller than that 

of plots with local authority paper (0.961 hectare) and the control counterpart (0.864 hectare). On 

the modes of land acquisition, most of the plots which landholders bought (mode of acquisition 

3) had a certificate from the local authority, implying that semi-formal land transfer is the routine 

practice, which is in line with a study by So (2009) as mentioned above. We cannot tell if the 

formal title was obtained through sporadic or systematic land registration, because there is no 

information about that in the data set.  Additionally, 2.1 percent of plots had been associated with 

conflict: 1.4 percent of plots with government title, 2.1 percent of plots with local authority 

papers and 2.4 percent of control plots.  

	

32 

Apart from this, we also control for household characteristics such as demographic 

situation, household assets, sources of income, and access to credit. For household assets, our 

main variable is the value of durable goods, and we also include dummy variables for some key 

assets that could be hypothesized to influence the adoption decision. These include television, 

mobile phone, hand-tractor, generator, motorbike and pumping machine. To control for 

heterogeneity across the villages, we include village dummies in the regressions. 

2.6. Empirical results and discussions 

2.6.1. Getting the matched samples 

As discussed, I use Propensity Score Matching (PSM) first to compare control plots to plots 

with a land certificate from the government, then again to compare control plots to plots with a 

certificate  from  the  local  authority,  then  again  to  compare  plots  with  a  certificate  from  the 

government to those with a certificate from the local authority, and finally again to compare control 

plots to plots with any kind of land certificate – either from the central government or from the 

local authority. As mentioned in the empirical methodology section, matching plots between the 

treatment and control groups can be done using one of a number of matching techniques. Our main 

purpose of using PSM is to get matched samples to reduce bias based on observable characteristics. 

I  choose  the  five  nearest-neighbor  matching  approach  because  it  is  the  common  matching 

technique,  and  it  also  has  quality  matching  for  all  comparisons.  For  example,  Asfaw,  Kassie, 

Simtowe, & Lipper (2012) used this approach to estimate the effects of adoption of pigeon pea 

technologies on poverty reduction using cross-sectional data from Tanzania. 

Comparisons  of  the  mean  and  median  of  standardized  bias,  Pseudo-R2  and  P-value  of 

likelihood ratio tests before and after matching are recommended as useful indications for checking 

matching quality (Caliendo & Kopeinig, 2008). The main purpose of the comparisons is to prove 

	

33 

that there are no systematic differences in the control variables across both groups after matching 

(Rosenbaum & Rubin, 1983). In our study, medians and means of absolute bias for all comparisons 

are below 3 and 5 percent respectively after matching (Table 3). These are the standard benchmarks 

suggested by Caliendo & Kopeinig (2008). Table 3 further shows that for all comparisons, the 

Pseudo-R2 values from the estimation of propensity score after matching become lower than that 

before  matching,  and  the  P-values  of  likelihood  ratio  tests  are  non-significant  after  matching, 

indicating there are no systematic differences in the distribution of observable covariates between 

both groups. 

Table 3: Summary of PSM quality indicators from 5 nearest-neighbor matching 

Land title from the government versus control group 

Before matching 

After matching 

12.5 
11.0 
0.062 
0.000 

10.3 
9.8 
0.033 
0.000 

7.7 
11.8 
0.06 
0.000 

10 
7.4 
0.032 
0.000 

1.4 
1.1 
0.001 
0.999 

0.8 
0.6 
0.000 
1.00 

1.8 
4.8 
0.002 
0.628 

0.7 
0.5 
0.001 
0.988 

Matching quality indicators 
Mean absolute bias (%) 
Median absolute bias (%) 
Pseudo R2  
P-value of LR 

Matching quality indicators 
Mean absolute bias (%) 
Median absolute bias (%) 
Pseudo R2  
P-value of LR 

Matching quality indicators 
Mean absolute bias (%) 
Median absolute bias (%) 
Pseudo R2  
P-value of LR 

Land title from local authority versus control group 

Before matching 

After matching 

Land title from the government versus land title from local authority 

Before matching 

After matching 

General land certificate versus control group 

Before matching 

After matching 

Matching quality indicators 
Mean absolute bias (%) 
Median absolute bias (%) 
Pseudo R2  
P-value of LR 
Source: Calculation based on CSES 2009 to 2014 
Note: LR: Livelihood ratio. 
 

	

34 

Furthermore,  the  visual  comparison  of  the  distribution  of  propensity  scores  between 

treatment and controls before and after matching also informs us about the quality of matching. 

Figures 2, 3, 4 and 5 of the appendix illustrate that there is bias in the distribution of propensity 

score between the treatment and comparison groups before matching. After matching, on the other 

hand, there is almost perfect overlapping in the distribution of estimated propensity score between 

the treatment and control plots for every set of comparisons, suggesting that I have good matches. 

Also, following each matching, I could identify the common support area (matched sample), and 

the number of plots that fell in off-support areas and in the matched samples for the four sets of 

comparisons.  The  plots  in  the  off-support  areas  are  not  included  in  the  matched  sample  for 

empirical analysis.  These are illustrated in Table 4. The details from test of matching quality can 

found in Tables 10, 11, 12 and 13 of the appendix. 

The figures after matching are weighted by the propensity score weights, and they are based 
on the common support observations that have propensity score weight. The good matches are 
given by the observations with propensity score weight (Khandker et al., 2010; Cerulli, 2015). The 
columns reporting the number of observations having a propensity score weight are the ones that 
are used in our regressions.   

 

 

 

 

 

 

 

	

35 

Table 4: Number of observations in common support region for each matching 

Sample 
Untreated 
Treated 
Total 

Sample 
Untreated 
Treated 
Total 

On support 

Off support 

0 
188 
188 

Land title from the government versus control group 
  Total 
9,434 
 
 
3,773 
  13,207 
Land title from local authority versus control group 
   Total 
 
9,434 
 
5,003 
  14,437 

9,434 
3,585 
13,019 

9,434 
4,753 
14,187 
Land title from the government versus land title from local authority 

9,434 
4,753 
14,187 

9,434 
3,585 
13,019 

0 
250 
250 

Off support 

On support 

On support and with 

propensity score weight 

On support and with 

propensity score weight 

Sample 
Semi-formal 
Formal 
Total 

On support 

Off support 

   Total 
5,003 
 
 
3,773 
8,776 
 
General land certificate versus control group 

5,003 
3,585 
8,588 

0 
188 
188 

5,003 
3,585 
8,588 

On support and with 

propensity score weight 

Off support 

Sample 
Untreated 
Treated 
Total 
Source: Calculation based on CSES 2009 to 2014 

   Total 
9,434 
 
 
8,776 
  18,210 

0 
438 
438 

On support 

9,434 
8,338 
17,772 

On support and with 

propensity score weight 

9,434 
8,338 
17,772 

2.6.2. Regressions on the matched samples 

This section focuses on the empirical findings and their discussion. Before proceeding to 
the regression, it is worth seeing the mean values of the outcome variables by group of plots for 
(1) the whole sample before matching and (2) the common support observations with propensity 
score weights, which are the ones our regressions rely on. These figures for the sample before 
matching are reported in Tables 5 and 6, respectively. 

 
 

	

36 

Table 5: Average outcome variables by group of plots (before matching) 

 

Land title from 
the government 

215.67 

Fertilizer expense4 
(‘000’ Riels/ha) 
Fertilizer adoption 
(dichotomous variable) 
Manure adoption 
(dichotomous variable) 
Yield (tons/ha) 
Source: Calculation based on CSES 2009 to 2014 

0.68 
2.23 

0.88 

land title from 
local authority 

181.48 

General 
land title 
196.18 

Control 
group 
157.73 

0.79 

0.58 
2.27 

0.83 

0.62 
2.25 

0.71 

0.54 
2.13 

Table 6: Average outcome variables by group of households (the common support observations 
with propensity score weights) 
 
Land title from 
the government 
versus control 

Land title from the 
government versus 
land title from local 

land title from 
local authority 
versus control 

General land 
title versus 
control group 

group 

group 

T 

C 

T 

C 
210.71  160.31  176.22  160.5
0 
0.73 

 
Fertilizer 
expense 
Fertilizer 
adoption 
Manure 
0.61 
adoption 
Yield  
2.20 
Source: Calculation based on CSES 2009 to 2014 

0.53 
2.16 

0.56 
2.14 

0.65 
2.24 

0.73 

0.78 

0.87 

authority 

Formal  Semi-formal 
215.71 

180.68 

0.88 

0.64 
2.29 

0.79 

0.61 
2.22 

T 

C 

192.4  161.57 

0.82 

0.74 

0.63 
2.21 

0.55 
2.16 

The details from each regression can be found in Table 14, 15, 16 and 17 of the appendix. 

Because the tables are long, I report only the results for the land title category. The results for other 

variables that I controlled for in the regressions can be found in the appendix.  Not only did these 

regressions use the matched samples, they are also weighted by the propensity weights obtained 

from propensity scores, as suggested by Khandker et al. (2010), so that treatment and comparison 

groups  are  comparable  in  terms  of  their  observable  characteristics.  Also,  the  standard  error  is 

clustered at village level, and I controlled for the heterogeneity across the villages using village 

																																																								
4 In real terms 

	

37 

fixed effects (village-level dummy variables). The standard error of each regression is clustered at 
village level so as to get the robust standard error for making inference.  

The regression results for the impact of land titling are reported in Table 7. It should be 

noted that the number of observations for probit regressions significantly dropped when I included 

village dummies. There are no village-specific variables in these regressions, because I already 

controlled for differences in village characteristics with the village dummies. The coefficients on 

the land titling for probit regressions are reported in terms of marginal probability effects, which 

is the common practice for interpreting the results from either probit or logit regression (Baum, 

2006).  With marginal probability effects, the interpretation becomes as simple as that for a linear 

regression. That is, a parameter estimate is the expected probability that a unit change in variable 
X will result in the outcome variable Y equaling 1, with other independent variables held fixed. 

Neither  semi-formal  nor  formal  land  certificate  has  a  statistically  significant  effect  on 

fertilizer  expenditure  (Table  7).  Also,  average  expenditure  on  fertilizer  of  the  plots  with 

government title is not significantly different from that of plots with semi-formal certificate issued 

by the local authority. Further, results presented in Table 7 show that a general land title (formal 

from  the  government  or  semi-formal  from  the  local  authority)  does  not  have  a  statistically 

significant  impact  on  fertilizer  expenditure.  Thus,  our  findings  reveal  that  land  titling  has  no 

significant effect on fertilizer expenditure. 

In contrast, for all sets of comparisons of land title vs. control, the point estimates for 

dichotomous  adoption  variables  are  statistically  significant.  As  can  be  seen  from  Table  7,  for 

instance, the average adoption rate of chemical fertilizer of plots with a formal land title is around 

7 percent higher than those of the control plots, though it is statistically significant at 10 percent 

level. Also, the impact for manure adoption rate is around 9 percent and it is statistically significant 

	

38 

at 5 percent level. In addition, estimates of the effect of semi-formal land title are approximately 

5 percent for fertilizer and 6 percent for manure.  

Table 7: Impacts of land titling 
 
Variables  

OLS  

 

Formal land title 
 
Observations 
R-squared 

Semi-formal ltitle 
 
Observations 
R-squared 

Formal land title  
 
Observations 
R-squared 

Fertilizer 
expense 

 

6.8680 
(10.5923) 
13,019 
0.5566 

5.6368 
(5.8038) 
14,187 
0.5668 

-4.8266 
(12.2486) 

8,588 
0.5634 

Land title from the government versus control group 

(Marginal effects) 

(Marginal effects) 

Probit regression 

Probit regression 

Fertilizer 
adoption 

0.0737* 
(0.0436) 
6,950 

Manure 
adoption 

0.0867** 
(0.0410) 
10,665 

Land title from local authority versus control group 
0.0634*** 
(0.0178) 
11,916 

0.0541** 
(0.0221) 
7,723 

 

 

 

 

0.0052 
(0.0418) 
4,017 

 

 

 

 

0.0556* 
(0.0337) 
6,650 

0.0652*** 
(0.0168) 
15,116 

OLS  
yield 

 

0.0491 
(0.0762) 
13,019 
0.5676 

0.0238 
(0.0324) 
14,187 
0.5888 

0.0407 
(0.0772) 
8,588 
0.6012 

0.0255 
(0.0301) 
17,772 
0.5575 

Land title from the government versus land title from local authority 

General land certificate versus control group 

General land title 
 
Observations 
R-squared 
Source: Calculation based on CSES 2009 to 2014 
Notes: (1) Clustered robust standard errors in parentheses 
            (2) *** p<0.01, ** p<0.05, * p<0.1 

5.5064 
(5.0396) 
17,772 
0.5400 

0.0527*** 
(0.0197) 
9,771 

The  third  set  of  comparisons  reveals  that  the  manure  adoption  rate  of  a  plot  with 

government  title  is  around  6  percent  higher  than  that  for  the  plot  with  certificate  from  local 

authority, at 10 percent significance level. This result indicates that formal land titling is likely to 

play a more important role than semi-formal titling in a longer term of soil improvement. This is 

because manure has benefits that last more than one season by adding organic substance to soil, 

but fertilizer has a shorter duration of benefits (Motavalli et al., 1994). 

	

39 

The empirical results show that both types of land title have positive and significant impact 

on the adoption of both chemical fertilizer and manure in Cambodia, and this is partially consistent 

with the previous studies by Markussen (2008) and Thin (2012), whose land title variable is only 

defined as a paper to certify land rights in general. Our results show that there is no significant 

effect of land titling on rice yield, which is not in line with findings by Markussen (2008) and Thin 

(2012) using the Cambodia-Socio-economic Survey (CSES) 2004. Regarding the impact of land 

titling  on  the  adoption  of  fertilizer  and  manure,  our  finding  is  similar  to  the  studies  in  other 

countries; for example, Li, Rozelle, & Brandt (1998) empirically showed a positive effect of the 

length of tenure and use of organic manure and phosphate fertilizer in maize production in rural 

northeast China. Additionally, Nguyen, Bauer, & Grote (2016) also found that land tenure security 

has a positive impact on manure use in the northern uplands of Vietnam.  

The insignificant effect of land titling on fertilizer expenditure indicates that although land 

titling  positively  affects  the  farmers’  decision  to  apply  the  productivity-enhancing  and  soil-

improving inputs, i.e. fertilizer and manure, it does not significantly influence the extent of the 

fertilizer expenditure. This finding seems to be in line with a qualitative study about the political 

economy  of  land  registration  in  Cambodia  by  So  (2009)  indicating  that  some  farmers  claim 

ownership by utilizing the land for farming, regardless of land titling, as long as they occupy the 

land. A possible explanation from the literature is that land rights play less important roles in short 

term investment such as expenditure for fertilizer (Fenske, 2011).  

Also, the study does not show any productivity effect and this can be explained by the 

insignificant impact of land titling on the extent of the fertilizer expenditure. The finding tends to 

be similar to that from a study in Indonesia by Suyanto, Tomich, & Otsuka (2001) as they found 

that land titling has no significant effect on the management efficiency of paddy production. Their 

	

40 

results suggested that with secure land ownership farmers cultivating long-term crops such as fruit 

trees are more likely than the ones cultivating short-term crops (e.g., rice) to invest in their land. 

Unfortunately, the sub-sample for long-term crops in our dataset is so small that it does not have 

enough statistical power for us to conduct the empirical analysis. This is because Cambodia’s 

agriculture is dominated by rice farming which accounts for the substantial share of cultivated 

land. Hence, this is a potential gap for further studies in Cambodia to look at the impact of land 

titling on the extent of the land investment for long-term or high-value crops, if the data about this 

kind of farming with sufficient sample size is available.  

2.7. Conclusion and policy implications  

This study provides evidence about the impact of land titling by its type, which differs 

according  to  the  formality  or  its  record  in  the  cadastral  register  of  the  Ministry  of  Land 

Management,  Urban  Planning  and  Construction.  Additionally,  this  article  aims  to  add  to  the 

literature on land titling and adoption of fertilizers by documenting the impact of land title on the 

adoption of chemical fertilizer and manure in Cambodia. 

I use propensity score matching (PSM) and regressions on matched samples by controlling 

village heterogeneity. The sample size of our dataset is 18,210 plots (12,142 households) in pooled 

cross-sectional data from a nationally representative household survey. The data come from the 

Cambodia Socio-Economic Survey (CSES), collected in 2009, 2010, 2011, 2012, 2013 and 2014 

by the National Institute of Statistics (NIS) of Cambodia. The empirical results suggest that both 

types of land title have a positive and significant impact on the adoption of chemical fertilizer and 

manure, but the impact of formal land titling on manure adoption is higher than that of semi-formal 

titling. Nonetheless, the productivity impact of land titling is insignificant. Overall, the impact of 

	

41 

formal  land  titling  is  comparable  to  that  of  informal  titling,  except  for  the  effect  on  manure 

adoption.  

This study used data for the entire country and did not address the possibility that land title 

has different impacts in different locations.  For example, it may be that a title makes a greater 

difference  in  locations  where  agriculture  is  more  commercialized  and  land  is  more  valuable, 

perhaps closer to cities.  This is an issue to address in subsequent studies. In addition, there is no 

information on the year of receiving the land title, which is an important variable indicating the 

length of tenure, which has a positive effect on the use manure and fertilizer in literature (Li, 

Rozelle, & Brandt, 1998). For instance, a plot with title issued ten years before the household 

survey would yield more productivity impact than one with title received a few months before that 

period. This is a limitation of this study that needs to be addressed in subsequent studies.  

The conceptual framework hypothesizes that land title from either the central government 

or  a  local  government  authority  increases  use  of  inputs  and  farm  investment,  leading  to  the 

improvements  in  soil  fertility  and  productivity.  The  empirical  evidence  confirms  that  the 

predictions are correct in a sense that land title increases probability of using fertilizer and manure. 

Also, the finding that formal land titling is likely to play a more important role than semi-formal 

titling in a longer term of soil improvement is consistent with what the conceptual framework 

hypothesized, because the former provides more secure land rights than the latter. However, the 

framework  does  not  accurately  predict  the  impact  of  land  titling  on  fertilizer  expenditure  and 

productivity. 

The insignificant effect of land titling on rice productivity implies that land tilting may not 

be essential for short-term crops such as rice. Additionally, the study suggests that land titling is 

not a problem for rice farmers, so the government has other priorities to improve rice productivity. 

	

42 

Since the focus of this study is the impact of land titling, the priorities for the government are 

beyond the scope of this essay and should be examined in further studies.  

These empirical findings lead to various policy implications. First, formal and semi-formal 

titles differ very little in their effect on input use and yields, so it is unlikely that most farmers will 

be willing to invest in a title under sporadic land titling. If farmers have to bear the cost of land 

titling they will more likely pursue semi-formal titling.  However, in the long term the formal land 

title will be more secure. Even if it has no productivity impact for rice farming, we should not 

conclude that it is not important, because it may still play an important role in a longer-term farm 

investment for high value crops such as pepper and rubber. Unfortunately, our data do not let us 

draw conclusions in this regard and this is a potential topic for further studies. Further, the long 

term  security  can  reduce  the  likelihood  of  land  grabbing,  which  would  severely  affect  the 

livelihood of the rural farm households (Markussen, 2008; So, 2009). 

Systematic land registration is the more likely approach to promote it due to its lower cost 

to farmers relative to sporadic land registration.  Alternatively, the government should continue 

reducing the cost of land transfer and simplifying the procedure in order for landholders to afford 

the land transfer and registration through the formal system, and it also allows central database of 

the cadastral offices to be updated, as suggested by Thin (2012) and So (2009). 

 

 

	

43 

APPENDICES

 44 

Appendix A: Extra tables and figures for the empirical analysis  
 
Table 8: Definitions of outcome variables and covariates 
Variables 
Outcome variables  
fertilizer 
adopt_fert 

Definitions 

cost of chemical fertilizer in '000' Riels/ha (in real terms) 
1 if fertilizer was applied on the plot in either the last completed dry 
season, wet season or both seasons. 
1 if manure was used on that plot in either the last completed dry 
season, wet season or both seasons. 
yield (tons/ha) defined as output divided by harvested area 

plot size in hectare (physical size) 
1: If plot is close to irrigation in both wet and dry seasons 
1: if plot was used for both wet- and dry-season rice  
1: if plot was for only dry-season rice 
1: if farmers have had conflict about this plot 
the number of years that plot has been acquired 

adopt_manure 
yield 
Covariates for Matching 
plot_size 
dum_bothirrigat 
dum_wetdryplot 
dum_dryplot 
dum_conflict 
year_acquisition 
dum_acquisition2  1: if the plot was acquired by inheritance 
dum_acquisition3  1: if the plot was bought from others 
dum_acquisition4  1: if the plot was acquired by clearing forest for free 
hh_age 
hh15_64 
dum_married 
dumhh_female 
dum_literate 
assetvalue 
dum_telev 
dum_cellp 
dum_motor 
dum_genar 
dum_htrac 
dum_wpump 
dum_electrif 
dum_notoilet 
dumwage 
dumlivestock 

Age of household head (years) 
household member aged from 15 to 64 
1 if head of household is married 
1 if head of household is female 
1: if head of household could read and write; 0 otherwise 
total asset value in '0000' riels (in real terms) 
1: if household has television  
1: if household has cell phone 
1: if household has motorbike 
1: if household has generator 
1: if household has hand tractor 
1: if household has water pump 
1: if household has access to public-provided electrify 
1: if there is no toilet for the household 
1: if one of the HH members involved in wage employment 
1: if HH engaged in livestock production in the past 12 months 

 45 

1: if HH engaged in non-farm act. in the past 12 months (e.g. tuk tuk 
driver) 
1: if HH received income from other sources such as remittance, 
pensions. 
1: if HH have outstanding loans 
1: if year= 2010 
1: if year= 2011 
1: if year= 2012 
1: if year= 2013 
1: if year= 2014 

dumother_inc 
dum_loan 
dumyear2 
dumyear3 
dumyear4 
dumyear5 
dumyear6 
Additional control variables for yield regression 
other_input 
fertilizer 
adopt_manure 

total input cost, excluding fertilizer and manure (in real terms) 
cost of chemical fertilizer in '000' Riels/ha (in real terms) 
1 if manure was used on that plot 

Table 8 (cont’d) 

dumnon_arg 

	
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 46 

Semi-formal 

181.476 
(185.404) 

Formal 
215.666 
(178.226) 

Table 9: Descriptive statistics for treatment and control plots  
Variables 
fertilizer 
 
adopt_fert 
 
adopt_manure 
 
yield 
 
plot_size 
 
dum_bothirrigate     
 
dum_wetdryplot 
 
dum_dryplot 
 
dum_conflict 
 
age_acquisition 
 
dum_acquisition2 
 
dum_acquisition3 
 
dum_acquisition4 
 
hh_age 
 
hh15_64 
 
dum_married 
 
dumhh_female 
 
dum_literate 
 
assetvalue 
 
dum_telev 

0.883 
(0.322) 
0.680 
(0.467) 
2.233 
(1.088) 
0.689 
(0.865) 
0.044 
(0.206) 
0.031 
(0.174) 
0.019 
(0.136) 
0.014 
(0.117) 
21.583 
(9.682) 
0.430 
(0.495) 
0.087 
(0.282) 
0.014 
(0.117) 
47.207 
(13.588)  
3.115 
(1.437) 
0.826 
(0.379) 
0.177 
(0.382) 
0.768 
(0.422) 
399.210 
(1421.518) 

0.793 
(0.405) 
0.579 
(0.494) 
2.265 
(1.226) 
0.981 
(1.206) 
0.054 
(0.226) 
0.028 
(0.164) 
0.110 
(0.312) 
0.020 
(0.139) 
20.034 
(10.154) 
0.368 
(0.482) 
0.210 
(0.407) 
0.027 
(0.162) 
47.423 
(13.347) 
3.191 
(1.474) 
0.836 
(0.371) 
0.167 
(0.373) 
0.735 
(0.441) 
372.507 
(1479.023) 

0.665         

0.642 

 47 

Control  Whole sample 
157.729 
(175.946) 

176.258 
(180.463) 

0.713 
(0.452) 
0.541 
(0.498) 
2.126 
(1.174) 
0.864 
(1.030) 
0.049 
(0.216) 
0.018 
(0.134) 
0.081 
(0.273) 
0.024 
(0.153) 
19.233 
(10.355) 
0.546 
(0.498) 
0.075 
(0.264) 
0.069 
(0.253) 
44.825 
(13.482) 
3.075 
(1.450) 
0.848 
(0.359) 
0.156 
(0.363) 
0.685 
(0.464) 
277.379 
(637.821) 

0.770 
(0.421) 
0.580 
(0.494) 
2.187 
(1.173) 
0.860 
(1.056) 
0.049 
(0.217) 
0.024 
(0.152) 
0.076 
(0.265) 
0.021 
(0.143) 
19.940 
(10.204) 
0.473 
(0.499) 
0.115 
(0.319) 
0.046 
(0.209) 
46.032 
(13.524) 
3.115 
(1.455) 
0.840 
(0.366) 
0.164 
(0.370) 
0.716 
(0.451) 
328.757 
(1110.481) 

0.532 

0.589 

(0.499) 
0.512 
0.512 
(0.500) 
0.521 
(0.500) 
0.022 
(0.148) 
0.148 
(0.355) 
0.170 
(0.419) 
0.149 
(0.356) 
0.706 
(0.455) 
0.411 
(0.492) 
0.874 
(0.332) 
0.224 
(0.417) 
0.539 
(0.498) 
0.427 
(0.495) 
9,434 

(0.492) 
0.556 
0.556 
(0.497) 
0.559 
(0.497) 
0.025 
(0.157) 
0.153 
(0.360) 
0.197 
(0.398) 
0.186 
(0.389) 
0.653 
(0.476) 
0.426 
(0.494) 
0.862 
(0.332) 
0.229 
(0.420) 
0.568 
(0.495) 
0.403 
(0.491) 
18,210 

Table 9 (cont’d) 
 
dum_cellp 
 
 
dum_motor 
 
dum_genar 
 
dum_htrac 
 
dum_wpump 
 
dum_electrif 
 
dum_notoilet 
 
dumwage 
 
dumlivestock 
 
dumnon_arg 
 
dumother_inc 
 
dum_loan 
 
Number of plots 
Source: Calculation based on CSES 2009 to 2014 
Note: standard deviations in parentheses 
	
	

(0.472) 
0.611 
(0.488) 
(0.488) 
0.596 
(0.491) 
0.028 
(0.165) 
0.145 
(0.352) 
0.227 
(0.419) 
0.241 
(0.428) 
0.570 
(0.495) 
0.467 
(0.499) 
0.879 
(0.326) 
0.241 
(0.428) 
0.583 
(0.493) 
0.418 
(0.493) 
3,773 

(0.480) 
0.598 
(0.490) 
(0.490) 
0.604 
(0.489) 
0.029 
(0.167) 
0.167 
(0.373) 
0.227 
(0.419) 
0.216 
0.411) 
0.615 
(0.487) 
0.423 
(0.494) 
0.825 
(0.380) 
0.228 
(0.419) 
0.612 
(0.487) 
0.347 
(0.476) 
5,003 

 48 

Figure 2: Distribution of propensity scores for matching between treatment group (land title 
from the government) and control group 

Kernel density estimate

Kernel density estimate

3

3

Before	
match	

After	match	

2

y
t
i
s
n
e
D

1

0

0

.2
.8
psmatch2: Propensity Score

.4

.6

2

y
t
i
s
n
e
D

1

1

0

0

.8
.2
psmatch2: Propensity Score

.4

.6

1

Kernel density estimate
kdensity _pscore

Kernel density estimate
kdensity _pscore

kernel = epanechnikov, bandwidth = 0.0224

kernel = epanechnikov, bandwidth = 0.0202

Source: Calculation based on CSES 2009 to 2014 
	
Figure 3: Distribution of propensity scores for matching between treatment group (land title 
from the local authority) and control group 

	

Kernel density estimate

Kernel density estimate

Before	match	

3

After	match	

2

y
t
i
s
n
e
D

1

1

0

0

.2
.8
psmatch2: Propensity Score

.6

.4

.2
.8
psmatch2: Propensity Score

.4

.6

3

2

y
t
i
s
n
e
D

1

0

0

Kernel density estimate
kdensity _pscore

Kernel density estimate
kdensity _pscore

kernel = epanechnikov, bandwidth = 0.0250

kernel = epanechnikov, bandwidth = 0.0225

Source: Calculation based on CSES 2009 to 2014 
	

 49 

1

	

Figure 4: Distribution of propensity scores for matching between treatment group (land title 
from the government) and land title from the local authority 

Kernel density estimate

Kernel density estimate

4

3

y
t
i
s
n
e
D

2

1

0

0

Before	match	

After	match	

4

3

y
t
i
s
n
e
D

2

1

.2

.4

psmatch2: Propensity Score

.6

.8

0

0

.2

psmatch2: Propensity Score

.4

.6

.8

Kernel density estimate
kdensity _pscore

Kernel density estimate
kdensity _pscore

kernel = epanechnikov, bandwidth = 0.0188

kernel = epanechnikov, bandwidth = 0.0182

Source: Calculation based on CSES 2009 to 2014 
	
Figure 5: Distribution of propensity scores for matching between treatment group ( general land 
title) and land title from the local authority 

	

Kernel density estimate

Kernel density estimate

3

3

Before	match	

After	match	

2

y
t
i
s
n
e
D

1

0

0

2

y
t
i
s
n
e
D

1

1

0

0

.2
.8
psmatch2: Propensity Score

.4

.6

.2
.8
psmatch2: Propensity Score

.4

.6

Kernel density estimate
kdensity _pscore

Kernel density estimate
kdensity _pscore

kernel = epanechnikov, bandwidth = 0.0197

kernel = epanechnikov, bandwidth = 0.0184

Source: Calculation based on CSES 2009 to 2014 

 50 

1

	

Table 10: Test of matching quality (land title from the government versus control group) 
T-stat 
Variables 

Sample 

Unmatched 

Unmatched 

plot_size 
 
dum_bothirrigat 
 
Unmatched 
dum_wetdryplot 
Matched 
 
Unmatched 
dum_dryplot 
Matched 
 
Unmatched 
dum_conflict 
Matched 
 
Unmatched 
age_acquisition 
 
Matched 
dum_acquisition2  Unmatched 
 
Matched 
dum_acquisition3  Unmatched 
 
Matched 
dum_acquisition4  Unmatched 
 
Matched 
Unmatched 
hh_age 
Matched 
 
Unmatched 
hh15_64 
 
Matched 
Unmatched 
dum_married 
Matched 
 
dumhh_female 
Unmatched 
 
Matched 
dum_literate 
 
assetvalue 
 
dum_telev 
 
dum_cellp 
 
dum_motor 
 
dum_genar 
 
dum_htrac 
 
dum_wpump 

Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 

Treated  Control  %bias  %reduc. 
 |bias| 
 
85.5 
 
61.5 
 
96.6 
 
98.3 
 
86.4 
 
92.2 
 
98.2 
 
52.7 
 
99.9 
 
96.9 
 
61.6 
 
74.9 
 
80.7 
 
92.5 
 
93.5 
 
92.1 
 
88.0 
 
78.2 
 
30.9 
 
-69.9 
 

0.68905  0.86371 
Matched   0.69602  0.72133 
0.04426  0.04918 
Matched   0.03905  0.04095 
0.03127  0.01823 
0.02538  0.02583 
0.01882  0.08098 
0.01953  0.02059 
0.01378  0.02406 
0.0145  0.01311 
21.583 
19.233 
21.233 
21.049 
0.43043  0.54622 
0.4516  0.44948 
0.08693  0.07537 
0.08619  0.09166 
0.01378  0.06879 
0.0145  0.01456 
44.825 
47.207 
46.641 
46.715 
3.0754 
3.1153 
3.1007 
3.116 
0.82613 
0.8481 
0.82706  0.83258 
0.17731  0.15624 
0.17573  0.17166 
0.68518 
0.759  0.76524 
399.21 
277.38 
320.06 
327.96 
0.66472  0.53159 
0.64798  0.65852 
0.61065  0.51166 
0.59414  0.60603 
0.59555  0.52078 
0.58243  0.59872 
0.02783  0.02226 
0.02678  0.02293 
0.14524  0.14787 
0.1431  0.14756 
0.22661  0.17013 

-18.4 
-2.7 
-2.3 
-0.9 
8.4 
-0.3 
-28.8 
-0.5 
-7.5 
1.0 
23.4 
1.8 
-23.3 
0.4 
4.2 
-2.0 
-27.9 
-0.0 
17.6 
0.5 
2.8 
-1.1 
-6.0 
-1.5 
5.7 
1.1 
18.7 
-1.4 
11.1 
-0.7 
27.4 
-2.2 
20.0 
-2.4 
15.1 
-3.3 
3.6 
2.5 
-0.7 
-1.3 
14.2 

Unmatched  076835 

-9.20*** 
-1.39 
-1.20 
-0.41 
4.62*** 
-0.12 
-13.35*** 
-0.32 
-3.71*** 
0.51 
12.00*** 
0.78 
-12.09*** 
0.18 
2.23** 
-0.81 
-12.82*** 
-0.02 
9.15*** 
0.23 
1.43 
-0.45 
-3.13*** 
-0.62 
2.97*** 
0.46 
9.54*** 
-0.62 
6.79*** 
-0.47 
14.06*** 
-0.94 
10.35*** 
-1.03 
7.81*** 
-1.40 
1.90* 
1.05 
-0.38 
-0.54 
7.55*** 

 51 

93.7 
 
87.1 
 
97.9 
 
93.7 
 
-3.5 
 
57.2 
 
95.3 
 
90.3 

-0.37 
12.74*** 
-1.23 
-15.13*** 
0.25 
5.95*** 
0.30 
0.83 
0.72 
2.09** 
-0.72 
4.61*** 
-0.18 
-0.96 
0.08 

Matched 

0.20335  0.20692 

-0.9 
0.14851  23.6 
0.22287 
-3.0 
0.70649  -28.6 
0.59548 
0.6 
0.41064  11.4 
0.45138 
0.7 
1.6 
0.87386 
1.7 
0.88184 
0.22398 
4.0 
-1.7 
0.22398 
8.9 
0.53922 
-0.4 
0.57612 
0.42739 
-1.9 
0.2 
0.41808 

Unmatched  0.24119 
Matched  0.21088 
Unmatched  0.57037 
Matched  0.59833 
Unmatched  0.46727 
Matched  0.45495 
Unmatched  0.87914 
Matched  0.88731 
Unmatched  0.24092 
Matched  0.23933 
Unmatched  0.58336 
Matched  0.57406 
Unmatched  0.41823 
Matched  0.41897 

Table 10 (cont’d) 
 
dum_electrif 
 
dum_notoilet 
 
dumwage 
 
dumlivestock 
 
dumnon_arg 
 
dumother_inc 
 
dum_loan 
 
Source: Calculation based on CSES 2009 to 2014.  *** p<0.01, ** p<0.05, * p<0.1 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 52 

Table 11: Test of matching quality (land title from local authority versus control group) 
Variables 

Treated  Control  %bias  %reduc.  

Sample 

T-stat 

6.14*** 
-0.66 
1.25 
0.29 
3.76*** 
0.57 
5.69*** 
-0.06 
-1.64 
0.32 
4.45*** 
-0.69 
-20.69*** 
0.06 
23.99*** 
0.21 
-10.59*** 
0.71 
11.06*** 
0.13 
4.53*** 
-0.55 
-1.95* 
0.07 
1.75* 
0.31 
6.28*** 
0.13 
5.38*** 
0.45 
12.80*** 
0.17 
9.99*** 
0.47 
9.58*** 
0.48 
2.41** 
-0.00 
3.07*** 
0.07 
8.32*** 

plot_size 
 
dum_bothirrigat 
 
dum_wetdryplot 
 
dum_dryplot 
 
dum_conflict 
 
age_acquisition 
 
dum_acquisition2 
 
dum_acquisition3 
 
dum_acquisition4 
 
hh_age 
 
hh15_64 
 
dum_married 
 
dumhh_female 
 
dum_literate 
 
assetvalue 
 
dum_telev 
 
dum_cellp 
 
dum_motor 
 
dum_genar 
 
dum_htrac 
 
dum_wpump 

Unmatched 

Unmatched 

0.98118 
Matched   0.93051 
0.05397 
Matched   0.05323 
0.02778 
0.02546 
0.10953 
0.10141 
0.01979 
0.02062 
20.034 
20.393 
0.36798 
0.38712 
0.21007 
0.17 
0.02698 
0.0284 
47.423 
47.328 
3.1909 
3.1828 
0.8357 
0.83169 
0.1675 
0.17105 
0.73536 
0.72649 
372.51 
337.31 
0.64181 
0.62866 
0.59844 
0.58489 
0.60384 
0.59205 
0.02878 
0.02672 
0.1673 
0.16074 
0.22706 

Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 

10.5 
-1.4 
2.2 
0.6 
6.4 
1.2 
9.7 
-0.1 
-2.9 
0.6 
7.8 
-1.4 
-36.4 
0.1 
39.2 
0.5 
-19.7 
1.1 
19.4 
0.3 
7.9 
-1.2 
-3.4 
0.1 
3.1 
0.6 
11.1 
0.3 
8.4 
0.5 
22.5 
0.4 
17.5 
1.0 
16.8 
1.0 
4.1 
0.0 
5.3 
0.2 
14.3 

0.86371 
0.94599 
0.04918 
0.05188 
0.01823 
0.02365 
0.08098 
0.10179 
0.02406 
0.01969 
19.233 
20.537 
0.54622 
0.38653 
0.07537 
0.1684 
0.06879 
0.02605 
44.825 
47.294 
3.0754 
3.1999 
0.8481 
0.83114 
0.15624 
0.16869 
0.68518 
0.72527 
277.38 
331.05 
0.53159 
0.62693 
0.51166 
0.58018 
0.52078 
0.58717 
0.02226 
0.02672 
0.14787 
0.16019 
0.17013 

 53 

|bias| 
 
86.8 
 
71.9 
 
81.1 
 
98.7 
 
78.3 
 
82.0 
 
99.7 
 
98.8 
 
94.4 
 
98.7 
 
85.2 
 
95.6 
 
79.1 
 
97.6 
 
93.4 
 
98.4 
 
94.6 
 
94.1 
 
100.0 
 
97.2 
 

3.0 
17.5 
1.8 
-19.4 
-0.6 
2.5 
0.9 
-13.6 
0.3 
0.9 
0.2 
14.7 
0.5 
-16.5 
-0.9 

79.1 
 
89.5 
 
97.1 
 
63.3 
 
98.0 
 
73.7 
 
96.9 
 
94.5 

1.42 
10.22*** 
0.86 
-11.19*** 
-0.27 
1.45 
0.45 
-7.95*** 
0.13 
0.50 
0.11 
8.39*** 
0.22 
-9.39*** 
-0.45 

Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 

0.20543 
0.21734 
0.14851 
0.21567 
0.196 
0.20303 
0.70649 
0.61523 
0.6332 
0.63055 
0.41064 
0.42315 
0.42083 
0.42542 
0.87386 
0.8253 
0.83535 
0.83631 
0.22398 
0.22766 
0.22668 
0.22765 
0.61183 
0.53922 
0.6032  060097 
0.42739 
0.34719 
0.3562 
0.36061 

Table 11 (cont’d) 
 
dum_electrif 
 
dum_notoilet 
 
dumwage 
 
dumlivestock 
 
dumnon_arg 
 
dumother_inc 
 
dum_loan 
 
Source: Calculation based on CSES 2009 to 2014.  *** p<0.01, ** p<0.05, * p<0.1 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 54 

Table 12: Test of matching quality (land title from the government versus local authority) 
T-stat 
Variables 

Treated  Control  %bias  %reduc. 

Sample 

Unmatched 
plot_size 
Matched  
 
Unmatched 
dum_bothirrigat 
Matched  
 
Unmatched 
dum_wetdryplot 
Matched 
 
Unmatched 
dum_dryplot 
Matched 
 
Unmatched 
dum_conflict 
Matched 
 
Unmatched 
age_acquisition 
 
Matched 
dum_acquisition2  Unmatched 
 
Matched 
dum_acquisition3  Unmatched 
 
Matched 
dum_acquisition4  Unmatched 
 
Matched 
Unmatched 
hh_age 
Matched 
 
Unmatched 
hh15_64 
 
Matched 
Unmatched 
dum_married 
Matched 
 
dumhh_female 
Unmatched 
Matched 
 
Unmatched 
dum_literate 
Matched 
 
assetvalue 
Unmatched 
Matched 
 
Unmatched 
dum_telev 
 
Matched 
Unmatched 
dum_cellp 
Matched 
 
Unmatched 
dum_motor 
 
Matched 
Unmatched 
dum_genar 
Matched 
 
dum_htrac 
Unmatched 
Matched 
 
dum_wpump 
Unmatched 

-27.8 
-4.2 
-4.5 
1.8 
2.1 
2.0 
-37.7 
-0.6 
-4.7 
-1.8 
15.6 
-2.3 
12.8 
1.8 
-35.2 
-0.1 
-9.4 
1.2 
-1.6 
-1.7 
-5.2 
0.4 
-2.6 
-0.9 
2.6 
0.2 
7.6 
-1.8 
1.8 
1.1 
4.8 
-4.0 
2.5 
-2.8 
-1.7 
-1.8 
-0.6 
-2.4 
-6.1 
-0.3 
-0.1 

0.68905  0.98118 
0.68687  0.73139 
0.04426  0.05397 
0.04184  0.03794 
0.03127 
.02778 
0.03096 
0.0275 
0.01882  0.10953 
0.01869  0.02008 
0.01378  0.01979 
0.01311  0.01545 
21.583 
20.034 
21.467 
21.692 
0.43043  0.36798 
0.4318  0.42315 
0.08693  0.21007 
0.08954  0.08993 
0.01378  0.02698 
0.0145  0.01278 
47.423 
47.207 
47.495 
47.26 
3.1909 
3.1153 
3.1166 
3.1101 
0.82613 
0.8357 
0.82315  0.82667 
0.17731 
0.1675 
0.17908  0.17819 
0.76835  0.73536 
0.75983  0.76764 
399.21 
372.51 
366.68 
350.77 
0.66472  0.64181 
0.65328  0.67219 
0.61065  0.59844 
0.59805  0.61177 
0.58996  0.60384 
0.60384 
0.599 
0.02783  0.02878 
0.02594  0.02996 
0.14524 
0.1673 
0.14282  0.14399 
0.22661  0.22706 

 55 

|bias| 
 
84.8 
 
59.8 
 
0.9 
 
98.5 
 
61.0 
 
85.5 
 
86.2 
 
99.7 
 
86.9 
 
-8.8 
 
91.4 
 
63.3 
 
90.9 
 
76.3 
 
40.4 
 
17.4 
 
-12.4 
 
-9.0 
 
-321.3 
 
94.7 
 

-12.63*** 
-2.68*** 
-2.07** 
0.84 
0.96 
0.87 
-16.69*** 
-0.43 
-2.14** 
-0.84 
7.22*** 
-0.98 
5.94*** 
0.74 
-15.92*** 
-0.06 
-4.24*** 
0.63 
-0.74 
-0.73 
-2.40** 
0.19 
-1.19 
-0.39 
1.21 
0.10 
3.53** 
-0.78 
0.85 
0.70 
2.23** 
-1.69* 
1.16 
-1.19 
-0.78 
-0.78 
-0.27 
-1.03 
-2.81*** 
-0.14 
-0.05 

-2.3 
6.1 
-1.6 
-9.1 
2.0 
8.9 
-2.0 
15.2 
0.7 
3.1 
-2.3 
-5.8 
1.6 
14.7 
-1.7 

-1978.4 
 
73.3 
 
77.9 
 
77.1 
 
95.6 
 
25.9 
 
71.8 
 
88.5 

-0.97 
2.83*** 
-0.68 
-4.24*** 
0.85 
4.12*** 
-0.86 
6.98*** 
0.30 
1.45 
-0.98 
-2.70*** 
0.69 
6.81*** 
-0.70 

Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 

0.21339  0.22282 
0.24119  0.21567 
0.2265  0.23331 
0.57037  0.61523 
0.59191  0.58198 
0.46727  0.42315 
0.45579  0.46589 
0.8253 
0.87914 
0.87364 
0.8713 
0.24092  0.22766 
0.23152  0.24134 
0.58336  0.61183 
0.59079  0.58276 
0.41823  0.34719 
0.3947  0.40285 

Table 12 (cont’d) 
 
dum_electrif 
 
dum_notoilet 
 
dumwage 
 
dumlivestock 
 
dumnon_arg 
 
dumother_inc 
 
dum_loan 
 
Source: Calculation based on CSES 2009 to 2014.  *** p<0.01, ** p<0.05, * p<0.1 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 56 

Table 13: Test of matching quality (general land title versus control group) 
Variables 

Treated  Control  %bias 

Sample 

 %reduc.  

Unmatched 
plot_size 
Matched  
 
Unmatched 
dum_bothirrigat 
Matched  
 
Unmatched 
dum_wetdryplot 
Matched 
 
Unmatched 
dum_dryplot 
Matched 
 
Unmatched 
dum_conflict 
Matched 
 
Unmatched 
age_acquisition 
 
Matched 
dum_acquisition2  Unmatched 
 
Matched 
dum_acquisition3  Unmatched 
 
Matched 
dum_acquisition4  Unmatched 
 
Matched 
Unmatched 
hh_age 
Matched 
 
Unmatched 
hh15_64 
 
Matched 
Unmatched 
dum_married 
Matched 
 
dumhh_female 
Unmatched 
Matched 
 
Unmatched 
dum_literate 
Matched 
 
assetvalue 
Unmatched 
Matched 
 
Unmatched 
dum_telev 
 
Matched 
Unmatched 
dum_cellp 
Matched 
 
Unmatched 
dum_motor 
 
Matched 
Unmatched 
dum_genar 
Matched 
 
dum_htrac 
Unmatched 
Matched 
 
dum_wpump 
Unmatched 

-0.8 
0.5 
0.3 
0.8 
7.3 
0.9 
-3.9 
-0.1 
-4.8 
-0.3 
14.4 
-0.3 
-30.7 
-2.7 
25.7 
1.9 
-23.0 
2.3 
18.6 
2.3 
5.7 
2.3 
-4.5 
0.2 
4.2 
-0.3 
14.3 
-0.6 
9.5 
1.5 
24.6 
-0.4 
18.6 
0.4 
16.1 
1.0 
3.9 
0.5 
2.8 
0.8 
14.3 

0.85559 
0.84274 
0.04979 
0.04713 
0.02928 
0.02423 
0.07053 
0.07232 
0.01721 
0.01751 
20.7 
20.811 
0.39483 
0.41521 
0.15713 
0.12797 
0.02131 
0.02159 
47.33 
47.145 
3.1584 
3.1544 
0.83159 
0.8297 
0.17172 
0.1727 
0.74954 
0.73915 
383.99 
337.08 
0.65166 
0.6354 
0.60369 
0.58947 
0.60027 
0.58779 
0.02837 
0.02663 
0.15782 
0.15363 
0.22687 

0.86371 
0.8379 
0.04918 
0.04533 
0.01823 
0.02288 
0.08098 
0.07258 
0.02406 
0.01792 
19.233 
20.844 
0.54622 
0.42871 
0.07537 
0.1218 
0.06879 
0.01679 
44.825 
46.84 
3.0754 
3.1208 
0.8481 
0.82881 
0.15624 
0.17395 
0.68518 
0.74181 
277.38 
320.02 
0.53159 
0.6374 
0.51166 
0.58746 
0.52078 
0.58285 
0.02226 
0.02579 
0.14787 
0.15076 
0.17013 

 57 

|bias| 
 
40.4 
 
-194.4 
 
87.8 
 
97.5 
 
94.1 
 
97.7 
 
91.1 
 
92.5 
 
89.9 
 
87.8 
 
59.6 
 
94.6 
 
91.9 
 
95.9 
 
84.0 
 
98.3 
 
97.8 
 
93.8 
 
86.3 
 
71.1 
 

T-stat 

-0.52 
0.31 
0.19 
0.55 
4.92*** 
0.57 
-2.66*** 
-0.07 
-3.24*** 
-0.20 
9.72*** 
-0.21 
-20.68*** 
-1.77* 
17.44*** 
0.229 
-15.40*** 
2.26** 
12.54*** 
1.47 
3.85*** 
1.47 
-3.04*** 
0.15 
2.82*** 
-0.21 
9.65*** 
-0.39 
6.48*** 
1.42 
16.58*** 
-0.27 
12.54*** 
0.26 
10.83*** 
0.65 
2.63*** 
0.34 
1.86* 
0.52 
9.63*** 

97.0 
 
95.6 
 
97.2 
 
97.5 
 
72.0 
 
63.7 
 
99.2 
 
98.1 

0.27 
13.60*** 
0.55 
-15.77*** 
0.40 
4.29*** 
0.10 
-4.96*** 
1.31 
1.51 
0.52 
8.23*** 
0.06 
-6.83*** 
-0.13 

0.21264 
0.22664 
0.20724 
0.59594 
0.61897 
0.44211 
0.43871 
0.84845 
0.86076 
0.23336 
0.23183 
0.59959 
0.59139 
0.37773 
0.38606 

0.21091 
0.14851 
0.20381 
0.70649 
0.61593 
0.41064 
0.43792 
0.87386 
0.85363 
0.22398 
0.22842 
0.53922 
0.59093 
0.42739 
0.38702 

0.4 
20.1 
0.9 
-23.4 
0.6 
6.4 
0.2 
-7.4 
2.1 
2.2 
0.8 
12.2 
0.1 
-10.1 
-0.2 

Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 

Table 13 (cont’d) 
 
dum_electrif 
 
dum_notoilet 
 
dumwage 
 
dumlivestock 
 
dumnon_arg 
 
dumother_inc 
 
dum_loan 
 
Source: Calculation based on CSES 2009 to 2014.  *** p<0.01, ** p<0.05, * p<0.1	
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 58 

Table 14: Impact of formal land title 

OLS 

Fertilizer exp. 

 
Variables 
 

Formal land title 
 
plot_size 
 
dum_bothirrigat 
 
dum_wetdryplot 
 
dum_dryplot 
 
dum_conflict 
 
age_acquisition 
 
hh_age 
 
hh15_64 
 
dum_married 
 
dumhh_female 
 
dum_literate 
 
assetvalue 
 
dum_telev 
 
dum_cellp 
 
dum_motor 
 
dum_genar 
 
dum_htrac 
 
dum_wpump 
 
dum_electrif 
 

 

6.8680 
(10.5923) 
-10.1827*** 

(2.2963) 
15.2472 
(11.5449) 
59.2753*** 
(17.5123) 

108.9169*** 

(25.7502) 
4.1914 
(16.0069) 
-0.2007 
(0.2419) 
-0.3163 
(0.2011) 
-0.7002 
(1.6595) 
2.1229 
(8.4352) 
-1.5129 
(9.0382) 
14.4746** 
(6.6905) 
-0.0052** 
(0.0026) 
5.3516 
(4.5463) 
2.8521 
(5.2912) 
8.6025* 
(4.6349) 
-11.1036 
(14.6692) 
6.9782 
(6.5587) 
10.4308** 
(5.2983) 
-13.1271 
(10.7553) 

Probit regression 
Manure adopt. 
(Marginal effects) 

0.0867** 
(0.0410) 
-0.0015 
(0.0086) 
0.0949** 
(0.0461) 
0.0260 
(0.0403) 
-0.3726*** 
(0.0731) 
-0.0500 
(0.0420) 
0.0017** 
(0.0008) 
-0.0004 
(0.0006) 
0.0160*** 
(0.0052) 
-0.0011 
(0.0316) 
-0.0297 
(0.0320) 
-0.0158 
(0.0132) 
0.0000 
(0.0000) 
0.0287* 
(0.0148) 
-0.0348** 
(0.0158) 
0.0144 
(0.0128) 
0.0020 
(0.0424) 
-0.0196 
(0.0186) 
0.0182 
(0.0244) 
-0.0118 
(0.0255) 

Probit regression 
Fertilizer adopt. 
(Marginal effects) 

0.0737* 
(0.0436) 
0.0179* 
(0.0106) 
-0.0697* 
(0.0415) 
0.2952*** 
(0.0923) 
0.3649*** 
(0.0405) 
0.0134 
(0.0559) 
-0.0001 
(0.0009) 
-0.0010 
(0.0006) 
-0.0059 
(0.0054) 
-0.0048 
(0.0297) 
-0.0536* 
(0.0305) 
-0.0020 
(0.0181) 
0.0000 
(0.0000) 
0.0186 
(0.0208) 
0.0144 
(0.0163) 
0.0487*** 
(0.0164) 
0.0605 
(0.0428) 
-0.0220 
(0.0237) 
0.0181 
(0.0171) 
-0.0568 
(0.0360) 

 59 

OLS 
yield 

 

0.0491 
(0.0762) 
-0.0893*** 
(0.0227) 
0.0391 
(0.0650) 
0.4405*** 
(0.0999) 
0.6236*** 
(0.1411) 
0.1940 
(0.1250) 
-0.0002 
(0.0018) 
-0.0007 
(0.0011) 
0.0003 
(0.0088) 
0.0338 
(0.0569) 
-0.0054 
(0.0616) 
0.0011 
(0.0321) 
0.0000* 
(0.0000) 
0.0206 
(0.0269) 
-0.0120 
(0.0286) 
0.0487 
(0.0316) 
0.0303 
(0.0746) 
0.0455 
(0.0395) 
0.0656** 
(0.0324) 
0.0350 
(0.0562) 

-14.9829*** 

-0.0253 
(0.0213) 
-0.0396** 
(0.0163) 
0.0150 
(0.0289) 
-0.0194 
(0.0194) 
0.0011 
(0.0182) 
0.0116 
(0.0140) 
0.0775** 
(0.0392) 
0.0766 
(0.0473) 
0.3333*** 
(0.0736) 
0.4185*** 
(0.0734) 
0.5268*** 
(0.0859) 

3.8643 
(6.3862) 
-1.3608 
(4.5148) 
4.1903 
(6.7801) 
7.5285 
(6.8104) 

Table 14 (cont’d) 
dum_notoilet 
 
dumwage 
 
dumlivestock 
 
dumnon_arg 
 
dumother_inc 
 
dum_loan 
 
dumyear2 
 
dumyear3 
 
dumyear4 
 
dumyear5 
 
dumyear6 
 
other_input 
 
fertilizer 
 
adopt_manure 
 
Constant 
 
Observations 
R-squared 
 
Notes: (1) Clustered robust standard errors in parentheses 
            (2) *** p<0.01, ** p<0.05, * p<0.1 

(5.3871) 
4.4620 
(4.1194) 
-14.4666 
(12.5851) 
14.9893 
(12.4383) 
71.7013*** 
(25.0404) 
86.3969*** 
(24.1509) 
74.1004*** 
(22.9751) 

386.7443*** 

(15.1653) 
13,019 
0.5566 

 
 
 
 
 
 

     (3) Village dummies are controlled for 

 
 
 
 
 
 
 
 
 
 

-0.0083 
(0.0153) 
-0.0166 
(0.0167) 
0.2718*** 
(0.0262) 
-0.0590*** 
(0.0160) 
0.0093 
(0.0196) 
-0.0398*** 
(0.0130) 
0.0197 
(0.0361) 
-0.0255 
(0.0321) 
-0.0175 
(0.0973) 
-0.0822 
(0.0805) 
-0.1025 
(0.0806) 

 
 
 
 
 
 
 
 

 

0.0363 
(0.0282) 
-0.0075 
(0.0267) 
0.0720* 
(0.0436) 
-0.0424 
(0.0334) 
0.0474 
(0.0339) 
-0.1190*** 
(0.0270) 
-0.0900 
(0.0924) 
-0.0893 
(0.0687) 
-0.2090 
(0.1669) 
-0.1699 
(0.1834) 
-0.0066 
(0.1424) 
0.0007*** 
(0.0001) 
0.0013*** 
(0.0001) 
0.0635** 
(0.0309) 
2.9986*** 
(0.1237) 
13,019 
0.5676 

6,950 

10,665 

 
 
 
 
 
 
 
 

 60 

 

-12.3181*** 

5.6368 
(5.8038) 

Table 15: Impact of semi-formal land title 
 
Variables 
 
Semi-formal title 
 
plot_size 
 
dum_bothirrigat 
 
dum_wetdryplot 
 
dum_dryplot 
 
dum_conflict 
 
age_acquisition 
 
hh_age 
 
hh15_64 
 
dum_married 
 
dumhh_female 
 
dum_literate 
 
assetvalue 
 
dum_telev 
 
dum_cellp 
 
dum_motor 
 
dum_genar 
 
dum_htrac 
 
dum_wpump 
 
dum_electrif 
 

(1.7220) 
29.2406** 
(13.6819) 
25.2349 
(22.3045) 
90.1600*** 
(13.1171) 
2.8245 
(8.2777) 
-0.3779* 
(0.1964) 
-0.1745 
(0.1360) 
-1.0501 
(1.2107) 
3.3411 
(7.9468) 
7.3551 
(7.2160) 
3.7288 
(3.6038) 
-0.0069*** 
(0.0023) 
6.3520* 
(3.8113) 
4.5242 
(4.4017) 
5.6534 
(3.8361) 
1.8350 
(8.9091) 
7.8880 
(5.5110) 
11.5671** 
(4.6246) 
6.9607 
(7.4956) 

0.0541** 
(0.0221) 
0.0090 
(0.0070) 
-0.0454 
(0.0645) 
0.3506*** 
(0.0938) 
0.3099*** 
(0.0496) 
-0.0326 
(0.0355) 
0.0005 
(0.0008) 
-0.0012** 
(0.0006) 
-0.0008 
(0.0044) 
0.0155 
(0.0291) 
0.0276 
(0.0293) 
0.0409*** 
(0.0140) 
-0.0000 
(0.0000) 
0.0165 
(0.0139) 
-0.0039 
(0.0156) 
0.0297** 
(0.0150) 
0.0368 
(0.0430) 
0.0063 
(0.0243) 
0.0287 
(0.0195) 
0.0258 
(0.0299) 

 61 

OLS 

Probit regression 
Fertilizer exp.  Fertilizer adopt.  Manure adopt. 
(Marginal effects) 

(Marginal effects) 

Probit regression 

 

OLS 
yield 

0.0238 
(0.0324) 
-0.0950*** 
(0.0119) 
-0.0786 
(0.0687) 
0.3392*** 
(0.0852) 
0.4781*** 
(0.0743) 
0.1238* 
(0.0633) 
0.0002 
(0.0013) 
-0.0018* 
(0.0010) 
0.0094 
(0.0077) 
-0.0133 
(0.0469) 
-0.0558 
(0.0457) 
0.0673*** 
(0.0242) 
0.0000 
(0.0000) 
-0.0009 
(0.0244) 
-0.0208 
(0.0267) 
0.0308 
(0.0245) 
0.0056 
(0.0703) 
0.0838*** 
(0.0317) 
0.0452 
(0.0317) 
0.0108 
(0.0512) 

0.0634*** 
(0.0178) 
0.0010 
(0.0055) 
-0.0277 
(0.0369) 
-0.0104 
(0.0456) 
-0.3782*** 
(0.0372) 
-0.0467 
(0.0371) 
0.0038*** 
(0.0008) 
-0.0009* 
(0.0005) 
0.0144*** 
(0.0046) 
0.0485* 
(0.0249) 
-0.0159 
(0.0243) 
-0.0126 
(0.0133) 
-0.0000 
(0.0000) 
0.0233** 
(0.0116) 
-0.0142 
(0.0142) 
0.0035 
(0.0128) 
0.0445 
(0.0349) 
-0.0317* 
(0.0170) 
0.0112 
(0.0163) 
-0.0343 
(0.0279) 

0.0151 
(0.0144) 
-0.0268** 
(0.0119) 
0.2502*** 
(0.0194) 
0.0002 
(0.0141) 
-0.0195 
(0.0150) 
-0.0334*** 
(0.0120) 
0.0111 
(0.0310) 
-0.0335 
(0.0312) 
0.0510 
(0.0659) 
-0.0090 
(0.0643) 
-0.1029 
(0.0922) 

 
 
 
 
 
 
 
 

 

0.0190 
(0.0264) 
-0.0251 
(0.0252) 
0.0179 
(0.0365) 
-0.0413 
(0.0283) 
0.0158 
(0.0318) 
-0.0755*** 
(0.0207) 
-0.1429** 
(0.0655) 
-0.0136 
(0.0728) 
0.0581 
(0.1468) 
-0.1218 
(0.1252) 
0.3030*** 
(0.1126) 
0.0009*** 
(0.0001) 
0.0012*** 
(0.0001) 
0.0632** 
(0.0270) 
3.2118*** 
(0.0997) 
14,187 
0.5888 

7,723 

11,916 

-0.0108 
(0.0149) 
-0.0252* 
(0.0131) 
0.0215 
(0.0201) 
0.0030 
(0.0155) 
-0.0130 
(0.0182) 
-0.0057 
(0.0125) 
0.0837*** 
(0.0301) 
0.1313*** 
(0.0335) 
0.2469*** 
(0.0807) 
0.3602*** 
(0.0879) 
0.4397*** 
(0.1244) 

-0.3477 
(4.2512) 
-4.6537 
(3.4590) 
2.4124 
(4.7986) 
4.8683 
(4.2979) 
-6.6939 
(4.9226) 
6.8850* 
(3.8435) 

Table 15 (cont’d) 
dum_notoilet 
 
dumwage 
 
dumlivestock 
 
dumnon_arg 
 
dumother_inc 
 
dum_loan 
 
dumyear2 
 
dumyear3 
 
dumyear4 
 
dumyear5 
 
dumyear6 
 
other_input 
 
fertilizer 
 
adopt_manure 
 
Constant 
 
Observations 
R-squared 
 
Notes: (1) Clustered robust standard errors in parentheses 
            (2) *** p<0.01, ** p<0.05, * p<0.1 

-28.3339*** 
(10.9431) 
-7.3284 
(12.0657) 
24.3958 
(36.0669) 
45.1636 
(41.8290) 
49.4223* 
(25.6067) 

399.9622*** 

(12.6403) 
14,187 
0.5668 

 
 
 
 
 
 

     (3) Village dummies are controlled for 

 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 

 62 

Table 16: Formal land title versus semi-formal land title 
Probit regression 

OLS 

Probit regression 
Fertilizer exp.  Fertilizer adopt.  Manure adopt. 
 
(Marginal effects) 

(Marginal effects) 

 
Variables 
 
Formal land title 
 
plot_size 
 
dum_bothirrigat 
 
dum_wetdryplot 
 
dum_dryplot 
 
dum_conflict 
 
age_acquisition 
 
hh_age 
 
hh15_64 
 
dum_married 
 
dumhh_female 
 
dum_literate 
 
assetvalue 
 
dum_telev 
 
dum_cellp 
 
dum_motor 
 
dum_genar 
 
dum_htrac 
 
dum_wpump 
 
dum_electrif 
 

-4.8266 
(12.2486) 
-15.4065*** 

(2.8502) 
22.8878 
(17.8864) 
38.4322 
(23.7801) 

123.1076*** 

(28.5167) 
-12.3067 
(18.1017) 
-0.4087 
(0.2899) 
-0.0665 
(0.2192) 
-1.6183 
(1.8266) 
3.2480 
(10.2623) 
2.8703 
(10.4246) 
11.8697* 
(6.8640) 
-0.0015** 
(0.0006) 
10.8084** 
(5.0098) 
2.3420 
(6.0019) 
5.3818 
(5.2341) 
-5.4824 
(15.3972) 
10.9882 
(7.5765) 
8.3647 
(6.2115) 
4.4593 
(12.8568) 

 

OLS 
yield 

0.0407 
(0.0772) 
-0.1101*** 
(0.0202) 
0.0560 
(0.0856) 
0.3551*** 
(0.1105) 
0.8042*** 
(0.1719) 
0.0505 
(0.1015) 
-0.0026 
(0.0020) 
0.0005 
(0.0012) 
0.0019 
(0.0109) 
0.0471 
(0.0622) 
0.0609 
(0.0652) 
0.0479 
(0.0382) 
0.0000 
(0.0000) 
0.0226 
(0.0305) 
-0.0088 
(0.0313) 
0.0473 
(0.0342) 
0.0626 
(0.0895) 
0.0747 
(0.0475) 
0.0322 
(0.0344) 
-0.0080 
(0.0627) 

0.0556* 
(0.0337) 
0.0105 
(0.0097) 
0.0813 
(0.0578) 
-0.0948* 
(0.0518) 
-0.3404*** 
(0.0751) 
-0.0869 
(0.0545) 
0.0024*** 
(0.0009) 
-0.0006 
(0.0007) 
0.0245*** 
(0.0066) 
0.0520 
(0.0356) 
-0.0137 
(0.0358) 
-0.0147 
(0.0178) 
-0.0000** 
(0.0000) 
0.0201 
(0.0161) 
-0.0513*** 
(0.0191) 
0.0395** 
(0.0165) 
0.0274 
(0.0498) 
-0.0511** 
(0.0235) 
0.0216 
(0.0215) 
-0.0366 
(0.0324) 

0.0052 
(0.0418) 
0.0146 
(0.0105) 
-0.0568 
(0.0705) 
0.2308** 
(0.1033) 
0.2960*** 
(0.0819) 
-0.0412 
(0.0606) 
0.0004 
(0.0011) 
-0.0008 
(0.0008) 
-0.0067 
(0.0065) 
-0.0296 
(0.0421) 
-0.0348 
(0.0423) 
0.0256 
(0.0222) 
-0.0000 
(0.0000) 
0.0073 
(0.0244) 
0.0131 
(0.0209) 
0.0381* 
(0.0210) 
0.0660 
(0.0461) 
0.0107 
(0.0371) 
0.0066 
(0.0220) 
0.0463 
(0.0505) 

 63 

0.0147 
(0.0171) 
-0.0265 
(0.0168) 
0.2598*** 
(0.0279) 
-0.0486*** 
(0.0180) 
-0.0425** 
(0.0208) 
-0.0271 
(0.0167) 
0.0667 
(0.0465) 
0.0020 
(0.0382) 
-0.2069* 
(0.1256) 
-0.1600 
(0.1169) 
-0.1179 
(0.1027) 

 
 
 
 
 
 
 
 

 

0.0490 
(0.0309) 
-0.0259 
(0.0327) 
0.0248 
(0.0457) 
-0.0591* 
(0.0325) 
0.0060 
(0.0367) 
-0.1103*** 
(0.0302) 
-0.0822 
(0.1091) 
-0.0228 
(0.0877) 
0.1478 
(0.2616) 
0.1752 
(0.2309) 
0.0827 
(0.2008) 
0.0006*** 
(0.0001) 
0.0013*** 
(0.0001) 
0.0784** 
(0.0347) 
2.4433*** 
(0.2288) 
8,588 
0.6012 

4,017 

6,650 

-0.0125 
(0.0234) 
-0.0519*** 
(0.0196) 
0.0209 
(0.0349) 
-0.0312 
(0.0213) 
-0.0031 
(0.0232) 
-0.0004 
(0.0181) 
0.0652 
(0.0479) 
0.0210 
(0.0456) 
0.4285*** 
(0.1320) 
0.5092*** 
(0.1424) 
0.2920* 
(0.1578) 

1.1623 
(6.3498) 
-4.7395 
(4.7959) 
3.9842 
(7.3735) 
2.9040 
(6.6563) 
-12.9994** 
(6.1361) 
10.3416** 
(4.5338) 
-11.5975 
(15.7007) 
18.8188 
(15.3295) 
32.1072 
(46.2708) 
58.4377 
(45.2337) 
25.7893 
(35.4716) 

Table 16 (cont’d) 
dum_notoilet 
 
dumwage 
 
dumlivestock 
 
dumnon_arg 
 
dumother_inc 
 
dum_loan 
 
dumyear2 
 
dumyear3 
 
dumyear4 
 
dumyear5 
 
dumyear6 
 
other_input 
 
fertilizer 
 
adopt_manure 
 
Constant 
 
Observations 
R-squared 
 
Notes: (1) Clustered robust standard errors in parentheses 
            (2) *** p<0.01, ** p<0.05, * p<0.1 
            (3) Village dummies are controlled for 

8,588 
0.5634 

105.6117*** 

(40.5321) 

 
 
 
 
 
 

 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 

 64 

-12.4644*** 

5.5064 
(5.0396) 

Table 17: Impact of general land title 
 
Variables 
 
Land title  
 
plot_size 
 
dum_bothirrigat 
 
dum_wetdryplot 
 
dum_dryplot 
 
dum_conflict 
 
age_acquisition 
 
hh_age 
 
hh15_64 
 
dum_married 
 
dumhh_female 
 
dum_literate 
 
assetvalue 
 
dum_telev 
 
dum_cellp 
 
dum_motor 
 
dum_genar 
 
dum_htrac 
 
dum_wpump 
 
dum_electrif 
 

(1.6086) 
19.6567* 
(10.2726) 
44.1728*** 
(16.8690) 
95.2694*** 
(12.8188) 
3.6357 
(9.8480) 
-0.2178 
(0.1812) 
-0.2601* 
(0.1391) 
-0.7575 
(1.1051) 
2.3089 
(6.3854) 
1.6029 
(6.1855) 
6.1536* 
(3.6241) 
-0.0049** 
(0.0020) 
4.1871 
(2.9776) 
3.2883 
(3.8840) 
7.7752** 
(3.3382) 
-2.6659 
(8.9723) 
8.0023* 
(4.5202) 

(4.0176) 
-2.9246 
(6.9123) 

11.4201*** 

0.0527*** 
(0.0197) 
0.0118** 
(0.0058) 
-0.0544 
(0.0458) 
0.3095*** 
(0.0796) 
0.3060*** 
(0.0431) 
-0.0208 
(0.0367) 
0.0006 
(0.0007) 
-0.0011** 
(0.0005) 
-0.0024 
(0.0037) 
0.0071 
(0.0240) 
-0.0114 
(0.0242) 
0.0249** 
(0.0127) 
-0.0000 
(0.0000) 
0.0166 
(0.0130) 
0.0036 
(0.0126) 
0.0382*** 
(0.0122) 
0.0261 
(0.0318) 
-0.0030 
(0.0189) 
0.0271* 
(0.0151) 
-0.0043 
(0.0241) 

 65 

OLS 

Probit regression 
Fertilizer exp.  Fertilizer adopt.  Manure adopt. 
(Marginal effects) 

(Marginal effects) 

Probit regression 

 

 

OLS 
yield 

0.0255 
(0.0301) 
-0.1010*** 
(0.0114) 
-0.0248 
(0.0588) 
0.4092*** 
(0.0801) 
0.5230*** 
(0.0710) 
0.1193** 
(0.0604) 
-0.0000 
(0.0011) 
-0.0012 
(0.0009) 
0.0032 
(0.0065) 
-0.0029 
(0.0428) 
-0.0419 
(0.0431) 
0.0424** 
(0.0204) 
0.0000** 
(0.0000) 
0.0079 
(0.0198) 
-0.0126 
(0.0218) 
0.0411* 
(0.0211) 
0.0093 
(0.0568) 
0.0734*** 
(0.0277) 
0.0440 
(0.0272) 
0.0081 
(0.0457) 

0.0652*** 
(0.0168) 
-0.0017 
(0.0049) 
0.0191 
(0.0307) 
-0.0061 
(0.0362) 
-0.3930*** 
(0.0336) 
-0.0570* 
(0.0293) 
0.0030*** 
(0.0006) 
-0.0008* 
(0.0004) 
0.0133*** 
(0.0037) 
0.0317 
(0.0214) 
-0.0200 
(0.0212) 
-0.0159 
(0.0106) 
-0.0000 
(0.0000) 
0.0253** 
(0.0103) 
-0.0186 
(0.0117) 
0.0044 
(0.0103) 
0.0475 
(0.0297) 
-0.0264* 
(0.0146) 
0.0165 
(0.0142) 
-0.0188 
(0.0211) 

0.0067 
(0.0118) 
-0.0246** 
(0.0101) 
0.2589*** 
(0.0170) 
-0.0269** 
(0.0118) 
-0.0095 
(0.0130) 
-0.0363*** 
(0.0100) 
0.0173 
(0.0272) 
-0.0340 
(0.0256) 
0.0091 
(0.0655) 
-0.0510 
(0.0575) 
-0.0925 
(0.0731) 

 
 
 
 
 
 
 
 

 

0.0163 
(0.0222) 
-0.0157 
(0.0204) 
0.0240 
(0.0320) 
-0.0534** 
(0.0237) 
0.0170 
(0.0261) 
-0.0895*** 
(0.0182) 
-0.1138* 
(0.0594) 
-0.0326 
(0.0580) 
-0.0602 
(0.1159) 
-0.1672 
(0.1092) 
0.1617 
(0.0997) 
0.0008*** 
(0.0001) 
0.0013*** 
(0.0001) 
0.0732*** 
(0.0229) 
3.1424*** 
(0.0892) 
17,772 
0.5575 

9,771 

15,116 

-11.4041*** 

1.2026 
(3.8797) 
-0.8536 
(2.9451) 
3.9118 
(4.2712) 
6.6230* 
(3.9584) 

-0.0185 
(0.0140) 
-0.0336*** 
(0.0116) 
0.0155 
(0.0187) 
-0.0092 
(0.0137) 
-0.0087 
(0.0145) 
-0.0007 
(0.0105) 
0.0746*** 
(0.0283) 
0.1096*** 
(0.0295) 
0.2314*** 
(0.0635) 
0.3394*** 
(0.0678) 
0.4294*** 
(0.0893) 

Table 17 (cont’d) 
dum_notoilet 
 
dumwage 
 
dumlivestock 
 
dumnon_arg 
 
dumother_inc 
 
dum_loan 
 
dumyear2 
 
dumyear3 
 
dumyear4 
 
dumyear5 
 
dumyear6 
 
other_input 
 
fertilizer 
 
adopt_manure 
 
Constant 
 
Observations 
R-squared 
Notes: (1) Clustered robust standard errors in parentheses 
            (2) *** p<0.01, ** p<0.05, * p<0.1 

(4.0144) 
6.9520** 
(3.2709) 
-21.6652** 
(9.7422) 
4.1838 
(9.7754) 
26.2090 
(29.6156) 
51.5167* 
(30.7990) 
57.6531*** 
(20.7487) 

394.2086*** 

(11.2584) 
17,772 
0.5400 

 
 
 
 
 
 

     (3) Village dummies are controlled for 

 
 
 
 
 
 
 
 
 
	

 
 
 
 
 
 
 
 

 

 66 

 Appendix B: Choosing the matching technique or algorithm  
 
 

Once the propensity score is predicted, matching plots between the treatment and control 

groups is done using one of a number of matching techniques, namely nearest-neighbor, caliper or 

radius matching, stratification or interval matching, or kernel matching. Zhao (2004) proposed that 

none of the matching techniques are necessarily superior to the others for all situations, and that it 

depends the data structure. According to Khandker et al. (2010), each matching technique has its 

pros and cons as follows. 

 

Nearest-neighbor (NN) matching is one of the most commonly used matching techniques. In 

this approach, each treatment unit is matched to the control unit with the closest propensity score, 

and the frequently chosen number of nearest neighbors is 5 (n=5). This can be conducted with or 

without replacement. The former means the same control unit can be used as a match for different 

treatment  units,  and  it  increases  the  average  quality  of  match  and  reduces  bias.  Nonetheless, 

efficiency of this technique declines as variance (distance between high-score treatment units and 

low-score control units) increases. 

 

Caliper  or  radius  matching  can  deal  with  the  high  variance  in  propensity  scores  for  a 

treatment unit and its closest control neighbor by imposing a threshold on the highest propensity 

score distance (caliper). Thus, this approach involves matching with replacement, only within a 

particular range of propensity score. However, this matching technique is sensitive to the width of 

the radius, and it is difficult know what choice for the threshold level is reasonable (Caliendo & 

Kopeinig, 2008).  

 

Stratification or interval matching is a matching technique that partitions the common support 

(matched sample where distribution of propensity score for treatment and control group overlap) 

into  various  intervals  (strata)  and  estimates  the  impact  of  intervention  within  each  interval  by 

 67 

taking  the  difference  in  outcome  variables  between  treatment  and  control  groups.  The  overall 

intervention impact is calculated by a weighted average of these interval impact estimates with the 

share of treated observations in each interval being the weights.  

 

The matching methods described so far have a common potential problem that “only a few 

observations from the comparison group are used to construct the counterfactual outcome of the 

treated ones” (Caliendo & Kopeinig, 2008, p. 43). Kernel matching, a non-parametric method, 

uses a weighted average of all control units to construct the counterfactual match for each treatment 

observation. Hence, a major merit of this technique is the lower variance, which is obtained by 

using more information (Caliendo & Kopeinig, 2008). The potential drawback of this approach is 

that we possibly use observations that are bad matches. 

	
	

	
	
	
	
	
	
	
	
	

 

 

 

 

 

 

 

 68 

REFERENCES

 69 

REFERENCES 

 

Abdulai, A., Owusu, V., & Goetz, R. (2011). Land tenure differences and investment in land 

improvement measures: Theoretical and empirical analyses. Journal of Development 
Economics, 96(1), 66–78. 

ADB-Asian Development Bank. (2014). Key Indicators for Cambodia, 1996-2013. Manila. 

Retrieved from www.adb.org/statistics 

Asfaw, S., Kassie, M., Simtowe, F., & Lipper, L. (2012). Poverty Reduction Effects of 

Agricultural Technology Adoption: A Micro-evidence from Rural Tanzania. Journal of 
Development Studies, 48(9), 1288–1305. 

Atwood, D. A. (1990). Land registration in Africa: The impact on agricultural production. World 

Development, 18(5), 659–671. 

Banerjee, A. V, Gertler, P. J., & Ghatak, M. (2002). Empowerment and Efficiency : Tenancy 

Reform in West Bengal. Journal of Political Economy, 110(2), 239–280. 

Baum, C. F. (2006). An Introduction to Modern Econometrics Using Stata (1 edition). College 

Station, Tex: Stata Press. 

Benjaminsen, T. A., Holden, S., Lund, C., & Sjaastad, E. (2008). Formalisation of land rights: 
Some empirical evidence from Mali, Niger and South Africa. Land Use Policy, 26(1), 28–
35. 

Besley, T. (1995). Property Rights and Investment Incentives: Theory and Evidence from Ghana. 

Journal of Political Economy, 103(5), 903. 

Blimpo, M. P. (2014). Team Incentives for Education in Developing Countries: A Randomized 

Field Experiment in Benin. American Economic Journal: Applied Economics, 6(4), 90–109. 

Boreak, S. (2000). Land Ownership, Sales and Concentration in Cambodia: A Preliminary 

Review of Secondary Data and Primary Data from Four Recent Surveys (No. 16). Phnom 
Penh: CDRI. Retrieved from http://www.cdri.org.kh/webdata/download/wp/wp16e.pdf 

Brasselle, A. S., Gaspart, F., & Platteau, J. P. (2002). Land tenure security and investment 

incentives: Puzzling evidence from Burkina Faso. Journal of Development Economics, 
67(2), 373–418. 

Caliendo, M., & Kopeinig, S. (2008). Some practical guidance for the implementation of 

propensity score matching. Journal of Economic Surveys, 22(1), 31–72. 

CDRI -Cambodia Development Resource Institute. (2013). Removing Constraints to Cambodia’s 

Agricultural Development. Cambodia Outlook Brief, (2), 4. Retrieved from 
http://www.cdri.org.kh/webdata/policybrief/ob13/ob2e.pdf 

	

70 

Cerulli, G. (2015). Econometric Evaluation of Socio-Economic Programs: Theory and 

Applications. New York: Springer. 

Chan, S., Tep, S., & Sarthi, A. (2001). Land Tenure in Cambodia: A Data Update. Retrieved 

from http://www.cdri.org.kh/webdata/download/wp/wp19e.pdf 

Davis, K., Nkonya, E., Kato, E., Mekonnen, D. a., Odendo, M., Miiro, R., & Nkuba, J. (2012). 
Impact of Farmer Field Schools on Agricultural Productivity and Poverty in East Africa. 
World Development, 40(2), 402–413. 

Deininger, K., & Jin, S. (2006). Tenure security and land-related investment : Evidence from 

Ethiopia. European Economic Review 50, 50(5), 1245–1277. 

Dillon, A. (2011). Do Differences in the Scale of Irrigation Projects Generate Different Impacts 

on Poverty and Production? Journal of Agricultural Economics, 62(2), 474–492. 

Do, Q., & Iyer, L. (2008). Land Titling and Rural Transition in Vietnam. Economic Development 

and Cultural Change, 56(3), 531–579. 

Duflo, E. (2004). Scaling up and evaluation. Annual World Bank Conference on Development 

Economics, 341–369. 

Duflo, E., Glennerster, R., & Kremer, M. (2006). Using Randomization in Development 

Economics: A Toolkit. Handbook of Development Economics, 4, 3895–3962. 

FAO-Food and Agriculture Organization. (2014). Cambodia country fact sheet on food and 

agriculture policy trends. Rome. 

Feder, G., Onchan, T., Chanlamwong, Y., & Hongladarom, C. (1988). Land policies and farm 

productivity in Thailand. Baltimore: Johns Hopkins University Press. 

Fenske, J. (2011). Land tenure and investment incentives: Evidence from West Africa. Journal 

of Development Economics, 95(2), 137–156. 

Galiani, S., & Schargrodsky, E. (2004). Effects of land titling on child health. Economics and 

Human Biology, 2(3), 353–372. 

Galiani, S., & Schargrodsky, E. (2010). Property rights for the poor: Effects of land titling. 

Journal of Public Economics, 94(9–10), 700–729. 

Goldstein, M., & Udry, C. (2008). The Profits of Power : Land Rights and Agricultural 

Investment in Ghana. Journal of Political Economy, 116(6), 981–1022. 

Guo, S., & Fraser, M. W. (2014). Propensity Score Analysis: Statistical Methods and 

Applications: Statistical Methods and Applications. Los Angeles, Calif: SAGE Publications. 

Heckman, J. J., Ichimura, H., & Todd, P. E. (1997). Matching Evidence Job As An Econometric 

Estimator: Evidence from Evaluating a Job Training Programme. Review of Economic 

	

71 

Studies, 64(4), 605–654. 

Heckman, J. J., Ichimura, H., & Todd, P. E. (1998). Matching as an econometric evaluation 

estimator. The Review of Economic Studies, 65(2), 261–294. 

Holden, S. T., Deininger, K., & Ghebru, H. (2009). Impacts of low-cost land certification on 

investment and productivity. American Journal of Agricultural Economics, 91(2), 359–373. 

Holden, S. T., & Otsuka, K. (2014). The roles of land tenure reforms and land markets in the 

context of population growth and land use intensification in Africa. Food Policy, 48, 88–97. 

Jacoby, H. G., Li, G., & Rozelle, S. (2002). Hazards of Expropriation : Tenure Insecurity and 

Investment in Rural China. The American Economic Review, 92(5), 1420–1447. 

Jacoby, H. G., & Minten, B. (2007). Is land titling in Sub-Saharan Africa cost-effective? 

Evidence from Madagascar. World Bank Economic Review, 21(3), 461–485. 

Jansen, K., & Roquas, E. (1998). Modernizing Insecurity: The Land Titling Project in Honduras. 

Development and Change, 29(1), 81–106. 

Kassie, M., Jaleta, M., Shiferaw, B., Mmbando, F., & Mekuria, M. (2013). Adoption of 

Interrelated Sustainable Agricultural Practices in Smallholder Systems : Evidence from 
Rural Tanzania. Technological Forcasting and Social Change, 80, 525–540. 

Keo, S., & Theng, V. (2013). The Impact of Agricultural Extension Services on Rice 

Production:Evidence from Panel Data of Nine Rural Villages in Cambodia. In Annual 
Development Review 2013-14 (pp. 96–114). Phnom Penh: CDRI. Retrieved from 
http://www.cdri.org.kh/webdata/download/adr/adr1314e.pdf 

Khandker, S. R., Koolwal, G. B., & Samad, H. A. (2010). Handbook on Impact Evaluation: 

Quantitative Methods and Practices. Learning. World Bank Publications. Retrieved from 
https://openknowledge.worldbank.org/bitstream/handle/10986/2693/520990PUB0EPI1101
Official0Use0Only1.pdf?sequence=1 

Kuwornu, J. K. M., & Owus, E. S. (2012). Irrigation access and per capita consumption 

expenditure in farm households: Evidence from Ghana. Journal of Development and 
Agricultural Economics, 4(3), 78–92. 

Li, G., Rozelle, S., & Brandt, L. (1998). Tenure , land rights , and farmer investment incentives 

in China. Agricultural Economics, 19, 63–71. 

Markussen, T. (2008). Property Rights, Productivity, and Common Property Resources: Insights 

from Rural Cambodia. World Development, 36(11), 2277–2296. 

Minten, B., & Barrett, C. B. (2008). Agricultural Technology, Productivity, and Poverty in 

Madagascar. World Development, 36(5), 797–822. 

Motavalli, P. P., Singh, R. P., & Anders, M. M. (1994). Perception and management of farmyard 

	

72 

manure in the semi-arid tropics of India. Agricultural Systems, 46(2), 189–204. 

Nguyen, T., Bauer, S., & Grote, U. (2016). Does Land Tenure Security Promote Manure Use by 

Farm Households in Vietnam? Sustainability, 8(2), 178. 

NIS-National Institute of Statistics. (2010). Cambodia Socio-Economic Survey 2009. Phnom 

Penh: NIS. 

NIS-National Institute of Statistics. (2013). Cambodia Socio-Economic Survey 2012. Phnom 

Penh: NIS. 

NIS-National Institute of Statistics. (2014). Cambodia Socio-Economic Survey 2013. Phnom 

Penh: NIS. 

NIS-National Institute of Statistics. (2015a). Cambodia Socio-Economic Survey 2014. Phnom 

Penh: NIS. Retrieved from http://www.nis.gov.kh/nis/CSES/CSES_2014_Report.pdf 

NIS-National Institute of Statistics. (2015b). Census of Agriculture of the Kingdom of Cambodia 

2013. Phnom Penh: NIS. Retrieved from 
http://www.nis.gov.kh/nis/CAC2013/CAC_2013_Final_Report_En.pdf 

NIS -National Institute of Statistics. (2011). National Accounts of Cambodia 1993-2010. Phnom 

Penh: NIS. 

Pender, J. L., & Kerr, J. M. (1999). The effects of land sales restrictions: Evidence from south 

India. Agricultural Economics, 21(3), 279–294. 

Place, F. (2009). Land Tenure and Agricultural Productivity in Africa: A Comparative Analysis 

of the Economics Literature and Recent Policy Strategies and Reforms. World 
Development, 37(8), 1326–1336. 

Place, F., & Hazell, P. (1993). Productivity Effects of Indigenous Land Tenure Systems in Sub-

Saharan Africa. American Journal of Agricultural Economics, 75(1), 10–19. 

Place, F., & Otsuka, K. (2002). Land Tenure Systems and Their Impacts on Agricultural 

Investments and Productivity in Uganda. Journal of Development Studies, 38(6), 105–128. 

Platteau, J. P. (1996). The Evolutionary Theory of Land Rights as Applied to Sub-Saharan 

Africa: A Critical Assessment. Development and Change, 27, 29–86. 

RGC-Royal Government of Cambodia. (2013). Rectangular Strategy-Phase III, 2014-2018. 

Phnom Penh. 

RGC-Royal Government of Cambodia. (2014). National Strategic Development Plan 2014-2018. 

Phnom Penh. 

Rosenbaum, P. aul R., & Rubin, D. B. (1983). The Central Role of the Propensity Score in 

Observational Studies for Causal Effects. Biometrika, 70(1), 41–55. 

	

73 

Rosenbaum, P. aul R., & Rubin, D. B. (1985). Constructing a Control Group Using Multivariate 

Matched Sampling Methods that Incorporate the Propensity Score. The American 
Statistician, 39(1), 33–38. 

So, S. (2009). Political Economy of Land Registration in Cambodia. Northern Illinois 

University. Retrieved from 
http://search.proquest.com.proxy1.cl.msu.edu/docview/304968261?pq-origsite=summon 

Sok, S., Chap, S., & Chheang, V. (2011). Cambodia ’ s Agriculture : Challenges and Prospects 

(No. 37). Phnom Penh. 

Suyanto, S., Tomich, T. P., & Otsuka, K. (2001). Land tenure and farm management efficiency : 

the case of paddy and cinnamon production in customary land areas of Sumatra. The 
Australian Journal of Agricultural Economics, 45(3), 411–436. 

Teklewold, H., Kassie, M., & Shiferaw, B. (2013). Adoption of Multiple Sustainable 

Agricultural Practices in Rural Ethiopia. Journal of Agricultural Economics, 64(3), 597–
623. 

Theng, V., Khiev, P., & Phon, D. (2014). Development of the Fertiliser Industry in Cambodia: 

Structure of the Market, Challenges in the Demand and Supply Sides, and the Way Forward 
(CDR Working Paper Series No. 91). Phnom Penh. Retrieved from 
http://www.cdri.org.kh/webdata/download/wp/wp91e.pdf 

Thin, K. (2012). Essays on Land Property Rights in Cambodia : Empirical Analysis. University 

of Hawai. Retrieved from http://scholarspace.manoa.hawaii.edu/handle/10125/101033 

Tong, K. (2010). Land Tenure and Paddy Productivity: Evidence from Rural Cambodia. Annual 

Development Review. Phnom Penh. 

Un, K., & So, S. (2011). Land rights in Cambodia: How neopatrimonial politics restricts land 

policy reform. Pacific Affairs, 84(2), 289–308. 

Yu, B., & Fan, S. (2011). Rice Production Response in Cambodia. Agricultural Economics, 

42(3), 437–450. 

Zhao, Z. (2004). Using Matching To Estimate Treatment Effects : Data Requirements, Matching 

Metrics Carlo Evidence. The Review of Economics and Statistics, 86(1), 91–107. 

 
 
 
 
 
 
 
 

	

74 

CHAPTER 3: ADOPTION OF INTERRELATED FARM TECHNOLOGIES  
                    IN CAMBODIA: THE CASE OF IMPROVED RICE  
                    VARIETIES 
	

3.1. Introduction  
 

The Green Revolution technologies, which includes fertilizer and improved crop 

varieties, has transformed crop production since the late 1960’s through productivity 

improvement (Khush, 1995; Ogada, Mwabu, & Muchai, 2014). High-yield varieties of seeds are 

the most important of these technologies, but sometimes it requires other inputs to maximize its 

productivity. A variety of empirical studies has shown that the adoption of agricultural 

technologies has a positive impact on livelihood improvement and poverty reduction through 

increasing income and capacity of farmers (Otsuka, 2000; Pingali, 2012; Babu, Gajanan, & 

Sanyal, 2014). Also, adopting productivity-enhancing technologies such as high-yield varieties 

of seed and fertilizer has a positive effect on smallholder market participation – an important step 

towards agricultural commercialization (Olwande et al., 2015). From a perspective of structural 

transformation, farm commercialization is instrumental to the sustained long-term growth of a 

country.  

Cambodia is an agrarian economy, and rice dominates Cambodia’s agriculture in terms of 

total agricultural output and cultivated land (Yu & Fan, 2011). However, this sector is 

constrained by low productivity and subsistence farming practices. In this regard, the 

government and development partners acknowledge that promoting the adoption of improved 

technology and farm practices is an important agenda for agricultural development through 

productivity improvement. Due to the importance of technology adoption in agriculture, 

conducting research studies about this area can contribute important insights for designing 

agricultural development policy. It can assist in the process of transforming agriculture to a more 

	

75 

productive and commercialized sector. Nonetheless, previous research studies on agricultural 

technology adoption in Cambodia have not been very extensive. Most of the available studies for 

Cambodia use descriptive statistics and case studies without rigorous empirical analysis (e.g., 

Mak, 2001; Ly et  al., 2012; Sumner, Christie, & Boulakia, 2017).  

In literature, some studies have shown that farmers tend to adopt multiple, 

complementary technologies to cope with various constraints to increasing their farm’s 

productive capacity. For instance, Ogada et al. (2014) empirically showed that improved maize 

variety seeds and chemical fertilizer are complementary in Kenya because farmers’ decisions to 

adopt both inputs are positively interdependent. In addition to modern input adoption, some 

studies also detect whether farm practices are complements or substitutes. For example, Kassie et 

al. (2013) studied the adoption of interrelated sustainable agricultural practices in rural Tanzania, 

and Teklewold et al. (2013) conducted a similar study in rural Ethiopia. Their results suggested 

that adopting complementary technologies better enhances farm productivity than adopting each 

of them separately. Thus, from a policy perspective, it is important for government to promote 

joint adoption of complementary farm technologies and practices, to the extent that it is 

practically feasible.  

Previous studies for Cambodia treated the improved rice varieties and chemical fertilizer 

as independent technologies (e.g., Mak, 2001; Wang et al., 2012; Nguyen et al., 2013). In other 

words, in their empirical analysis those studies did not take into account the complementarity 

between improved rice varieties and chemical fertilizer. Another limitation of those previous 

studies is that none of them used panel data, so the empirical results tend to suffer bias caused by 

unobserved factors that potentially affect the outcomes.  To the best of our knowledge, there 

have been no studies about improved farm technologies in Cambodia using plot-level panel data. 

	

76 

One of the key advantages of panel data is that it enables us to remove unobservable, time-

constant factors from the regressions, thus improving the precision of the estimators. 

Therefore, aiming at contributing to agricultural development policies through the 

promotion of the improved farm technology adoption in Cambodia, this paper employs empirical 

methodology and longitudinal plot-level data to understand the adoption of interrelated farm 

technologies. More specifically, it addresses two empirical questions. First, this paper examines 

whether farmers’ decisions to adopt improved rice varieties and chemical fertilizer are 

interrelated (interdependent). The interdependence means that farmers’ decisions to adopt each 

improved technology is not mutually independent, and the statistical method to test for that is 

spelt out in the empirical methodology section. Second, it detects the determinants of the 

probability of adopting those improved farm technologies. Hence, the contribution of this study 

is twofold:  firstly, it contributes to agricultural development policies by taking into account the 

interdependence of adoption; and secondly, it uses plot-level panel data. 

The remainder of this paper is organized as follows. Following this introduction, section 

3.2 explores the relevant empirical literature on the adoption of improved farm technologies. The 

conceptual framework is spelt out in section 3.3 and section 3.4 discusses the empirical 

methodology including the qualitative fieldwork. Section 3.5 describes the data sources and 

descriptive statistics. Section 3.6 presents and discusses empirical findings, and the last section 

concludes and provides some policy implications.  

3.2. Literature review 

In general, empirical research on determinants of improved technology adoption has 

emphasized that risk and uncertainty associated with expected profits are the common constraints 

to farmers’ decisions to adopt agricultural technology (Feder & Umali, 1993; Smale, Just, & 

	

77 

Leathers, 1994). Besides, there are important factors that are related to overcoming problems 

related to risk and uncertainty. These  factors, which also determine a household’s decision to 

adopt agricultural technologies, include availability of complementary inputs, access to 

agricultural extension services, social networking, and wealth of households, credit and market 

accessibilities, land rights, off-farm income, socio-demographic characteristics and agro-

ecological locations of households (Feder, Just, & Zilberman, 1985; Nkonya, Schroeder, & 

Norman, 1997; Matuschke & Qaim, 2008). 

Smale et al. (1994) studied Malawian smallholders’ decisions regarding land allocation to 

improved maize seed, and their findings suggest that decisions to adopt high-yield varieties and 

fertilizer are interrelated. In line with this finding, the study by Nkonya et al. (1997) reveals that 

there is a significant effect of improved seed adoption on the adoption of fertilizer, implying that 

the improved rice varieties and fertilizer are complementary inputs. Marenya & Barrett (2009) 

empirically detect the determinants of farmers’ decisions to adopt and utilize fertilizer based on 

soil fertility status in Western Kenya. Their findings show that the use of hybrid seed has a 

positive and significant effect on the probability of adopting fertilizer. However, their study 

ignores the interdependence of decisions to adopt fertilizer and hybrid seed, so the empirical 

results may suffer endogeneity because the use of hybrid seed is treated as an independent 

variable in the regression equations. 

In literature, a number of recent studies have taken into account the interdependence of 

farmers’ decisions to adopt improved farm technologies (complements and/or substitutes) by 

using an econometric approach and treating adoption of each improved technology as a 

dependent variable (e.g., Kassie et al., 2013; Teklewold et al., 2013; Ogada et al., 2014). 

However, as mentioned in the introduction, this kind of study has not been conducted in 

	

78 

Cambodia. For instance, an empirical study by Wang et al. (2012) used an econometric model to 

examine the determinants of farmers’ decision to adopt improved rice varieties. Yet, relying on 

cross-sectional data and ignoring the interdependence of agricultural technologies, their results 

may have endogeneity bias (Ogada et al., 2014). 

Apart from economic aspects, social networks and learning also play an important role in 

the adoption of agricultural technology due to information asymmetries. Sometimes, farmers are 

not able to apply agricultural technologies properly due to the complexity of the technologies. To 

overcome this constraint, the farmers can engage in learning-by-doing, experimenting with the 

new technology to better understand the technology or determine the sensitivity of the 

technology to local conditions (Barham & Udry, 1999). Alternatively, they might learn from 

others – either from other farmers who are familiar with the improved technology, or from the 

extension agents in their neighborhood. In this regard, an empirical study by Foster & 

Rosenzweig (1995) suggest that imperfect information is a significant constraint to adoption, and 

farmers become increasingly receptive of the applicability and profitability of the high-yield 

seeds from their own experiences and those of their neighbors. Also, Conley & Udry (2010) 

shows that social learning plays an important role in the adoption of fertilizer in pineapple 

production in Ghana. In some cases, farmers realize the profitability of new technologies from 

the experience of other farmers in their groups or associations. For example, most of the farmers 

who adopted pearl millet hybrids in India, learned from their fellow farmers via their social 

network about the advantages of this crop variety (Matuschke & Qaim, 2008). 

3.3. Conceptual framework   

This section briefly conceptualizes farm households’ decisions to adopt interrelated 

technologies and the factors that are the potential determinants of agricultural technology 

	

79 

adoption, based on the literature. A household chooses a set of improved farm technologies to 

maximize the expected utility of profit conditional on the adoption decision (Feder, 1982; Smale, 

Just, & Leathers, 1994; Feder & Umali, 1993; Ogada, Mwabu, & Muchai, 2014). In general, 

farm households maximize their expected utility through the dichotomous choice of whether or 

not to adopt the technology (Feder & Umali, 1993).  In our study, a household chooses to apply 

interrelated inputs together to a plot if the expected utility of profit with this adoption is higher 

than that of adopting only one of the inputs or neither. The study tests the hypothesis that a farm 

household’s decisions to use chemical fertilizer and improved rice varieties on a plot are 

complementary.  

Regarding the determinants of technology adoption, the first category is farmers’ access 

to information and their knowledge related to rice production technology. It is worth observing 

that communication – a process in which farmers generate and share information – is a main 

element of the diffusion of innovations  through certain channels such as media and interpersonal 

interactions (Rogers, 2003). This idea is in line with the recently emerging studies that focus on 

learning through social networks. Under certain circumstances, farmers are not able to apply 

agricultural technologies properly because of the complexity of the technologies, which Barham 

& Udry (1999) attribute to tacit elements of a technology, sensitivity, and institutional context of 

some agricultural technologies. To overcome the first two issues of new improved farm 

technologies, farmers engage in learning-by-doing, experimenting with the new improved farm 

technology to reveal tacit elements of the technology. Besides, farmers might learn either from 

other farmers or others such as extension agents or their neighbor. This category of determinants 

implies the importance of sources from which farmers acquired knowledge about improved farm 

technology. 

	

80 

Another important group of determinants of technology adoption is socio-economic 

characteristics of farm households such as wealth of the household, access to credit, market 

accessibility, land titling, sources of income, and households’ socio-demographic characteristics 

such as the educational attainment, age, gender of household head, household size, and family 

labor supply (Marenya & Barrett, 2007; Matuschke & Qaim, 2008; Teklewold et al., 2013; 

Kassie et al., 2013; Gajanan, 2014). For instance, a farmer with good education tends to have a 

better ability to think analytically and use information about improved farm technologies more 
effectively.  

Furthermore, institutional factors such as appropriate policy intervention and availability 

of an agricultural development program in the community are also significant determinants of 

adoption (Feder et al., 1985; Feder & Umali, 1993). In contrast, weak regulatory enforcement 

and unclear roles between the state institutions of responsibilities in agricultural modern input 

trade may negatively influence the adoption rates. For instance, the complexity and high costs of 

the fertilizer licensing process in Cambodia may hinder some farmers’ decision to use fertilizer 

(Theng, Khiev, & Phon, 2014). 

Further, good infrastructure is also supportive to the adoption of improved farm 

technology. For example, accessible and reliable irrigation system has a positive linkage with 

farmers’ decision to adopt a new improved farm technology because it can reduce the risk and 

uncertainties associated with farm profitability through the reduction in the negative impacts of 

rainfall variations. On the other hand, underdeveloped irrigation infrastructure in an area of 

unreliable rainfall patterns makes it difficult for farmers to adopt modern varieties because the 

likelihood of crop failure tends to be high.  

	

81 

Besides, economic factors (i.e. markets, input and output prices) also influence farmers’ 

decision to adopt improved farm technology. For example, high fertilizer prices may prevent 

farmers from applying fertilizer to crops in sufficient amount. In Cambodia, most villages are not 

large, so we can assume that prices of agricultural inputs and outputs do not vary much within a 

village. Hence, in the absence of information on input and output prices, controlling for village 

dummies (village fixed effects) can account for village-level differences in prices.  

Also, the agro-ecological locations of households and soil quality play an important role 

in farmers’ decisions on input use (Marenya & Barrett, 2009). For example, farmers whose 

cultivated land is not fertile are less likely than those with fertile plots to adopt modern inputs 

such as high-yield varieties and fertilizer. In our study, we use plot-level panel data which allows 

us to control for time-invariant unobservable plot characteristics, particularly soil fertility of a 

plot, because this factor would not change over the four-year period of our data collection (2012-

2106). Additionally, in practice, households’ locations including district or zones can be also be 

alternative indicators for the agro-ecological locations of households (Kassie et al., 2013), so our 

study can account for this factor by controlling for village dummies. By doing so, we can also 

control for village fixed effects, and do not need to control for village-level characteristics, which 

were not collected in the baseline survey. 

3.4. Empirical methodology 

3.4.1. Multivariate probit model 
			

To test for simultaneous adoption decisions, the farmer’s choice of interrelated farm 

technology is modelled using the multivariate probit (MVP) model (Dorfman, 1996). The 

multivariate probit regression allows us to simultaneously run more than one univariate 

(ordinary) probit model (Cappellari & Jenkins, 2003), each of which has only one binary 

	

82 

dependent variable. This approach can statistically show whether adoption decisions on the 

improved farm technology are interdependent. Adapted from the empirical studies by Kassie et 

al. (2013) and Teklewold et al. (2013), the econometric model of bivariate probit regression, 

characterized by two binary dependent variables for our study, is expressed in the following 

equation. 

 

 

 

12=>∗ =42=>5=+92>         (equation 4), 

Where i represents plot (i=1,2…. N), and t represents time t= 1, 2 (or 2012 and 2016 for this 

study). Also, k denotes binary dependent variables which are described as follows:  

(1) Improved  varieties  (V):  1  if  improved  rice  variety  was  used  on  a  plot  in  the  last  rice 

growing completed season (dry or wet-season), 0 otherwise. 

(2) Chemical fertilizer (F): 1 if chemical fertilizer was used on a plot in the last rice growing 

completed season (dry or wet-season), 0 otherwise. 

The unobserved choices in equation (4) can be represented by the observed binary 

outcome equation for each improved farm technology adoption as follows. 

12=>= 1		)?		12=>∗ >0

0	%*ℎ&$A)(&             (equation 5) 

 

 

 
 

 

 

Equation (4) and (5) imply that the net benefit 12=>∗ 	from the adoption of the kth improved 
farm technology is a latent variable determined by observed household characteristics (42>) and 
the error term (92>).  

In the multivariate regression model, the adoption of multiple technologies is possible, so 

the error terms of equation (4) jointly follow a multivariate normal distribution with zero 

conditional mean and variance normalized to unity, with systematic covariance matrix given by:  

 

	

83 

 Ω=	 1											CDE
CED											1			          (equation 6) 

 

 
 where   V: improved varieties; F: chemical fertilizer 

The specification with non-zero off-diagonal elements in the covariance matrix C, 

symbolizes the unobservable correlation between the error terms of the two univariate probit 

equations. In other words, this method allows the error terms of each equation to be freely 

correlated, which implies whether different technologies are complement (positive correlation) 

or substitute (negative correlation) for each other in the adoption decision. 

3.4.2. Correlated random effects (CRE) model 

Our study uses plot-level panel data. Thus, to examine the determinants of improved farm 

technology adoption, the fixed effect (FE) estimator would be the most feasible approach. It 

enables us to control for confounding effects of time-constant unobservable factors such as a 

plot’s soil fertility and a farmer’s ability and intelligence (Wooldridge, 2008), which might be 

correlated with the explanatory variables in the regression. In other words, FE can address the 

omitted variable bias which might be caused by unobserved heterogeneity of plots and 

households.  

However, with panel data, FE estimators are not consistent in the case of nonlinear 

models such as probit, logit and Tobit (Wooldridge, 2010). Hence, to deal with the problem of 

unobserved heterogeneity with panel data, Wooldridge (2010) recommends the correlated 

random effects (CRE) model proposed by Mundlak (1978) and Chamberlain (1984). This model, 

which is also known as the Mundlak-Chamberlain device, controls for the correlation between 

unobserved, time-invariant heterogeneity and the explanatory variables in the nonlinear 

regression framework by including the averages of time-varying regressors in the model. The 

CRE technique has been used in a number of recent studies with panel data to handle the 

	

84 

be summarized in the following form of CR 

 

 

aforementioned challenge (e.g., Smale & Olwande, 2014; Mathenge, Smale, & Olwande, 2014; 

Olwande, Smale, Mathenge, Place, & Mithofer, 2015). The probit regression equation can also 

takes the value of one if an improved farm technology was used on plot i at time t (2012 and 2016). 

F12>=1 =F	12>∗>0 =G2H+	42>5+-2+92>         (equation 7) 

Where 12> is the observed binary outcome for each improved farm technology that is adopted. It 
And, the improved farm technology is adopted when the net benefit 12>∗ of the adoption, which is 
a latent variable, is positive. G2 is a vector of observed time-constant factors5 expected to influence 
adoption of improved farm technology on plot i. Also, 42> is a vector of observed time-varying 
characteristics  of  plot  i  at  time  t, -2  is  a  variable  that  captures  the  unobserved  time-invariant 
attributes that influence adoption decision, and 92> is the error term that changes across time, plots, 
and households. The CRE framework allows us to model the relationship between -2 and observed 
characteristics, 42> as follows: 
Where 4J is the mean of 42 across time (2012 and 2106), I is constant, L2 is the error term which 
is assumed to have zero mean. Equation (8) implies that -2 is correlated with the time-varying 
regressors  through  its  average  level  over  time.  Plugging -2  in  equation  (7),  we  obtain  probit 
F12>=1 =F	12>∗>0 =G2H+	42>5+I+4JK+L2+92>    (equation 9) 

 -2=I+4JK+L2   (equation 8) 

regression in the form of CRE as shown in (9). 

 

 

Equation (9) means that we run the probit regression with the CRE framework by 

including the time-averages of time-varying independent variables for each plot and household, 

along with time dummies and time-invariant observed variables. Doing this ensures that our 

																																																								
5 This also includes observable characteristics of household, because a plot i belongs to a household. 

	

85 

independent variables (G2	L'M	42>) are not correlated with the time-invariant attributes captured 
in the error terms (Wooldridge, 2010). With balanced panel data, 5 of equation (9) and the 

estimate provided from pooled cross-section regression are comparable (Wooldridge, 2010). We 

report the coefficients for probit CRE regression in terms of average marginal (or partial) effects 

because it implies a linear relationship between dependent and independent variables, which is 

convenient for the interpretation. Thus, we use pooled probit CRE regression to estimate 

equation (9), because this method is recommended when we focus on the marginal effects 

(Wooldridge, 2010). There are three dependent variables for the pooled probit CRE estimation. 

They are a dichotomous adoption variable (1 for adoption at plot-level; 0 otherwise) for 

improved rice varieties, chemical fertilizer and the combination of the two types of technology. 

To control for village heterogeneity (or village fixed effects), we also include village 

dummies in our regression model. Since there are 84 villages, there will be 83 village dummies 

included in the regressions. Also, the standard error in each regression estimation is clustered at 

village to ensure that it is robust for making inference. 

3.4.3. Qualitative fieldwork 

25 semi-structured qualitative interviews were conducted in late August 2017 with some 

of the surveyed households in order to assist with the interpretation and enrich the discussion of 

the empirical findings.  These households were selected first via key informants who could 

identify households who had participated in the survey, and then through snowball sampling 

based on the knowledge of those who participated in the qualitative interviewing. The 

information gathered in the semi-structured interviews includes questions specifically about 

improved rice varieties and chemical fertilizer use. I was the interviewer assisted by a note taker; 

	

86 

I did this fieldwork in three provinces of HARVEST program (Battambang, Siem Reap and 

Kampong Thom). 

The purpose of the interviews is to supplement the discussion of the empirical results. 

What I expected to get through the interviews are in-depth answers to the questions related to 

why farmers used the modern inputs together, the important sources of information about 

improved farm technologies, and the constraints they have faced. The interviews were auto-

recoded and summarized for the analysis. And, to protect identity of the participants, any 

personal information such as names and positions, as well as other information that I recorded 

during the data collection, was strictly confidential.  

For data analysis, I recorded the conversations and took written notes. After doing my 

statistical analysis, I explored the notes and the recordings to find discussions that helped 

interpret the statistical findings. Besides, the notes taken during the interviews were checked 

against the audio records to ensure internal validity of the qualitative data and trustworthiness.  

3.5. Data and descriptive statistics  

The study is based on the HARVEST household panel survey (2012-2016) in four 

provinces in Tonle Sap region – one of the four agro-ecological zones in Cambodia. The baseline 

survey of 2100 households across 84 villages was carried out by the Cambodia Development 

Resource Institute (CDRI) in four provinces from late August to early September 2012. 

HARVEST, standing for Helping Address Rural Vulnerabilities and Ecosystem Stability, was a 

five-year program, financially supported by the United States Agency for International 

Development (USAID). The Cambodia HARVEST program set strategic goals to improve food 

security, strengthen natural resource management and resilience to climate change, and increase 

	

87 

the capacity of the public and private sectors and civil society to support agricultural 

competitiveness.  

CDRI was also responsible for data collection in the end-term survey, which was 

scheduled for the same month as the baseline survey, four years later. The survey instrument was 

designed by the HARVEST program evaluation team, mainly involving international consultant 

team from the Michigan State University and local researchers of Cambodia Development 

Resource Institute (CDRI). The questionnaire was adapted from the standard population-based 

survey (PBS) such as Cambodia Socio-Economic Survey (CSES) with some modification to 

reflect local context. The data has information at plot- and household levels, including 

demography, sources of income, consumption expenditure, rice and vegetable production, 

adoption of improved farm technologies and nutrition.  The purpose of data collections (baseline 

and end-line surveys) is for the impact assessment of the HARVEST program. 

The sample size of the end-term survey is 2,328 households, meaning that some new 

households were included in this period. Additionally, 179 households (8.5 percent) from the 

baseline dropped out and were not included in the end-term survey, resulting in a balanced panel 

data of 1921 households in each period. In this study, we focus on rice production at the plot 

level, so in our empirical analysis I excluded the plots that did not grow rice. With this condition, 

there were 5,890 plots (belonging to 1,982 households) in 2012, and 7,884 plots (2,048 

households) in 2016, engaging in rice production. These are the numbers prior to constructing 

the balanced plot-level data, so the number of plots and households in the two time periods are 

not the same. It is worth noting that 1,738 among the 5,890 plots from the baseline were not 

involved in rice production. Thus, after data cleaning and merging, we obtain a balanced panel 

	

88 

data of 4,112 rice plots (1,598 households) in each period for our empirical analysis. As a result, 

the pooled sample consists of 8,224 observations (Table 18).  

The CDRI data collection team confirmed that the data is plot-level panel. There are 

various operational aspects of identifying the baseline plots in the end-line survey. First, the 

name of each plot was collected and recorded in Khmer (the local language) during the baseline 

survey, so it is convenient for the farmers as well as the data collectors to track the same plots 

over time. Second, during the end-line survey, for each household, the CDRI team prepared a list 

of plots with names and IDs. Since there are 2,100 households in baseline, they prepared 2,100 

lists accordingly with the printed lists distributed to the data collectors according to the villages 

where they were assigned to. Third, based on the household IDs, the data collectors listed all of 

the plot names corresponding to the IDs in the questionnaire. When conducting the interview, the 

farmers were asked whether or not each plot was available. If a plot was no longer used by a 

household, it was excluded from the questionnaire, and thus that particular household has no 

information about that plot in the end-line survey. If a household acquired a new plot after the 

baseline survey, on the other hand, the new ID was given so as to have no any duplicated plot 

IDs over the two time periods.  

The dependent and independent variables at plot- and household levels are defined and 

summarized in Table 18. If some plots were planted in both the wet and dry seasons, I collapsed 

the two observations to one observation at the plot level to have a unique plot ID for empirical 

analysis. There are three dependent variables which are reported at plot-level. They are a 

dichotomous adoption variable (1 for adoption at plot-level; 0 otherwise) for improved rice 

varieties, chemical fertilizer and the combination of the two types of technology.  For instance, 

fertilizer adoption was defined as 1 if this input was either in dry season, wet season or both 

	

89 

seasons. This definition also applies to adoption of improved rice varieties. To further deal with 

seasonal dimension, the combination variable was defined before I collapsed the data to the plot-

level. Thus, in a scenario where an improved variety was planted without fertilizer in the wet 

season (for example), and an unimproved variety was planted with fertilizer in the dry season, 

my combination variable would be zero. In other words, the combination equals 1 one if 

improved variety and fertilizer were used in the same season.		

An important plot characteristic is whether a plot was used for dry-season rice, so I use a 

dummy variable to indicate if the data is for the dry season, which is a proxy for irrigation. That 

is, a dry-season rice plot is defined as 1 farmer grew rice in the dry season or in both dry and wet 

seasons on that plot; it equals 0 otherwise. By controlling for this variable in the regression 

equations, I do not do the empirical estimations by season, because the proportion of the dry-

season plots is less than 15 percent of the pooled sample, which is quite small for statistical 

analysis. 

The average adoption rates of improved rice varieties, chemical fertilizer, and the 

combination of the two types of inputs were 57 percent, 72 percent and 45 percent, respectively 

in 2012. And, these rates respectively increased by 2.1, 7.1 and 3.3 percentage points in 2016. 

The explanatory variables are plot and household characteristics that determine decisions to 

adopt the improved farm technologies. After reconciling the available information from survey 

data and the literature on technology adoption (e.g., Bandiera & Rasul, 2006; Doss, 2006; Kassie 

et al., 2013), we choose a number of the independent variables to include in the model based on 

what other authors have done. Below is the brief descriptive statistics for some of those 

variables, which can be found in Table 18. 

	

90 

 In the baseline, 14.7 percent of the 4,112 plots were used for dry season rice farming, 

and this rate decreased to 8.9 percent in 2016. On average, the plot size increased from 0.793 

hectare in 2012 to 0.809 hectare in 2016, implying that some households expanded their plots 

over this period. This is consistent with what I learnt from some data collectors who engaged in 

the end-line survey. I have four dichotomous variables for the households’ access to 

information/or agricultural extension services about improved rice varieties and chemical 

fertilizer application from various sources – their neighbors, NGOs, HARVEST program and 

provincial department of agriculture (PDA). These are the common sources of information about 

improved farm technologies, according to previous studies in Cambodia (Theng et al., 2014). On 

average, in each period around 21 percent of households in the sample acquired knowledge about 

improved technologies from their neighbors. However, the proportion of households drawing on 

other information sources, especially the HARVEST program and PDA, declined over the period 

2012-2016. For example, in 2016, around 2 percent of households received information about the 

improved farm technologies from HARVEST during the previous 12 months, compared to about 

24 percent in 2012 (Table 18). The reduced proportion might be because in some areas the 

HARVEST program completed its work approximately one year before the end-line survey was 

conducted.  

 

 
 

	

91 

Table 18: Variable definitions and summary statistics by year 

Definition 

 

 

 

Variable 

Dependent variables 
Improved variety 

Fertilizer 

Combination 

1 if improved rice variety was used on a plot in the last rice growing 
completed season (dry or wet-season), 0 otherwise. 
1 if a plot with chemical fertilizer used in the last rice growing 
completed season (dry or wet-season), 0 otherwise. 
1 if a plot with both chemical fertilizer and improved rice variety used in 
the last rice growing completed season, 0 otherwise. 

Independent variables (plot level) 
Dry-season rice plot 

Plot size 

1 if a plot is used for dry season rice (irrigated rice production) 
Plot size in hectare 

Independent variables (HH level) 
Extension from 
neighbors 
Extension from NGOs 

Extension (HARVEST) 

Extension from PDA 

1 if HH had access to extension knowledge about the improved rice or 
fertilizer from the neighbors in the last 12 months; 0 otherwise. 
1 if HH had access to extension knowledge about the improved rice or 
fertilizer from the NGOs in the 12 last months; 0 otherwise. 
1 if HH had access to extension knowledge about the improved rice or 
fertilizer from the HARVEST in the 12 last months; 0 otherwise. 
1 if HH had access to extension knowledge about the improved rice or 
fertilizer from the provincial department of agriculture (PDA) 

2012 

2016 

Pooled 

Mean 

Std.  Mean 

Std.  Mean 

Std. 

 

 

 

 

 

 

0.570 

0.495 

0.591 

0.492 

0.580 

0.494 

0.723 

0.447 

0.794 

0.404 

0.759 

0.428 

0.451 

0.516 

0.561 

0.500 

0.484 

0.500 

0.147 
0.793 

0.354 
1.027 

0.089 
0.807 

0.285 
0.991 

0.118 
0.800 

0.322 
1.009 

0.205 

0.404 

0.209 

0.407 

0.207 

0.405 

0.234 

0.424 

0.176 

0.381 

0.205 

0.404 

0.241 

0.428 

0.021 

0.142 

0.131 

0.337 

Household size 
sex of HH head: 1 if male 
Age of household head (years) 
1 if HH can read or write, and 0 otherwise 
1 if household head has primary education (grade 1 to 6) 
1 if household head has secondary education (grade 7 to 9) 
1 if HH own a TV; 0 otherwise 

Household’s landholding  Total land holding of a household in hectare 
Household size 
Household head (male) 
Age of household head 
Literate 
Primary education 
Secondary education 
Dummy of TV 
Dummy of mobile phone  1 if household owns a mobile phone, and 0 otherwise 
Dummy of tractor 
Dummy of motorbike 
Asset value 
Consumption 
Wage income 
Remittances 
Source: Calculation based on HARVEST data 2012-2016. Notes: in the regressions, Plot size, Household’s landholding, Asset value and Consumption are converted into 
logarithm.  Std.: standard deviation. There are on 4,122 observations for plot-level variables in each period and 1,598 observations for household-level variables in each period. 

1 if household owns a hand-tractor/tractor, and 0 otherwise 
1 if household owns a motorbike, and 0 otherwise 
Asset value in ‘0000’ in real terms in real terms in '0000' Riels 
Daily consumption per capita in real terms in Riels 
Monthly wage income per year in real terms in '0000' Riels 
Income from remittances per year in real terms in ‘0000’ Riels 

0.315 
0.029 
3.152 
2.947 
2.035 
6.370 
0.373 
0.786 
12.88 
56.525 
0.464 
0.713 
0.500 
0.558 
0.339 
0.150 
0.472 
0.715 
0.448 
0.859 
0.464 
0.449 
0.491 
0.755 
817.2 
525.84 
2,125 
4,433 
68.81 
150.66 
151.25  133.83 

0.111 
2.861 
5.703 
0.834 
54.648 
0.688 
0.493 
0.133 
0.666 
0.723 
0.313 
0.594 
409.84 
4,483 
48.946 
39.68 

0.256 
3.143 
2.131 
0.393 
12.789 
0.458 
0.499 
0.349 
0.462 
0.407 
0.486 
0.469 
797.31 
2,667 
294.72 
310.10 

0.169 
3.135 
2.174 
0.410 
12.63 
0.452 
0.497 
0.358 
0.451 
0.349 
0.498 
0.430 
772.8 
3,116 
374.1 
406.1 

0.070 
2.904 
6.037 
0.810 
55.59 
0.701 
0.525 
0.142 
0.690 
0.791 
0.381 
0.675 
467.9 
4,458 
99.6 
86.81 

	

92 

3.6. Empirical results 

First, this section presents the empirical results estimated by the multivariate probit 

model. Overall, the likelihood ratio test [chi2(1) = 153.94, P<0.01] of the null hypothesis that the 

covariance of error terms across the equations are jointly equal to zero is rejected (Table 19). 

This result implies that farm households’ decisions to adopt improved rice varieties and chemical 

fertilizer at plot level are interdependent in the four provinces. The binary correlations between 

error terms of the three adoption equations are shown in Table 19. The statically significant and 

positive coefficient suggests that the two technologies are complements. This finding is 

consistent with some studies in other countries (e.g., Kassie et al., 2013; Teklewold et al., 2013; 

Ogada, Mwabu, & Muchai, 2014). Also, it is in line with what I learned from the qualitative data 

which shows that most of them have commonly used chemical fertilizer with improved rice 

varieties in their rice farming. The main reason for using these inputs together is that the farmers 

would not obtain good output by using improved rice varieties alone.  

The qualitative findings also indicate that some farmers purchased the improved variety 

seeds every year while the others re-used the seeds they produced in the previous season. The 

main reason for the multiplication of the improved seed is that some farmers could not afford to 

buy the improved variety seeds for all of their plots. For instance, a farmer in Battambang used 

one fourth of his land to produce the improved variety seeds for the next farming cycle. He 

indicated that amount of seeds needed per hectare is around 190 kilograms and the seed price 

was around 3,200 riels (USD 0.8).   

 

 

 

 93 

Table 19: Correlation coefficients for MVP regression equations 

	

!"#$%&'()	'+%"(,-	
Likelihood ratio test of: ρ2345676834_6:;4<=3>	=?4635@	=0 

!.(%,"/"0(%	

0.165 (0.0197) *** 

chi2(1) = 153.94 ***      (significance levels: * 10 percent; ** 5 percent; *** 1 percent) 
(clustered robust standard errors in the parenthesis) 

   Source: Calculation based on HARVEST data 2012-2016.  

The second objective of the study is to analyze the factors that influence the adoption of 

improved rice varieties, chemical fertilizer and the combination of both improved technologies 

using Correlated Random Effects probit regressions. The coefficients in Table 20 are reported in 
terms of average marginal (partial) effects for a convenient interpretation. That is, a parameter 

estimate is the expected probability that a unit change in variable X will result in in the outcome 
variable Y equaling 1, with other independent variables held fixed. Coefficients on the averages 

of time-varying variables that I controlled for are not in the table, because I do not interpret them 

(Wooldridge, 2010). There are no village-specific variables in these regressions, because I 

already control for differences in village characteristics with the village dummies. 

It is worth noting that the number of observations is lower in the regressions, particularly 

for the chemical fertilizer adoption. The reason for this reduction is that I control for the village 

dummies in the regressions to take into account the village fixed effects. Then, STATA 

automatically dropped the plots in the villages where all or most of the observations have the 

same value of dependent variables. For instance, if fertilizer was used on all of or most of the 

plots in a village, STATA excludes those observations from the regression, because it perfectly 

predicts that the outcome is equal to 1 for that village. This reason also applies to the regression 

for improved rice variety. 

 

 

 94 

Improved varieties  Chemical fertilizer 
(Marginal effects) 
(Marginal effects) 

Combination 

(Marginal effects) 

Table 20: Correlated random effects (CRE) estimation results of the determinants of adoption of 
agricultural technology 
Variables  
 
Dry-season rice plot 
 
Plot size (in logarithm) 
 
Extension from neighbors 
 
Extension from NGOs 
 
Extension (HARVEST) 
 
Extension from PDA 
 
Landholding (logarithm) 
 
Household size 
 
Household head (male) 
 
Age of household head 
 
Literate 
 
Primary education 
 
Secondary education 
 
TV 
 
Dummy of mobile phone 
 
Dummy of tractor 
 
Dummy of motor 
 
Asset value (logarithm) 
 
Consumption (logarithm) 
 
Wage income 
 

0.23092*** 
(0.02743) 
0.00374 
(0.02615) 
0.06696*** 
(0.02350) 
0.00066 
(0.02303) 
0.02941 
(0.02913) 
0.01255 
(0.04085) 
-0.00818 
(0.02700) 
-0.00402 
(0.01523) 
-0.02579 
(0.02157) 
-0.00224 
(0.00307) 
0.00937 
(0.02210) 
-0.00151 
(0.02045) 
0.04549* 
(0.02543) 
0.07802*** 
(0.02416) 
-0.00188 
(0.03068) 
-0.01352 
(0.03093) 
0.04683 
(0.03256) 
-0.00150 
(0.01304) 
-0.02631 
(0.02862) 
0.00002 
(0.00003) 

0.28390*** 
(0.02757) 
-0.02289 
(0.01827) 
-0.00158 
(0.01747) 
0.01601 
(0.01601) 
0.01689 
(0.02205) 
0.03002 
(0.02286) 
-0.03590* 
(0.01935) 
0.00422 
(0.01109) 
-0.01549 
(0.01624) 
0.00377* 
(0.00212) 
-0.00242 
(0.01885) 
0.00929 
(0.01704) 
0.03538* 
(0.02103) 
0.01698 
(0.01734) 
-0.02582 
(0.02300) 
0.00267 
(0.02222) 
0.01019 
(0.02296) 
0.00601 
(0.00917) 
0.00288 
(0.02023) 
0.00001 
(0.00002) 

0.15964*** 
(0.02806) 
0.03718 
(0.02845) 
0.09540*** 
(0.02641) 
0.01367 
(0.02642) 
-0.00628 
(0.03206) 
0.02185 
(0.04702) 
0.00209 
(0.02900) 
-0.00710 
(0.01630) 
-0.02381 
(0.02194) 
-0.00272 
(0.00306) 
0.00597 
(0.02280) 
0.01568 
(0.02134) 
0.04045 
(0.02670) 
0.08705*** 
(0.02708) 
0.01995 
(0.03187) 
-0.02429 
(0.03297) 
0.01032 
(0.03557) 
-0.00793 
(0.01310) 
-0.01939 
(0.03043) 
0.00002 
(0.00004) 

 95 

Table 20 (cont’d) 
Remittances  
 
Dummy of year 2016 
 
Pseudo R-squared 
Observations 
Source: Calculation based on HARVEST data 2012-2016.  *** p<0.01, ** p<0.05, * p<0.1 
Notes: (1) Clustered robust standard errors in parentheses (2) We also controlled for village fixed effects   
            (3) The coefficients are reported as the marginal effects for linear interpretation 

0.000002 
(0.00004) 
0.05468*** 
(0.01509) 
0.3701 
7,496 

0.00008* 
(0.00004) 
0.03353 
(0.02273) 
0.1799 
8,218 

0.00007* 
(0.00004) 
0.06885*** 
(0.02161) 
0.2371 
8,218 

The probability of the adoption of improved farm technologies is greater on the plots of 

dry season rice, which is always irrigated. On average, the dry-season rice plots are around 16 

percent more likely than the rainy-season plots to be associated with the adoption of improved 

rice varieties, holding other factors fixed. Also, the probability of adoption of chemical fertilizer 

and the combination of both improved rice and fertilizer for the dry-season plots is higher than 

rain-fed plots by 28.4 percent and 23 percent, respectively, at 1 percent significance level. This 

finding is in line with that of previous empirical studies in other countries (e.g., Feder & Umali, 

1993; Doss, 2006), and in Cambodia (e.g., Wang, Pandey, & Velarde, 2012; Ly et al., 2012). In 

general, modern inputs such as chemical fertilizers and improved seed varieties work well with 

irrigation. That is why the underdeveloped irrigation infrastructure in an area of unreliable 

rainfall patterns is a constraint for farmers to adopt improved farm technology.  

This finding suggests that, on average, the likelihood that farmers used improved inputs 

on a dry-season rice plot is higher than a rain-fed plot, because dry-season rice relies on 

irrigation. However, there are also farmers who use both improved seeds and chemical fertilizer 

in the dry season and also use fertilizer on their rain-fed land in the wet season. This finding 

solely confirms the important roles of irrigation in the adoption of the improved farm 

technologies. Also, this finding is in line with what we observed during the in-depth interviews, 

 96 

because farmers who adopted improved rice varieties and fertilizers said that sufficient water 

plays an important role in their rice farming. 

In addition, the decision to adopt improved rice varieties and the combination of fertilizer 

and improved rice varieties is positively and significantly associated with knowledge about them 

that farmers acquired from their neighbors. On average, farm households with access to this type 

of knowledge from their neighbors are 9.5 percent and 6.7 percent more likely to adopt improved 

rice varieties and the combination of both improved rice and fertilizer, respectively (Table 20). 

We can explain this finding from the perspective of social learning that farmers can learn from 

others about the use of improved seed and other farm technology. That is, farmers become more 

receptive of the applicability and profitability of the farm technology from the experience of their 

neighbors. This result is reinforced by some empirical studies in other countries. For instance, 

Mathenge, Smale, & Olwande (2014) show that the proportion of village households that use 

hybrid maize seed has a positive and significant influence on a farmer’s decision to adopt that 

improved technology in Kenya. Similarly, a seminal study of pineapple production in Ghana by 

Conley & Udry (2010) empirically shows that farmers adjust their inputs after observing the 

harvest of their neighbors who are also using improved inputs.  Our finding in this paper 

suggests that social learning is also important in the context of agricultural technology adoption 

in Cambodia. Also, the qualitative interviews reinforce this empirical finding. For instance, some 

farmers adopted the improved varieties after observing that some of their neighbors switched 

their seeds from traditional varieties to the improved ones which have a shorter duration and 

higher yield.  

The total landholding of a household has a negative relationship with the probability of 

chemical fertilizer adoption (Table 20). For each hectare of additional land to the total land that a 

 97 

household has, the probability of adopting chemical fertilizer decreases by around 3.5 percent. 

Even though this result is not consistent with those of some studies (e.g., Feder & Mara, 1981; 

Marenya & Barrett , 2007), it is not an exception to all of the literature (e.g., Nkonya et al., 1997; 

Croppenstedt, Demeke, & Meschi, 2003; Kassie et al., 2013). For example, our finding is 

consistent with an empirical study by Kassie et al. (2013) which shows  that households with less 

land are more likely to adopt chemical fertilizer. A possible explanation for this inverse 

relationship is that the limited farm land may induce farm households to intensify their rice 

production by using productivity-enhancing inputs for a particular plot. Also, it is one of the 

essential drivers of the inverse farm-size-productivity, which is commonly found around the 

world. 

Similar to the findings by Olwande, Sikei, & Mathenge (2009) and Teklewold et al. 

(2013), I find that the probability of chemical fertilizer adoption increases by around 3.8 percent 

for each extra year of the household head’s age. This coefficient is statistically significant at 10 

percent level. I checked for a nonlinear relationship, and found that the coefficient on age 

squared is not statistically significant. What we can explain from this finding is that older 

farmers may have more exposure than younger ones to farm technologies and environments, and 

better social capital. On the other hand, age is negatively related with adoption of improved rice 

varieties, but it is not statistically significant.  Some studies found that age is associated with a 

reduction in the probability of adopting some improved farm technologies. Their reason for this 

inverse relationship is that older farmers are more risk averse and have shorter planning 

prospects than their younger counterparts (Mbaga-Semgalawe & Folmer, 2000). Given the 

conflicting arguments, Kassie et al. (2013) hypothesize that linkage between age of household 

head and the probability of adopting improved farm technologies is indeterminate. 

 98 

Regarding educational attainment, which is an indicator of human capital, a household 

head with secondary education is associated with 3.5 percent and 4.5 percent increase in the 

probability of adopting chemical fertilizer and the combination of both improved technologies, 

respectively. The coefficients are statistically significant at 10 percent level. Usually, farmers 

with higher education seem to be better aware of how to apply new knowledge or techniques and 

also better aware of the potential benefits of the improved technologies. The finding is aligned 

with some empirical work in other countries. For instance, Olwande et al. (2009) found that 

secondary education by the household head increases the probability of adopting fertilizer by 

11.2 percent in Kenya. Also, a study in Madagascar by Moser & Barrett (2006) empirically 

shows that an additional year schooling of household head increases probability of adopting 

System of Rice Intensification (SRI) by 1 percent. 

Moreover, on average, a household that owns a television has around 8.9 percent and 7.9 

percent higher probability of adopting improved rice varieties and the combination of both 

improved farm technologies, respectively. This variable is an indicator for a good means of 

getting information because there is extension effort prepared by the Agricultural Extension 

Department via the TV in Cambodia. It is one of the good sources for information about 

improved farm technologies. This linkage is reinforced by the idea of diffusion of innovations 

theory, which hypothesizes that a certain level of an individual’s uncertainty and perceived risk 

can decline by having more access to information. The dissemination of information about 

improved farm technologies and new knowledge depends on communication channels including 

media (Rogers, 2003). In other words, other things equal, the differences across farmers in 

information and knowledge through the television about improved farm technologies can 

positively influence adoption decisions. 

 99 

In addition, TV ownership can be an indicator of wealth, which influences risk aversion.  

That is, a wealthy farmer is less risk-averse than a less wealthy counterpart when it comes to 

adoption of new farming practices. In general, other things equal, wealthier households have 

more financial resources to afford the adoption of improved farm technologies. However, it 

should be noted that other important variables of household wealth in the regressions have no 

significant relationship with the adoption of improved inputs. Those variables include asset 

value, consumption expenditure, motorbike, and hand tractor.  

Furthermore, on average, each 1 million Riels (USD 250) of additional remittances per 

year that a household receives is associated with 0.7 percent increase in the probability of 

adopting the combination of both improved technologies, at 10 percent significance level. In the 

literature, income from non-farm activities has been shown to have some influence on a farm’s 

decision to adopt better farm practices or new variety of crops, because it provides budget-

constrained farmers with liquidity to purchase farm inputs and make other investments 

(Haggblade, Hazell, & Reardon, 2010). In line with this, a recent study by Roth & Tiberti (2016) 

found that migration and remittances reduce poverty headcount rate in Cambodia by 3 to 7 

percentage points.   

The year 2016 dummy captures the dynamics of farm technology adoption in our sample 

over the period 2012-2016. On average, with other factors in the regression equations held 

constant, the result indicates that the probability of adopting chemical fertilizer and the 

combination of both inputs increased by 5.5 percent and 6.8 percent, respectively over that 

period.  Controlling for year allows us to better capture the effects on technology adoption of our 

main factors of interest.  

 100 

3.7. Conclusions and policy implications  

In this article, I have used longitudinal plot-level data of the HARVEST program (2012-

2106) to address two empirical questions. First, we have examined whether farmers’ decisions to 

adopt improved rice varieties and chemical fertilizer are interrelated (interdependent). Second, 

we have investigated the factors that facilitate or impede the probability of adopting those 

improved farm technologies. We have applied a bivariate probit model to test for the 

interdependence of technology adoption. To apply a nonlinear probit model to panel data, we use 

the correlated random effects (CRE) framework, which controls for the correlation between 

unobserved, time-invariant heterogeneity and observed factors in the nonlinear regression model 

by including the means of time-varying independent variables. And, we have used this 

estimation strategy to examine the determinants of adoption by also controlling for village fixed 

effects and using clustered robust standard errors. The main findings of this study can be 

summarized as follows. 

First, our results suggest that adopting the improved rice variety and chemical fertilizer at 

the plot level are complementary. This provides a basis for the government to design policies that 

promote multiple technology adoptions at once, because farmers tend to adopt both improved 

inputs together. Nonetheless, in practice, there may be important constraints to the whole 

package, and those constraints include the lack of good irrigation system, limited access to 

information and knowledge about improved farm technologies, low education and liquidity 

constraint. It implies that we should reduce the constraints to adopting the whole package of 

improved technologies to the extent possible, but also that it is important to have options for the 

ones who cannot adopt everything.  

 101 

Further, there is empirical evidence that the probability of adopting improved farm 

technologies is positively influenced by various factors: irrigation, social learning in the form of 

information from neighbors, age of household head, secondary education, TV ownership (as a 

means of accessing the media), and remittances. At the same time, our results show that the 

probability of adoption is negatively associated with land size of a household.  

The significant role of irrigation on adoption suggests the need to further develop reliable 

irrigation, which helps reduce risk and uncertainty by relying less on the rainfall. Also, this helps 

make Cambodia’s farming system less vulnerable to climate variability and shocks, especially 

late rainfall and drought. Thus, interventions intended to improve market conditions alone may 

not be sufficient to encourage adoption of improved farm technologies. Additionally, the 

importance of social learning calls for further strengthening the rural or local institutions to 

accelerate and sustain technology adoption. In the rural areas, where information asymmetry is 

common and input and output market is incomplete, the local institutions such as farmer 

cooperatives or groups can provide farmers with timely information and inputs.  

The positive linkage between mean of media in the form TV ownership and the 

probability of adoption suggest that promoting agricultural technologies via television program 

might be an effective way of communication. Thus, expansion of the media coverage for 

agricultural extension services, especially via the television would contribute to encouraging the 

adoption of improved rice varieties and the combination of both inputs in our study area as well 

as Cambodia. Further, our results indicate that further investment in public education (the 

secondary one) will also be supportive of the adoption of the interrelated technologies and 

chemical fertilizer. 

 102 

Finally, adoption of improved farm technologies can be affected by other important 

factors such as rainfall, expected profit, and particularly the market participation. Hence, further 

research should take into account those factors when information of that kind is available. This is 

because adoption of improved farm technology is not economically sustainable and attractive to 

the farmers if their farm produce is not marketable.  

 

 

 

 

 

 

 

 

 

 

 

 103 

REFERENCES

 104 

REFERENCES 

 

Babu, S. C., Gajanan, S. N., & Sanyal, P. (2014). Implications of Technological Change, Post-

Harvest Technology, and Technology Adoption for Improved Food Security. In Food 
Security, Poverty and Nutrition Policy Analysis: Statistical Methods and Applications (2nd 
ed., pp. 56–87). Waltham, MA: Academic Press. 

Bandiera, O., & Rasul, I. (2006). Social Networks and Technology Adoption in the Zambezia 

province of Northern Mozambique. The Economic Journal, 116(1957), 869–902. 

Barham, P., & Udry, C. (1999). Technological Progress and Learning. In Development 

Microeconomics (pp. 152–167). New York: Oxford University Press. 

Cappellari, L., & Jenkins, S. P. (2003). Multivariate probit regression using simulated maximum 

likelihood. The Stata Journal, 3(3), 278–294. 

Chamberlain, G. (1984). Handbook of Econometrics. Amsterdam, North Holland: Elsevier. 

Conley, T. G., & Udry, C. R. (2010). Learning About a New Technology : Pineapple. American 

Economic Review, 100(1), 35–69. 

Croppenstedt, A., Demeke, M., & Meschi, M. M. (2003). Technology Adoption in the Presence 

of Constraints : the Case of Fertilizer Demand in Ethiopia. Review of Development 
Economics, 7(1), 58–70. 

Dorfman, J. H. (1996). Modeling Multiple Adoption Decisions in a Joint Framework. American 

Journal of Agricultural Economics, 78(3), 547–557. 

Doss, C. R. (2006). Analyzing Technology Adoption Using Microstudies: Limitations, 

Challenges, and Opportunities for Improvement. Agricultural Economics, 34(3), 207–219. 

Feder, G. (1982). Adoption of Interrelated Agricultural Innovations: Complementarity and the 

Impacts of Risk, Sclae and Credit. American Journal of Agricultural Economics, 64(1), 
255–298. 

Feder, G., Just, R. E., & Zilberman, D. (1985). Adoption of Agricultural Innovations in 

Developing Countries: A Survey. Economic Development and Cultural Change, 33(2), 
255–298. 

Feder, G., & Mara, G. T. O. (1981). Farm Size and the Diffusion of Green Revolution 

Technology *. Economic Development and Cultural Change, 30(1), 59–76. 

Feder, G., & Umali, D. L. (1993). The Adoption of Agricultural Innovations A Review. 

Technological Forcasting and Social Change, 43, 215–239. 

Foster, A. D., & Rosenzweig, M. R. (1995). Learning by Doing and Learning from Others : 

 105 

Human Capital and Technical Change in Agriculture Learning by Doing and Learning from 
Others : Human Capital and Technical Change in Agriculture. The Journal of Political 
Economy, 103(6), 1176–1209. 

Haggblade, S., Hazell, P., & Reardon, T. (2010). The Rural Non-farm Economy: Prospects for 

Growth and Poverty Reduction. World Development, 38(10), 1429–1441. 

Kaliba, A. R. M., Verkuijl, H., & Mwangi, W. (2000). Factors Affecting Adoption of Improved 
Maize Seeds and Use of Inorganic Fertilizer for Maize Production in the Intermediate and 
Lowland Zones of Tanzania. Journal of Agricultural and Applied Economics, 32(1), 35–47. 

Kassie, M., Jaleta, M., Shiferaw, B., Mmbando, F., & Mekuria, M. (2013). Adoption of 

Interrelated Sustainable Agricultural Practices in Smallholder Systems : Evidence from 
Rural Tanzania. Technological Forcasting and Social Change, 80, 525–540. 

Khush, G. S. (1995). Modern varieties - Their real contribution to food supply and equity. 

GeoJournal, 35(3), 275–284. 

Ly, P., Jensen, L. S., Bruun, T. B., Rutz, D., & de Neergaard, A. (2012). The System of Rice 
Intensification: Adapted practices, reported outcomes and their relevance in Cambodia. 
Agricultural Systems, 113, 16–27. 

Mak, S. (2001). Continued innovation in a Cambodian rice-based farming system: Farmer testing 

and recombination of new elements. Agricultural Systems, 69(1–2), 137–149. 

Marenya, P. P., & Barrett, C. B. (2007). Household-level determinants of adoption of improved 

natural resources management practices among smallholder farmers in western Kenya. 
Food Policy, 32(4), 515–536. 

Marenya, P. P., & Barrett, C. B. (2009). Soil quality and fertilizer use rates among smallholder 

farmers in western Kenya. Agricultural Economics, 40(5), 561–572. 

Mathenge, M. K., Smale, M., & Olwande, J. (2014). The impacts of hybrid maize seed on the 

welfare of farming households in Kenya. Food Policy, 44, 262–271. 

Matuschke, I., & Qaim, M. (2008). Seed Market Privatisation and Farmers’ Access to Crop 

Technologies: The Case of Hybrid Pearl Millet Adoption in India. Journal of Agricultural 
Economics, 59(3), 498–515. 

Mbaga-Semgalawe, Z., & Folmer, H. (2000). Household adoption behaviour of improved soil 

conservation: the case of the North Pare and West Usambara Mountains of Tanzania. Land 
Use Policy, 17(4), 321–336. http://doi.org/10.1016/S0264-8377(00)00033-8 

Moser, C. M., & Barrett, C. B. (2006). The Complex Dynamics of Smallholder Technology 

Adoption : Agricultural Economics, 35(607), 373–388. 

Mundlak, Y. (1978). On the Pooling of Time Series and Cross Section Data. Econometrica, 

46(1), 69–85. 

 106 

Nguyen, Y. T. B., Kamoshita, A., Araki, Y., & Ouk, M. (2013). Water availability, management 
practices and grain yield for deepwater rice in Northwest Cambodia. Field Crops Research, 
152, 44–56. 

Nkonya, E., Schroeder, T., & Norman, D. (1997). Factors Affecting Adoption of Improved 

Maize Seed and Fertiliser in Northern Tanzania. Journal of Agricultural Economics, 48(1–
3), 1–12. 

Ogada, M. J., Mwabu, G., & Muchai, D. (2014). Farm technology adoption in Kenya: a 

Simultaneous Estimation of Inorganic Fertilizer and Improved Maize Variety Adoption 
Decisions. Agricultural and Food Economics, 2(1), 1–18. 

Olwande, J., Sikei, G., & Mathenge, M. (2009). Agricultural Technology Adoption: A Panel 

Analysis of Smallholder Farmers’ Fertilizer Use in Kenya (No. AfD-0908). CEGA Working 
Paper Series. Retrieved from 
https://www.researchgate.net/publication/237718845_Agricultural_Technology_Adoption_
A_Panel_Analysis_of_Smallholder_Farmers’_Fertilizer_use_in_Kenya 

Olwande, J., Smale, M., Mathenge, M. K., Place, F., & Mithofer, D. (2015). Agricultural 

marketing by smallholders in Kenya: A comparison of maize, kale and dairy. Food Policy, 
52, 22–32. 

Otsuka, K. (2000). Role of agricultural research in poverty reduction: Lessons from the Asian 

experience. Food Policy, 25(4), 447–462. 

Pingali, P. (2012). Green Revolution: Impacts, Limits, and the path ahead. Proceedings of the 

National Academy of Science, 109(32), 12302–12308. Retrieved from 
http://www.pnas.org/content/109/31/12302.full.pdf 

Rogers, E. M. (2003). Diffusion of innovations (5th ed.). New York: Free Press. 

Roth, V., & Tiberti, L. (2016). Economic Effects of Migration on the Left-Behind in Cambodia. 

The Journal of Development Studies, 0(0), 1–19. 

Smale, M., Just, R. E., & Leathers, H. D. (1994). Land Allocation in HYV Adoption Models: An 

Investigation of Alternative Explanations. American Journal of Agricultural Economics, 
76(3), 535–546. 

Smale, M., & Olwande, J. (2014). Demand for maize hybrids and hybrid change on smallholder 

farms in Kenya. Agricultural Economics, 45(4), 409–420. 

Sumner, D., Christie, M. E., & Boulakia, S. (2017). Conservation agriculture and gendered 

livelihoods in Northwestern Cambodia: decision-making, space and access. Agriculture and 
Human Values, 34(2), 347–362. 

Teklewold, H., Kassie, M., & Shiferaw, B. (2013). Adoption of Multiple Sustainable 

Agricultural Practices in Rural Ethiopia. Journal of Agricultural Economics, 64(3), 597–
623. 

 107 

Theng, V., Keo, S., Nou, K., Sum, S., & Khiev, P. (2014). Impact of Farmer Organisations on 

Food Security : The Case of Rural Cambodia (No. 95). Phnom Penh. Retrieved from 
http://www.cdri.org.kh/webdata/download/wp/wp95e.pdf 

Theng, V., Khiev, P., & Phon, D. (2014). Development of the Fertiliser Industry in Cambodia: 

Structure of the Market, Challenges in the Demand and Supply Sides, and the Way Forward 
(CDR Working Paper Series No. 91). Phnom Penh. Retrieved from 
http://www.cdri.org.kh/webdata/download/wp/wp91e.pdf 

Wang, H., Pandey, S., & Velarde, O. (2012). Pattern of Adoption of Improved Rice Varieties and 

Its Determinants in Cambodia. Procedia Economics and Finance, 2, 335–343. 

Wooldridge, J. M. (2008). Introductory Econometrics: A Modern Approach (4th ed.). Mason, 

OH: Cengage Learning. 

Wooldridge, J. M. (2010). Econometric Analysis of Cross Section and Panel Data. Cambridge, 

Mass: MIT Press. 

Yu, B., & Fan, S. (2011). Rice Production Response in Cambodia. Agricultural Economics, 

42(3), 437–450. 

	
	
	
	
	
	
	
	
	
	
	
	
	
	
	
	
	
	
	
	
	
	

 108 

CHAPTER 4: THE LINKAGE BETWEEN AGRICULTURAL CREDIT  
                    AND ADOPTION OF INTERRELATED MODERN INPUTS IN  
                    CAMBODIA’S RICE PRODUCTION  
	

4.1. Introduction  
 

Improving productivity and production practices at the farm level is of particular 

importance for agrarian economies. An approach to this agenda has centered on promoting 

adoption of improved farm technology, which is expected to have a significant impact on 

livelihood improvement and poverty reduction through better income generation (Minten & 

Barrett, 2008; Ali & Abdulai, 2010). In this regard, a body of literature shows that access to 

credit plays an important role in encouraging adoption of improved farm technologies and 

diversifying farm investment for cash-constrained households  (Barslund & Tarp, 2008; Herath 

& Jayasuriya 2010; Abebaw & Haile, 2013). 

In Cambodia, however, there is limited empirical evidence of the relationship between 

credit and adoption of agricultural technology. For instance, Theng & Kem (2009) used in-depth 

interviews to show that households demanded more loans to smooth their food consumption and 

repay existing debts during a crisis. Lun (2013) used panel household data from nine rural 

villages to empirically examine the impacts of the global financial and economic downturn on 

rural households’ demand for loans. The findings reveal that demand for a loan used for farm 

investment and debt repayment increased during a crisis, and demand for a formal loan was 

higher than an informal one. The most recent study by Roth, Araar, Sry, & Phann (2017) used 

difference-in-differences estimator and panel data of eleven villages in Cambodia to evaluate the 

impact of micro-credit access on expenditure on inputs of paddy production and paddy income. 

They found that the impact of microcredit access is positive and statistically significant.  

 109 

Also, research studies on improved farm technologies in Cambodia have not been very 

extensive. Among the few examples, there is a case study by Ngo & Chan (2010) providing an 

overview of agricultural extension services in two provinces, and a qualitative analysis by 

Chhim, Theng, & Nou (2013) that explores some of the key constraints to application of 

agricultural knowledge disseminated by extension agents in Takeo province. Regarding the 

studies on adoption of modern inputs, Mak (2001) studied the innovation process in rice-based 

farming systems with a focus on adoption of modern rice varieties, and Theng, Khiev, & Phon 

(2014) studied the regulatory and market constraints on farmers’ demand for fertilizer. However, 

those studies also rely on descriptive statistics and case study, which does not provide a rigorous 

measure of the factors which are negatively or positively associated with adoption decisions. 

Thus, it is worth studying credit access in the context of joint adoption of modern inputs which I 

first introduced in the second essay of this dissertation. 

An empirical problem in adoption research is that from the econometric perspective, 

some important potential explanatory variables such as access to agricultural extension, 

participation in a program, and access to credit are subject to endogeneity, which causes 

empirical results to be biased. Dealing with more than one endogenous variable in a regression 

equation is not common or even can be infeasible in empirical work(Wooldridge, 2008). If I can 

only use one, it is wise to choose a factor that I think is likely to have the strongest effect on 

adoption decision as our variable of interest to estimate its effect on outcome variable, while also 

using the feasible and appropriate method to minimize the bias. In our study, I use credit for 

agricultural activities as the variable of interest because of its signification relationship with 

likelihood of technology adoption in the literature (Tadesse, 2014; Yokouchi & Saito, 2016), and 

because this linkage has not yet been empirically examined in Cambodia. Hence, with the 

 110 

intention of filling the literature gap, the main objective of this study is to examine the causal 

linkage between credit and the adoption the interrelated inputs (high-yield variety, chemical 

fertilizer, and pesticide, herbicide or fungicide) in Cambodia. Building on what I learned in the 

second paper of this dissertation, this study will also examine whether adoption decisions by 

farm households for high-yield varieties, chemical fertilizer and pesticide are interdependent.  

To test for interrelated adoption decisions, we apply the multivariate probit (MVP) 

regression. For the linkage between credit and the probability of adoption, I use propensity score 

matching (PSM) to reduce the selection bias of credit, with the household and village 

characteristics included as the covariates. Then, I run regressions on the matched samples with 

propensity score weights to estimate the effect of credit for agricultural activities on the adoption 

of modern inputs by controlling for household and village characteristics as well as province 

fixed effects. This study is based on a cross-sectional survey from the Census of Agriculture of 

Cambodia (CAC) 2013, conducted by Cambodia’s National Institute of Statistics (NIS). Thus, to 

the best of our knowledge, this is the first empirical study in Cambodia that uses the data from 

this census of agriculture. Also, it should be noted that this essay and the second essay of this 

dissertation are based on analysis using different datasets. 

This paper is organized as follows. Following this introduction, section 4.2 contains a 

literature review about agricultural technology adoption and the effect of credit on the adoption. 

The Cambodian context of agriculture and rural credit is presented in section 4.3. The conceptual 

framework and empirical methodology are explained in section 4.4. Section 4.5 describes the 

data source and descriptive statistics. Section 4.6 focuses on the findings and discussion. The last 

section concludes the study and provides some policy implications. 

 111 

4.2. Literature review 

One of the well-known early works on agricultural technology adoption is Griliches's 

(1957) study of aggregate adoption of hybrid corn in mid-west US. The study focused on the 

effect of heterogeneity of local conditions on adoption, and the expected profits were also 

believed to influence the rates of adoption. In this connection, it is worth noting that risk and 

uncertainty associated with the expected profits are the common constraints to farmers’ decisions 

to adopt agricultural technology (Smale, Just, & Leathers, 1994; Feder & Umali, 1993; Feder, 

Just, & Zilberman, 1985). 

In addition, a flourishing literature reveals that one of the important determinants of 

modern input adoption is credit access. Drawing on an extensive review from experiences of 

Asian economies, Herath & Jayasuriya (2010) concluded that subsidized credit programs can be 

an important mechanism to promote the adoption of agricultural technology, particularly at the 

early phase. Tadesse (2014) used panel data on 278 households (5,700 plots) from rural Ethiopia 

to assess the impact of credit access and safety nets on fertilizer adoption. The findings from the 

instrumental variable approach suggest that having access to credit significantly increases the 

probability and intensity of fertilizer use. In line with this, by using the instrumental variable 

approach to address the endogeneity problem, Teklewold, Kassie, & Shiferaw's (2013) empirical 

study shows that credit constraint is a factor significantly impeding the adoption of the multiple 

sustainable farming practices, including improved seed varieties and fertilizer in Ethiopia. 

Applying the treatment-effects model and cross-sectional data of 404 households, Simtowe, 

Zeller, & Diagne (2009) found credit constraints have a negative impact on the amount of land 

allocated to hybrid maize in rural  Malawi. A recent study by Yokouchi & Saito (2016) also 

found that using credit for agricultural purposes is positively associated with the adoption of a 

 112 

newly introduced rice variety (called NERICA) in Benin. However, the main objective of that 

study is to investigate the linkage between participation in farmer group and the adoption 

decision, so the endogeneity of credit use is not addressed.  

Concerning the use of credit for agricultural activities, it is worth noting that there are 

various factors which negatively affect rural credit markets. The first major issue is credit 

default, which is caused by income shocks such as weather fluctuations (drought or flood) and 

commodity price shock (Besley, 1998). Hence, farmers who contracted a loan for agricultural 

production may be unable to repay it because of the shocks, which has a severe effect on credit 

supply. The second issue is that the majority of farmers mistakenly expect that defaulting on a 

loan repayment is normal, and oftentimes it happens with a credit program supported by the 

government (Tadesse, 2014). Third, the enforcement of contracts is also a problem for credit 

markets. For instance, in practice, it is problematic to use or sell the land seized from the 

defaulters, due to politics. Another far-reaching problem is the underdevelopment in 

complementary institutions, such as low levels of literacy, poor communication systems (storage 

of credit histories), and the limited complementary markets, especially insurance (Besley, 1998). 

4.3. Country context 

4.3.1. Cambodia’s agriculture 

Agriculture is a pillar of Cambodia’s economy, and rice is the dominant crop in terms of 

cultivated area and total production (FAO, 2014). Additionally, the majority of rural livelihoods 

depend on agriculture despite an increasing trend of labor migration (MAFF, 2015). This sector 

is characterized by smallholding and subsistence farm activities with most of households owning 

less than two hectares of land (FAO, 2014). A recent study by the World Bank suggests that 

growth in Cambodia’s agriculture was pro-poor, because 60 percent of poverty reduction from 

 113 

2007 to 2011 was attributed to larger rice production and increased farm wages (World Bank, 

2014). The study further indicates that agricultural development will continue to be a contributor 

to poverty reduction given its share in the labor force, especially the fact that most of the poorest 

and more vulnerable households are involved in this sector (World Bank, 2015). 

There are two main seasons of rice production. In the wet season, rice is planted in June 

or July and it is harvested in November or December. It is the main season for rice farming 

because it accounts for approximately 77 percent of annual total rice production and 84 of the 

rice cultivated area in 2014 (MAFF, 2015). Recently, there has been an expansion of early wet 

season rice cultivation, which is from June or July to August or September. Dry season rice 

production, which depends on irrigation, begins in December or January and the harvest is made 

in April or May. Wet season rice yield was 2.82 tons per hectare while the dry season rice yield 

was 4.43 tons per hectare in 2014 (MAFF, 2015). The aggregate average rice yield declined by 

2.67 percent from 3.163 tons per hectare in 2013 to 3.079 tons per hectare in 2014, due to 

draught and pests, according to the annual report of the Ministry of Agriculture. This is a case in 

point of the fact that Cambodian farming system is vulnerable to the extreme climatic shocks 

such as drought (RGC, 2013), because farmers mainly rely on rain-fed farming. 

Low productivity remains one of the major challenges of Cambodia’s agriculture because 

of an underdeveloped irrigation system, lack of agricultural knowledge among farmers about 

appropriate production techniques, and inadequate use of modern inputs such as fertilizers and 

seeds (Yu & Fan, 2011; RGC, 2013). Inadequate input use results in part from constraints and 

challenges facing input markets that increase the costs of agricultural input use. Those challenges 

include (1) complicated licensing procedures and regulations, (2) and lack of clarity of roles and 

responsibilities between government institutions in charge of regulating fertilizer trade (Theng, 

 114 

Khiev, et al., 2014). Fertilizer adoption is more common than high-yield varieties (HYV), 

especially rice production (Yu & Fan, 2011). The limited body of literature about HYV adoption 

in Cambodia’s rice production has found that rainfall uncertainty is one of the reasons for low 

adoption, because most of Cambodia’s rice farming systems are rainfed (Mak, 2001; Wang et al., 

2012). 

According to the National Institute of Statistics, Cambodia has four agro-ecological 

zones: Tonle Sap, Plains, Plateau/Mountain and Coastal region. Rice is mainly planted in the 

Tonle Sap Zone, which is mostly in two zones: the northwest around Tonle Sap Lake, bordering 

Thailand, and the Plains Zone, which is in the southeast and borders Vietnam (Yu & Fan, 2011; 

NIS, 2015). Dry season rice (irrigated rice) is commonly produced in the Plains Zone which 

accounts for approximately 70 percent of the total cultivated area of this type of rice in 

Cambodia (Wang et al., 2012). Since dry season rice requires more fertilizer, the proportion of 

farm households using fertilizers was the highest in the Plains Zone, followed by the Tonle Sap 

Zone (NIS, 2015).  

4.3.2. Cambodia’s rural credit 

The government of Cambodia considers credit essential for rural livelihood 

improvements (RGC, 2013), particularly through improving agricultural productivity of 
liquidity-constrained households. At the same time, limited access to rural credit has been 
identified as one of key constraints on agricultural development in Cambodia (Sok, Chap, & 
Chheang, 2011). There are two main sources of credit in rural Cambodia. According to the 
previous studies in Cambodia, the formal sources are banks and microfinance institutions (MFIs) 
while informal sources include friends and relatives, moneylenders or farmer 
groups/cooperatives (Lun, 2013). Loans from moneylenders generally have very high interest 

 115 

rates compared to others (Lun, 2013). A poverty dynamics study in nine rural villages showed 
that the average monthly interest rate charged by the moneylenders was 6.65 percent in 2011, 
while the rate charged by NGOs and farmer groups was 2.49 percent and that for MFIs was 2.6 
percent (CDRI, 2012). Further, the share of informal loans from moneylenders has declined over 

the period 2001-2011, due to the expansion of the services provided by formal sources, 

particularly MFIs. Moneylenders, NGOs and farmer groups are lumped together as informal 

credit in those previous studies, because most of the credit participating households got loans 

from more than one source. Then, categorizing loans by formal or informal status is the feasible 

way in the analysis, and it is also applied to our study. Additionally, in our dataset, there is no 

information on the amount of loans and the interest rates.  

In 2011, I had a chance to conduct some key informant interviews and focus group 

discussions with some members of the self-help groups and the agricultural cooperatives in four 
provinces. We found that their primary motive to participate in these groups is to build savings 
and to access short-term emergency loans at lower interest rate rates so that they can reduce their 
dependence on moneylenders (Theng, Keo, Nou, Sum, & Khiev, 2014). Some the agricultural 
cooperatives and self-help groups also provide in-kind credit to their members such as seeds, 
fertilizers, chicks and ducklings. 

Nonetheless, shortage of capital for credit is a major obstacle to those farmer 

organizations. The quantitative results of the study confirmed that 83 percent of 330 member 
households reported that their cooperatives or groups did not have enough capital to extend loans 
to the members who need to buy agricultural inputs or invest in their farm production (Theng, 
Keo, et al., 2014). Also, this finding was reinforced by what I observed from the key informant 
interviews and focus group discussions. In other words, the capital savings of those cooperatives 

 116 

or self-help groups were so limited that the demands of their members could not be met. Some of 
the cooperatives and farmer groups got the initial capital from their supporting agencies such as 
NGOs and the provincial department of agriculture. This implies that increasing loan funds in 
agricultural cooperatives or self-help groups has a potential to promote the adoption of modern 
inputs or other agricultural technology.  This is a characteristic of the economic success story of 
South Korean agriculture between 1961 and 1975 (Herath & Jayasuriya, 2010).  

4.4. Conceptual framework and empirical methodology 

4.4.1. Conceptual framework 

4.4.1.1. Adoption of interrelated inputs 

The common characteristic of models conceptualizing farm households’ decisions to 

adopt agricultural technology is that a household chooses a set of improved farm technologies to 

maximize the expected utility of profit conditional on the adoption decision (Feder, 1982; Smale, 

Just, & Leathers, 1994; Feder & Umali, 1993; Ogada, Mwabu, & Muchai, 2014). “Given a 

perceived output risk related to agricultural technology, and farmer risk aversion, farmers 

maximize their expected utility through the dichotomous choice of whether or not to adopt the 

technology” (Feder & Umali, 1993, p. 217).  In our study, a household chooses to adopt 

interrelated inputs together, if the expected utility of profit with this adoption is higher than that 

of adopting only one of the inputs or neither, and it is expressed in equation (10). 

[
UE

]
(
)
p !

with

[
UE

(
p

without

]
)

               (equation 10) 

The study hypothesizes that a household’s decisions to adopt fertilizer and high-yield 

variety seeds and pesticide (or herbicide or fungicide6 ) are complementary.  

																																																								
6	Pesticide, herbicide and fungicide are represented by the same dichotomous variable in this study.	

 117 

4.4.1.2. Use of credit in agricultural activities 

The fundamental framework for explaining credit demand is utility maximization 

(Barslund & Tarp, 2008). According to this framework, households demand credit if they expect 

that the utility from contracting a loan (A"/&+B) is larger than that without using any loan 
(A"B&/&+B). In other words, agricultural households demand credit for their agricultural activities 
if the utility gained from contracting loan (A"/&+B−A"B&/&+B) is greater than 0, and this utility 

gain is a latent variable which is a function of observed characteristics of households and 

villages. 

 Using credit to support agricultural activities has been empirically found to have a 

signification relationship with likelihood of technology adoption, particularly the adoption of 

modern inputs (Shakya & Flinn, 1985; Herath & Jayasuriya, 2010; Yokouchi & Saito, 2016). 

That is, the liquidity-constrained farmers contract loans from either formal and informal sources 

to relax their liquidity problems in farm investment (Feder, Lau, Lin, & Luo, 1990; Simtowe, 

Zeller, & Diagne, 2009). In general, this idea implies that credit plays a more important role in 

the adoption of agricultural technology that requires more cash costs. 

In our study, the modern inputs must be purchased, so farmers need capital to obtain 

them. Contracting a loan for agricultural activities reduces capital constraints on farm 

households, therefore improving the financial ability of households to get the needed modern 

inputs (Barslund & Tarp, 2008; Herath & Jayasuriya, 2010). Further, I hypothesize that the high-

yield variety, fertilizer and pesticide are complementary; the joint adoption of these interrelated 

inputs requires higher cash costs. Therefore, in our study I hypothesize that contracting loans for 

agricultural production increases the probability of adoption of modern inputs, i.e. high-yield 

variety, fertilizer and pesticide (also herbicide and fungicide) of cash-constrained households.  

 118 

4.4.2. Empirical methodology 

4.4.2.1. Simultaneous adoption  
	

To test for simultaneous adoption decisions, the farmer’s choice of interrelated inputs is 

modeled using the multivariate probit (MVP) model (Dorfman, 1996). The multivariate probit 

regression allows us to simultaneously run more than one univariate (ordinary) probit model, 

each of which has only one binary dependent variable. This approach can statistically show 

whether adoption decisions on the modern inputs are interdependent. Adapted from the empirical 

studies by Kassie et al. (2013) and Teklewold et al. (2013), the econometric model of 

multivariate probit regression, characterized by the binary dependent variables for our study, is 

expressed as follows. 

Y

hi X
*
=

'
hi

eb +
hi

i

(11) 

Y
hi

=

1
ì
í
0
î

>

Yif
0
*
hi
otherwise

(12) 

where h represents household level, and i=S, F, and P denotes the input choices (high-yield 

variety seeds (S), fertilizer (F), pesticide, herbicide or fungicide (P)). Since there are three input 

choices (three binary dependent variables) which is more than two, this model is called 

multivariate probit regression. 

Equation (12) implies that a household (hth)’s latent variable *

hiY ,	which is the net benefit 

that farm households generates from the adoption of the ith input, is associated with observed 

household and plot characteristics (

hiX ) and the unobserved characteristics in the error term (

hie ).  The error terms of equation (12) jointly follow a multivariate normal distribution with 

zero conditional mean and variance normalized to unity, with systematic covariance matrix given 

by:  

 119 

Ω= 1
!H.
1
!.H
!IH !I.

!HI
!.I
1     (equation 13) 

 

 

The specification with non-zero off-diagonal elements in the covariance matrix 

SFr , 

SPr 	and 

FPr 	symbolizes  the  unobservable  correlation  between  the  error  terms  of  the  three  univariate 

probit equations. In other words, this method allows the error terms of each equation to be freely 

correlated, which implies whether different inputs complement (positive correlation) or substitute 

(negative correlation) for each other in the adoption decision. Fundamentally, not only does this 

method enable us to test for simultaneity of adoption decisions, it also reports the empirical results 

of each univariate probit regression (Cappellari & Jenkins, 2003).  

4.4.2.2. Propensity Score Matching (PSM)  

In the absence of an experimental design, a household’s use of credit is not random in 

practice, and there is very likely to be a selection bias problem. As this study uses cross-sectional 

data, the potential methods to address selection bias are instrumental variables (IV) and 

propensity score matching (PSM). In principle, the former is preferred to the latter because one 

can address unobservable selection bias such as differences in motivation to contract loans for 

agricultural activities if an instrumental variable is available. 

 However, finding a valid instrumental variable that is correlated with getting credit 

without having a direct effect on adoption of modern inputs is a difficult task empirically. One 

approach to finding an instrumental variable (IV) is if there are specific eligibility criteria to 

determine who can obtain credit and who cannot.  For example, if holding more than two 

hectares of land were required in order for a farmer to be eligible for credit, this could be a 

potential instrumental variable. On the other hand, this kind of criteria, which was used as an 

 120 

instrumental in an Pitt & Khandker's (1998) empirical study on the effect of credit programs in 

Bangladesh, is not available in Cambodia.  Another approach to identifying an IV, which is 

based on the program placement perspective, includes distance to a source of credit – a bank or 

an MFI. In practice, nonetheless, most MFI dispatch their credit promotion officials to the 

villages, so this variable is not a valid IV, from that perspective. 

Therefore, to examine the causal linkage between credit use and adoption of modern 

inputs, I use PSM to get the matched samples so as to minimize the selection bias of credit. 

Then, I run regressions on the samples to estimate the effect of credit on the adoption of modern 

inputs. The covariates for PSM will include attributes of households and villages, which 

potentially affect the probability of contracting a loan for agricultural purposes. With this 

method, I can minimize selection bias by reducing the differences in observable characteristics of 

the households that used credit to support their agricultural activities and those that did not.  

I have four treatments characterizing different types of farmers who took loans for 

agricultural purposes during the last 12 months.  They will be compared to a group of control 

households defined as those that did not contract a loan for agricultural activities during the last 

12 months.  The four treatment categories are defined as follows: (1) households that contracted 

loans (from any sources); (2) households with credit only from formal sources such as banks or 

micro-finance institutions (MFI); (3) households with loans only from informal sources 

including a friend, relative or money lender or agricultural cooperatives; and (4) household with 

loans from both types of credit – formal and informal. The four outcome variables for this study 

are dichotomous adoption variables (1 for adoption; 0 otherwise) for 1) high-yield variety seeds, 

2) chemical fertilizer, 3) pesticide/herbicide/or fungicide, and 4) the combination of the three 

types of modern input. 

 121 

The limitation of PSM is that it is based on the observed characteristics, so it does not 

deal with unobserved factors affecting the probability that households contract a loan for 

agricultural activities. However, as indicated by Kuwornu & Owus (2012), after controlling 

observed covariates, PSM can help reduce (though not eliminate) selection bias resulting from 

unobserved confounding factors. Further, the sample size for our study (almost 70,000 

households across more than 8,000 villages) is exceptionally large and nationally representative.  

This is a considerable advantage for the PSM approach. More importantly, the sample size of our 

comparison group is large relative to the treatment group, which enables good matching, 

therefore improving the precision of our estimates (Heckman, Ichimura, & Todd, 1997; 1998).  

4.4.2.3. Regressions on the matched samples  

After matching, I run regressions on the matched samples to estimate the effects of credit. 

By doing so, the regressions are weighted by propensity score weights from PSM to minimize 

the selection bias of credit, and the standard error is clustered at village level to ensure that it is 

robust for making inference. The propensity score weights are defined as 1/KLMKNOPQRS	PTMLN 
for treatment group and as 1/(1−KLMKNOPQRS	PTMLN) for comparison group (Khandker et al., 

2010). It should be noted that there are so many village dummies, and most the villages have a 

small number of households. Thus, I decided to control for village-specific characteristics and 

province dummies instead of using village dummies in the regressions.   

To examine the effects of land titling on the adoption of the modern inputs, I use a probit 

model which can be derived from introducing a latent variable V"∗, which is defined as   
 In practice, we cannot observe V"∗, but instead we observe V", which takes the values of 1 if V"∗>
0, and V"=0 if  V"∗≤0.  

V"∗=X"Y+["  (equation 15) 

 122 

4.5. Data and descriptive statistics 

4.5.1. Background on Census of Agriculture of Cambodia (CAC) 2013 

The study is based on a survey from the Census of Agriculture of Cambodia (CAC) 2013, 

which has 6 modules. The first two modules (A and B) were on census basis, and their main 

purpose is to provide a sampling frame for agricultural surveys. The data in these two modules 

were collected from approximately 2.6 million households across 14,073 villages, 1,621 

communes, and 24 provinces, and this census was conducted from April 14 to May 31, 2013. 

Four other household-level supplementary modules were on a sample basis with detailed data on 

growing crops (Form C), raising livestock and poultry (Form D), aquaculture activity (Form E), 

and socio-economic activities of agricultural households (Form F). Module G, the fifth 

supplementary module, is the village survey.    

Our study uses the CAC survey data (supplementary modules C to G). The survey 

sampling design was a stratified two-stage sampling one in which the first stage units (FSU) 

were villages in a commune, and households were the second stage units (SSU). FSUs were 

selected by systematic Probability Proportional to Size with size being the number of households 

with agricultural holdings. And, the SSUs were systematically selected with a random starting 

point. The survey involved 98,883 agricultural households across 10,218 villages throughout 

Cambodia, and it was conducted over the period of November 1-22, 2013. Module G is the 

village survey which included some village characteristics such as soil types, proneness to any 

disaster, economic activities, distance to a national road, communication facilities, and presence 

of NGOs. NIS  (2015) explains further details about sampling design and methodology.  

I managed the sample for our empirical analysis with a focus on the purpose of our study. 

Its main objective is to examine the effects of agricultural credit on the adoption of modern 

 123 

inputs in rice production. However, the adoption variables were reported at household level, and 

only adoption of the rice high-yield variety is exactly for rice production. The other modern 

inputs (chemical fertilizer, and pesticide or herbicide) can be used for other crops besides rice. 

Due to this limitation, it is worth dropping the households that did not grow rice, for two 

important reasons. First, by doing so, the empirical results will mainly have implications for rice 

production which is the most common crop in the sample and Cambodia’s agriculture, and the 

focus of this paper. Second, excluding the households which did not engage in rice production 

ensures that I have more comparable data for all observations in the sample. By doing this so, I 

can also minimize the chance of comparing apples and oranges, which is very important for the 

impact estimates. 

After data cleaning and merging, I obtain 69,229 rice farming households across 8,733 

villages. There are 23,644 households that contracted loans for agricultural activities during the 

last 12 months, whereas the other 45,585 households that did not get any agricultural loans 

within the same period are included in the comparison group. Of the 23,644 households with 

agricultural loans, 11,012 households got loans from only formal sources (bank or micro-finance 

institutions –MFI); 6,153 households took loans from only informal sources such as friend, 

relative, money lender or agricultural cooperative; the rest (6,479 households) used credit from 

both formal and informal sources. And, it should be noted that these three categories of 

household with agricultural loans are mutually exclusive in our sample. In addition, I 

acknowledge that land title is an important covariate for contracting agricultural loans, but the 

information about land titling was not captured in the CAC survey data (supplementary modules 

C to G). 

 124 

4.5.2. Definitions and summary of outcome variables and covariates 

There are four outcome variables for our study. They are a dichotomous adoption 

variable (1 for adoption; 0 otherwise) for high-yield variety of rice, chemical fertilizer, 

pesticide/herbicide/or fungicide, and the combination of the three types of modern inputs. For the 

covariates of PSM, I control for 39 variables including characteristics of households and villages.  

Definitions of all variables are reported in Table 27 of the appendix (it is in an appendix because 

it is very long). 

Tables 28 and 29 of the appendix present descriptive statistics for all variables before 

matching, and I choose to describe only some main variables. The average adoption rates of 

high-yield variety for rice, chemical fertilizer, PHF7 (pesticide, herbicides or fungicide) and the 

combination of the three types of inputs were 26 percent and 78 percent, 37 percent and 13.7 

percent, respectively. Households with both types of credit for agricultural purposes had higher 

adoption rates than those with only formal credit, only informal credit, or no credit (the control 

households) (Table 29). The average percentage of rice farms with irrigation is higher for 

households with credit from both sources (37.7 percent) than for households with credit from 

only formal sources (35 percent), those with loans from only informal sources (26 percent), and 

the control households (25 percent). Additionally, 14 percent of households participated in a 

social protection program8 during the last 12 months: 39 percent households with both types of 

credit, 20 percent of households with only informal credit, 18 percent of those with only formal 

credit, and only 1 percent of control households (Table 29).   

																																																								
7 Pesticide, herbicide and fungicide were reported by the same dichotomous variable in the survey dataset we got 
from National Institute of Statistics. 
8 This variable is defined as social protection program in general such as cash transfer or work-for-food program. 
The survey was not particularly designed for any social protection program, because it is a part of the Census of 
Agriculture.	

 125 

Furthermore, I also control for household characteristics such as demographic situation, 

livestock assets, sources of income, some modern agricultural tools and access to agricultural 

extension service by sources such as government, NGO and community-based organizations 

(CBO). I also include dummy variables for access to information about improved farming 

practices that could be hypothesized to affect the adoption decision. These include information 

from television, radio, peers and farmer organizations. The village-specific variables that I 

control for include distance from the village the nearest national road, seasonal labor movement, 

flood or drought in the last five years, soil type and agro-ecological zones. 

4.6. Empirical results 

4.6.1. Simultaneous adoption of the modern inputs 

This section presents the empirical results estimated with the multivariate probit model. 

)3(2c =3575.51, P<0.001) of the null hypothesis that the 

Overall, the likelihood ratio test (
covariance of error terms across the three univariate probit regressions are jointly equal to zero is 
rejected (Table 21). This result implies that the modern input adoption decisions are not mutually 
independent (Cappellari & Jenkins, 2003; Teklewold et al., 2013; Kassie et al., 2013). Further, 
the correlations between error terms of the three adoption equations are shown in Table 21, and 
their statically significant and positive coefficients suggest that the high-yield rice variety, 
fertilizer and pesticide or herbicide or fungicide are all complements.  

!_`a 

!.(%,"/"0(% 

- 

Table 21: Correlation coefficients for MVP regression equations 

!.(%,"/"0(% 
!I_. 
Likelihood ratio test of: !.(%,"/"0(%__`a= !I_.__`a= !I_._.(%,"/"0(%=0 

0.165 (0.0197)*** 
0.15 (0.0176)*** 

0.45 (0.016)*** 

 

chi2(3) =  3406.62***           (Significance levels: * 10 percent; ** 5 percent; *** 1 percent) 

   Source: Calculation based on CAC 2013. Clustered robust standard errors in the parenthesis 

 126 

As mentioned in the section presenting the descriptive statistics, on average 26 percent of 

the 69,229 rice farming households adopted the high-yield variety for rice, which is known as 
short-duration variety. Seventy-eight percent adopted chemical fertilizer, 37 percent adopted 
PHF9 (pesticide, herbicides or fungicide), and 13.7 percent adopted the combination of rice, 
fertilizer and PHF. It is worth noting that fertilizer is very common, but HYV adoption is low, 
and PHF is in the middle. These adoption rates are consistent with the fact that rice yield increase 
from 1997 and 2008 was mainly attributed to better access to fertilizer and irrigation expansion, 
rather than the high-yield varieties (Yu & Fan, 2011). One of the reasons for the low adoption of 
improved rice varieties is that most of Cambodia’s rice farming is rain-fed, which is vulnerable 
to drought (Wang et al., 2012). 

4.6.2. Getting the matched samples 

I use propensity score matching (PSM) to get the matched samples to reduce the selection 

bias of contracting loans based on observable characteristics. There are four treatments 
characterizing different types of farmers who took loans for agricultural purposes during the last 
12 months. As mentioned in section 5, the four treatment categories are defined as follows: (1) 
households that contracted loans (from any sources); (2) households with credit only from formal 
sources such as banks or micro-finance institutions (MFI); (3) households with loans only from 
informal sources including a friend, relative, money lender or agricultural cooperative10; and (4) 
households with loans from both types of credit – formal and informal. Then, in terms of the 
outcome variables, we compare each of them to a group of control households defined as those 
that did not contract a loan for agricultural activities during the last 12 months. Given that it is 
only 12 months, we are looking at short-term credit, represented by the available information 

																																																								
9 They were reported by the same dichotomous variable in the dataset I got from National Institute of Statistics. 
10 This categorization is based on the previous studies in Cambodia.	

 127 

about agricultural credit in our data. Additionally, there is no information that enables us to 
distinguish farmers who are rich enough so that they do not need short-term credit from those 
who are very poor and need it but do not take it. In other words, this lack of distinction between 
farmers regarding their non-use of credit does not enable us to construct a comparison group 
which should be comprised of the credit-constrained farmers who needed a loan but could not 
get it. We use the five-nearest-neighbor matching technique because it has good quality matching 
as shown in Table 22. 

The useful indications for checking the matching quality are the comparisons of the mean 
and median of standardized bias, Pseudo-R2 and P-value of likelihood ratio tests before and after 
matching. As reported in Table 22, the medians and means of absolute bias for all comparisons 
are below 3 and 5 percent, respectively. These are the standard benchmarks proposed by 
Caliendo & Kopeinig (2008). Additionally, the Pseudo-R2 from the estimation of propensity 
score after matching becomes significantly lower after matching. Also, P-value of likelihood 
ratio tests become insignificant after matching, implying PSM can minimize the systematic 
differences in the distribution of observable characteristics between the matched groups. The 
details about the test of matching for each matched sample can be found in Table 30, 31, 32 and 
33 of the appendix. This test informs us about the bias reduction in the observable characteristics 
after the two groups are matched. 

 
 
 
 
 

 128 

Before matching 

After matching 

Before matching 

After matching 

After matching 

12.1 
8.1 
0.086 
0.000 

9.8 
7.8 
0.061 
0.000 

Table 22: Summary of PSM quality indicators from 5 nearest-neighbor matching 
HHs with any source of credit for agriculture versus Control Group 
0.9 
0.9 
0.001 
0.493 
HHs with only formal credit for agriculture versus Control Group 
0.7 
0.8 
0.000 
1.00 
HHs with only informal credit for agriculture versus Control Group  
0.6 
0.4 
0.000 
1.000 
HHs with both sources of credit for agriculture versus Control Group 
1.4 
1.3 
0.002 
0.597 

Matching quality indicators 
Mean absolute bias (%) 
Median absolute bias (%) 
Pseudo R2  
P-value of LR 
Matching quality indicators 
Mean absolute bias (%) 
Median absolute bias (%) 
Pseudo R2  
P-value of LR 
Matching quality indicators 
Mean absolute bias (%) 
Median absolute bias (%) 
Pseudo R2  
P-value of LR 
Matching quality indicators 
Mean absolute bias (%) 
Median absolute bias (%) 
Pseudo R2  
P-value of LR 
Source: Calculation based on CAC 2013 
Note: LR: Livelihood ratio 

12.2 
9.6 
0.073 
0.000 

17.8 
9.0 
0.161 
0.000 

Before matching 

Before matching 

After matching 

Moreover, we also have a visual comparison of the distribution of the propensity score 
between treatment and control groups before and after matching as it also shows the quality of 
matching. Figures 6, 7, 8 and 9 of the appendix illustrate that there is almost perfect overlap in 
the distribution of estimated propensity scores between the treatment and control households for 
every matching.  

The figures after matching are weighted by the propensity score weights, and they are 

based on the common support observations that have propensity score weight. The best matches 
are provided by the observations that are on common support  (Khandker et al., 2010; Cerulli, 

 129 

2015). The columns reporting the number of observations with a propensity score weight are the 
ones that provide the best match, and they are ones that are used in our regressions.  

Table 23: Number of observations in common support region for each matching 

    HHs with any source of credit for agriculture versus Control Group 

Off support 

On support 

On support and with 

propensity score weight 

 
 
 

45,585 
22,462 
68,047 

    HHs with only formal credit for agriculture versus Control Group 

Off support 

On support 

On support and with 

propensity score weight 

Sample 

Untreated 
Treated 
Total 

Sample 
Untreated 
Treated 
Total 

Sample 
Untreated 
Treated 
Total 

  Total 
  45,585 
  23,644 
  69,229 

   Total 
  45,585 
  11,012 
  56,597 

   Total 
  45,585 
 
6,153 
  51,738 

0 

1,182 
1,182 

0 
550 
550 

0 
307 
307 

45,585 
22,462 
68,047 

45,585 
10,462 
56,047 

45,585 
5,846 
51,431 

    HHs with only informal credit for agriculture versus Control Group 

Off support 

On support 

On support and with 

propensity score weight 

45,585 
10,462 
56,047 

45,585 
5,846 
51,431 

    HHs with both formal and informal credit for agriculture versus Control Group 

Sample 
Untreated 
Treated 
Total 
Source: Calculation based on CAC 2013 

   Total 
  45,585 
 
6,479 
  52,064 

Off support 

On support 

0 
323 
323 

45,585 
6,156 
51,741 

On support and with 

propensity score weight 

45,585 
6,156 
51,741 

As mentioned in the descriptive statistics section, most farm households have no credit 
and only a small percentage have the different types of credit. After using PSM, the number of 
households with credit decreases because some of them are not on common support, and this 
applies to the four sets of comparisons (Table 23).  

4.6.3. Regressions on the matched samples  

This section presents and discusses the empirical findings. Before proceeding to the 

regression, it is worth seeing the mean values of the outcome variables by group of households 

 130 

for (1) the whole sample before matching and (2) the common support observations with 
propensity score weights, which are the ones our regressions rely on. The figures for the 
sample before matching are reported in Table 24, and those after matching are in Table 25. 

Table 24: Average modern input adoption rates by group of households (before matching) 
Control 

 

Any source 
of credit 

HYV 
Fertilizer 
PHF 
Package of 
input 
Source: Calculation based on CAC 2013 

0.32 
0.85 
0.51 
0.20 

Formal 
credit 
0.32 
0.82 
0.46 
0.18 

Informal 
credit 
0.29 
0.84 
0.48 
0.19 

Both formal and 
informal credit 

0.34 
0.89 
0.61 
0.24 

0.24 
0.75 
0.37 
0.11 

 

Table 25: Average modern input adoption rates by group of households (the common support 
observations that have propensity score weights) 
Only formal 
credit versus 
control group 
T 
C 
0.25 
0.29 
0.76 
0.79 
0.41 
0.32 
0.12 
0.15 

Only informal 
credit versus 
control group 
T 
C 
0.24 
0.25 
0.75 
0.80 
0.41 
0.31 
0.11 
0.14 

control group 
T 
C 
0.25 
0.31 
0.75 
0.86 
0.48 
0.32 
0.12 
0.20 

Any source of 
credit versus 
control group 
T 
C 
0.26 
0.29 
0.77 
0.82 
0.44 
0.34 
0.13 
0.16 

 
HYV 
Fertilizer 
PHF 
Package 
of input 
Source: Calculation based on CAC 2013 

informal credit versus 

Both formal and 

The regression results of the impact of credit for agricultural activities on the outcome 
variables are summarized in Table 26. The details from each regression are in Table 34, 35, 36 
and 37 of the appendix. As the tables are long, we report only the results for the credit category 
in the main paper, as shown in Table 26. It is worth noting the objective of this study is to 
examine the impact of credit on the adoption of high-yield variety, fertilizer, pesticide or 
herbicide, so we only interpret and discuss the results of this linkage. The regressions are also 
weighted by the propensity weights obtained from PSM, as suggested by Khandker, Koolwal, & 
Samad (2010) so that the treatment and control households are comparable in terms of 

 131 

observable characteristics. The standard error of each regression is clustered at village level so as 
to get the robust standard error for making inference.  

The coefficients for probit regressions are reported in terms of marginal probability 

effects because it is convenient for interpretation. In other words, this approach is the common 
practice when it comes to the interpretation of the results from either a logit or probit model, 
because the coefficients imply a linear relationship between dependent and independent variables 
(Baum, 2006). That is, a parameter estimate is the expected probability that a unit change in 
variable X will result in in the outcome variable Y equaling 1, with other independent variables 
held fixed. For example, marginal effects of the probit model for fertilizer show changes in the 
probability of adopting fertilizer for additional unit change in independent variables. In each 
regression, we also include province dummies to control for the differences in characteristics 
across the 24 provinces.  

Table 26: Impacts of credit for agriculture 
 
Variables  

HYV adoption  Fertilizer adoption  PHF adoption  Package adoption 

Probit regression (marginal effects) 

Credit 
 
Observations 

Credit 
 
Observations 

HHs with any source of credit for agriculture versus Control Group 

0.0155** 
(0.0077) 
68,047 

0.0315*** 
(0.0056) 
68,047 
HHs with only formal credit for agriculture versus Control Group 

0.0748*** 
(0.0070) 
68,047 

0.0225*** 
(0.0055) 
68,047 

0.0271*** 
(0.0089) 
56,047 

0.0240*** 
(0.0063) 
56,047 
HHs with only informal credit for agriculture versus Control Group 

0.0651*** 
(0.0088) 
56,047 

0.0099 
(0.0070) 
56,047 

Credit  
 
Observations 

-0.0138 
(0.0104) 
51,431 

0.0145** 
(0.0071) 
51,431 
HHs with both types of credit for agriculture versus Control Group 

0.0524*** 
(0.0094) 
51,431 

0.0171** 
(0.0077) 
51,431 

Credit 
 
Observations 
Source: Calculation based on CAC 2013 
Notes: (1) Clustered robust standard errors in parentheses.  (2) *** p<0.01, ** p<0.05, * p<0.1 

0.0483*** 
(0.0085) 
51,741 

0.1078*** 
(0.0121) 
51,741 

0.0235* 
(0.0130) 
51,741 

0.0571*** 
(0.0104) 
51,741 

 132 

Our empirical results show that, on average, credit for agricultural activities increases the 

probability of adopting high-yield variety of rice by 1.5 percent, but it is only statistically 
significant at 10 percent level (Table 26). However, when looking at credit by source, it is only 
the formal credit and both types of credit that have a positive and significant effect on high-yield 
rice variety adoption, the effect of formal credit is more robust because it is at the 10 percent 
significance level. On average, formal credit and both types of credit increase the probability of 
adoption by almost 3 percent.  

The positive impact of credit on adoption of a high-yield rice variety seems to be in line 
with findings by Simtowe, Zeller, & Diagne (2009) that credit access increases the likelihood of 
adoption of improved seed varieties. They found that after correction for endogeneity, credit 
constraints have negative and significant effect on adoption of hybrid maize in Malawi. From our 
data, it is difficult to know the direction that causality goes, because our study draws on cross-
sectional data. What we can do is to justify the causality by hypothesizing that contracting 
agricultural credit reduces capital constraints on farm households, thus increasing the probability 
of adoption of modern inputs. This idea is conceptualized based on the conceptual framework 
from the previous studies, as we explained in that section of this chapter. In this regard, a cross-
sectional study by Abebaw & Haile (2013) has a similar way to justify the direction of causal 
relationship between participation in agricultural cooperative and adoption of modern inputs. For 
example, one of their findings shows that, on average, cooperative membership increases the 
probability of fertilizer adoption by around 10 percent. 

We acknowledge that it is worth looking into the cost that farm households spent on 

those modern inputs. However, our data has no information about this. From the key informant 
interviews and focus group discussions I conducted for a study about farmer organizations by 

 133 

Theng, et al. (2014), we found that most paddy input expenditure was for fertilizer and 
pesticides, followed by improved seed varieties. Hence, it implies that fertilizer and PHF requires 
more capital than the HYV seeds.  

Our empirical findings further reveal that credit for farm activities has a positive and 

significant impact on the adoption of fertilizer, but this does not hold true when households only 
contracted loans from the formal sources. On average, the credit from any source increases the 
probability of adoption of fertilizer by approximately 2 percent, and it is statistically significant 
at 1 percent level (Table 26). Also, the informal loans increase the fertilizer adoption rate by 
around 2 percent at 5 percent significance level. The impact of contracting loans from both 
formal and informal sources is the most robust, because it increases the probability of fertilizer 
adoption by around 5 percent at 1 percent significance level. This finding is consistent with the 
study by Tadesse (2014) which shows that access to credit has a positive and significant effect 
fertilizer use.  

For all sets of regressions, the impacts of credit on adoption of pesticide or herbicide or 

fungicide are positive and statistically significant at 1 percent level. Getting loans for farm 
activities from any source increases the probability of adopting pesticide or herbicide by around 
8 percent. A possible explanation is that farmers need capital to buy PHF and this type of input 
also accounts for a significant share of paddy input expenditure, so agricultural credit helps 

reduce financial constraints on farm households to get this type of modern inputs (Barslund & 
Tarp, 2008; Herath & Jayasuriya, 2010). The effects of formal credit and informal credit are 
comparable (around 6 percent). Getting credit from both formal and informal sources yields the 
most robust impact on adoption of pesticide or herbicide. It increases the probability of adoption 
by around 11 percent at the 1 percent significance level. This is a much bigger effect than on 

 134 

other inputs. A possible explanation that we can speculate on regarding why credit has a 
significantly higher effect PHF use than other modern inputs is that some farmers may have re-
used the improved variety seeds that they produced in the previous season, because they could 
not afford buy the seeds for all of their plots, as indicated in the second essay.   

Regarding the adoption of the combination of the three types of modern input, the 
adoption impact of credit is statistically significant and positive for all sets of comparison. 
Nonetheless, the impact of formal credit is only statistically significant at 1 percent level. On 
average, credit for farm activities increases probability of joint adoption of the three types of 
modern input by around 3 percent (Table 26). Using credit from both formal and informal 
sources tends to represent the strongest impact on the adoption of the three complementary 

inputs (6 percent). This finding tends to be consistent with a recent study by Roth et al. (2017), 

which empirically shows that multiple-source borrowing helps ease cash constraints, inducing 

credit-participating households to spend more on paddy inputs compared to non-borrowers. The 

study further reveals that most multiple borrowing households, whose first loan was from formal 

sources, contracted the second loan from the informal sources. This raises the question of why 
would a farmer go for both sources of credit if the formal credit is less expensive? A possible 
explanation is that after contracting a formal loan, a borrower may need more loan to cope with 
further farm expenditure. Then, it may not be easy for the borrower, who is already in debt, to 
get another loan from a formal institution such as MFI or bank, which has more demanding 
conditions. However, due to the lack of information in our data to reinforce this explanation, we 
leave it for further studies. 

The empirical evidence of our study shows that using credit from both formal and 

informal sources yields the strongest impact on adoption of modern inputs. At the same time, we 

 135 

should interpret our empirical results with caution, especially when it comes to informal loans. 
This is because informal loans from moneylenders are not promoted by the government because 
of their high interest rates. For instance, when farmers borrow from moneylenders they are 
charged with an interest rate of more than 6 percent per month, so it is costly for farmers (Lun, 
2013). On the other hand, access to credit from informal sources (i.e. friends or relatives) implies 
that a farmer has good social capital or a larger social network, which is can be a good predictor 
of technology adoption. As mentioned in section 4.3, a possible mechanism through which the 
government can promote informal credit for agriculture is the agricultural cooperative. 
Nonetheless, how to effectively continue promoting affordable credit for farmers through 
agricultural cooperatives is beyond the scope of this study. 

4.7. Conclusions 

In literature, contracting loans to support agricultural activities is hypothesized to 

encourage more farm investment including adoption of modern inputs for cash-constrained 
farmers. Meanwhile, the government of Cambodia considers credit essential for rural livelihood 
improvements. However, little is known about the causal relationship between credit and 
adoption of modern inputs especially in rice production, the major crop of Cambodia’s 
agriculture. More specifically, none of the previous studies in Cambodia focused on the credit 
access in the context of joint adoption of modern inputs which we introduced in the second 
essay. Thus, the main objective of this study is to measure the impact of credit on the probability 
of modern input adoption. Also, we examine whether farm households’ decisions to adopt 
modern inputs (high-yield variety, chemical fertilizer, pesticide and herbicide) are 
complementary. To this end, this study draws on the household survey from the Census of 
Agriculture of Cambodia (CAC) 2013. To test for complementarity of modern inputs, we use 

 136 

multivariate probit regressions. And, to assess the impact of credit, we apply PSM and 
regressions using the matched samples.  

Our results indicate that the high-yield rice variety, fertilizer and pesticide or herbicide or 

fungicide are all complements. Further, credit for agricultural activities has a statistically 
significant and positive impact on adoption of high-yield rice variety, fertilizer, pesticide (or 
herbicide/fungicide) and the combination of the three types of modern input. However, the 
impact on high-yield rice variety adoption is not robust. The impact of credit on adoption of 
pesticide (or herbicide/fungicide) is the strongest, particularly when farm households use both 
formal and informal loans. Also, credit yields a positive and statistically significant impact on the 
adoption of the combination of the three types of modern inputs. Overall, empirical evidence 
implies that credit helps overcome the budget constraint to use modern inputs. 

Therefore, our empirical findings suggest that government should continue expanding 
affordable credit for farm investment including loans for modern agricultural inputs. Promoting 
access to credit for agriculture, particularly rice production, can be done through both formal and 
informal credit systems. Formal credit access though bank and microfinance institutions is what 
the government has been promoting. One of the considerable mechanisms of promoting access to 
informal loans is through the agricultural cooperative that provides smaller loans to the members 
who have built savings for lending.  There are fewer restrictions to get these loans, so people 
would use them, especially for smaller loans (Theng, et al., 2014). However, how to improve 
capital savings for lending to the members of agricultural cooperatives is beyond the scope of 
this study. 

Though these empirical results are difficult to explain, our study empirically documents 
the effect of credit for agricultural activities and the probability of modern input adoption, which 

 137 

contributes to the limited body of literature on the linkage between credit and agricultural 
technology adoption in the Cambodian context. Some empirical studies came up with estimated 
results that are not easy to explain, and suggest that further studies be conducted to learn more 
about the pathways of their empirical findings. For instance, Abebaw & Haile (2013) find that 
cooperative membership has a positive and significant impact of fertilizer adoption, but the 
impact on improved seeds and pesticides is not statistically significant. Instead of giving some 
explanations for the estimation results, this study leaves this work for further research. Likewise, 
for our study, we expect that further research is needed to answer various interesting questions 
such as, “Why is formal credit significant for seed but not fertilizer, but informal credit is 
significant for fertilizer but not seed?” And, “Why does having both sources of credit make the 
most difference?” 

 

 

 

 

 

 

 

 

 

 

	
	

 138 

APPENDIX 

 139 

Appendix: Extra tables and figures the empirical analysis  
	
Table 27: Definitions of outcome variables and covariates 

Variable 

Treatment Variable 
agriculture_credit 

Formal_credit11 

Informal_credit 

both_types 

Outcome variables 
highyield_var 

fertilizer 
PHF 

package_input  

Definitions 

1 if HH contracted loans to support agricultural activities in the last 12 
months (from every source) 
 1 if HH contracted loans from only formal sources: bank or MFI for 
agricultural activities during the last 12 months 
1 if HH contracted loans from only informal sources: relative, friend, 
lender, cooperative or   self-help group for agricultural activities during 
the last 12 months 
1 if HH contracted loans from both types of credit at the same time for 
agricultural activities during the last 12 months 

1 if HH used high-yield variety in crop production during the last 12 
months 
1 if household used fertilizer in crop production during the 12 months 
1 if household used either pesticides, herbicides or fungicides in crop 
production during the 12 months 
1 if household used all the three types of input in crop production  
during the 12 months 

Covariates (household and village characteristics) 
irrigated_farm 
n_farm 
cult_area 
market_produce 
social_program 
power_tiller 
Tractor 
rice_miller 
water_pump 
harvester 
ext_govt  

percentage of rice farm with irrigation of a HH 
number of rice farms/plots of HH 
rice planted area per household in hectares 
1 if HH sold agricultural produce in the last 12 months 
1 if HH participated in social protection program: last 12 months 
1 if HH owned power tiller 
1 if HH owned tractor 
1 if HH owned rice miller 
1 if HH owned water pump 
1 if HH owned harvester 
1 HH received agricultural extension service from govt. during the last 
12 months 
1 HH received agricultural extension service from community-based 
organization during the last 12 months 
1 HH received agricultural extension service from NGOs during the last 
12 months 
 

 
 
																																																								
11 It is worth noting that the 3 sub-categories – formal credit only, informal credit only and both types of credit – are 
mutually exclusive. 

ext_CBO 

ext_NGO 

 140 

Table 27 (cont’d) 
info_radio 

1 HH received information on improved farm from radio in the last 12 
months 
1 HH received information on improved farm from TV in the last 12 
months 
1 HH received information on improved farm from neighbor/other 
farmers in the last 12 months 
farmer_org 
1 if one of HH members is a member of FO 
num_largelivestock  number of large livestock: cattle and buffalo 
num_smalllivestock  number of small livestock: pig/goat and so on. 
Female 
Age 
Educ 
Married 
Hhsize 
dep_ratio 
n_agriactivity 

1 if HH head is female 
age of HH head 
years of education of household head 
1 HH head is married 
household size 
dependency ratio: number of dependents/labor force 
number of HH members whose primary act. in the last 12 months is 
agriculture 
number of HH members whose primary act. in the last 12 months is in 
other industry 
number of HH members whose primary act. in the last 12 months is in 
services 
Distance from the village to the nearest national road (Km) 
1 if there is seasonal labor movement in the village 
1 if there is NGO in the village 
1 if a village experienced flood or draught in the last five years 
1 if there is health center in the village 
1 there is middle school in the village 
1 there is primary school in the village 
1 if soil type of the village is fine sandy 
1 if household located in plain region 
1 if household located in Tonlesap region 
1 if household located in mountainous or plateau region 

info_tv 

info_peer 

n_otherindustry 

n_service 

distance_road 
labor_move 
vil_NGO 
Disaster 
healthcenter 
vil_midschool 
vil_primschool  
soil_finesandy 
Plain 
Tonlesap 
Mountain_plateau 

 
 
 
 
 

 141 

Table 28: Descriptive statistics for households with agricultural credit 
Variables 
highyield_var 
 
fertilizer 
 
PHF 
 
package_input 
 
irrigated_farm 
 
n_farm 
 
cult_area 
 
market_produce 
 
social_program 
 
power_tiller 
 
Tractor 
 
rice_miller 
 
water_pump 
 
Harvester 
 
ext_govt   
 
ext_CBO 
 
ext_NGO 
 
info_radio 
 
info_tv 
 
info_peer 
 
farmer_org 
 

Control 
0.236 
(0.425) 
0.747 
(0.435) 
0.301 
(0.458) 
0.105 
(0.306) 
25.785 
(42.522) 
2.044 
(1.450) 
2.608 
(5.126) 
0.571 
(0.495) 
0.092 
(0.290) 
0.238 
(0.426) 
0.022 
(0.146) 
0.038 
(0.192) 
0.156 
(0.362) 
0.015 
(0.123) 
0.262 
(0.440) 
0.092 
(0.290) 
0.136 
(0.343) 
0.574 
(0.494) 
0.490 
(0.500) 
0.204 
(0.403) 
0.055 
(0.229) 

Agricultural credit 

0.319 
(0.466) 
0.848 
(0.359) 
0.505 
(0.500) 
0.199 
(0.399) 
33.505 
(46.217) 
2.032 
(1.486) 
3.510 
(5.517) 
0.721 
(0.448) 
0.244 
(0.429) 
0.345 
(0.475) 
0.040 
(0.196) 
0.042 
(0.201) 
0.215 
(0.411) 
0.023 
(0.150) 
0.435 
(0.496) 
0.140 
(0.347) 
0.177 
(0.382) 
0.703 
(0.457) 
0.614 
(0.487) 
0.263 
(0.440) 
0.142 
(0.349) 

Whole sample 

0.264 
(0.441) 
0.781 
(0.413) 
0.370 
(0.483) 
0.137 
(0.344) 
28.422 
(43.971) 
2.040 
(1.463) 
2.916 
(5.280) 
0.622 
(0.485) 
0.144 
(0.351) 
0.274 
(0.446) 
0.028 
(0.165) 
0.040 
(0.195) 
0.176 
(0.381) 
0.018 
(0.133) 
0.321 
(0.467) 
0.109 
(0.311) 
0.150 
(0.357) 
0.618 
(0.486) 
0.532 
(0.499) 
0.224 
(0.417) 
0.085 
(0.279) 

 142 

Table 28 (cont’d) 
(4.741) 
 
1.361 
num_smallivestock 
(4.741) 
 
0.175 
female 
(0.380) 
 
46.281 
age 
(12.649) 
 
4.879 
educ 
(3.672) 
 
0.859 
Married 
(0.348) 
 
4.550 
Hhsize 
(1.696) 
 
0.523 
dep_ratio 
(0.574) 
 
2.634 
n_agriactivity 
(1.424) 
 
0.188 
n_otherindustry 
(0.602) 
 
0.284 
n_service 
(0.755) 
 
9.645 
distance_road 
(14.563) 
 
0.762 
labor_move 
(0.426) 
 
0.293 
vil_NGO 
(0.455) 
 
0.691 
Disaster 
(0.462) 
 
0.171 
Healthcenter 
(0.376) 
 
0.180 
vil_midschool 
(0.385) 
 
0.651 
vil_primschool 
(0.477) 
 
0.526 
soil_finesandy 
(0.499) 
 
0.621 
plain 
(0.485) 
 
0.056 
Tonlesap 
(0.230) 
 
0.287 
Mountain_plateau 
 
(0.452) 
Source: Calculation based on CAC 2013.  Standard deviations in parentheses.	

(5.178) 
1.495 
(5.178) 
0.164 
(0.371) 
45.802 
(12.098) 
5.026 
(3.638) 
0.883 
(0.321) 
4.559 
(1.679) 
0.508 
(0.559) 
2.791 
(1.439) 
0.150 
(0.538) 
0.219 
(0.648) 
9.019 
(12.458) 
0.698 
(0.459) 
0.286 
(0.452) 
0.727 
(0.445) 
0.160 
(0.367) 
0.156 
(0.363) 
0.676 
(0.468) 
0.450 
(0.498) 
0.650 
(0.477) 
0.069 
(0.253) 
0.251 
(0.434) 

 143 

(4.895) 
1.407 
(4.895) 
0.171 
(0.377) 
46.118 
(12.465) 
4.929 
(3.661) 
0.867 
(0.339) 
4.553 
(1.690) 
0.518 
(0.569) 
2.687 
(1.431) 
0.175 
(0.582) 
0.262 
(0.721) 
9.431 
(13.883) 
0.740 
(0.439) 
0.291 
(0.454) 
0.703 
(0.457) 
0.167 
(0.373) 
0.172 
(0.378) 
0.660 
(0.474) 
0.500 
(0.500) 
0.631 
(0.483) 
0.060 
(0.238) 
0.275 
(0.446) 

Formal only 

Informal only 

Both types 

Table 29: Descriptive statistics for households by different category of agricultural credit 
Control 
Variables 
0.236 
highyield_var 
 
(0.425) 
0.747 
Fertilizer 
(0.435) 
 
0.301 
PHF 
 
(0.458) 
0.105 
package_input 
(0.306) 
 
irrigated_farm 
25.785 
(42.522) 
 
2.044 
n_farm 
(1.450) 
 
cult_area 
2.608 
(5.126) 
 
0.571 
market_produce 
 
(0.495) 
0.092 
social_program 
(0.290) 
 
0.238 
power_tiller 
 
(0.426) 
0.022 
Tractor 
(0.146) 
 
rice_miller 
0.038 
(0.192) 
 
0.156 
water_pump 
(0.362) 
 
Harvester 
0.015 
(0.123) 
 
0.262 
ext_govt   
 
(0.440) 
0.092 
ext_CBO 
(0.290) 
 
ext_NGO 
0.136 
(0.343) 
 
0.574 
info_radio 
(0.494) 
 
info_tv 
0.490 
(0.500) 
 
0.204 
info_peer 
 
(0.403) 
0.055 
farmer_org 
 
(0.229) 

0.338 
(0.473) 
0.895 
(0.306) 
0.607 
(0.488) 
0.238 
(0.426) 
37.733 
(47.896) 
1.933 
(1.419) 
3.571 
(5.569) 
0.793 
(0.405) 
0.392 
(0.488) 
0.362 
(0.481) 
0.033 
(0.179) 
0.052 
(0.223) 
0.211 
(0.408) 
0.023 
(0.149) 
0.547 
(0.498) 
0.233 
(0.423) 
0.217 
(0.412) 
0.797 
(0.402) 
0.710 
(0.454) 
0.320 
(0.467) 
0.225 
(0.417) 

0.294 
(0.456) 
0.842 
(0.365) 
0.478 
(0.500) 
0.189 
(0.392) 
26.032 
(42.819) 
2.109 
(1.499) 
3.649 
(4.988) 
0.696 
(0.460) 
0.200 
(0.400) 
0.386 
(0.487) 
0.038 
(0.191) 
0.034 
(0.182) 
0.166 
(0.372) 
0.021 
(0.142) 
0.372 
(0.483) 
0.121 
(0.327) 
0.151 
(0.358) 
0.679   
(0.467) 
0.583 
(0.493) 
0.292 
(0.455) 
0.099 
(0.298) 

0.322 
(0.467) 
0.823 
(0.382) 
0.461 
(0.498) 
0.182 
(0.386) 
35.194 
(46.562) 
2.048 
(1.514) 
3.396 
(5.761) 
0.693 
(0.461) 
0.181 
(0.385) 
0.311 
(0.463) 
0.045 
(0.208) 
0.040 
(0.197) 
0.245 
(0.430) 
0.025 
(0.155) 
0.404 
(0.491) 
0.096 
(0.294) 
0.168 
(0.374) 
0.662 
(0.473) 
0.574 
(0.494) 
0.214 
(0.410) 
0.117 
(0.321) 

 144 

Table 29 (cont’d) 
 
num_smallivestock 
 
female 
 
age 
 
educ 
 
Married 
 
Hhsize 
 
dep_ratio 
 
n_agriactivity 
 
n_otherindustry 
 
n_service 
 
distance_road 
 
labor_move 
 
vil_NGO 
 
Disaster 
 
Healthcenter 
 
vil_midschool 
 
vil_primschool 
 
soil_finesandy 
 
plain 
 
Tonlesap 
 
Mountain_plateau 
 
Source: Calculation based on CAC 2013.  Standard deviations in parentheses.  

(4.860) 
1.548 
(5.301) 
0.164 
(0.371) 
45.875        
(11.887)      
5.144 
(3.573) 
0.888 
(0.315) 
4.585 
(1.691) 
0.500 
(0.553) 
2.812 
(1.416) 
0.191 
(0.608) 
0.227 
(0.668) 
9.377 
(12.598) 
0.723 
(0.448) 
0.301 
(0.459) 
0.712 
(0.453) 
0.155 
(0.362) 
0.158 
(0.365) 
0.650 
(0.477) 
0.478 
(0.500) 
0.663 
(0.473) 
0.060 
(0.237) 
0.243 
(0.429) 

(5.725) 
1.783 
(5.802) 
0.169 
(0.375) 
46.100 
(12.180) 
5.223 
(3.807) 
0.883 
(0.322) 
4.471 
(1.659) 
0.482 
(0.545) 
2.826 
(1.462) 
0.117 
(0.486) 
0.236 
(0.673) 
8.868 
(12.174) 
0.697 
(0.460) 
0.271 
(0.445) 
0.724 
(0.447) 
0.173 
(0.379) 
0.160 
(0.366) 
0.693 
(0.461) 
0.437 
(0.496) 
0.601 
(0.490) 
0.072 
(0.259) 
0.300 
(0.458) 

(5.232) 
1.098 
(4.126) 
0.160 
(0.367) 
45.359 
(12.372) 
4.607 
(3.539) 
0.875 
(0.331) 
4.605 
(1.675) 
0.550 
(0.584) 
2.717 
(1.453) 
0.112 
(0.443) 
0.185 
(0.580) 
8.537 
(12.483) 
0.654 
(0.476) 
0.274 
(0.446) 
0.759 
(0.428) 
0.156 
(0.362) 
0.149 
(0.356) 
0.707 
(0.455) 
0.412 
(0.492) 
0.677 
(0.468) 
0.081 
(0.272) 
0.215 
(0.411) 

 145 

(4.741) 
1.361 
(4.741) 
0.175 
(0.380) 
46.281 
(12.649) 
4.879 
(3.672) 
0.859 
(0.348) 
4.550 
(1.696) 
0.523 
(0.574) 
2.634 
(1.424) 
0.188 
(0.602) 
0.284 
(0.755) 
9.645 
(14.563) 
0.762 
(0.426) 
0.293 
(0.455) 
0.691 
(0.462) 
0.171 
(0.376) 
0.180 
(0.385) 
0.651 
(0.477) 
0.526 
(0.499) 
0.621 
(0.485) 
0.056 
(0.230) 
0.287 
(0.452) 

Table 30: Test for matching quality (impact of credit from any sources) 
Variables 

Treated  Control  %bias  % reduc. 

Sample 

Unmatched 
irrigated_farm 
 
Matched  
Unmatched 
n_farm 
Matched  
 
Unmatched 
cult_area 
Matched 
 
Unmatched 
market_produce 
Matched 
 
Unmatched 
social_program 
Matched 
 
Unmatched 
power_tiller 
Matched 
 
Unmatched 
Tractor 
Matched 
 
Unmatched 
rice_miller 
Matched 
 
Unmatched 
water_pump 
Matched 
 
Unmatched 
Harvester 
Matched 
 
Unmatched 
ext_govt   
Matched 
 
Unmatched 
ext_CBO 
Matched 
 
Unmatched 
ext_NGO 
Matched 
 
Unmatched 
info_radio 
Matched 
 
Unmatched 
info_tv 
Matched 
 
Unmatched 
info_peer 
Matched 
 
Unmatched 
farmer_org 
 
Matched 
num_largelivestock  Unmatched 
 
Matched 
num_smallivestock  Unmatched 
Matched 
 
female 
Unmatched 
Matched 
 
age 
Unmatched 

33.50 
32.96 
2.032 
2.034 
3.510 
3.347 
0.721 
0.708 
0.244 
0.209 
0.345 
0.332 
0.041 
0.036 
0.042 
0.042 
0.215 
0.210 
0.023 
0.022 
0.435 
0.408 
0.140 
0.129 
0.177 
0.173 
0.703 
0.670 
0.614 
0.599 
0.263 
0.253 
0.142 
0.112 
3.167 
3.218 
1.496 
1.450 
0.164 
0.164 
45.802 

25.785 
33.069 
2.044 
2.042 
2.608 
3.457 
0.571 
0.7083 
0.093 
0.206 
0.238 
0.339 
0.022 
0.038 
0.038 
0.041 
0.156 
0.207 
0.015 
0.024 
0.262 
0.406 
0.092 
0.125 
0.136 
0.175 
0.574 
0.696 
0.490 
0.603 
0.204 
0.260 
0.055 
0.107 
4.15 
3.250 
1.361 
1.503 
0.175 
0.162 
46.281 

17.4 
-0.2 
-0.8 
-0.5 
16.9 
-2.1 
31.9 
0.0 
41.4 
1.0 
23.7 
-1.6 
10.6 
-1.6 
1.9 
0.7 
15.4 
0.8 
5.6 
-1.4 
37.0 
0.4 
14.8 
1.3 
11.2 
-0.5 
27.1 
-1.4 
25.0 
-0.8 
14.0 
-1.3 
29.3 
1.9 
-17.1 
-0.6 
2.7 
-1.1 
-2.7 
0.7 
-3.9 

 146 

|bias| 
 
98.6 
 
31.7 
 
87.9 
 
99.9 
 
97.7 
 
93.5 
 
85.3 
 
65.8 
 
94.8 
 
74.4 
 
98.8 
 
91.0 
 
95.8 
 
94.7 
 
96.7 
 
90.7 
 
93.5 
 
96.8 
 
60.9 
 
75.0 
 

T-stat 

21.98*** 
-0.25 
-0.99 
-0.57 
21.39*** 
-1.85 
39.18*** 
0.04 
54.98*** 
0.93 
30.13 
-1.57 
13.89*** 
-1.51 
2.45** 
0.70 
19.57*** 
0.80 
7.25*** 
-1.40 
47.00*** 
0.45 
19.06*** 
1.36 
14.18*** 
-0.48 
33.36*** 
-1.55 
31.06*** 
-0.88 
17.73*** 
-1.34 
39.07*** 
1.92* 
-20.82*** 
-0.67 
3.43*** 
-1.08 
-3.37*** 
0.73 
-4.79*** 

-1.0 
4.0 
0.6 
7.3 
0.2 
0.5 
-1.5 
-2.5 
0.1 
11.0 
-0.9 
-6.6 
0.2 
-9.3 
1.1 
-4.6 
-1.1 
-14.5 
1.5 
-1.7 
-0.2 
8.0 
0.3 
-2.8 
-1.2 
-6.5 
-0.6 
5.3 
-0.9 
-15.3 
0.8 
5.9 
1.2 
5.1 
-2.0 
-8.1 
-0.1 

74.8 
 
85.4 
 
97.2 
 
-199.1 
 
96.3 
 
92.1 
 
97.3 
 
88.2 
 
77.0 
 
89.8 
 
88.6 
 
96.6 
 
56.7 
 
91.3 
 
82.7 
 
94.7 
 
80.4 
 
61.1 
 
99.0 

-1.04 
5.03*** 
0.63 
8.94*** 
0.22 
0.64 
-1.64 
-3.14*** 
0.10 
13.75 
-0.91 
-8.07*** 
0.20 
-11.30*** 
1.26 
-5.63*** 
-1.16 
-18.35*** 
1.52 
-2.10** 
-0.20 
9.98*** 
0.29 
-3.50*** 
-1.29 
-8.02*** 
-0.61 
6.60*** 
-0.98 
-19.08*** 
0.86 
7.31*** 
1.23 
6.48*** 
-1.98** 
-10.07*** 
-0.09 

45.836 
5.026 
4.992 
0.883 
0.881 
4.559 
4.559 
0.508 
0.512 
2.791 
2.781 
0.150 
0.153 
0.219 
0.224 
9.019 
9.076 
0.698 
0.704 
0.286 
0.288 
0.727 
0.721 
0.160 
0.163 
0.156 
0.156 
0.677 
0.672 
0.450 
0.455 
0.650 
0.652 
0.069 
0.069 
0.251 
0.249 

Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 

Table 30 (cont’d) 
45.957 
 
4.879 
educ 
4.970 
 
0.859 
Married 
0.880 
 
4.550 
Hhsize 
4.585 
 
0.523 
dep_ratio 
0.511 
 
2.634 
n_agriactivity 
2.793 
 
0.188 
n_otherindustry 
0.152 
 
0.284 
n_service 
0.216 
 
9.645 
distance_road 
9.220 
 
0.762 
labor_move 
0.697 
 
0.293 
vil_NGO 
0.289 
 
0.691 
Disaster 
0.719 
 
0.171 
Healthcenter 
0.167 
 
0.181 
vil_midschool 
0.158 
 
0.651 
vil_primschool 
0.676 
 
0.526 
soil_finesandy 
0.451 
 
0.621 
plain 
0.647 
 
0.0562 
Tonlesap 
0.074 
 
0.287 
Mountain_plateau 
 
0.248 
Source: Calculation based on CAC 2013.  *** p<0.01, ** p<0.05, * p<0.1	
 
	
 
 
 
	

 147 

Table 31: Test for matching quality (impact of getting only formal credit) 
Variables 

Treated  Control  %bias  % reduc.  

Sample 

Unmatched 
irrigated_farm 
 
Matched  
Unmatched 
n_farm 
Matched  
 
Unmatched 
cult_area 
Matched 
 
Unmatched 
market_produce 
Matched 
 
Unmatched 
social_program 
Matched 
 
Unmatched 
power_tiller 
Matched 
 
Unmatched 
Tractor 
Matched 
 
Unmatched 
rice_miller 
Matched 
 
Unmatched 
water_pump 
Matched 
 
Unmatched 
Harvester 
Matched 
 
Unmatched 
ext_govt   
Matched 
 
Unmatched 
ext_CBO 
Matched 
 
Unmatched 
ext_NGO 
Matched 
 
Unmatched 
info_radio 
Matched 
 
Unmatched 
info_tv 
Matched 
 
Unmatched 
info_peer 
Matched 
 
Unmatched 
farmer_org 
 
Matched 
num_largelivestock  Unmatched 
 
Matched 
num_smallivestock  Unmatched 
 
Matched 
Unmatched 
female 
 
Matched 
Unmatched 
age 

35.194 
33.453 
2.048 
2.049 
3.396 
3.253 
0.693 
0.681 
0.181 
0.159 
0.311 
0.302 
0.045 
0.038 
0.040 
0.039 
0.245 
0.227 
0.025 
0.022 
0.404 
0.377 
0.096 
0.099 
0.168 
0.168 
0.662 
0.648 
0.574 
0.560 
0.214 
0.218 
0.117 
0.095 
2.997 
3.049 
1.548 
1.515 
0.165 
0.163 
45.875 

25.785 
33.849 
2.044 
2.045 
2.608 
3.179 
0.571 
0.685 
0.093 
0.164 
0.238 
0.306 
0.022 
0.038 
0.038 
0.040 
0.156 
0.231 
0.015 
0.022 
0.262 
0.380 
0.092 
0.101 
0.136 
0.167 
0.574 
0.654 
0.490 
0.568 
0.204 
0.223 
0.056 
0.090 
4.15 
3.105 
1.361 
1.524 
0.175 
0.160 
46.281 

21.1 
-0.9 
0.3 
0.3 
14.5 
1.3 
25.6 
-0.8 
26.0 
-1.6 
16.6 
-0.8 
13.1 
0.5 
1.1 
-0.5 
22.5 
-1.1 
6.7 
0.1 
30.6 
-0.8 
1.1 
-0.9 
8.8 
0.3 
18.1 
-1.3 
16.9 
-1.8 
2.3 
-1.2 
22.1 
2.0 
-20.7 
-1.0 
3.7 
-0.2 
-2.7 
0.8 
-3.3 

 148 

|bias| 
 
95.8 
 
-22.1 
 
90.7 
 
97.0 
 
94.0 
 
95.1 
 
96.3 
 
49.9 
 
94.9 
 
98.6 
 
97.5 
 
17.7 
 
97.2 
 
93.1 
 
89.6 
 
47.9 
 
90.8 
 
95.1 
 
94.7 
 
71.7 
 

T-stat 

20.45*** 
-0.62 
0.25 
0.23 
14.13*** 
0.96   
23.63*** 
-0.57 
26.90*** 
-1.04 
16.04*** 
-0.57 
13.86*** 
0.33 
1.01 
-0.38 
22.33*** 
-0.78 
6.80*** 
0.07 
29.80*** 
-0.54 
1.00 
-0.61 
8.52*** 
0.17 
16.79*** 
-0.92 
15.87*** 
-1.27 
2.22** 
-0.87 
23.24*** 
1.42 
-18.2*** 
-0.87 
3.63*** 
-0.13 
-2.53** 
0.56 
-3.06*** 

-0.3 
7.3 
-0.5 
8.9 
-1.4 
2.1 
-0.5 
-3.9 
0.8 
12.6 
-0.0 
0.5 
0.8 
-8.0 
0.0 
-2.0 
0.5 
-9.0 
0.5 
1.6 
0.8 
4.5 
1.1 
-4.2 
-0.3 
-5.9 
0.1 
-0.3 
0.9 
-9.6 
0.2 
8.7 
-0.3 
1.5 
0.2 
-10.0 
0.2 

90.6 
 
93.6 
 
83.8 
 
76.6 
 
80.2 
 
99.6 
 
-50.1 
 
99.9 
 
75.9 
 
94.2 
 
51.7 
 
76.1 
 
93.0 
 
97.6 
 
-166.7 
 
97.8 
 
96.2 
 
83.6 
 
97.5 

-0.23   
6.86*** 
-0.34 
8.10*** 
-1.09 
1.95* 
-0.35 
-3.66*** 
0.57 
11.82*** 
-0.03 
0.50 
0.56 
-7.27*** 
0.00 
-1.78* 
0.35 
-8.65*** 
0.37 
1.52 
0.56 
4.25*** 
0.79 
-3.90*** 
-0.22 
-5.51*** 
0.11 
-0.32 
0.65 
-9.01*** 
0.15 
8.10*** 
-0.24 
1.38 
0.17 
-9.29*** 
0.18 

45.912 
5.145 
5.079 
0.889 
0.886 
4.586 
4.585 
0.501 
0.506 
2.812 
2.778 
0.191 
0.194 
0.227 
0.235 
9.377 
9.479 
0.723 
0.730 
0.301 
0.30 
0.712 
0.708 
0.155 
0.157 
0.158 
0.158 
0.650 
0.651 
0.479 
0.482 
0.663 
0.659 
0.060 
0.061 
0.243 
0.247 

Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 

Table 31 (cont’d) 
45.951 
 
4.879 
educ 
5.096 
 
0.859 
Married 
0.859 
 
4.550 
Hhsize 
4.593 
 
0.523 
dep_ratio 
0.502 
 
2.634 
n_agriactivity 
2.778 
 
0.188 
n_otherindustry 
0.189 
 
0.284 
n_service 
0.235 
 
9.645 
distance_road 
9.415 
 
0.762 
labor_move 
0.728 
 
0.293 
vil_NGO 
0.296 
 
0.691 
Disaster 
0.703 
 
0.171 
Healthcenter 
0.158 
 
0.181 
vil_midschool 
0.157 
 
0.651 
vil_primschool 
0.647 
 
0.526 
soil_finesandy 
0.481 
 
0.621 
plain 
0.660 
 
0.056 
Tonlesap 
0.060 
 
0.287 
Mountain_plateau 
 
0.246 
Source: Calculation based on CAC 2013.  *** p<0.01, ** p<0.05, * p<0.1 
	
	
	
	
	

 149 

Table 32: Test for matching quality (impact of getting only informal credit) 
Variables 

Treated  Control  %bias  % reduc.  

Sample 

Unmatched 
irrigated_farm 
 
Matched  
Unmatched 
n_farm 
Matched  
 
Unmatched 
cult_area 
Matched 
 
Unmatched 
market_produce 
Matched 
 
Unmatched 
social_program 
Matched 
 
Unmatched 
power_tiller 
Matched 
 
Unmatched 
Tractor 
Matched 
 
Unmatched 
rice_miller 
Matched 
 
Unmatched 
water_pump 
Matched 
 
Unmatched 
Harvester 
Matched 
 
Unmatched 
ext_govt   
Matched 
 
Unmatched 
ext_CBO 
Matched 
 
Unmatched 
ext_NGO 
Matched 
 
Unmatched 
info_radio 
Matched 
 
Unmatched 
info_tv 
Matched 
 
Unmatched 
info_peer 
Matched 
 
Unmatched 
farmer_org 
 
Matched 
num_largelivestock  Unmatched 
 
Matched 
num_smallivestock  Unmatched 
Matched 
 
female 
Unmatched 
Matched 
 
age 
Unmatched 

26.032 
26.545 
2.109 
2.101 
3.650 
3.452 
0.696 
0.681 
0.200 
0.174 
0.386 
0.363 
0.038 
0.031 
0.034 
0.036 
0.166 
0.170 
0.021 
0.020 
0.372 
0.357 
0.121 
0.123 
0.151 
0.155 
0.679 
0.665 
0.583 
0.567 
0.292 
0.273 
0.099 
0.090 
3.181 
3.282 
1.098 
1.126 
0.160 
0.164 
45.359 

25.785 
26.044 
2.044 
2.095 
2.608 
3.455 
0.571 
0.677 
0.093 
0.174 
0.238 
0.361 
0.022 
0.032 
0.038 
0.035 
0.156 
0.169 
0.015 
0.019 
0.262 
0.354 
0.092 
0.116 
0.136 
0.150 
0.574 
0.660 
0.490 
0.568 
0.204 
0.271 
0.056 
0.089 
4.15 
3.269 
1.361 
1.144 
0.175 
0.163 
46.281 

0.6 
1.2 
4.4 
0.4 
20.6 
-0.1 
26.2 
0.7 
30.9 
-0.2 
32.3 
0.3 
9.6 
-0.5 
-2.1 
0.7 
2.9 
0.2 
3.9 
0.3 
23.8 
0.5 
9.4 
2.2 
4.2 
1.6 
21.7 
0.9 
18.6 
-0.2 
20.4 
0.5 
16.3 
0.2 
-16.9 
0.2 
-5.9 
-0.4 
-3.9 
0.2 
-7.4 

 150 

|bias| 
 
-102.5 
 
90.3 
 
99.7 
 
97.2 
 
99.5 
 
99.1 
 
95.2 
 
67.2 
 
92.7 
 
92.8 
 
97.8 
 
76.0 
 
62.7 
 
95.8 
 
98.8 
 
97.4 
 
98.9 
 
98.6 
 
92.9 
 
94.9 
 

T-stat 

0.43 
0.63 
3.31*** 
0.23 
15.01*** 
-0.03 
18.80*** 
0.41 
26.06*** 
-0.08 
25.13*** 
0.15 
7.91*** 
-0.24 
-1.53 
0.38 
2.19** 
0.11 
3.07*** 
0.15 
18.22*** 
0.28 
7.25*** 
1.16 
3.14*** 
0.82 
15.63*** 
0.50 
13.65*** 
-0.12 
15.80*** 
0.28 
13.40*** 
0.09 
-11.69*** 
0.14 
-4.14*** 
-0.25 
-2.86*** 
0.11 
-5.38*** 

45.392 
4.607 
4.615 
0.875 
0.871 
4.605 
4.586 
0.550 
0.547 
2.717 
2.720 
0.112 
0.117 
0.185 
0.193 
8.537 
8.674 
0.654 
0.665 
0.274 
0.279 
0.759 
0.754    
0.156 
0.160 
0.149 
0.154 
0.707 
0.699 
0.412 
0.427 
0.677 
0.670 
0.081 
0.081 
0.215 
0.220 

Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 

Table 32 (cont’d) 
-0.1 
45.401 
 
-7.5 
4.879 
educ 
0.0 
4.614 
 
4.7 
0.859 
Married 
-0.6 
0.873 
 
3.2 
4.550 
Hhsize 
-0.2 
4.589 
 
0.523 
4.8 
dep_ratio 
0.546        0.1 
 
2.634 
5.8 
n_agriactivity 
0.4 
2.714 
 
-14.3 
0.188 
n_otherindustry 
0.1 
0.116 
 
0.284 
-14.6 
n_service 
0.1 
0.193 
 
-8.2 
9.645 
distance_road 
8.839 
-1.2 
 
-23.9 
0.762 
labor_move 
-0.0 
0.665 
 
-4.3 
0.293 
vil_NGO 
0.282 
-0.6 
 
15.4 
0.691 
Disaster 
1.0 
0.749 
 
0.171 
-4.1 
Healthcenter 
2.2 
0.151 
 
-8.5 
0.181 
vil_midschool 
1.1 
0.150 
 
0.651 
11.9 
vil_primschool 
0.5 
0.697 
 
-23.0 
0.526 
soil_finesandy 
0.427 
0.1 
 
11.7 
0.621 
plain 
-0.6 
0.673 
 
0.056 
9.7 
Tonlesap 
0.8 
0.079 
 
-16.8 
0.287 
Mountain_plateau 
 
0.219 
0.2 
Source: Calculation based on CAC 2013.  *** p<0.01, ** p<0.05, * p<0.1 
	
	
	
	
	

 151 

99.0 
 
99.6 
 
86.5 
 
95.0 
 
97.2 
 
93.0 
 
99.1 
 
99.7 
 
85.1 
 
99.9 
 
85.4 
 
93.7 
 
45.8 
 
87.6 
 
96.2 
 
99.5 
 
94.5 
 
91.5 
 
98.6 

-0.04 
-5.48*** 
0.02 
3.39*** 
-0.35 
2.36** 
-0.09   
3.54*** 
0.07 
4.28*** 
0.22 
-9.52*** 
0.08 
-9.84*** 
0.03 
-5.69*** 
-0.69 
-18.39*** 
-0.02 
-3.12*** 
-0.34 
11.01*** 
0.54 
-2.99*** 
1.23 
-6.07*** 
0.59 
8.63*** 
0.25 
-16.83*** 
0.06 
8.46*** 
-0.35 
7.63*** 
0.42 
-11.95*** 
0.13 

Table 33: Test for matching quality (impact of getting both formal and informal credit) 
Variables 

Treated  Control  %bias  % reduc. 

Sample 

Unmatched 
irrigated_farm 
 
Matched  
Unmatched 
n_farm 
Matched  
 
Unmatched 
cult_area 
Matched 
 
Unmatched 
market_produce 
Matched 
 
Unmatched 
social_program 
Matched 
 
Unmatched 
power_tiller 
Matched 
 
Unmatched 
Tractor 
Matched 
 
Unmatched 
rice_miller 
Matched 
 
Unmatched 
water_pump 
Matched 
 
Unmatched 
Harvester 
Matched 
 
Unmatched 
ext_govt   
Matched 
 
Unmatched 
ext_CBO 
Matched 
 
Unmatched 
ext_NGO 
Matched 
 
Unmatched 
info_radio 
Matched 
 
Unmatched 
info_tv 
Matched 
 
Unmatched 
info_peer 
Matched 
 
Unmatched 
farmer_org 
 
Matched 
num_largelivestock  Unmatched 
 
Matched 
Unmatched 
num_smallivestock 
 
Matched 
Unmatched 
female 
 
Matched 
Unmatched 
age 

37.733 
38.218 
1.933 
1.94 
3.571 
3.446 
0.793 
0.782 
0.392 
0.361 
0.362 
0.350 
0.033 
0.031 
0.052 
0.052 
0.211 
0.214 
0.023 
0.022 
0.547 
0.524 
0.233 
0.205 
0.217 
0.209 
0.797 
0.787 
0.710 
0.698 
0.320 
0.304 
0.225 
0.185 
3.444 
3.503 
1.783 
1.642 
0.169 
0.169 
46.100 

26.4 
25.785 
40.176 
-4.3 
-7.7 
2.044 
1.981 
-2.9 
18.0 
2.608 
-1.7 
3.536 
49.1 
0.571 
1.4 
0.776 
74.7 
0.092 
1.2 
0.355 
27.5 
0.238 
-1.5 
0.360 
7.0 
0.022 
-1.3 
0.034 
6.8 
0.038 
2.2 
0.047 
14.3 
0.156 
0.3 
0.213 
5.3 
0.015 
-1.5 
0.024 
60.7 
0.262 
1.3 
0.518 
38.7 
0.092 
-1.1 
0.208 
21.2 
0.136 
-3.3 
0.222 
49.4 
0.574 
-2.1 
0.796 
45.9 
0.490 
0.4 
0.696 
26.6 
0.204 
-1.3 
0.310 
50.3 
0.055 
0.184 
0.2 
4.150         -11.8 
3.498 
0.1 
8.0 
1.361 
0.0 
1.641 
0.175 
-1.6 
-0.6 
0.171 
46.281 
-1.5 

 152 

 |bias| 
 
83.6 
 
62.7 
 
90.7 
 
97.3 
 
98.4 
 
94.4 
 
81.0 
 
67.3 
 
97.7 
 
71.0 
 
97.8 
 
97.2 
 
84.5 
 
95.8 
 
99.1 
 
95.1 
 
99.5 
 
99.3 
 
99.8 
 
62.0 
 

T-stat 

20.82*** 
-2.26** 
-5.79*** 
-1.65* 
14.00*** 
-0.79   
34.52*** 
0.82 
70.29*** 
0.57 
21.71*** 
-0.81 
5.75*** 
-0.68 
5.46*** 
1.19 
11.32*** 
0.17 
4.34*** 
-0.78 
48.01 
0.70 
34.16*** 
-0.53 
17.15*** 
-1.68* 
34.65*** 
-1.28 
33.42*** 
0.23 
21.25*** 
-0.69 
49.05*** 
0.11 
-8.64*** 
0.05 
6.51*** 
0.01 
-1.17   
-0.33   
-1.08 

0.6 
9.2 
0.3 
7.0 
0.3 
-4.7 
0.6 
-7.3 
1.6 
13.3 
-1.7 
-12.9 
0.4 
-6.7 
1.7 
-5.8 
2.0 
-14.7 
2.8 
-4.9 
0.4 
7.2 
1.1 
0.7 
-3.2   
-5.6 
-1.2 
8.8 
0.5 
-17.8 
-0.2 
-4.1 
2.1 
6.6 
-5.0 
2.8 
0.9 

55.5 
 
97.2 
 
96.3 
 
87.4 
 
78.1 
 
87.5 
 
97.3 
 
73.9 
 
65.4 
 
80.7 
 
92.3 
 
84.7 
 
-366.2 
 
78.0 
 
94.8 
 
98.8 
 
49.7 
 
24.6 
 
67.2 

0.36 
7.04*** 
0.14 
5.13*** 
0.15 
-3.54*** 
0.34 
-5.39*** 
0.91 
10.13*** 
-0.91 
-9.04*** 
0.22 
-4.81*** 
1.03 
-4.09*** 
1.19   
-11.4*** 
1.52 
-3.66*** 
0.21 
5.38*** 
0.62 
0.51 
-1.72* 
-4.11*** 
-0.69 
6.57*** 
0.26 
-13.4*** 
-0.12 
-3.14***   
1.16 
5.22*** 
-2.52 
2.11** 
0.51 

46.11 
5.223 
5.155 
0.883 
0.881 
4.471 
4.474 
0.482 
0.484 
2.826 
2.818 
0.117 
0.118 
0.236 
0.237 
8.868 
8.922 
0.697 
0.699 
0.271 
0.272 
0.724 
0.717 
0.173 
0.176 
0.160 
0.159 
0.693 
0.690 
0.437 
0.434 
0.601 
0.620 
0.072 
0.072 
0.300 
0.281 

Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 
Unmatched 
Matched 

Table 33 (cont’d) 
46.029 
 
4.879 
educ 
5.145 
 
0.859 
Married 
0.880 
 
4.550 
Hhsize 
4.464 
 
0.523 
dep_ratio 
0.475 
 
2.634 
n_agriactivity 
2.842 
 
0.188 
n_otherindustry 
0.116 
 
0.284 
n_service 
0.225 
 
9.645 
distance_road 
8.654 
 
0.762 
labor_move 
0.686 
 
0.293 
vil_NGO 
0.270 
 
0.691 
Disaster 
0.712 
 
0.171 
Healthcenter 
0.188 
 
0.180 
vil_midschool 
0.164 
 
0.651 
vil_primschool 
0.688 
 
0.526 
soil_finesandy 
0.435 
 
0.621 
plain 
0.609 
 
0.056 
Tonlesap 
0.085 
 
0.287 
Mountain_plateau 
 
0.277 
Source: Calculation based on CAC 2013.  *** p<0.01, ** p<0.05, * p<0.1	
	
 
 
 
 
	

 153 

Figure 6: Distribution of propensity scores for matching between treatment group (HHs with any 
category of agricultural credit) and control group 

Kernel density estimate

Kernel density estimate

4

3

y
t
i
s
n
e
D

2

1

0

0

.2
.8
psmatch2: Propensity Score

.6

.4

4

3

y
t
i
s
n
e
D

2

1

1

0

0

.2
.8
psmatch2: Propensity Score

.6

.4

1

Kernel density estimate
kdensity _pscore

Kernel density estimate
kdensity _pscore

kernel = epanechnikov, bandwidth = 0.0201

kernel = epanechnikov, bandwidth = 0.0183

Source: Calculation based on CAC 2013.   
	
Figure 7: Distribution of propensity scores for matching between Treatment Group (HHs with 
only formal credit for agriculture) and control group 

 

Kernel density estimate

Kernel density estimate

6

6

4

y
t
i
s
n
e
D

2

0

0

4

y
t
i
s
n
e
D

2

.8

0

0

.2

.4

psmatch2: Propensity Score

.6

.2

.4

.6

psmatch2: Propensity Score

Kernel density estimate
kdensity _pscore

Kernel density estimate
kdensity _pscore

kernel = epanechnikov, bandwidth = 0.0164

kernel = epanechnikov, bandwidth = 0.0139

Source: Calculation based on CAC 2013.  	
	

 154 

.8

	

Figure 8: Distribution of propensity scores for matching between Treatment Group (HHs with 
only informal credit for agriculture) and control group 

Kernel density estimate

Kernel density estimate

0
1

8

6

y
t
i
s
n
e
D

4

2

0

0

0
1

8

6

y
t
i
s
n
e
D

4

2

.8

0

0

.2

.4

psmatch2: Propensity Score

.6

.2

psmatch2: Propensity Score

.4

.6

.8

Kernel density estimate
kdensity _pscore

Kernel density estimate
kdensity _pscore

kernel = epanechnikov, bandwidth = 0.0162

kernel = epanechnikov, bandwidth = 0.0135

Source: Calculation based on CAC 2013.   
	
Figure 9: Distribution of propensity scores for matching between Treatment Group (HHs with 
both types of credit for agriculture) and control group 

 

Kernel density estimate

Kernel density estimate

0
1

8

6

y
t
i
s
n
e
D

4

2

0

0

.2
.8
psmatch2: Propensity Score

.6

.4

0
1

8

6

y
t
i
s
n
e
D

4

2

0

1

0

.2
.8
psmatch2: Propensity Score

.4

.6

Kernel density estimate
kdensity _pscore

Kernel density estimate
kdensity _pscore

kernel = epanechnikov, bandwidth = 0.0297

kernel = epanechnikov, bandwidth = 0.0263

Source: Calculation based on CAC 2013.   

 155 

1

 

Probit regression12 (Marginal effects) 

HYV adoption 
0.0155** 
(0.0077) 
0.0007*** 
(0.0001) 
0.0143*** 
(0.0025) 
0.0030*** 
(0.0007) 
0.0606*** 
(0.0094) 
0.1172*** 
(0.0134) 
0.0035 
(0.0082) 
0.0288* 
(0.0170) 
0.0436** 
(0.0183) 
0.0222** 
(0.0098) 
0.0399*** 
(0.0152) 
0.0663*** 
(0.0097) 
0.0271** 
(0.0132) 
0.0049 
(0.0110) 
-0.0755*** 
(0.0108) 
0.0382*** 
(0.0103) 
0.0521*** 
(0.0103) 
0.0410*** 
(0.0136) 

Table 34: Impact of credit for agriculture (from any sources) 
 
Variables 
treated 
 
irrigated_farm 
 
n_farm 
 
cult_area 
 
market_produce 
 
social_program 
 
power_tiller 
 
tractor 
 
rice_miller 
 
water_pump 
 
harvester 
 
ext_govt 
 
ext_CBO 
 
ext_NGO 
 
info_radio 
 
info_tv 
 
info_peer 
 
farmer_org 
 
num_largelivestock  0.0006 
 
(0.0006) 
num_smalllivestock  -0.0001 
																																																								
12 This regression is based on the sample of common support observations having propensity score weights. This is 
also the case for Table 35, 36, and 37 

Pest adoption  Package adoption 
0.0748*** 
(0.0070) 
0.0018*** 
(0.0001) 
0.0081*** 
(0.0021) 
0.0028*** 
(0.0006) 
0.0264*** 
(0.0084) 
0.1137*** 
(0.0116) 
0.0186** 
(0.0076) 
0.0247* 
(0.0148) 
-0.0782*** 
(0.0157) 
0.0977*** 
(0.0086) 
0.0346** 
(0.0162) 
0.0344*** 
(0.0091) 
0.1100*** 
(0.0119) 
-0.0140 
(0.0099) 
0.0066 
(0.0108) 
0.0567*** 
(0.0098) 
0.1001*** 
(0.0091) 
-0.0323*** 
(0.0122) 
0.0007 
(0.0006) 
0.0009** 

0.0315*** 
(0.0056) 
0.0009*** 
(0.0001) 
0.0087*** 
(0.0017) 
0.0023*** 
(0.0005) 
0.0294*** 
(0.0072) 
0.0813*** 
(0.0089) 
0.0138** 
(0.0060) 
0.0389*** 
(0.0113) 
-0.0066 
(0.0126) 
0.0434*** 
(0.0065) 
0.0247** 
(0.0100) 
0.0373*** 
(0.0070) 
0.0440*** 
(0.0094) 
-0.0079 
(0.0077) 
-0.0417*** 
(0.0080) 
0.0306*** 
(0.0074) 
0.0420*** 
(0.0071) 
0.0166* 
(0.0094) 
0.0003 
(0.0004) 
-0.0002 

Fertilizer  
0.0225*** 
(0.0055) 
0.0006*** 
(0.0001) 
0.0157*** 
(0.0021) 
0.0001 
(0.0005) 
0.0382*** 
(0.0063) 
0.0373*** 
(0.0121) 
0.0310*** 
(0.0067) 
-0.0412*** 
(0.0116) 
-0.0331** 
(0.0136) 
0.0324*** 
(0.0074) 
0.0109 
(0.0153) 
0.0465*** 
(0.0071) 
0.0029 
(0.0108) 
0.0101 
(0.0082) 
-0.0133* 
(0.0070) 
0.0201*** 
(0.0070) 
0.0251*** 
(0.0076) 
0.0092 
(0.0094) 
-0.0014*** 
(0.0004) 
0.0003 

 156 

Table 34 (cont’d) 
 
female 
 
age 
 
educ 
 
married 
 
hhsize 
 
dep_ratio 
 
n_agriactivity 
 
n_otherindustry 
 
n_service 
 
distance_road 
 
labor_move 
 
vil_NGO 
 
disaster 
 
healthcenter 
 
vil_midschool 
 
vil_primschool 
 
soil_finesandy 
 
plain 
 
Tonlesap 
 
Mountain_plateau 
 
Observations 
Source: Calculation based on CAC 2013.   
Notes: (1) Clustered robust standard errors in parentheses. (2) *** p<0.01, ** p<0.05, * p<0.1 

(0.0004) 
-0.0040 
(0.0057) 
0.0003** 
(0.0001) 
0.0019*** 
(0.0007) 
0.0025 
(0.0059) 
-0.0044*** 
(0.0016) 
-0.0011 
(0.0032) 
0.0021 
(0.0022) 
0.0158*** 
(0.0043) 
-0.0011 
(0.0036) 
-0.0007** 
(0.0003) 
-0.0230*** 
(0.0089) 
-0.0007 
(0.0075) 
0.0019 
(0.0079) 
0.0061 
(0.0101) 
-0.0056 
(0.0105) 
0.0011 
(0.0076) 
-0.0136* 
(0.0074) 
0.0008 
(0.0165) 
0.0151 
(0.0219) 
-0.0487*** 
(0.0172) 
68,047 

(0.0005) 
0.0021 
(0.0081) 
0.0006*** 
(0.0002) 
0.0019** 
(0.0009) 
0.0043 
(0.0081) 
0.0082*** 
(0.0022) 
-0.0192*** 
(0.0044) 
-0.0130*** 
(0.0031) 
-0.0317*** 
(0.0059) 
-0.0275*** 
(0.0048) 
0.0002 
(0.0004) 
-0.0209* 
(0.0125) 
0.0060 
(0.0112) 
0.0331*** 
(0.0116) 
0.0140 
(0.0149) 
-0.0159 
(0.0153) 
-0.0132 
(0.0111) 
0.0153 
(0.0107) 
0.0219 
(0.0321) 
0.0246 
(0.0370) 
-0.0385 
(0.0340) 
68,047 

(0.0004) 
-0.0085 
(0.0072) 
0.0006*** 
(0.0002) 
0.0038*** 
(0.0008) 
0.0046 
(0.0075) 
-0.0028 
(0.0022) 
-0.0048 
(0.0041) 
0.0009 
(0.0029) 
-0.0009 
(0.0051) 
0.0005 
(0.0046) 
0.0006 
(0.0003) 
-0.0105 
(0.0110) 
-0.0069 
(0.0099) 
0.0582*** 
(0.0100) 
0.0030 
(0.0126) 
-0.0063 
(0.0137) 
0.0128 
(0.0100) 
-0.0339*** 
(0.0096) 
0.0528** 
(0.0251) 
0.0739** 
(0.0306) 
0.0175 
(0.0270) 
68,047 

(0.0003) 
-0.0175*** 
(0.0060) 
0.0001 
(0.0002) 
0.0010 
(0.0006) 
-0.0088 
(0.0060) 
0.0020 
(0.0016) 
-0.0080** 
(0.0033) 
-0.0035 
(0.0023) 
-0.0062 
(0.0041) 
-0.0052 
(0.0033) 
0.0003 
(0.0003) 
-0.0316*** 
(0.0088) 
-0.0095 
(0.0081) 
0.0437*** 
(0.0084) 
0.0113 
(0.0105) 
-0.0110 
(0.0114) 
-0.0032 
(0.0085) 
-0.0147* 
(0.0077) 
0.0382 
(0.0233) 
0.0360 
(0.0269) 
-0.0223 
(0.0250) 
68,047 

 157 

Table 35: Impact of credit for agriculture (from only formal source) 

	

Probit regression (Marginal effects)	

Variables	

HYV adoption	
0.0271*** 
(0.0089) 
0.0007*** 
(0.0001) 
0.0153*** 
(0.0028) 
0.0034*** 
(0.0008) 
0.0928*** 
(0.0106) 
0.1301*** 
(0.0163) 
0.0049 
(0.0096) 
0.0127 
(0.0188) 
0.0281 
(0.0270) 
0.0099 
(0.0110) 
0.0324 
(0.0199) 
0.0606*** 
(0.0111) 
0.0364** 
(0.0168) 
-0.0185 
(0.0121) 
-0.0811*** 
(0.0127) 
0.0245** 
(0.0120) 
0.0690*** 
(0.0119) 
0.0226 
(0.0148) 

treated 
 
irrigated_farm 
 
n_farm 
 
cult_area 
 
market_produce 
 
social_program 
 
power_tiller 
 
tractor 
 
rice_miller 
 
water_pump 
 
harvester 
 
ext_govt 
 
ext_CBO 
 
ext_NGO 
 
info_radio 
 
info_tv 
 
info_peer 
 
farmer_org 
 
num_largelivestock  0.0003 
 
(0.0008) 
num_smalllivestock  -0.0003 
(0.0006) 
 
female 
-0.0092 

Fertilizer 	 Pest adoption	 Package adoption	
0.0099 
(0.0070) 
0.0007*** 
(0.0001) 
0.0160*** 
(0.0027) 
0.0004 
(0.0006) 
0.0371*** 
(0.0077) 
0.0185 
(0.0138) 
0.0372*** 
(0.0084) 
-0.0400*** 
(0.0146) 
-0.0547*** 
(0.0205) 
0.0303*** 
(0.0086) 
0.0126 
(0.0188) 
0.0424*** 
(0.0086) 
0.0129 
(0.0139) 
0.0153 
(0.0100) 
-0.0115 
(0.0092) 
0.0267*** 
(0.0091) 
0.0116 
(0.0098) 
0.0094 
(0.0111) 
-0.0024*** 
(0.0006) 
0.0009* 
(0.0005) 
-0.0084 

0.0651*** 
(0.0088) 
0.0019*** 
(0.0001) 
0.0096*** 
(0.0026) 
0.0034*** 
(0.0008) 
0.0214** 
(0.0097) 
0.0663*** 
(0.0145) 
0.0204** 
(0.0094) 
0.0178 
(0.0174) 
-0.1035*** 
(0.0234) 
0.1053*** 
(0.0098) 
0.0313 
(0.0201) 
0.0214** 
(0.0107) 
0.1294*** 
(0.0163) 
0.0091 
(0.0115) 
-0.0137 
(0.0144) 
0.0609*** 
(0.0132) 
0.0862*** 
(0.0115) 
-0.0358*** 
(0.0135) 
0.0001 
(0.0008) 
0.0011** 
(0.0005) 
-0.0118 

0.0240*** 
(0.0063) 
0.0010*** 
(0.0001) 
0.0098*** 
(0.0018) 
0.0026*** 
(0.0004) 
0.0395*** 
(0.0078) 
0.0598*** 
(0.0102) 
0.0055 
(0.0069) 
0.0319*** 
(0.0122) 
-0.0209 
(0.0164) 
0.0421*** 
(0.0070) 
0.0153 
(0.0132) 
0.0233*** 
(0.0079) 
0.0510*** 
(0.0126) 
-0.0127 
(0.0085) 
-0.0438*** 
(0.0097) 
0.0247*** 
(0.0088) 
0.0471*** 
(0.0079) 
0.0173* 
(0.0095) 
-0.0003 
(0.0006) 
0.0000 
(0.0004) 
-0.0302*** 

 158 

Table 35 (cont’d) 
 
age 
 
educ 
 
married 
 
hhsize 
 
dep_ratio 
 
n_agriactivity 
 
n_otherindustry 
 
n_service 
 
distance_road 
 
labor_move 
 
vil_NGO 
 
disaster 
 
healthcenter 
 
vil_midschool 
 
vil_primschool 
 
soil_finesandy 
 
plain 
 
Tonlesap 
 
Mountain_plateau 
 
Observations 
Source: Calculation based on CAC 2013.   
Notes: (1) Clustered robust standard errors in parentheses. (2) *** p<0.01, ** p<0.05, * p<0.1 
 
 

(0.0092) 
0.0001 
(0.0002) 
0.0043*** 
(0.0010) 
-0.0033 
(0.0099) 
-0.0085*** 
(0.0027) 
-0.0020 
(0.0055) 
0.0087** 
(0.0035) 
0.0073 
(0.0059) 
0.0084 
(0.0053) 
0.0014*** 
(0.0004) 
-0.0108 
(0.0133) 
-0.0017 
(0.0108) 
0.0489*** 
(0.0119) 
-0.0035 
(0.0144) 
-0.0073 
(0.0156) 
0.0275** 
(0.0117) 
-0.0322*** 
(0.0111) 
0.0578* 
(0.0330) 
0.0730* 
(0.0403) 
0.0092 
(0.0352) 
56,047 

(0.0078) 
0.0001 
(0.0002) 
0.0012 
(0.0008) 
-0.0010 
(0.0085) 
-0.0070*** 
(0.0021) 
0.0026 
(0.0045) 
0.0051* 
(0.0027) 
0.0197*** 
(0.0050) 
0.0046 
(0.0046) 
-0.0001 
(0.0003) 
-0.0150 
(0.0107) 
-0.0059 
(0.0090) 
0.0028 
(0.0096) 
0.0029 
(0.0122) 
-0.0013 
(0.0129) 
0.0111 
(0.0089) 
-0.0141 
(0.0088) 
0.0192 
(0.0193) 
0.0205 
(0.0270) 
-0.0442** 
(0.0203) 
56,047 

(0.0067) 
0.0001 
(0.0002) 
0.0011 
(0.0007) 
-0.0199*** 
(0.0076) 
-0.0017 
(0.0020) 
-0.0047 
(0.0041) 
0.0039 
(0.0025) 
-0.0020 
(0.0043) 
0.0009 
(0.0035) 
0.0001 
(0.0003) 
-0.0350*** 
(0.0108) 
0.0025 
(0.0082) 
0.0443*** 
(0.0094) 
0.0029 
(0.0113) 
-0.0151 
(0.0118) 
0.0090 
(0.0090) 
-0.0166** 
(0.0081) 
0.0107 
(0.0257) 
-0.0193 
(0.0295) 
-0.0419 
(0.0284) 
56,047 

 

 

(0.0098) 
0.0008*** 
(0.0003) 
0.0027** 
(0.0011) 
-0.0105 
(0.0105) 
0.0060** 
(0.0027) 
-0.0155*** 
(0.0055) 
-0.0065* 
(0.0036) 
-0.0280*** 
(0.0066) 
-0.0259*** 
(0.0055) 
-0.0002 
(0.0004) 
-0.0205 
(0.0142) 
0.0177 
(0.0122) 
0.0220* 
(0.0131) 
0.0071 
(0.0165) 
-0.0181 
(0.0168) 
-0.0032 
(0.0122) 
0.0195* 
(0.0117) 
-0.0294 
(0.0328) 
-0.0421 
(0.0391) 
-0.0665* 
(0.0351) 
56,047 

 159 

Table 36: Impact of credit for agriculture (from only informal source) 

 

Probit regression (Marginal effects) 

Variables 

HYV adoption 
-0.0138 
(0.0104) 
0.0006*** 
(0.0001) 
0.0101*** 
(0.0033) 
0.0039*** 
(0.0008) 
0.0260** 
(0.0111) 
0.1206*** 
(0.0166) 
0.0032 
(0.0103) 
0.0289 
(0.0234) 
0.0035 
(0.0182) 
0.0532*** 
(0.0147) 
0.0070 
(0.0196) 
0.0828*** 
(0.0123) 
0.0124 
(0.0164) 
0.0047 
(0.0144) 
-0.0421*** 
(0.0144) 
0.0533*** 
(0.0140) 
0.0291** 
(0.0122) 
0.0469*** 
(0.0163) 

treated 
 
irrigated_farm 
 
n_farm 
 
cult_area 
 
market_produce 
 
social_program 
 
power_tiller 
 
tractor 
 
rice_miller 
 
water_pump 
 
harvester 
 
ext_govt 
 
ext_CBO 
 
ext_NGO 
 
info_radio 
 
info_tv 
 
info_peer 
 
farmer_org 
 
num_largelivestock  0.0011 
 
num_smalllivestock  0.0011 
 
female 

(0.0012) 
-0.0042 

(0.0009) 

0.0524*** 
(0.0094) 
0.0020*** 
(0.0001) 
0.0074** 
(0.0029) 
0.0034*** 
(0.0008) 
0.0432*** 
(0.0105) 
0.0732*** 
(0.0152) 
0.0167* 
(0.0099) 

Fertilizer  Pest adoption  Package adoption 
0.0171** 
(0.0077) 
0.0005*** 
(0.0001) 
0.0213*** 
(0.0034) 
-0.0004 
(0.0006) 
0.0323*** 
(0.0076) 
0.0446*** 
(0.0135) 
0.0338*** 
(0.0080) 
-0.0471***  0.0019 
(0.0155) 
-0.0292** 
(0.0132) 
0.0238* 
(0.0139) 
0.0245 
(0.0185) 
0.0630*** 
(0.0092) 
-0.0135 
(0.0146) 
0.0117 
(0.0110) 
-0.0150* 
(0.0091) 
0.0069 
(0.0093) 
0.0423*** 
(0.0091) 
0.0168 
(0.0134) 
-0.0018***  0.0009 
(0.0007) 
-0.0005 
(0.0007) 
-0.0082 

0.0145** 
(0.0071) 
0.0010*** 
(0.0001) 
0.0090*** 
(0.0021) 
0.0027*** 
(0.0005) 
0.0110 
(0.0080) 
0.0887*** 
(0.0107) 
0.0191*** 
(0.0073) 
0.0235 
(0.0157) 
-0.0254* 
(0.0138) 
0.0477*** 
(0.0087) 
0.0027 
(0.0127) 
0.0503*** 
(0.0083) 
0.0233** 
(0.0108) 
-0.0035 
(0.0092) 
-0.0257** 
(0.0109) 
0.0374*** 
(0.0097) 
0.0252*** 
(0.0080) 
0.0041 
(0.0108) 
-0.0002 
(0.0006) 
-0.0009 
(0.0006) 
-0.0178** 

(0.0200) 
-0.0626*** 
(0.0210) 
0.0848*** 
(0.0121) 
0.0330 
(0.0202) 
0.0418*** 
(0.0112) 
0.0583*** 
(0.0155) 
-0.0201 
(0.0133) 
0.0242* 
(0.0133) 
0.0761*** 
(0.0125) 
0.0935*** 
(0.0110) 
-0.0213 
(0.0170) 

(0.0008) 
0.0004 
(0.0010) 
0.0008 

 160 

Table 36 (cont’d) 
 
age 
 
educ 
 
married 
 
hhsize 
 
dep_ratio 
 
n_agriactivity 
 
n_otherindustry 
 
n_service 
 
distance_road 
 
labor_move 
 
vil_NGO 
 
disaster 
 
healthcenter 
 
vil_midschool 
 
vil_primschool 
 
soil_finesandy 
 
plain 
 
Tonlesap 
 
Mountain_plateau 
 
Observations 
Source: Calculation based on CAC 2013.   
Notes: (1) Clustered robust standard errors in parentheses. (2) *** p<0.01, ** p<0.05, * p<0.1 
	
	

(0.0106) 
0.0006* 
(0.0003) 
0.0012 
(0.0011) 
0.0143 
(0.0111) 
-0.0005 
(0.0029) 
-0.0110* 
(0.0063) 
-0.0008 
(0.0039) 
-0.0160** 
(0.0079) 
-0.0090 
(0.0073) 
-0.0000 
(0.0004) 
0.0121 
(0.0133) 
-0.0103 
(0.0125) 
0.0559*** 
(0.0128) 
0.0120 
(0.0150) 
-0.0183 
(0.0162) 
-0.0022 
(0.0121) 
-0.0475*** 
(0.0116) 
0.0680** 
(0.0298) 
0.0863** 
(0.0350) 
0.0003 
(0.0322) 
51,431 

(0.0083) 
0.0005** 
(0.0002) 
0.0002 
(0.0009) 
0.0068 
(0.0095) 
-0.0005 
(0.0024) 
-0.0018 
(0.0047) 
-0.0020 
(0.0030) 
0.0301*** 
(0.0067) 
-0.0131*** 
(0.0048) 
-0.0014*** 
(0.0003) 
-0.0251** 
(0.0101) 
0.0027 
(0.0092) 
-0.0049 
(0.0094) 
0.0043 
(0.0121) 
0.0012 
(0.0119) 
0.0020 
(0.0094) 
-0.0174* 
(0.0097) 
-0.0643*** 
(0.0203) 
-0.0352 
(0.0256) 
-0.0908*** 
(0.0209) 
51,431 

(0.0079) 
-0.0002 
(0.0002) 
0.0004 
(0.0007) 
-0.0044 
(0.0087) 
0.0062*** 
(0.0021) 
-0.0166*** 
(0.0048) 
-0.0080*** 
(0.0028) 
-0.0145*** 
(0.0055) 
-0.0146*** 
(0.0043) 
0.0005* 
(0.0003) 
-0.0323*** 
(0.0089) 
-0.0175* 
(0.0097) 
0.0426*** 
(0.0104) 
0.0199 
(0.0121) 
-0.0198 
(0.0125) 
-0.0045 
(0.0092) 
-0.0283*** 
(0.0087) 
0.0744*** 
(0.0227) 
0.0677** 
(0.0269) 
0.0071 
(0.0243) 
51,431 

	

(0.0116) 
-0.0000 
(0.0003) 
0.0008 
(0.0012) 
0.0064 
(0.0128) 
0.0094*** 
(0.0031) 
-0.0213*** 
(0.0067) 
-0.0127*** 
(0.0041) 
-0.0388*** 
(0.0090) 
-0.0217*** 
(0.0070) 
0.0008* 
(0.0004) 
-0.0242* 
(0.0144) 
-0.0187 
(0.0136) 
0.0489*** 
(0.0143) 
0.0246 
(0.0181) 
-0.0203 
(0.0178) 
-0.0099 
(0.0133) 
0.0045 
(0.0133) 
0.0674* 
(0.0382) 
0.0672 
(0.0450) 
0.0080 
(0.0403) 
51,431 

 161 

Table 37: Impact of credit for agriculture (from both sources) 

Probit regression (marginal effects) 

(0.0126) 

(0.0151) 

(0.0120) 
0.0815*** 
(0.0241) 
-0.0708*** 
(0.0208) 

(0.0131) 
0.0326 
(0.0332) 

 
Variables 
treated 
 
irrigated_farm 
 
n_farm 
 
cult_area 
 
market_produce 
 
social_program 
 
power_tiller 
 
tractor 
 
rice_miller 
 
water_pump 
 
harvester 
 
ext_govt 
 
ext_CBO 
 
ext_NGO 
 
info_radio 
 
info_tv 
 
info_peer 
 
farmer_org 
 
num_largelivestock 
 
num_smalllivestock  0.0011 
 
female 

HYV adoption  Fertilizer   Pest adoption  Package adoption 
0.0235* 
(0.0130) 
0.0005*** 
(0.0002) 
0.0091* 
(0.0051) 
0.0029*** 
(0.0009) 
0.0083 
(0.0136) 
0.0922*** 
(0.0185) 
0.0112 
(0.0121) 
0.0112 
(0.0286) 
0.0195 
(0.0321) 
0.0104 
(0.0159) 
0.0644** 
(0.0315) 
0.0730*** 
(0.0131) 
0.0162 
(0.0177) 
0.0409** 
(0.0186) 
-0.0889*** 
(0.0157) 
0.0680*** 
(0.0144) 
-0.0194 
(0.0160) 
0.0847*** 
(0.0181) 
-0.0003 
(0.0010) 

0.0483***  0.1078*** 
(0.0085) 
0.0005***  0.0018*** 
(0.0001) 
(0.0001) 
0.0106***  0.0062* 
(0.0037) 
(0.0029) 
0.0018** 
0.0001 
(0.0005) 
(0.0009) 
0.0389***  0.0151 
(0.0087) 
0.0524***  0.1741*** 
(0.0176) 
0.0220***  0.0162 
(0.0085) 
0.0029 
(0.0211) 
-0.0342** 
(0.0143) 
0.0385***  0.0966*** 
(0.0127) 
-0.0271 
(0.0243) 
0.0407***  0.0480*** 
(0.0128) 
(0.0103) 
0.1091*** 
0.0015 
(0.0168) 
(0.0129) 
-0.0335** 
0.0081 
(0.0122) 
(0.0163) 
-0.0223**  0.0326** 
(0.0106) 
(0.0153) 
0.0251***  0.0352*** 
(0.0096) 
0.0152 
(0.0110) 
-0.0072 
(0.0120) 
0.0003 
(0.0004) 
0.0005 
(0.0006) 
0.0190** 

0.0571*** 
(0.0104) 
0.0006*** 
(0.0001) 
0.0021 
(0.0034) 
0.0027*** 
(0.0007) 
0.0009 
(0.0111) 
0.0944*** 
(0.0134) 
0.0279*** 
(0.0090) 
0.0507*** 
(0.0196) 
-0.0161 
(0.0262) 
0.0362*** 
(0.0113) 
0.0502** 
(0.0211) 
0.0538*** 
(0.0101) 
0.0413*** 
(0.0127) 
-0.0017 
(0.0120) 
-0.0413*** 
(0.0138) 
0.0504*** 
(0.0119) 
-0.0129 
(0.0114) 
0.0235* 
(0.0135) 
-0.0014* 
(0.0007) 
-0.0004 
(0.0007) 
-0.0070 

(0.0133) 
0.1012*** 
(0.0129) 
-0.0633*** 
(0.0168) 
-0.0010 
(0.0012) 
0.0001 
(0.0010) 
-0.0303** 

(0.0121) 

(0.0008) 
0.0379*** 

 162 

Table 37 (cont’d) 
 
age 
 
educ 
 
married 
 
hhsize 
 
dep_ratio 
 
n_agriactivity 
 
n_otherindustry 
 
n_service 
 
distance_road 
 
labor_move 
 
vil_NGO 
 
disaster 
 
healthcenter 
 
vil_midschool 
 
vil_primschool 
 
soil_finesandy 
 
plain 
 
Tonlesap 
 
Mountain_plateau 
 
Observations 
Source: Calculation based on CAC 2013.   
Notes: (1) Clustered robust standard errors in parentheses. (2) *** p<0.01, ** p<0.05, * p<0.1 
	

(0.0146) 
0.0008** 
(0.0004) 
0.0014 
(0.0015) 
0.0317** 
(0.0144) 
0.0133*** 
(0.0041) 
-0.0280*** 
(0.0087) 
-0.0229*** 
(0.0053) 
-0.0207* 
(0.0122) 
-0.0416*** 
(0.0082) 
0.0007 
(0.0005) 
-0.0130 
(0.0193) 
0.0138 
(0.0174) 
0.0451*** 
(0.0158) 
0.0354 
(0.0224) 
-0.0052 
(0.0229) 
-0.0314* 
(0.0188) 
0.0117 
(0.0159) 
0.1693*** 
(0.0480) 
0.1882*** 
(0.0527) 
0.0812 
(0.0495) 
51,741 

(0.0089) 
0.0006** 
(0.0003) 
0.0040*** 
(0.0011) 
0.0123 
(0.0105) 
-0.0045* 
(0.0026) 
-0.0048 
(0.0046) 
0.0030 
(0.0032) 
0.0089 
(0.0094) 
0.0017 
(0.0055) 
-0.0007** 
(0.0003) 
-0.0248*** 
(0.0096) 
-0.0071 
(0.0096) 
0.0021 
(0.0111) 
0.0076 
(0.0116) 
-0.0201 
(0.0126) 
-0.0194** 
(0.0097) 
-0.0015 
(0.0091) 
0.0008 
(0.0229) 
0.0118 
(0.0282) 
-0.0396* 
(0.0236) 
51,741 

(0.0154) 
0.0012*** 
(0.0004) 
0.0033** 
(0.0015) 
0.0026 
(0.0144) 
-0.0036 
(0.0044) 
-0.0037 
(0.0083) 
0.0020 
(0.0056) 
0.0142 
(0.0122) 
-0.0011 
(0.0086) 
0.0002 
(0.0005) 
-0.0230 
(0.0171) 
-0.0036 
(0.0173) 
0.0581*** 
(0.0142) 
-0.0253 
(0.0240) 
0.0097 
(0.0223) 
-0.0035 
(0.0176) 
-0.0115 
(0.0162) 
-0.0141 
(0.0319) 
0.0097 
(0.0386) 
-0.0157 
(0.0339) 
51,741 

(0.0136) 
0.0002 
(0.0003) 
0.0009 
(0.0012) 
-0.0033 
(0.0116) 
0.0007 
(0.0037) 
-0.0028 
(0.0075) 
-0.0009 
(0.0042) 
0.0121 
(0.0096) 
-0.0081 
(0.0066) 
0.0008** 
(0.0004) 
-0.0221 
(0.0144) 
-0.0061 
(0.0155) 
0.0411*** 
(0.0126) 
-0.0023 
(0.0193) 
0.0086 
(0.0213) 
-0.0154 
(0.0171) 
-0.0051 
(0.0133) 
0.1173*** 
(0.0419) 
0.1621*** 
(0.0450) 
0.0465 
(0.0433) 
51,741 

 

 163 

REFERENCES

 164 

REFERENCES 

 

Abebaw, D., & Haile, M. G. (2013). The impact of cooperatives on agricultural technology 

adoption: Empirical evidence from Ethiopia. Food Policy, 38(1), 82–91. 

Ali, A., & Abdulai, A. (2010). The Adoption of Genetically Modified Cotton and Poverty 

Reduction in Pakistan. Journal of Agricultural Economics, 61(1), 175–192. 

Barslund, M., & Tarp, F. (2008). Formal and Informal Rural Credit in Four Provinces of 

Vietnam. Journal of Development Studies, 44(4), 485–503. 

Baum, C. F. (2006). An Introduction to Modern Econometrics Using Stata (1 edition). College 

Station, Tex: Stata Press. 

Besley, T. (1998). How do market failures justify interventions in credit markets? In C. K. Eicher 

& J. M. Staatz (Eds.), International Agricultural Development (3 edition, pp. 370–389). 
Baltimore, MD: Johns Hopkins Univ. Press. 

Caliendo, M., & Kopeinig, S. (2008). Some practical guidance for the implementation of 

propensity score matching. Journal of Economic Surveys, 22(1), 31–72. 

Cappellari, L., & Jenkins, S. P. (2003). Multivariate probit regression using simulated maximum 

likelihood. The Stata Journal, 3(3), 278–294. 

CDRI -Cambodia Development Resource Institute. (2012). Understanding Poverty Dynamics : 

Evidence from Nine Villages in Cambodia (No. 69). Phnom Penh. Retrieved from 
http://www.cdri.org.kh/webdata/download/wp/wp69e.pdf 

Cerulli, G. (2015). Econometric Evaluation of Socio-Economic Programs: Theory and 

Applications. New York: Springer. 

Chhim, C., Theng, V., & Nou, K. (2013). Case Study: Rice Marketing Value Chains in Takeo 

Province. Cambodia Development Review, 17(2), 9–14. Retrieved from 
http://www.cdri.org.kh/webdata/cdr/2013/cdr13-2e.pdf 

Dorfman, J. H. (1996). Modeling Multiple Adoption Decisions in a Joint Framework. American 

Journal of Agricultural Economics, 78(3), 547–557. 

FAO-Food and Agriculture Organization. (2014). Cambodia country fact sheet on food and 

agriculture policy trends. Rome. 

Feder, G. (1982). Adoption of Interrelated Agricultural Innovations: Complementarity and the 

Impacts of Risk, Sclae and Credit. American Journal of Agricultural Economics, 64(1), 
255–298. 

Feder, G., Just, R. E., & Zilberman, D. (1985). Adoption of Agricultural Innovations in 

 165 

Developing Countries: A Survey. Economic Development and Cultural Change, 33(2), 
255–298. 

Feder, G., Lau, L. J., Lin, J. Y., & Luo, X. (1990). Credit Relationship Agriculture : Productivity 

The Disequilibrium. American Journal of Agricultural Economics, 72(5), 1151–1157. 

Feder, G., & Umali, D. L. (1993). The Adoption of Agricultural Innovations A Review. 

Technological Forcasting and Social Change, 43, 215–239. 

Griliches, Z. (1957). Hybrid Corn: An Exploration in the Economics of Technological Change. 

Econometrica, 25(4), 501–522. 

Heckman, J. J., Ichimura, H., & Todd, P. E. (1997). Matching Evidence Job As An Econometric 

Estimator: Evidence from Evaluating a Job Training Programme. Review of Economic 
Studies, 64(4), 605–654. 

Heckman, J. J., Ichimura, H., & Todd, P. E. (1998). Matching as an econometric evaluation 

estimator. The Review of Economic Studies, 65(2), 261–294. 

Herath, G., & Jayasuriya, S. (2010). Adoption of HYV Technology in Asian Countries : The 

Role of Concessionary Credit Revisited. Asian Survey, 36(12), 1184–1200. 

Kassie, M., Jaleta, M., Shiferaw, B., Mmbando, F., & Mekuria, M. (2013). Adoption of 

Interrelated Sustainable Agricultural Practices in Smallholder Systems : Evidence from 
Rural Tanzania. Technological Forcasting and Social Change, 80, 525–540. 

Khandker, S. R., Koolwal, G. B., & Samad, H. A. (2010). Handbook on Impact Evaluation: 

Quantitative Methods and Practices. Learning. World Bank Publications. Retrieved from 
https://openknowledge.worldbank.org/bitstream/handle/10986/2693/520990PUB0EPI1101
Official0Use0Only1.pdf?sequence=1 

Kuwornu, J. K. M., & Owus, E. S. (2012). Irrigation access and per capita consumption 

expenditure in farm households: Evidence from Ghana. Journal of Development and 
Agricultural Economics, 4(3), 78–92. 

Lun, P. (2013). The Role of Income Diversification during the Global Financial Crisis: Evidence 

from nine villages in Cambodia (No. 79). Phnom Penh. 

MAFF -Ministry of Agriculture Forestry and Fisheries. (2015). Annual Report 2014-2015. 

Phnom Penh. 

Mak, S. (2001). Continued innovation in a Cambodian rice-based farming system: Farmer testing 

and recombination of new elements. Agricultural Systems, 69(1–2), 137–149. 

Ngo, S., & Chan, S. (2010). Agriculture Sector Financing and Services for Smallholder Farmers. 

Phnom Penh. 

NIS-National Institute of Statistics. (2015). Census of Agriculture of the Kingdom of Cambodia 

 166 

2013. Phnom Penh: NIS. Retrieved from 
http://www.nis.gov.kh/nis/CAC2013/CAC_2013_Final_Report_En.pdf 

Pitt, M. M., & Khandker, S. R. (1998). The impact of Group - Based Credit Programs on Poor 
Households in Bangladesh : Does the Gender of Participants Matter? Journal of Political 
Economy, 106(5), 958–996. 

RGC-Royal Government of Cambodia. (2013). Rectangular Strategy-Phase III, 2014-2018. 

Phnom Penh. 

Roth, V., Araar, A., Sry, B., & Phann, D. (2017). The dynamics of microcredit borrowings in 

Cambodia (No. 17). London. 

Shakya, P. B., & Flinn, J. C. (1985). Adoption of Modern Varieties and Fertilizer. Journal of 

Agricultural Economics, 36(3), 409–419. 

Simtowe, F., Zeller, M., & Diagne, A. (2009). The impact of credit constraints on the adoption of 

hybrid maize in Malawi. Revue dEtudes En Agriculture et Environnement, 90(1), 5–22. 

Smale, M., Just, R. E., & Leathers, H. D. (1994). Land Allocation in HYV Adoption Models: An 

Investigation of Alternative Explanations. American Journal of Agricultural Economics, 
76(3), 535–546. 

Sok, S., Chap, S., & Chheang, V. (2011). Cambodia ’ s Agriculture : Challenges and Prospects 

(No. 37). Phnom Penh. 

Tadesse, M. (2014). Fertilizer adoption, credit access, and safety nets in rural Ethiopia. 

Agricultural Finance Review, 74(3), 290–310. 

Teklewold, H., Kassie, M., & Shiferaw, B. (2013). Adoption of Multiple Sustainable 

Agricultural Practices in Rural Ethiopia. Journal of Agricultural Economics, 64(3), 597–
623. 

Theng, V., & Kem, S. (2009). Rapid Assessment of Impacts of Global Economic Crisis on 

Cambodian Households: Effectsa nd Coping Strategies. Cambodia Development Review, 
13(3), 5–9. Retrieved from http://cdri.org.kh/webdata/cdr/2009/cdr09-3e.pdf 

Theng, V., Keo, S., Nou, K., Sum, S., & Khiev, P. (2014). Impact of Farmer Organisations on 

Food Security : The Case of Rural Cambodia (No. 95). Phnom Penh. Retrieved from 
http://www.cdri.org.kh/webdata/download/wp/wp95e.pdf 

Theng, V., Khiev, P., & Phon, D. (2014). Development of the Fertiliser Industry in Cambodia: 

Structure of the Market, Challenges in the Demand and Supply Sides, and the Way Forward 
(CDR Working Paper Series No. 91). Phnom Penh. Retrieved from 
http://www.cdri.org.kh/webdata/download/wp/wp91e.pdf 

Wang, H., Pandey, S., & Velarde, O. (2012). Pattern of Adoption of Improved Rice Varieties and 

Its Determinants in Cambodia. Procedia Economics and Finance, 2, 335–343. 

 167 

Wooldridge, J. M. (2008). Introductory Econometrics: A Modern Approach (4th ed.). Mason, 

OH: Cengage Learning. 

World Bank. (2014). Where Have All The Poor Gone ? Cambodia Poverty Assessment 2013. 

Washington, D.C. 

World Bank. (2015). Cambodian Agriculture in Transition: Opportunities and Risks. 

Washington, D.C. Retrieved from http://www-
wds.worldbank.org/external/default/WDSContentServer/WDSP/IB/2015/08/17/090224b083
07affe/1_0/Rendered/PDF/Cambodian0agri0ortunities0and0risks.pdf 

Yokouchi, T., & Saito, K. (2016). Factors affecting farmers’ adoption of NERICA upland rice 
varieties: the case of a seed producing village in central Benin. Food Security, 8(1), 197–
209. 

Yu, B., & Fan, S. (2011). Rice Production Response in Cambodia. Agricultural Economics, 

42(3), 437–450. 

 
	
	
	
	
	
	
	
	
	
	

 168