PROFIT EFFICIENCY AMONG RICE PRODUCERS IN EASTERN AND NORTHERN UGANDA

BY HYUHA THEODORA SHUWU BA ECONOMICS (SIMON FRASER UNIVERSITY, CANADA) MSC. AGRIC ECONOMICS (UNIVERSITY OF ALBERTA, CANADA)

A THESIS SUBMITTED TO THE SCHOOL OF GRADUATE STUDIES FOR THE AWARD OF THE DEGREE OF DOCTOR OF PHILOSOPHY OF MAKERERE UNIVERSITY

January, 2006

DECLARATION I, THEODORA SHUWU HYUHA, DO HERE DECLARE THAT THIS Thesis is my own work and has not been submitted for a Degree Course in any other University.

Signature-----------------------------------

Date-----------------------------------------

This thesis has been submitted with our approval as University Supervisors

Signed

........................................................................ Dr. Bernard Bashaasha

Date

........................................................................ Dr. Ephraim Nkonya

Date

........................................................................ Professor David Kraybill

Date

ii

 2006, Theodora S. Hyuha All rights reserved. No part of this thesis may be reproduced, stored in any retrieval system, or transmitted in any form by any means, electronic, mechanical, photocopying, recording or otherwise without prior written permission of the author or Makerere University.

iii

DEDICATION

To my late father Donozious Shuwu, my mother Mary Nambozo Shuwu and my daughter Hanifa Hyuha.

iv

ACKNOWLEDGEMENT

It is not possible to enumerate the names of all the people who helped me during the period I was working on this thesis. Nevertheless it would be ungracious of me if I did not thank some of them by name and others in general terms. The first people I wish to express my gratitude to are those rice farmers who allowed me to ply into their private lives by asking them endless questions regarding their rice enterprises. They were really wonderful for they gave me their time freely. The second category of people to thank is the district and village level officials. These officials also gave me unreserved cooperation for they knew my success in the research would in some way or another lead to the betterment of rice farmers livelihoods.

As a researcher, I designed the data collection instrument. However, during the implementation stage, I was assisted by a number of student researchers drawn from Makerere University. These were: Eria Hisali, John Kasembeli, Hon. Ahabwe Godfrey (then lecturer at the department of Agricultural Economics and Agribusiness, Makerere University) Ema Mugalazi, Madina Guloba, Brenda Piloya, Monica Atube and Agea Jacob.

At Makerere University, I received a lot of encouragement from colleagues and friends from the department of Agricultural Economics & Agribusiness, the faculty of Agriculture at large and outside it. I am grateful to all of them. However, special mention first goes to my Supervisors Dr Bernard Bashaasha, Dr Ephraim Nkonya (currently working at International Food Policy Research Institute) and Professor David Kraybill a v

Fulbright scholar currently working in the department. Second, Mr Jairus Mhehe provided assistance in computer work. Mary Busingye and Joyce Oyella provided the initial secretarial work. Thank you for being patient with me. Third, I am grateful to Mr William Ekere, Mr Bernard Tayebwa, Professor Marion Okot and Mr Paul Kabasa for their constant encouragement when the energy ran very low. Mr William Ekere has amazing heart, he never at any time complained about my endless consultation about computer gymnastics. Fourth, I am grateful to Dr Barnabas Kiiza for providing some editorial work. Finally, I thank Professor Elly Sabiiti (former Dean of the faculty of Agriculture) and the current Dean, Professor Matete Bekunda and their respective executive for their constant encouragement and support.

The analysis of data and the write up of the first draft thesis was undertaken at the University of Dar es Salaam, Economic Research Bureau. The Bureau gave me an office, and unlimited access to internet and other facilities for serious writing. For this, my gratitude goes particularly to Dr Godwin Mjema, the Director of the Bureau for his generosity and for integrating me in their Bureau. While there, I presented a paper based on my work in their regularly held seminars. I am grateful to the members of staff for the constructive comments received from them. Out side the seminar room, I received useful comments from Professor Robert Mabele, Dr Innocent Karamagi and Dr Micheal Ndashau . The two secretaries of the Bureau namely, Mwanaisha Kassanga and Grace Kiwia were great. External to the department, but within the University of Dar es Salaam, Professor Letticia Rutasobya and Professor Fred Kaijage mentored me a great deal.

vi

The financial support was provided by African Economic Research Consortium (AERC), International Food Policy Research Institute (IFPRI) and Faculty of Agriculture. I am indeed grateful to these organizations.

My final thanks go to my husband Professor Mukwanason Hyuha and the family for standing by me during the lengthy gestation period of this document.

In spite of all these numerous assistance from supervisors, colleagues, friends, family, the errors and shortcomings remaining in the document are all mine.

vii

ACRONYMS APC CRS FAO GDP LDC MAAIF NAARI Agricultural Policy Committee The Chinese Rice Study Team Food and Agriculture Organization Gross Domestic Product Less Developed Countries Ministry of Agriculture Animal Industry and Fisheries Namulonge Agricultural and Animal Production Research Institute NARO National Agricultural Research Organization UBOS Uganda Bureau of Statistics

viii

ABSTRACT Uganda is implementing the Plan for Modernization of Agriculture as one of the ways to eradicate poverty in rural areas. Consequently, it becomes critical to access technical information on strategic commodities, such as rice, which has become a major cash earner. The main objective of this study was to determine profit efficiency in rice production with a view to isolating factors leading to variation in farm-specific inefficiencies.

The study relied on cross-sectional data collected in 2001 from three districts (Tororo, Pallisa and Lira) of Eastern and Northern Uganda. Two models, namely a profit translog stochastic frontier model and a firm-specific inefficiency model were used. The parameters were estimated simultaneously, using FRONTIER 4.1 computer programme.

Results showed area under rice and capital had a positive influence on profit levels while cost of family labor and other inputs had a negative effect. The analysis also showed that all farmers were not operating on the profit frontier and scored a mean profit efficiency of 66 percent with about 70 percent of the farmers scoring at least 61 percent. The efficiency levels at the district level were 75, 70 and 65 percent, respectively for Pallisa , Lira and Tororo, respectively.

Further analysis showed rice farmers were losing income due to allocative and technical inefficiency. The established sources of inefficiency were: limited access to extension services, low education, limited non-farm employment opportunities and lack of experience in rice growing. Among these, lack of education, limited access to education ix

and limited access to extension services were the major constraints to increasing profit efficiency in rice enterprises. Based on elasticity estimates, the study also established further that improving efficiency would require expansion of the area under cultivation, which would have the greatest positive impact on profits.

TABLE OF CONTENTS DECLARATION.................................................................................................................ii DEDICATION....................................................................................................................iv ACRONYMS...................................................................................................................viii ABSTRACT.......................................................................................................................ix LIST OF TABLES............................................................................................................xiv LIST OF FIGURES .........................................................................................................xvi CHAPTER I ........................................................................................................................1 INTRODUCTION...............................................................................................................1 1.1 Background...................................................................................................................1 1.2 Problem Statement.......................................................................................................2 1.3 The Objectives of the Study.........................................................................................3 1.4 Hypotheses....................................................................................................................4 1.5 Organization of the Study.............................................................................................4 CHAPTER II.......................................................................................................................5 LITERATURE REVIEW AND THEORETICAL FRAMEWORK...................................5 2.1 Meaning of Efficiency...................................................................................................5 2.2 Theoretical Basis for Measurement of Efficiency.........................................................8 2.2.1 Technical, Allocative and Economic Efficiency........................................................8 P............9 D............9 Q............9 R............9 B............9 xi

D'...........9 2.2.2 Profit Function..........................................................................................................10 2.3 Profit Inefficiency Model............................................................................................13 2.4. Technical Efficiency: Empirical Studies....................................................................15 2.5 Profit Function Analysis: Empirical Studies ..............................................................23 CHAPTER III....................................................................................................................28 METHODOLOGY............................................................................................................28 3.0 Introduction..................................................................................................................28 3.1 Approaches to Measuring Efficiency..........................................................................28 Deterministic Versus Stochastic Frontier Models ............................................................29 3.3 Theoretical Profit Function and Stochastic Frontier Model .....................................31 3.4 Empirical Models........................................................................................................33 3.4.1 Translog Stochastic Frontier Profit Function Model................................................34 3.4. 2 Definition of Variables and Estimation of Profit Frontier Function .......................36 3.4.3 Variables Included in the Inefficiency Model..........................................................40 3.5 Study Area, Data and Sources.....................................................................................44 3.5.1 Description of the Study Area .................................................................................44 3.5.2 The Data....................................................................................................................46 3.5.3 Data Reliability and Validity ...................................................................................48 Data Analysis/Model Implementation...............................................................................54 CHAPTER IV....................................................................................................................56 Results and Discussion......................................................................................................56 4.0 Introduction..................................................................................................................56 4.1 Socio Demographic and Socio Economic Characteristics ..........................................56 4.2 Testing for the Appropriateness of C-D Model...........................................................60 xii

4.3 Estimation of Frontier Profit Function: Translog Model.............................................65 4.4 Profit Efficiency Score Estimates: Translog Model ...................................................71 4.5 Determinants of Firm-Specific Profit Inefficiency in Rice-Translog Model .............74 4.6 Key Constraints to Profit Efficiency in Rice Production............................................77 4.6 Summary .....................................................................................................................82 CHAPTER V.....................................................................................................................84 SUMMARY, CONCLUSIONS AND POLICY RECOMMENDATIONS......................84 5.0 Introduction.................................................................................................................84 Summary............................................................................................................................87 ..............87 5.2 Conclusions and Policy Recommendations................................................................90 5.3 Recommendations for Further Research....................................................................92 REFERENCES..................................................................................................................93 APPENDIX A..................................................................................................................100 APPENDIX B..................................................................................................................120 APPENDIC C: MAP SHOWING STUDY DISTRICTS ...............................................130

xiii

LIST OF TABLES Table 3.1: Variables Included in the Frontier Profit Function Models and their Descriptions........................................................................................................38 Table 3. 2: Variables Included in the Inefficiency Model and Descriptions.....................40 Table 3. 3: Skewness and Normality Variables (Unstandardized)- Translog Model.......51 Table 3. 4: Skewness and Normality Variables (Standardized)- Translog Model..........52 Table 3.5 Effect Magnitude Measures for the MLE result Estimates...............................53 Table 3. 6: Tests of Significance using Two-Sample Kolmogorov-Smirnov Test...........54 Table 4.1a: Selected Socio-economic Characteristics of Farmers in the study area ........58 Table 4.1b: Other Household Characteristics in the Study Area.......................................60 Table 4.2: Hypotheses Testing for the Models and its Inefficiency Effects......................64 Table 4.3a: Frontier Profit Function among Rice Producers in selected Districts ..........66 Table 4.3b: Frontier Profit Function among Rice Producers in Tororo District...............68 Table 4.3c: Frontier Profit Function among Rice Producers in Pallisa District................69 Table 4.3d: Frontier Profit Function among Rice producers in Lira District....................70 Table 4.4: Frequency Distribution of Farm- Specific Profit Efficiency Index in ............71 Studied Areas-Translog Model..........................................................................................71 Table 4.5 Comparison of mean Profit loss per hectare as a result of Profit Efficiency by Districts...............................................................................................................73 Table 4.6 Tests of Significance of Mean Profit loss .......................................................73 Table 4.7 Estimated Profit Elasticities in the studied Area...............................................73 Table 4.8: Determinants of Farm-Specific Inefficiency in Rice Production in the Sampled Districts...............................................................................................................75 Table 4.9a Profit Loss in Rice Production in Tororo District by Key Constraints............79 xiv

Table 4.9b Profit Loss in Rice Production in Pallisa District by Key Constraints............80 Table 4.9c Profit Loss in Rice Production in Lira District by Key constraints.................81

xv

LIST OF FIGURES Figure 1: Stochastic Production Frontier.............................................................................9 Figure 1: Stochastic Production Frontier.............................................................................9 Figure 2: Frontier MLE and OLS Stochastic Profit Function...........................................13 ...............118 ..............120

xvi

CHAPTER I INTRODUCTION 1.1 Background Agriculture plays an important role in Ugandas economy. About 74.8% of the people in Uganda not only live in the rural setting, but also depend on agriculture for their livelihood. Of these 68.1% depend on subsistence agriculture (Uganda Bureau of Statistics UBOS, 2005). In 2003, the agricultural sector contributed 40% of Gross

Domestic Product (GDP), (UBOS, 2004).

Rice is one of the emerging crops grown currently in Uganda. It plays an important role both as a food and a cash crop in the country (Sabiiti, 1995; Ochollah et al., 1997). In 1997 it ranked first in terms of returns per labor day among major crops grown in the country (Agricultural Policy Committee APC, 1997) and in 2005, a study by Jagwe et al., (2005) confirmed this for Kabarole distribution. The crop ranked fourth among the cereal crops, occupying a total of 80 thousand hectares of land with an estimated output of 120, 000 tonnes (UBOS, 2004). It is becoming a staple food countrywide, especially in urban areas (World Bank, 1993). Available figures show that Uganda consumed an average of 7,877 tons per annum over the four-year period of 1994-1997, and imported rice worth 184.5 million Uganda Shillings (US $ 174,386 thousand) for the same period. In 2003, the countrys rice import requirements were estimated at 50 thousand tons (FAO, 2004). Uganda is therefore a net importer of the commodity and will continue to do so in the near future unless there is an improvement in domestic production. This is feasible as the country has 70,000 hectares of land with ideal agronomic conditions for rice production (CRS, 1982). However, the crop ranks low in terms of research among the cereal crops within the National Agricultural Research Organization (NARO), an organization 1

charged with agricultural research in the country. It is only recently (1998) that the crop has attracted the attention of agricultural research (personal communication with the cereal Program Leader at Namulonge Agricultural and Animal Production Research Institute (NAARI). Even then the emphasis is on upland rice and looking at agronomic factors. Limited knowledge exists, particularly on socio-economics. The present study thus makes a contribution to the empirical research in this field.

It should also be noted that, the current policy thrust with respect to agriculture in Uganda is modernization of the sector (Ministry of Agriculture Animal Industry and Fisheries MAAIF, 2000). This calls for increased research, on how best to increase productivity and inform policy. It is hoped that the transformation of the sector can make a significant contribution to poverty reduction efforts. The case of rice profit efficiency therefore becomes interesting. However, the crop has faced a declining trend in yield in the last five years (2000-2004). This trend needs to be reversed and hence the importance of this study.

1.2 Problem Statement Rice is largely grown as a cash crop in Eastern and Northern Uganda. Production of the crop is therefore motivated by the economic objective of earning a positive economic return. Meeting this objective requires efficient utilization of scarce resources. However, there could be intervening variables which may hinder agents to realize this objective. Thus, there is a need to examine profit efficiency in rice production in Eastern and Northern Uganda and to identify factors that influence efficiency in this sector. An approach that can be used to solve the problem of efficient utilization of scarce resources focuses on two questions: first, whether farmers are economically (technically 2

and allocatively) efficient in rice production and second, what factors determine their level of efficiency? Answers to these two questions provide a clue on how we can assist farmers to be efficient in utilizing their resources employed in rice production.

To date, there is only one known study that has addressed efficiency and management practices of Ugandan rice farmers (Ssenteza, 1993). Ssenteza (1993) estimated elasticities using a Cobb-Douglas (C-D) production function of Kibimba rice scheme. Other related studies include Yilma (1996) and Appleton and Balihuta (1996). Yilma (1996) estimated productive efficiency of coffee and bananas in Masaka district while Appleton and Balihuta (1996) focused on the impact of education on agricultural productivity in Uganda. Thus there is a need to examine profit efficiency among rice producers in the country with the view to providing answers to the aforementioned questions.

1.3 The Objectives of the Study The main objective of this study is to examine the profit efficiency of rice production at farm level in Eastern and Northern regions of Uganda.

The specific objectives of this study include the following:1) To characterize rice production system in Eastern and Northern Uganda. 2) To estimate the rice frontier profit function for Eastern and Northern Uganda and determine factors influencing profit. 3) To determine farm specific factors that influence the observed variability of profit efficiency levels among rice producers.

1.4 Hypotheses In view of the problem and objectives, the following hypotheses are tested:

1)

Rice farmers in Eastern and Northern Uganda are not operating on efficient profit frontier.

2)

There is no variability in the level of profit inefficiency among rice farmers in Eastern and Northern Uganda.

3)

Factors such as non-farm employment, education, access to extension services and credit, experience, employment and degree of specialization in rice production influence the observed level of profit inefficiency among rice farmers in Eastern and Northern Uganda.

1.5 Organization of the Study Chapter two begins with a discussion of the concept of economic efficiency. The rest of the chapter covers issues concerning model development, and factors associated with measurement of economic inefficiency. The chapter concludes with a review of empirical studies concerned with measuring allocative and technical efficiency.

Chapter three provides a detailed discussion of the methodology adopted, conceptual model, empirical model used in the study, and describes data sources. Both descriptive and econometric results are discussed in chapter 4 while summary and conclusions are presented in chapter 5.

CHAPTER II LITERATURE REVIEW AND THEORETICAL FRAMEWORK

2.1 Meaning of Efficiency The analysis of efficiency dates back to Knight (1933), Debrew (1951) and Koopmans (1951). Koopmans (1951) provided a definition of technical efficiency while Debrew (1951) introduced its first measure of the coefficient or resource utilization. Following on Debrew in a seminal paper Farrell (1957), provided a definition of frontier production functions, which embodied the idea of maximality. Farrell (1957) distinguished three types of efficiency: 1) technical efficiency and 2) price or allocative efficiency and 3) economic efficiency which is the combination of the first two.

Technical efficiency is an engineering concept referring to the input-output relationship. A firm is said to be efficient if it is operating on the production frontier (Ali and Byerlee, 1991). On the other hand, a firm is said to be technically inefficient when it fails to achieve the maximum output from the given inputs, or fails to operate on the production frontier. Mbowa (1996) in his study on the sugarcane industry in South Africa defined an efficient farm as that which utilizes fewer resources than other farms to generate a given quantity of output. Yilma (1996), while studying efficiency among the smallholder coffee producers in Uganda, defined an efficient farm as that which produces more output from the same measurable inputs than that one which produces less. Fan (1999) referred to technical inefficiency as a state in which actual or observed output from a given input mix is less than the maximum possible.

Price or allocative efficiency has to do with the profit maximizing principle. Under competitive conditions, a firm is said to be allocatively efficient if it equates the marginal returns of factor inputs to the market price of output (Fan, 1999). Akinwumi and Djato (1996) in their study of relative efficiency of women farm managers in Cote dIvoire define allocative efficiency as the extent to which farmers make efficient decisions by using inputs up to the level at which their marginal contribution to production value is equal to factor costs. Failure to equate revenue product of some or all factors to their marginal cost is at the very core of economic theory (Timmer, 1971). Similarly, Ali and Byerlee (1991) agree with this definition in their review of economic efficiency of small farmers in a changing world. They contend that allocative inefficiency is failure to meet the marginal conditions for profit maximization. Thus allocative inefficiency is failure of a farmer to equate marginal returns of factor inputs to its price.

Economic efficiency is distinct from the other two even though it is the product of technical and allocative efficiency (Farrell, 1957). A firm that is economically efficient should by definition be both technically and allocatively efficient. However, this is not always the case as Akinwumi and Djato (1997) pointed out. It is possible for a firm to have either technical or allocative efficiency without having economic efficiency. The reason may be that the farmer, in this case, is unable to make efficient decisions as far as the use of inputs is concerned. In some cases, a farmer might fail to equate marginal input cost to marginal value of product. If technical and allocative efficiency occur together they are both a necessary and a sufficient condition for economic efficiency. This assumes that the farmer has made right decision to minimize costs and maximize profits implying operating on the profit frontier. However, one needs to recognize that in least developed countries (LDCs) there are inherent market failures due to a number of 6

reasons such as unwarranted government interventions, lack of information on the markets and poor infrastructure. Notwithstanding this phenomenon, this study adopts a definition of efficiency, which encompasses technical and allocative efficiency, in essence economic efficiency.

Apart from these definitions, literature on efficiency distinguishes many other forms of efficiency and these are productive, scale (economies of size) and economies of scope and x-efficiency. Production is said to be efficient if it is not possible to produce more of one good without taking resources away from production of another good (Binger and Hoffman, 1998). From the discussion by Wang et al., (1996b) production efficiency is equivalent to economic efficiency because it combines two components, that is, technical and allocative efficiency. Scale efficiency can also arise from spreading the cost of production, particularly fixed costs over a large output. Taking an example of an assembly line, it would not be cost effective if the firm opts to produce a few cars a year when it is capable of producing a large number of cars to achieve low per unit cost. The assembly reaps economies of scale when it experiences substantial cost savings at relatively high output (Binger and Hoffman, 1998). But the firm can experience diseconomies of scale due to coordination problems. According to Sadoulet and Alain de Janvry (1995) the presence of economies of scale in agriculture is not conclusive.

Economies of scope exist when a firm decides to put two separate enterprises under one management. The enterprises share the same factors of production such as labour and in the process cut down on costs. In the process of sharing the factors, the management saves on costs and as such it is able to reap economies of scope. X-efficiency is realized through motivating staff who in turn work hard to produce maximum output. 7

2.2 Theoretical Basis for Measurement of Efficiency 2.2.1 Technical, Allocative and Economic Efficiency

Measurement of economic efficiency requires an understanding of the decision making behaviour of the producer. A rational producer, producing a single output from a number of inputs, x = x1xn, that are purchased at given input prices, w = w1..wn and operating on a production frontier will be deemed to be efficient. But if the producer is using a combination of inputs in such a way that it fails to maximize output or can use less inputs to attain the same output, then the producer is not economically efficient. A given combination of input and output is therefore economically efficient if it is both technically and allocativelly efficient; that is, when the related input ratio is on both the isoquant and the expansion path. These contentions are best illustrated in the figure 1. In figure 1, AB is an isoquant, representing technically efficient combinations of inputs, x1 and x2, used in producing output Q. AB is also known as the best practice1production frontier. DD' is an iso-cost line, which shows all combinations of inputs x1 and x2 such that input costs sum to the same total cost of production. However, any firm intending to maximize profits has to produce at Q', which is a point of tangency and representing the least cost combination of x1 and x2 in production of Q. At point Q' the producer is economically efficient.

Coelli (1995) indicates that the production function of the fully efficient firm best practice is not known in practice, and thus it must be estimated from the sample of the industry concerned.

Figure 1:

Stochastic Production Frontier

x2/y

>
B Q
'

D' O

x1/y Turning to measurement of technical, allocative and economic efficiency, the same figure 1 is employed. Suppose a farmer is producing its output depicted by isoquant AB with input combination level of (X1and X2) in figure1. At this point (P) of input combination the production is not technically efficient because the level of inputs needed to produce the same quantity is Q on isoquant AB. In other words, the farmer can produce at any point on AB with fewer inputs (X1 and X2) in this case at Q in an input-input space. The degree of technical efficiency of such a farm is measured as OQ/OP. OQ/OP is the proportional reduction of all inputs that could theoretically be achieved without any reduction in output. In figure 1, DD' represent input price ratio or iso-cost line, which gives the minimum expenditure for which a firm intending to maximize profit should adopt. The same farm using (X1 and X2) to produce output P would be allocatively inefficient in relation to R. Its level of allocative efficiency is represented by OR/OQ, since the distance RQ

represents the reduction in production costs if the farmer using the combination of input (X1 and X2) was to produce at any point on D D', particularly R instead of P.

The overall (economic) efficiency is measured as the product of OQ/OP and OR/OQ, which is OR/OP. This follows from interpretation of distance RP as the reduction in costs if a technically and allocatively inefficient producer at P were to become efficient (both technically and allocatively) at Q'. These forms reflect alternative behavioral objectives (i.e. profit maximization or cost minimization) and can account for multiple outputs (Coelli, 1995).

2.2.2

Profit Function

A profit function is an extension and formalization of the production decisions taken by a farmer. According to production theory, a farmer is assumed to choose a combination of variable inputs and outputs that maximize profit subject to technology constraint (Sadoulet and De Janvry, 1995). The underlying production function can be generalized as h (q, x, z) = 0 where q is a vector of output, x is a vector of variable inputs, z is a vector of fixed inputs and h is a technology. Assuming the technology to be homogeneous across farms, restricted profit function is specified as follows: Max p.q-wx,., s.t. h(q,x,z) = 0 1

Where: p is a vector of prices of outputs and w is a vector of prices of variable inputs

10

Considering a set of inputs and outputs the profit maximizing input demand and output supply functions are generally respectively expressed as: X Q = = x (p, w, z) q (p, w, z) 2 3

Substituting equation 2 and 3 into1 gives a profit function which is the maximum profit that the farmer can obtain given prices of p and w, availability of fixed factors z and production technology h(.). The profit function can be written as = p'q( p,w,z) - w'x(p,w,z) 4

This study uses the normalized profit function outlined in equation 5 given the fact that the study is dealing with a single output, that is, rice (Sadoulet and De Janvry, 1995). Hence for rice, we have: i = (Pij, Zik). exp (i ) 5

This makes profit non-linear in its error term. However, the profit function can be loglinearized to obtain the form : ln i = lnf(.) + ei. where: i
=

normalized profit on firm i defined as gross revenue minus variable cost divided by the output price.

Pij Zik
i

prices of variable input j on firm i divided by the output price. level of fixed input on firm i where k are a number of fixed inputs.
1,..,

n number of farms in the sample.

11

error term assumed to behave in a manner consistent with the frontier

concept (Ali and Flinn, 1989).

Figure 2 shows the stochastic profit frontier function adopted from Ali and Flinn, (1989). The stochastic profit frontier function is an extension of incorporating farm level prices and input use in the frontier production function. The incorporation of the farm specific level prices leads to the profit function approach formulation (Ali and Flinn, 1989; Wang et al., 1996a). A production approach to measure efficiency may not be appropriate when farmers face different prices and have different factor endowment (Ali and Flinn, 1989). Hence the uses of stochastic profit function to estimate farm specific efficiency directly (Ali and Flinn, 1989; Ali et al., 1994; Wang et al., 1996a). The profit function approach combines the concepts of technical, allocative and scale inefficiency in the profit relationships and any errors in the production decision translate into lower profits or revenue for the producer (Rahman, 2003). Profit efficiency is defined as the ability of a farm to achieve highest possible profit given the prices and levels of fixed factors of that farm and profit inefficiency in this context is defined as the loss of profit from not operating on the frontier (Ali and Flinn, 1989).

12

Figure 2:

Frontier MLE and OLS Stochastic Profit Function

$ Normalized Profit

D M E F P . M E OL D E

Normalized input price given fixed resources P/Zj

Source: Ali and Flinn (1989)

In the context of frontier literature, DD in figure 2 represents profit frontier of farms in the industry (the best practice firm in the industry with the given technology). EE is the average response function (profit function) that does not take into account the farm specific inefficiencies. All farms that fall below DD are not attaining optimal profit given the prevailing input and output prices in the product and the input markets. They are producing at allocativelly inefficient point F in relation to M in Figure 2. Profit inefficiency is defined as profit loss of not operating on the frontier. In Figure 2, a firm operating at F, is not efficient and its profit inefficiency is measured as FP/MP (Ali and Flinn, 1989; Sadoulet and Janvry, 1995).

2.3 Profit Inefficiency Model The issue of whether a farmer in a developing country is responsive to economic incentive is now a mute point. The attention has shifted to how the whole system works 13

(Ali and Byerlee, 1991). From an engineering point of view, a system is said to be efficient if maximum output is generated from the given input keeping other factors constant. If this ideal position does not obtain, it is said to be inefficient and the sources of inefficiency could either be internal or external.

In agriculture, a farmer has to pay attention to relative prices of the inputs such that the production is undertaken at the point where the isoquant is tangent to isocost line (Figure 1). If that is not done, economic efficiency is not achieved. The farmer may be able to achieve technical efficiency but not allocative efficiency. This inefficiency could arise from a number of sources, which include access to appropriate information in a timely manner or lack of skills to take advantage of modern agricultural inputs. Basically, what is being referred to here is the managerial ability of the farmer. The farmer should be able to make decisions that lead to optimal utilization of resources and this requires accurate information on availability of the new varieties, the inputs, and access to markets.

Besides, the farmers inability to make optimal decisions may be due to external factors, which lie outside his/her prevue. These include untimely input supply, bad weather, nonconducive policies and other random shocks such as wars, floods, pests and diseases, droughts, and statistical errors. Inefficiency could also arise from the introduction of new varieties without adequate provision of back-up packages to the farmers. In this learning stage, a farmer could appear inefficient while he/she is not, due to the fact that he/she is unfamiliar with the new variety. One has to recognize that it takes time to learn new agronomic practices. During this learning stage, production functions among the farmers would differ. Therefore, it is unrealistic to attribute all inefficiency to farmers own inability to make rational decisions. 14

2.4. Technical Efficiency: Empirical Studies This section presents a review of some of the technical efficiency studies. Lingard et al., (1983) applying a two-component model to panel data estimated a bias free agricultural production function for the Philippine rice farmers in Luzon district. The study showed that area was dominant in earlier years when the technology was introduced, while other variables (such as irrigation, fertilizers and chemicals) became significant overtime, reflecting full adoption of the technology.

The farm-specific efficiency was established through the 32 farm intercept terms on a series of variables for the year, 1979. The results showed that the variables most highly associated with farm-specific technical efficiency were soil type, credit access, education and land tenure differences. However, the authors did not carry out a second stage analysis of establishing factors affecting technical efficiency. Nevertheless, they concluded that the managerial efficiency is an important factor in rice production in the Philippines.

Belbase and Grabowski (1985) used corrected ordinary least squares (COLS) technique to measure technical efficiency of farmers in Nuwakot District in Nepal. The appropriately adjusted (removing the outliers) results showed that the Nepalese farmers were operating close to the technical frontier. The factors contributing positively to technical efficiency were: nutrition levels, family incomes and education. The structure (farm size) of the farms was taken as given, yet as noted by Mbowa (1996), the variable

15

bears a significant influence on technical efficiency. Further, the Belbase and Grabawoskis study did not deal with allocative inefficiency.

Taylor and Shonkwiler (1986) used both deterministic and stochastic frontier production function to study the impact of agricultural credit programs on farmers in South eastern Brazil. They used both models (deterministic and stochastic) to test the effectiveness of the programme. The results showed that both groups (participants and non participants in the programs) consistently had higher technical efficiency from stochastic than deterministic frontier specifications. Yet in applying both specifications (deterministic and stochastic) to analyse the effect of the credit programme to both participants and nonparticipants, the results showed that the participants had higher technical efficiency than the non-participants in deterministic specification. The two specifications, therefore, yielded conflicting inferences regarding the effectiveness of the programs. In their conclusion the authors noted that such results place considerable importance on the subjective beliefs of the researcher. Any definitive inference must, in the end, rest in the gray area of determining which specification is the most realistic on both theoretical and empirical grounds; Taylor and Shonkwiler (1986).

A deterministic model attributes any deviation from the frontier as resulting solely from inefficiency measured by . The major weakness with this model is that any measurement error and any other source of variations in the dependent variable is embedded in and cant be separated. As a result, outliers may have profound effects on the estimates and also, any shortcomings in the specification of the model could translate into inefficiency estimates.

16

A stochastic model on the other hand, takes into account random factors, which are outside the control of the farmer. The model addresses the noise problem characterizing deterministic frontier. In other words, the model enables the researcher to provide more explanation of the inefficiency observed than before. This was not possible before because of the violation of certain maximum likelihood regularity conditions. The coefficients estimated this way are expected to be more efficient parameters and its popularity by researchers may be due to this held view (Thiam et al., 2001). However, the model has its own shortcomings. It lacks apriori justification for the selection of a particular distributional form for the one sided inefficiency term . In this study a stochastic approach is adopted due to the reasons given earlier: provision of better explanation on observed inefficiency at farmer level and getting more efficient parameter estimates.

Unlike the previous studies reviewed, Kalirajan and Shand (1988) estimated technical efficiency for multiple crops (rice-corn- rice)2 and multiple outputs using stochastic translog production frontier for a sample of farmers operating in rain-fed areas of India. The results showed that levels of crop-specific and farm-specific efficiency varied widely among small farmers, but on the whole only 24% of the sample was found to be technically efficient in growing all the crops. The causes of variation in technical efficiency at farm level were found to differ across crops. In the case of rice, farming experience and extension officials visits were found to be important whereas financial availability was the most crucial to maize production. The authors concluded that a mere

Rice-corn-rice implies crop rotation of rice in the first season followed by corn in the next season and rice in the third season.

17

choice of high yielding technology is not sufficient to increase the production of rice, what is important is the proper use or application of the technology.

Acknowledging the fact that measurement of efficiency is sensitive to methodology used, the data, period, and sample, Dawson and Lingard (1991) employed two different approaches to estimate technical efficiency of rice farms in The Philippines (Luzon Province) namely, covariance analysis (CA) and stochastic frontier approaches. The CA was used to pull together the cross-sectional and panel data. The CA produced biased results, confirming observation made by Timmer (1971). Further, the estimation using cross-sectional data set produced a wide range of coefficients while those of panel data had better results. The range of technical efficiency ratings was narrower, with 14 farms being over 90% or more efficient. The major strength of this study was that it succeeded in showing that efficiency estimates are sensitive to methodology, type of data, and sample size. More crucial to the current study, is the fact that Dawson and Lingard (1991) pointed out that there was a need to go beyond identification and develop methodologies that explain sources of differences in farm efficiency.

Besides stochastic approaches, the non-stochastic3 frontier approach is one of the methods that can be employed to estimate relative efficiency. Ali and Byerlee (1991) revealed that the studies which used non-stochastic approaches, concentrated on nonconventional inputs, such as education. Most of the studies reviewed showed that education had positive and significant effect on productivity. The variable contributed up to 9.5% increase in productivity in a modernizing (countries that have undergone green
33

Non-stochatic frontier approaches attribute all deviation from the frontier as due to to technical inefficiency.

18

revolution) agriculture. However, the unsolved puzzle is about which level of education matters in farm management. Some studies have indicated that basic formal education of up to 4 years is essential, particularly the farm manager, irrespective of gender (Appleton and Balihuta, 1996; Weier, 1999).

Building on other studies, and still in quest to find a better and efficient approach to measure efficiency, other researchers such as Phillips (1994) and Thiam et al., (2001) reviewed studies on technical efficiency using a method called meta-analysis. The method uses empirical estimates of some indicator from several studies; the average technical efficiency in this case serves as the dependent variable, and attempts to explain the variation of the estimates based on differences across studies as explanatory variables in a regression model. Thiam et al. (2001) was the first to use this approach to analyze technical efficiency.

The issue that empirical measures of efficiency depend on the choice of the model adopted still remains controversial. Thiam et al., (2001) conclude, that despite this array of applied work, the extent to which empirical measures of efficiency are sensitive to the choice of methodology remains a matter of controversy. But this is not limited to efficiency studies. Many other results are sensitive to methodology; data and time the data were taken. Even though the study by Thiam et al,.(2001) covered 32 frontier studies, using farm level data from 15 different developing countries, none of the method was superior to the other in terms of robustness. Thiam et al,.(2001) concluded that the groundwork remained a challenge to researchers.

19

Indeed, earlier on, Kalirajan and Obwona (1994) investigated the appropriateness of the use of stochastic frontier methodology to estimate efficient utilization of the inputs given the technology. The authors argued that by nature of its generic assumptions, that is, the potential frontier being a neutral shift frontier from the realized production function, input-specific technical efficiency measures could not be obtained without contradictions. They further argued that even with the same identical levels of inputs, outputs would differ due to differences in the methods of application of inputs by the farmer. Thus it would be necessary to use a different model, which reflects the method of application of inputs by individual farmers.

Kalirajan and Obwona (1994) therefore went ahead and developed a Cobb-Douglas type of production function to address these concerns and computed actual response coefficients for individual observations. The result results obtained showed that there were variations in the farm-specific and input-specific actual response coefficients across sample farms. The authors therefore rejected the assumptions of the conventional stochastic frontier production function approach and used a non-stochastic frontier CobbDouglas (C-D) type of modeling of the production behavior of farmers. The authors concluded that the developed model was appealing because it made it easy to draw policy recommendations for the estimated results indicate the type of input being over or underutilized. Secondly, it was possible to set targets for different inputs to produce a given level of output. A latter study by Sharma et al., (1999) on Hawaiian swine farmers, applying the parametric stochastic and nonparametric approach Data Envelope Analysis (DEA)4 method produced results that were also robust. The method differs from the parametric method in that the researcher does not have to make arbitrary assumptions
4

A DEA is a non-parametric mathematical programming approach to frontier estimation.

20

about the functional form of the frontier and distributional form of the . Additionally, DEA does not make assumptions about the efficiency of farms since it measures relative efficiency of farms, given a set of inputs DEA also differs from Farrells single-input, single output efficiency analysis to multi-input, multi-output efficiency analysis. However, according to Coelli (1995) the method also suffers from the same weaknesses as that of the deterministic model covered before.

Studies estimating technical efficiency in Africa are limited. A review of some of them follows. Yilma (1996) used three different approaches to estimate smallholder efficiency in coffee and bananas namely, deterministic parametric, stochastic frontier approaches and DEA in Masaka district, Uganda. The deterministic parametric approach showed differences in mean scores of efficiencies in coffee and generally food production. The coefficients estimated under deterministic parametric frontier model showed lower efficiency than the stochastic frontier model, agreeing with many earlier studies (Kalirajan and Obwona, 1994 and Lingard et al., 1983). Nevertheless, irrespective of the approach used, all farmers were found not to be producing on the frontier.

Mbowa (1996) used DEA to examine resource use farm efficiency on small and largescale farms in sugarcane production in Kwazulu-Natal. The study results showed that small-scale farmers were technically inefficient than large-scale producers and concluded that the size of farm operation affects level of efficiency attainable.

21

Seyoum et al., (1998) used a two step procedure 5proposed by Coelli (1996a)6 to estimate separate stochastic frontier production functions for the two groups of maize farmers (within and outside Sasakawa Global (SG)2000 project) in Ethiopia. Empirical results of the study showed that technical efficiency levels for the participating group were higher than that of the non-participants. Furthermore; the participating group also registered higher mean frontier output than those outside the project. The inefficiency model showed that there exists technical inefficiency in the production of maize. The contributing factors were education, age, and extension contact. Education and extension services had a negative influence on technical inefficiency for those participating in the project. The authors therefore concluded that in order to promote agricultural productivity the government could introduce projects such as SG2000 which appear to have had a positive impact on technical efficiency.

Appleton and Balihuta (1996) studied the impact of education on agricultural productivity. Using the national household survey data, they found that education of at least 4 years of formal schooling of the farm manager appears to raise production by 7 percent. They also found that there were large effects of education on other farmers in the neighborhood. However, the analysis was limited in scope since they focused on agricultural production only. This study intends to focus on estimating profit function which takes into consideration prices of outputs and inputs.

The procedure involves estimating frontier profit function first and then using residuals of this function to estimate the inefficiency effects
6

The version 4.1 estimates the variance terms of

2 s

2 v

and

2 s

where

has the

value of 0 to 1.When it is close to 1 the observed variation of the parameter is attributed to the farmer.

22

Weier (1999) examined the benefits of schooling upon farmer productivity and efficiency employing both average production function and a two-stage-stochastic frontier production functions in rural Ethiopia. As in the case of Appleton and Balihuta (1996), the authors used household budget data set. The analysis revealed that at household level farmers in rural Ethiopia were not operating on the frontier. The results suggested that education has a role to play in increasing agricultural production.

2.5 Profit Function Analysis: Empirical Studies This section reviews some of the studies conducted in Asia and Africa using profit analysis approach. The review begins with studies were conducted on African continent. Saleem (1988) tested the Marshallian theory in Sudan to find out which system (sharecropping or fixed rent) operating on irrigated cotton farms was efficient. Marshallian theory on sharecropping states that farmers operating under fixed rent should be more efficient than those under sharecropping (Saleem, 1988). According to this theory, farmers who have to pay rent tend to equate the marginal value of product (MVP) of a variable factor to market price of that factor whereas those on share cropping have no incentive to do so. They may instead equate a fraction of MVP to its market price. This may happen when the input is subsidized. This is not true in many developing countries, where subsidies dont exist.

Applying Lau and Yotopolous (L-Y) profit function model to the data, Saleem (1988) found out that the two farmer groups both growing medium staple cotton, respectively in Gezira (shared) and Rahad (fixed) schemes were equally technically and price efficient, in other words economically efficient. The results therefore did not support Marshallian theory on rent and suggested that both groups had identical profit functions. The author 23

was not able to estimate farm-specific differences in profit efficiency among these farmers. In light of this, this study uses an identical profit function for the farmers in Doho and Olweny rice schemes. Farmers in Doho either rent or own plot(s) whereas in Olweny Scheme, farmers use plots without paying rent because it is a government owned scheme.

The relationship between firm size and efficiency has major implications to policy options for agricultural development in Africa. The previous studies carried out in Asia gave conflicting results on which size is more efficient -large or small (Akinwumi and Djato, 1996). But, in the case of Africa, the superiority (more efficient) of large-scale farms was apparently artificial because (according to the authors), the large-scale farmers had in the past been given preferential treatment over the small scale-farmers. For instance, the latter group tended to have restricted access to certain markets for their commodities and as such they could not demonstrate their efficiency in growing the crop in question. Indeed, in their study of rice farms in Ivory Coast, Akinwumi and Djato (1996) found no significant differences in economic efficiency between small and large rice farmers in Cote dIvoire. But in general, they found absolute allocative inefficiency within the rice sector.

Yotopoulos and Lau (1973) extended their 1971 study with the major purpose of isolating the causes of observed differences in economic efficiency between large and small scale farmers. Using a stochastic profit frontier approach, the authors reaffirmed their earlier findings that small farms had relatively higher economic efficiency than large farms and that both groups of farms succeeded in maximizing profits. Nevertheless, small farmers were found to be more technically efficient than large scale farmers. But, the authors 24

were not able to provide an explanation for the observed differences in technical efficiency (Yotopoulos and Lau, 1973).

Even though it is very instructive to have knowledge on farm efficiency levels, under different production system, what is equally important, from the policy point of view, is to pinpoint the causes of the observed efficiency levels. Ali and Flinn (1989) adopting the theoretical model formulated by Yotopoulos and Lau (Y-L), and using techniques proposed by Jondrow et al., (1982) to estimate farm-specific efficiency among the Basmati rice producers in Pakistan Punjab, found that farmers exhibited a wide range of profit inefficiency ranging from 87% less than maximum profit to 5% less than maximum profit. The results also showed that the variance ratio parameter 7 proposed by Battese and Corra (1977) was statistically greater than zero, implying that the variation in actual profit from maximum profit between farms arose from differences in farmers practice rather than random variability.

The study also found that the major determinants for profit loss among the Basmati rice producers were both socio-economic and institutional factors. The former accounted for 52% profit loss and, of these, education alone accounted for 31%. The latter accounted for 25% of profit loss. However, factors associated with resource base were not significant in explaining the profit loss and only explained 12% while institutional factors explained 25%. Rahman (2002, 2003) estimated a stochastic profit function for Bangladesh rice farmers. The results showed that there existed a high level of inefficiency in rice farming because was close to one. The average profit efficiency
7

is the likelihood- ratio test statistic which is = -2{log [Likelihood (H0)]-Log [Likelihood (H1)]} and has approximately 2 distribution with equal to the number of parameters assumed to be zero in the null hypothesis (Rahman, 2002). is bounded and lies between 0 and 1. When the null hypothesis is rejected, it implies that the inefficiency exist in the estimated model and are stochastic.

25

scores were 60%, which implied that the farmers could improve their profitability by as much as 40%.8 The farmers also exhibited a lot of profit inefficiency. The farm-specific factors responsible were poor access to input markets, unfavorable tenancy arrangements, and off farm employment.

A slightly different but as yet unsettled proposition is whether female farmers are as efficient as male farmers in agricultural operations. Quoting FAO (1985), Akinwumi and Djato (1997) observed that the argument used to discriminate against female farmers in projects is that they are not efficient. Yet, Moock (1976) showed that female managers in Vihiga district in Kenya were as efficient as men. Nkonya et al., (2004) found Ugandan women farmers to be more efficient. Other studies give mixed results. This is attributed to reliance on the production function, which suffers from simultaneous bias (Akinwumi and Djato, 1997). The authors used profit function approach for rice farmers in Cote dIvoire to settle this debate empirically. They found that the relative degree of economic efficiency of women rice farmers is similar to that of men rice farmers in Cote dIvoire. The authors concluded that there was no economic rational for biasing rice development strategies towards male farmers in Cote dlvoire because when the two groups have equal access to inputs; they would exhibit equal levels of economic efficiency.

A review of literature by Ali and Byerlee (1991) on economic efficiency of small farmers in a changing world revealed a need for sharpening conceptual and methodological problems to derive useful policies. On methodological level, the authors suggested widening the specification of production function to include environmental factors such as soil types and rainfall when measuring economic inefficiency. These factors, they
8

Wide variations in profit efficiency were observed, but were skewed to the right.

26

argued, are normally left out by economists, which lead to misspecification of the model. Yet, including environmental factors helps to reduce the usually overestimated technical inefficiency. On the conceptual level, the authors argued that the distinction drawn by economists between allocative and technical inefficiency is not meaningful because it may be dictated by aggregation of purchased inputs and level of application. Most of the studies reviewed by the authors were from Asia, efforts are required to generate information on Africa. As Akinwumi and Djato (1996) observed, studies in Asia cannot be used directly to inform an agrarian policy in Africa.

27

CHAPTER III METHODOLOGY 3.0 Introduction This chapter elaborates on approaches to measure efficiency, discusses theoretical advances to efficiency models explains, justifies and discusses the implementation of the translog model adopted in this study. The chapter concludes by describing the study area, data sources and discusses tests for data reliability and validity.

3.1 Approaches to Measuring Efficiency Following Farrells (1957) work, there has been a proliferation of studies in the field of measuring efficiencies in all fields. But in the field of agriculture, the modeling and estimation of stochastic function, originally proposed by Aigneir et al., (1977) and Meeusen and van den Broeck (1977), has proved to be invaluable. A critical narrative of the frontier literature dealing with farm level efficiency in developing countries conducted by Battese (1992), Bravo-Ureta and Penheiro (1993), Coelli (1995) and Thiam et al., (2001), indicated that there were wide-ranging theoretical issues that had to be dealt with in measuring efficiency in the context of frontiers and these included selection of functional forms and relevant approaches (parametric as opposed to non-parametric).

Parametric and non parametric models differ in two ways. First, the two models differ on assumptions of the distribution of the error term that represents inefficiency. Second, they differ in the way the functional form is imposed on the data. Parametric methods impose

28

functional and distributional forms on the error term whereas the non-parametric methods do not. Nevertheless, parametric models suffer from the same criticism as the frontier deterministic models, in a sense that they do not take into account the possible influence of measurement errors and other noises in the data as do stochastic frontier models (Thiam et al., 2001). The results can also be misleading because they do not allow for random error as in stochastic parametric approaches. Besides, non-parametric methods also lack statistical tests that would tell us about the confidence of the results. For this reason, this study adopts the stochastic parametric model and profit function frontier for rice farmers. Deterministic Versus Stochastic Frontier Models According to Taylor and Shonkwiler (1986), Afriat (1972) was the first to propose the formulation and application of a deterministic production frontier model. The basic structure of the model is: Y= (, ) -...............................................................................................................(6)

where (, ) denotes the frontier production function and is a one-sided non-negative distribution term. This model imposes constraint of 0, which implies output is less than the potential or it is equal to the potential, within the given input and output prices. According to Taylor and Shonkwiler (1986), the model is in full agreement with production theory, but the main criticism against it is that all the observed variations are accounted for by the management practices as pointed out in section 3.1.No account is taken of statistical noise such as random errors, omitted variables and shocks.

29

Stochastic models begin with Aigner and Chu (1968) who proposed a composed error term, and since their work much effort has been exerted to finding an appropriate model to measure technical efficiency. The result was the development of a stochastic frontier model (Aigner, et al., 1977, Meeusen and van den Broeck, 1977, Battese and Corra, 1977). The model addressed the weaknesses of the deterministic model by introducing into the deterministic model to form a composed error term model (stochastic frontier). The error term of the stochastic model is assumed to have two additive components: a symmetric component accounting for pure random factors and a one-sided component that captures the effects of inefficiency relative to stochastic frontier. The model is specified as follows: (, ) -........................................................................................................................(7)

where ( , ), is as defined in (6) and - is error term, represents factors external to the farmer and are assumed to be independently and identically distributed (iid) as (0,2); is half-normal distribution or exponential distribution. The model addresses the weaknesses of the deterministic model. It is also possible to estimate standard errors and test for hypotheses that the observed inefficiency is not due to farmers practices only as suggested in deterministic model (Thiam et al., 2001). Jondrow et al., (1982) provided an explicit formula to separate the two component error term for both half normal distribution and exponential distribution cases. Though this was an improvement over the deterministic model, it was still constrained by lack of a priori justification for the selection of a particular distributional form for the one-sided inefficiency term (Thiam et al., 2001).

30

3.3 Theoretical Profit Function and Stochastic Frontier Model A profit function under mild regularity conditions is a logical extension of the production function (Sadoulet and Alain de Janvry, 1995). Regularity conditions require that the function must be non-negative, monotonically increasing in output, convex and homogeneous of degree zero in all prices. To estimate the profit function, in the neoclassical theory, it is assumed that the farmer is operating on the frontier and the price of inputs and outputs are known. But in reality some of the farmers operate below and some above the frontier.

Furthermore, Junanker (1989) observed that farmers do not always operate in competitive input and output markets in developing countries and this violates the neoclassical assumptions. Since Junankers observation, there have been a number of developments to respond to this criticism. First, the assumption of output and input competitive markets is not needed in defining the firms profit function, especially in developing countries. What is needed is the output and input prices to be exogenous to the farm but be competitively determined (Sevilla-Siero, 1991). Secondly price variation can be handled by including district dummies (Lau and Yotopolous, 1971; Akinwumi and Djato, 1996). Third, it is currently possible to incorporate institutional and environmental factors referred to earlier such as quality of soils and rainfall as shown by (Ali and Flinn, 1989; Coelli ,1995). Fourth, profit function does not suffer from simultaneous equation bias problems as in production function. Fifth, the function has been used before in African context (Saleem, 1988; Akinwumi and Djato , 1996 and 1997). Thus, a stochastic profit function approach is deemed appropriate for this study. This study adopts the Ali and Flinns model specified in equation 8: = (P, Z, D).exp e ....................................................................................................(8) 31

Where = normalized profit of th farm defined as gross revenue less variable cost, divided by commodity prices from farm j. P Z D e = = = = prices of the variable inputs on jth farm, kth fixed factors on jth farm and exogenous variables on jth farm, an error term, and = 1,.n, is the number of farms in the sample.

If equation 8 is estimated using the Ordinary Least Squares (OLS) procedures, an average, instead of best practice frontier is shown by an envelope curve EE (figure 2) given in chapter 2. To attain best practice frontier, an appropriate error structure is appended to equation 8. Following Kmenta (1986), the study by Ali and Flinn (1989) proved that the same error term as that used in production function frontier analysis was relevant to profit frontier. Thus the following error term specified in equation 9 was used: e = - ......................................................................................................................................................................................................(9) where and are random error terms and inefficiency effects of the farm , respectively. When = 0, the firm lies on the frontier but if >0 the farm is profit inefficient and incurring losses.

The inefficiency effects () in equation (9) which are non-negative random variables are assumed to be identically and independently distributed such that is defined by the truncation (at zero) of the normal distribution with a mean of j = 0 + d wd + and variance where
2

d =1

are the variable representing socio-economic

characteristics of farm j to explain inefficiency and 0 and d are the unknown parameters to be

32

estimated. The profit efficiency of the farm in the context of stochastic frontier is given by: j
= E [ exp(- )| e ] = E[exp(- 0+ d dj )] ...........................................................(10) i

=1

where j is profit efficiency of farmer j and lies between 0 and 1 and is inversely related to the level of profit inefficiency. E is the expectation operator. This is achieved by obtaining the expressions for the conditional expectation upon observed value of j.

As pointed out by a number of researchers including Akinwumi and Djato (1996, 1997), a profit function is much superior to production function because first it permits straight forward derivation of own-price and cross-price elasticities and output supply and input demand functions, second, the indirect elasticity estimates via profit function have a distinct advantage of statistical consistency, third, it avoids problems of simultaneity bias because input prices are exogenously determined. Akinwumi and Djato (1997) quoting Quismbing(1994) confirm that problems of endogeneity can be avoided by estimating the profit or cost function instead of the production function. Besides, the profit function is extensively used in literature (Yotopoulos and Lau, 1973; Saleem, 1988; Akinwumi and Djato, 1996 and 1997; Abdulai and Huffman, 2000).

3.4 Empirical Models In this section, a discussion of the empirical models that are used in estimating the profit function, the profit inefficiency model and the determinants of profit inefficiency are presented.

33

3.4.1

Translog Stochastic Frontier Profit Function Model

A number of functional forms exist in literature for estimating the profit function which includes the Cobb-Douglas (C-D) and flexible functional forms, such as normalized quadratic, normalized translog and generalized Leontif. The C-D functional form is popular and is frequently used to estimate farm efficiency despite its known weaknesses (Saleem, 1988; Kalirajan and Obwona, 1994; Dawson and Lingard, 1991; Yilma, 1996; Nsanzugwanko et al., 1996; Battesse and Safraz, 1998). The translog model has its own weaknesses as well, but it has also been used widely (Ali and Flinn, 1989; Wang et al., 1996b). The main drawbacks of the translog model are its susceptibility to multicollinearity and potential problems of insufficient degrees of freedom due to the presence of interaction terms. The interaction terms of the translog also dont have economic meaning (Abdulai and Huffman, 2000).

A solution to these problems would be to estimate both (C-D and Translog) and then use the results of the values of the Loglikelihood at the set critical value to reject or accept one model over the other. Battesse and Safraz (1998) tried both models and found that the C-D production function model was an adequate representation of the data. This study runs both the C-D and translog frontier profit function models. Both of these models have been widely used in Asia and Africa (Ali and Flinn, 1989; Saleem, 1988; Abdulai and Huffman, 2000 and Rahman, 2002, 2003) as earlier noted. Suppressing the subscript j of the farm, the flexible translog profit equation (11) and the inefficiency equation (12) estimated in this study are respectively presented as follows:9

The model is adopted from Rahman (2002, 2003) with some modifications. 34

ln ' = 0 + i ln p i +
i =1
2 2

1 2

i =1

ri k ln p i ln p k + il ln pi ln z l + l ln z l +
k =1 i =1 l =1 l =1

1 2

l =1 q =1

lq

ln z l ln z q + .........................................................................................(11)

where:

= 0 + d wd + ...................................................................................................(12)
d =1

rik = rki for all k , i

'

restricted normalized profit computed for jth farm defined as gross revenue less variable costs divided by farm specific rice price p j .

ln
pi

= =

natural log price of variable inputs normalized by price of output where (for i =1, 2, and 3) so that:

p1 p2
p3
zl

= = = =

the cost of hired labor normalized by price of rice ( p y ) the cost of other inputs normalized by price of rice ( p y ) Imputed cost of family labor normalized by the price of rice ( p y ) the quantity of fixed input ( = 1, 2)

where :
z1 z2
10

= =

land under rice (hectares under rice) for each farm j capital used in farm j (sum of total cost of hoes and pangas)10.

The items were assumed to be used up in one production year therefore no depreciation is necessary 35

= = = = =

inefficiency effects truncated random variable constant in equation 12 variables explaining inefficiency effects and are defined as follows: non-farm employment education extension services credit access experience degree of specialization
, ,

1
2
3

4
5

0, i , rik il , l

lq , 0 and d, are the parameters to be estimated.

3.4. 2 Definition of Variables and Estimation of Profit Frontier Function Table 3.1 shows a list of variables included in profit frontier function (model 11). The variables were picked based on the literature earlier reviewed. Labor is included in the model because it is one of the primary factors of production. It has been disaggregated into cost of hired labor ( p1 ) and imputed cost of family labor ( p 3 ) as done in a number of profit efficiency studies (Ali and Flinn, 1989; Saleem 1988). However, some other studies treat labor differently by aggregating all the labor and normalizing it with the output prices (Lau and Yotopoulos, 1971; Abdulai and Huffman, 2000, Akinwumi, and Djato, 1996 and 1997; Sharma et al., 1999). Ali and Flinn, (1989) treat family labor as

36

fixed factor and hired labor as variable factor. Both cost of hired and imputed cost of family labor is treated as variable cost in this study.

In addition, there is a controversy as to whether men and women should be assigned the same weight when valuing labor. The basic argument is that a females hour of work is not equivalent to a mans hour. The practice has therefore been to value womens hour differently. In a number of cases, it has been put at 75% of mens. Basing on the prevailing average wage rate, Abdulai and Huffman (2000) treated females and childrens labor as equivalent to half of the mans. Moock (1976) in his study in Vihiga district in Kenya and Akinwumi and Djato (1997) in their study on rice in Ivory Cost gave the same weight to both men and women. Similarly, Ali and Flinn (1989) in their study on rice in Pakistan assigned the same value to men and women. Since a consensus is building up in favor of not discriminating between the two, this study treated labor of men and women to be equivalent to 1 person unit; childrens labor was 50% of the adult labor. Variable p 2 (fertilizer and insecticide) use in production enhance productivity

37

Table 3.1: Variable

Variables Included in the Frontier Profit Function Models and their Descriptions Descriptions Expected sign Normalized profit of the j th farm defined as gross revenue less variable cost divided by farm specific price (dependent). Note the j is suppressed.

'

Variables
p1 p2

Normalized cost of hired labor divided by price of rice. Normalized cost of other inputs (fertilizer and pesticides) divided by price of rice.

-ve

-ve

Normalized imputed cost of family labor on the

p3

farm x the prevailing wage rate divided by price -ve of rice.

Fixed factors
z1 z2

Land under rice in hectares on farm j Cost of capital (hoes and pangas) used in farm j

+ve -ve

Fertilizer used on the farm is a variable factor of production. Weir (1999) found fertilizer to have a positive and significant impact on output. However, Rahman (2002) found a weak relationship of fertilizer use and profit efficiency among the Bangladesh farmers. Abdulai and Huffman (2000) registered negative sign for rice farmers in Northern Ghana. In this study, preliminary analysis of the results showed that few people in Uganda used fertilizer and pesticides in rice production. Therefore, other inputs was arrived at by 38

multiplying the quantity of each input by their respective prices and treated as ( p 2 ) (Table 3.1). It is hypothesized that the cost of inputs affects profit efficiency negatively.

Land ( z1 ) is defined as net area covered by rice and was treated as fixed input in line with (Lau and Yotopoulos, 1971). The authors argued that given the periodic nature of agricultural technology, it was reasonable to treat land as a fixed factor in the short run and hypothesized to effect profit efficient positively. Capital ( z 2 ) in this study was derived as the sum total of the cost (using the prevailing prices) of hoes and pangas. It was also assumed that these items would be used up in one season as Akinwumi and Djato (1996 and 1997) assumed for the case of rice farmers in Cote dIvoire.

39

3.4.3

Variables Included in the Inefficiency Model

The variables included the Inefficiency model in equation 12 are presented in Table 3.2. Table 3. 2: Variables Included in the Inefficiency Model and Descriptions. Descriptions Inefficiency effects Intercept term Non farm employment 1= employment = 0 other wise have non-farm ve Expected sign

List of Variables

1
2
3

Education level of a respondent in years -ve Extension service visits to farm j 1= received -ve extension visits 0=otherwise Credit access by farmer j 1=access 0 = otherwise -ve Experience measured by years in rice production -ve by farmer j Degree of specialization in rice (acreage in -ve rice/total crop acreage) in farm j

4
5
6

1 One of the variables included in the model is non-farm employment ( ). It was

included to capture access to extra income, which can then be used to buy, among other items, agricultural inputs and thereby possibly reduce inefficiency. Rahman (2002) included the variable to capture unemployment situation in Bangladesh. However, engaging in non-farm employment may deprive the farm of valuable time to make timely decisions. In this regard, the variable was expected to have a positive impact on inefficiency (Rahman 2002, 2003 and Abdulai and Huffman, 2000, Ali and Flinn, 1989). In this study it is hypothesized to have an indeterminate influence on inefficiency.

40

Through education ( 2 ), the quality of labor is improved and with it the propensity to adopt new technologies. However, education has varying impacts depending on the environments, and has been proposed to be more effective in a rapidly changing technological or economic environment (Shultz, 1964 and 1975). Furthermore, the issue of a threshold is very important in determining what level of education the country should give to its people to raise productivity. Appleton and Balihuta (1996) in their study in Masaka district in Uganda showed the education threshold to be 4 years. In the same district, Yilma (1996) in his study on smallholder efficiency in coffee and food-crop production found it to be 10 years. Earlier, Jamison and Moock (1984) in Nepal found it to be 7 years.

The other education related question pertinent to this study is: whose education matters to agricultural productivity? Many studies capture the education of the head of the household, while others take it for all members of the household. In others, the community education is taken. Appleton and Balihuta (1996) used education of the entire household and found that total years of the farm workers were significant. Weir (1999) found a positive impact of average years of education in the village placement of production frontier. Appleton and Balihuta (1996) did not find community education to be significant.

41

Education is hypothesized to affect inefficiency negatively, and it is captured for the respondent11. Weier (1999) treated years of schooling of the household head separately from years of schooling of other adults in the household. Education is perceived to enhance allocative ability. According to Abdulai and Huffman (2000), this stems from the fact that response to changes in economic conditions requires first, perceiving that change has occurred, second collecting, retrieving, and analyzing useful information, third, drawing valid conclusions from the available information, and fourth, acting quickly and decisively.

Access to extension services ( ) is a conduit for the diffusion of new technology 3 among farmers. Thus it should reduce inefficiency levels among rice farmers through improvement in managerial ability. Ali and Byerlee (1991) review of a number of studies on economic efficiency reported negative influence of extension services on inefficiency. Bravo-Ureta and Rieger (1991) reported a positive relationship between extension services and economic efficiency for the dairy farms in New England, U.S.A. Similarly, Seyoum et al., (1998) studying the impact of SG 2000 project on participating maize producers in Eastern Ethiopia, also reported a negative influence on extension services. Rahman (2002, 2003) reported negative results for the variable, implying improvement among those rice farmers who had contact with extension officers in Bangladesh. Therefore, access to extension services was hypothesized to have a negative effect on profit inefficiency.

11

The debate on education concerns what type of education and whose education This debate is beyond the scope of this report The interested reader can consult (Weier, 1999)

42

Adoption of new methods to increase efficiency does not depend only on availability of technologies; it also depends on whether the farmer has the cash to purchase the recommended inputs. Therefore, credit ( 4 ) should play a crucial role in inefficiency improvement and should have a negative relationship with profit inefficiency. Lingard et al., (1983) found a negative relationship between credit access and inefficiency level in Central Luzzon, Phillipines. Ali and Flinn (1989) reported similar results for Basmati rice farmers in Pakistani. Results of a study by Abdulai and Huffman (2000) results on rice farmers in Northern Ghana also found credit access to be negatively related to profit inefficiency. Thus in this study credit was hypothesized to be negatively related to profit inefficiency.

Experience ( ) in rice production should have a direct relationship with profit 5 inefficiency. As one gets proficient in the methods of production, optimal allocation of resources at his/her disposal should be achieved. Thus the more experienced one is the higher the profit and the lower the profit inefficiency. Bravo-Ureta and Rieger (1991) recorded positive relationship between economic efficiency and experience in a study of dairy farms in New England. Wilson et al., (1998) also found a positive relationship between experience and inefficiency in potato production in UK, implying that farmers with fewer years of experience achieved higher levels of efficiency. The reason may be that those with little experience are likely to seek out for new technology, unlike those with experience. Rahman (2002) also reported similar results for Bangladesh rice farmers. However, the same author registered a negative relationship between inefficiency and experience for the same farmers (Rahman, 2003). In this study we hypothesized a negative relationship between experience and profit inefficiency. 43

Classical economic theory recognizes specialization as a key determinant of efficiency. Specialization implies optimal allocation of resources (time, money and human) in the enterprise to improve productivity. In this case specialization in rice production by rice farmers should lead them to seek better methods of production and hence improvement in
6 profits efficiency. Hence it was hypothesized that specialization ( ) in rice production

had negative impact on profit inefficiency levels.

3.5 Study Area, Data and Sources 3.5.1 Description of the Study Area

The data used in this study were collected from three districts in Uganda, namely Tororo, Pallisa and Lira, in 2001(see Map Appendix C). Tororo and Pallisa are located in Eastern Uganda and Lira is found in Northern Uganda. Tororo district occupies an area of 2,608.7 sq kms of which 8.87% is permanent wetlands. The district had a population of 536,888 with a sex ratio of 94.9.912 in 2002 population census and has one of the highest population growth rates (3.05%). It had a population density of 346 people per sq km, far above the national average of 126 people per sq km (UBOS, 2005). In Tororo district, the data were collected from Doho rice scheme and the surrounding villages.

The Doho rice scheme, which was set up with the help of the Chinese Government, is located in two of the six sub-counties of Bunyole County, that is, Kachonga and Mazimasa sub counties, of Tororo district. The rest of the counties are Budumba, Busaba

12

Sex Ratio = no of males per 100 females or proportion of females in a given population.

44

and Butaleja. River Mpologoma, whose source lies on the slopes of Mt Elgon, drains the area. The slopes have volcanic soils and through soil erosion, river Mpologoma carries them down to the plains of Bunyole to form the wetlands on which rice fields are located. The scheme is located in these wetlands and occupies an area of 828 hectares of land of which 140 hectares is under cultivation. Most of the current rice farmers were allocated the plots free of charge at the time when the scheme was set up while the government provided technical services. The two sub-counties traditionally grew finger millet and cassava as their main food crops and cotton as a cash crop. However, after the collapse of the cotton industry, the two sub counties switched to rice. The rest of the residents of the sub-county resorted to trading in what were formally food crops, such as sweet potatoes, millet and groundnuts.

Pallisa district has an area of 1,991.7 sq. km, of which 337.6sq. kms (16.9%) is permanent wetlands. These wetlands have not been surveyed to establish their potential for rice production. A number of activities are carried out in these wetlands, such as grazing animals, brick making, rice production and vegetable production. Many of these activities are carried out at a micro scale and in an unorganized fashion. With the provision of proper drainage and infrastructure, these wetlands could be turned into major rice fields and could boost rice production for the country.

Until 1991, Pallisa district was part of the greater district called Bukedi. Bukedi constituted the present three districts, namely Tororo, Pallisa and Busia districts. The district had a population of 520,578 in 2002 population census with a sex ratio of 93.0 (UBOS, 2005). As in the case of Tororo, the district is experiencing the highest (3.24%) 45

population growth rates. The district traditionally cultivated finger millet and cassava for food and cotton as a cash crop.

In Lira district, the study concentrated on Olweny rice scheme. Olweny rice scheme, which was opened up much later, is located in Agwata sub-county, Dokolo County, in Lira district. Lira district has an area of 7,200.7sq kms of which 4.13% are permanent wetlands. In 2002 population census, the district had a population of 741,240 persons with a sex ratio of 96.0. The district has one of the lowest population densities of 124 people per sq km. The main food crops were finger millet and sorghum and the cash crop was cotton.

Olweny rice scheme was estimated to have a reclaimable land of 5,000 hectares of which 3,500 hectares is suitable for rice cultivation (CRS, 1982). The main food crops are finger millet, sorghum and maize. The traditional cash crop is cotton. As in Tororo district, cotton also lost out as a cash crop and the majority of the farmers switched to trading in food crops such as sweet potatoes, millet, and simsim.

3.5.2

The Data

A structured questionnaire was used to collect primary quantitative data in the selected households. The three districts (Tororo, Pallisa, Lira) were selected mainly because they have been the major producers of rice in the country and for the period of 8 years (19932000) accounted for 67 percent of the national output (UBOS, 2005). In addition, Lira was specifically included in the sample for two reasons. First, to compare the 46

performance across the three districts and second, it is also an area, which is in search of a viable cash crop, after cotton lost out as pointed out earlier.

The mode of selection of sample size in the three districts was dictated by the presence of rice schemes in Tororo and Lira districts. In the two districts (Tororo and Lira), there was a register of participating rice farmers kept at the schemes office that served as a sampling frame. In case of Tororo district, large-scale farmers, who have opted to divide up their farms in acre plots for rent to small-scale farmers, supplemented the scheme register. Since these large farms are in the same locality near the scheme and in some cases adjacent to the scheme, we felt it prudent to include farmers outside the scheme in the sample. In this case, the landlords keep a register of the participating farmers in their offices and this was used as a sampling frame for these types of farmers.

Olweny scheme rice register had mainly female farmers. This was due to the fact that the scheme had initially, aimed at female farmers. However, at the time of the study, the scheme had started to recruit male farmers, but these were not many. Therefore, in addition to using random methods, purposive random sampling methods were employed to include male farmers13.

In Pallisa, a different approach was adopted since there was no official register in existence of small-scale farmers as in the case of Tororo and Lira districts. The assistance of government officials, namely agricultural officers, was therefore enlisted. The major rice growing sub-counties were identified and within these sub-counties major rice
13

The project recruited only female farmers at the beginning.

47

growing villages were purposively selected. The sub-counties selected were: Butebo, Ikiki and Budaka. The villages selected were Lyama, Katiira in Ikiki sub-county; Nabwali and Bokora in Budaka sub-county. Once the villages were identified, a village register was used to draw the required sample. Where the village register did not exist, the chairman/secretarys assistance was solicited and a fresh register was compiled. The sample was then randomly drawn using random numbers. In total, the survey covered a sample of 297 farmers of which 253 were used in estimating the Translog model. The 44 farmers not included in the model were eliminated as outliers. The sample is distributed as follows: Tororo (138), Pallisa (104) and Lira (55). The unevenness of the sample was dictated by the availability of the required information each variable required to run the translog model.

Data were collected on socio-demographics such as age, non-farm employment, education, extension service visits, credit access, years in rice farming and degree of specialization. For the stochastic frontier profit function, the relevant data collected were: hired labor used in production of rice, fertilizers, and pesticides as other variable inputs and their market prices. Family labor, land and capital were treated as fixed inputs. For output, the relevant data collected included quantity of rice produced, sales, and prices at which rice was sold for one season.

3.5.3

Data Reliability and Validity

In order to control for data reliability and validity, measurement and sampling errors; a number of measurements were effected. The first measure taken was to pretest the questionnaire in two of the districts. The instrument was tested in Tororo and Pallisa 48

districts. This was to ensure that the right questions were asked during the actual field survey. The data obtained during the pre testing exercise were coded and analyzed to gauge the accuracy of the questions. Second, the enumerators were then trained for one week on how to administer the questionnaire through role-play. Third, while in the field, the author participated in data gathering as well as supervising the field team.

Fourth, once data were captured, a number of tests were carried out to ensure getting unbiased estimates. These tests included testing for normality of residuals using the One Sample Kolmogorov-Smirnov test. The results suggest that, some variables did not conform to the assumption of the regression analysis such as normality of the data. The data that violated the normality assumption were transformed by use of logs. The results are presented in Appendix A figures 1a-7b.

Outliers whose observations had large residuals were removed from the analysis such that cases with studentized residuals greater than absolute value of 2 were excluded. In this respect 13 observations violated this criterion and were therefore left out. Observations whose central leverage values exceeded (2k +1)/n were deleted from the analysis (see Appendix A for the print out). Out of a total of 297 cases 17 violated this assumption and were deleted. Observations found to have an absolute value greater than n/4 were deleted from the analysis. A total of 14 observations were deleted using this criterion. When all abnormal cases were deleted 253 observations remained as the final data set.

Curve estimation procedure was employed for each predictor against the response variable to verify for linearity; it was observed that all predictors were significantly and linearly related to the dependent variable (F > 3.84; p < 0.000; R squared > 30.0 %).The 49

Variance Inflation Factor (VIF) method was used to detect multicollinearity and was preferred over the correlation coefficient method which does not give conclusive results. (Pindyck and Rubinfield,1981). No collinearity among the independent variables was detected by the test since their specific values was less than 10.

The test for homogeneity of variance was conducted using Breusch-Pagan/CookWeisberg test for heteroskedasticity , (www.stata.com, accessed 2nd April ,2004) and the null hypothesis of constant variances of the residuals was accepted (p > 0.000).The Ramsey test was conducted to test for omitted variables. The null hypothesis of no omitted variables was accepted (F = 0.761).The Durbin-Watson test for independently and identically distributed errors (iid) was calculated and from its value of less than 2 led the conclusion that there was no problem with iid errors. In addition to the KomogrovSmirnov (K-S) test, the variables were corrected to normality using the skewness test as shown in Table 3.4. The K-S test results are shown in Table3.3 and 3.4.Table 3.3 shows unstadardized statistics while Table 3.4 shows standardized statistics. As the results show in table 3.4 the assumption of normality of the data was upheld as shown by the K-S test which states that the computed figure should not exceed 3. Other than hired wage rate, all the variables included in the model have figures on Kurtosis of less than three (Table 3.4).

The traditional method of increasing reliability of estimates is to increase sample size. Increasing the sample size has its own problems as reported by (Bakan, 1966). After using a very large sample size (60,000 cases) the author observed that all statistical tests were significant. He therefore concluded that there was a possibility that in the analysis conducted, the significant values obtained could be attributed to either sampling errors or 50

the size of the sample. In order to test for this, effect magnitude measures was proposed by (Maxwell & Delaney, 1990). The effect magnitude size is defined as the degree to which the null hypothesis is false (Cohen, 1988). The null hypothesis is that the effect of sample size is equal to zero. The decision is, if the Cohen d statistic is less or equal to 0.2, the null hypothesis is rejected implying observed significance of estimated coefficients were due to large size and sampling errors. Table 3. 3: Statistics Skewness and Normality Variables (Unstandardized)- Translog Model Profit Hired Family Other Rice Capital Experie wage wage inputs Hectare nce rate s (years) 772.6 204.9 18.2 114.3 3879420 12.7 193.5 -311.5 31187.0 313.7 98427 3.4 14.9 .00 2239.3 465.2 29.7 512.4 262597 3.9 25.4 10.0 4860.0 222.0 13.2 223.7 50038 3.4 16.9 10.0 1929.4 1.3 9.9 1.7 2.9 6.1 56.7 .05 20.5 25491.2 1982.2 34160.9 116696724 7 7.6 88.7 00 46000 11.7 .54 9.1 83.3 1.4 2.3 1.0 51.0

Mean S.E. Std Variance in coefficient Skewness Kurtosis Minimum Maximum

Source: computed from Field survey

51

Table 3. 4: Skewness and Normality Variables (Standardized)- Translog Model Source: computed from Field survey Statistics Profit Hired Family Other Rice Capital Experience wage Mean S.E. S.D Variance Coefficient Skewness Kurtosis Minimum Maximum 6.09 .06 1.09 in 1.19 -0.16 1.17 2.3 10.4 1.51 -1.01 3.81 -1.29 7.71 1.14 -0.68 0.85 2.3 8.49 0.81 -0.26 0.07 2.3 7.56 1.04 -0.46 0.45 -3.0 3.02 0.77 0.06 0.48 6.91 13.04 0.75 -0.44 -0.24 0 3.93 rate 4.95 0.08 1.22 wage 5.67 0.06 1.06 inputs 5.02 0.05 0.89 Hectares -0.17 0.05 1.02 9.75 0.05 0.87 (years) 2.14 0.05 0.86

The results of the computed presented Cohens d statistic using results from the MLE (translog model) estimates of equation 11 and 12 are presented in Table 3.5. The test results show that the Cohen static is greater than 0.2 for most of the variables therefore the null hypothesis is accepted for almost all the variables included in equation 11 and12. It is concluded that the probability of the values observed on the estimated coefficients were neither due to chance nor sample size influence, that is, they are true values.

52

Table 3.5 Effect Magnitude Measures for the MLE result Estimates Variables Coefficient t-ratio Cohen d statistic Constant 7.65 15.2 30.4 Cost of hired labor 0.12 1.71 23.42 Other inputs -0.02 -3.27 -6.54 Imputed cost of family labor -0.28 -25.5 -0.51 Rice hectares 0.05 7.26 14.52 Capital 0.13 4.22 8.44 Cost of hired labor x other inputs -0.03 -1.62 -3.24 Cost of hired labor x Imputed cost of family -0.01 -1.71 -3.42 labor Cost of hired labor x rice hectarage 0.02 1.19 2.38 Cost of hired labor x capital 0 0.43 0.86 Other inputs x Imputed cost of family 0 0.04 0.09 labor Other inputs x rice hecatrage 0.13 3.24 6.48 Other inputs x capital 0.03 1.74 3.48 Imputed cost of family labor x rice hectarage -0.04 -1.52 -3.04 Cost of family labor x capital 0.04 2.72 5.44 Rice hecatarage x Capital -0.05 -2.68 -5.36 Cost of hired labor2 0.01 1.84 3.68 Other inputs2 -0.02 -1.23 -2.46 2 Imputed cost family labor -0.01 -1.12 -2.24 2 Rice hectarage -0.01 -4.33 -8.66 Capital2 -0.03 -4.03 -8.06 Constant -2.75 -7.91 -15.82 Non-farm employment -0.32 -1.82 -3.64 Education 0.16 2.67 5.34 Experience 0.32 7.52 15.04 Credit access 0.37 2.37 4.74 Extension services 0.24 1.86 3.72 Degree of specialization 0.23 3.67 7.34 Source: Computed from Survey data The K-S test suggested by Armstrong and Eperjesi (2000, 2002) like the t test, has the p value representing the probability that the observed difference between the two data sets (e.g. Tororo and Pallisa) could have arisen by chance. The criterion is the value lies p < 0.05 or p < 0.10 is used to reject or accept the null hypothesis. Table 3.6 shows that samples from the three districts (Tororo Pallisa and Lira) are independent and do not have the same distribution. 53

Table 3. 6: Tests of Significance using Two-Sample Kolmogorov-Smirnov Test District Kolmogorov-Smirnov Z P values Tororo- Pallisa 2.134 .000 Tororo- Lira 2.842 .000 Pallisa- Lira 2.010 .001 Source: computed from survey data

Data Analysis/Model Implementation The stochastic profit frontier function equation 11 and the Inefficiency equation 12 were estimated using FRONTIER 4.1 computer package. The estimation responded to objectives 2, 3, and 4. The program combines the two-stage procedure into one and effects the maximum likelihood estimates of the parameters of a stochastic profit frontier function. The procedure is as follows: 1) A two-phase grid search of is conducted, with the parameters (excepting 0) set the OLS values and the 0 and 2 parameters adjusted according to the corrected ordinary least squares formula as set in Coelli (1995). Any other parameters (, or s) are set to zero in this grid search. 2) The values selected in the grid search are used as starting values in an iterative procedure (using the Davidon-Flecher-Powell Quasi-Newton method) to obtain the final likelihood estimates (Coelli, 1996a).
2 2 The likelihood function is expressed in terms of the variance parameters, 2 = +

2 and = / 2 (Rahman, 2002, 2003).

This procedure has been hailed as being more superior to the two-stage procedure by a number of researchers because it does not violate the assumptions made about the 54

distribution of inefficiency effects (error terms and ) that they are independently and identically distributed (Battesse and Coelli, 1995; Coelli, 1996b; Abdulai and Huffman, 2000 and Rahman, 2002) .

55

CHAPTER IV RESULTS AND DISCUSSION 4.0 Introduction The purpose of this chapter is to present socio-economic characteristics of the farmers in study area, the econometric results from the frontier profit function and the results of the inefficiency model. The first part of the chapter presents socio-demographic results which respond to objective number one of this study. This is followed by a discussion of results from the log likelihood test presented (LL) presented in Table 4.2. The results show that the C-D model is not an adequate representation of the data and hence, necessitated estimation of a frontier translog model. The presentation of the findings from the CobbDouglas (C-D) model is therefore presented in the Appendix A for any reader interested in this part of the results.

The discussion of log likelihood results is followed by results from the translog models (Equation 11 and the inefficiency model (Equation 12). This was designed to respond to objectives number 2 and 3 .Objective number 2 was designed to estimate a translog profit frontier while objective number 3 examined specific factors influencing inefficiency levels among individual rice farmers.

4.1 Socio Demographic and Socio Economic Characteristics Table 4.1a shows socio-demographic characteristics of the households studied in the three districts. The means of the relevant variables in the three districts were tested to see whether they significantly different. The superscripts indicate their level of significance. Where the superscripts are the same, for example age, which has the same superscript for 56

the three districts, implies that there are no statistical differences. However, if the superscripts are different, this implies that the means tested are significantly.

On the whole, the respondents lie in similar age categories (with the mean of 41 years for both Tororo and Lira and 40 years for Pallisa), but are not significantly different. Although Tororo district has the largest average household size of 9 persons as compared to 8 for Pallisa and Lira districts, again these are not significantly different.

The average land holding for the three districts is 1.77, 2.88 and 2.02 hectares for Tororo, Pallisa and Lira, respectively. Out of this, 70 percent is under crop cultivation in Tororo and Pallisa districts and 58 percent in Lira. The low level of crop cultivation in Lira district could be explained by lack of oxen, which used to be the major mode of cultivation before the ongoing civil unrest in the region. Lira has the smallest area under rice (0.21 ha) as compared to Tororo (0.57 ha) and Pallisa (0.72 ha) as expected and are significantly different from the other two districts under study. This is in line with the fact that Lira is a new entrant into rice production.

The use of family labor in rice production is highest in Pallisa district (219.32 person days/ha) followed by Tororo district (178.27 person days/ha), and Lira district (218.13 person days/ha). This may reflect the level of commercialization in the latter two districts. Indeed, Tororos labor cost per hectare is far higher than that in the other two districts. Input costs that only reflect expenditure on fertilizers, chemicals, and pesticides are again highest in Tororo, with the average of Ushs 55,677 per hectare as opposed to Ushs 41,115 per hectare in Pallisa. 57

Generally, there was at least one plot of rice per household, but Tororo district had slightly more (2) plots than the other two districts that had (1) each (Table 4.1a). This was expected because Tororo was one of the initial districts that benefited from Governments policy of promoting rice production in the country. This is also reflected in experience levels. Tororo district had the highest number of years (15) in rice production followed by Pallisa (12 years) and Lira (5 years) and are significantly different.

Table 4.1a:

Selected Socio-economic Characteristics of Farmers in the study area

Tororo Pallisa Lira Mean Mean Mean a a Age of household head 41.09 40.02 41.22a 1 (years) (1.22) (1.39) (1.43 ) Household size 9.37 a 8.41 a 8.36 a (0.47) (0.49) (0.66) a bc Number of children 6.42 5.36 5.40 ac (0.35) (0.31) (0.38) a b Size of land (Ha) 1.77 2.87 2.02 a (0.12) (0.34) (0.20) Total (Ha) 1.25 a 2.14 b 1.13 a (0.08) (0.20) (0.10) Rice (Ha) 0.57 a 0.72 b 0.18 c (0.04) (0.09) (0.02) Imputed family labour cost 137475.47a 101518.50b 117682.50c Ha (Ush) (44.57) (94.59) (22.86) Hired Labour cost per Ha 57280.66 a 46994.06 a 10374.81 b (Ush) (6035.95) (9010.90) (1575.41) other input cost per Ha 4897.57 a 2557.73 a 2895.13 a (Ushs) (1255.92) (926.12) (852.01) a b Experience (years) 14.71 12.16 4.75 c (0.81) (0.96) (0.30 ) a b Degree specialization 3.23 3.90 3.56c (0.09) (0.11) (0.15) a bd Number of plots 1.72 1.21 1.12 cd (0.09) (0.05) (0.05) n 138 104 55 Source: Computed from Field Survey Data. Superscripts with the different letters are significantly different at the 0.10 level 1 () figures in brackets are standard errors

Characteristics

58

Table 4.1b shows results of some other variables, which could not be presented in the preceding Table because of its categorical nature. Table 4.1b reveals that Tororo district had the lowest level of education with 51 percent of the respondents having attended primary education as compared to Lira, which had 60 percent in the same category, although not significantly different. Credit access by rice farmers in all three district was poor; with maximum of 56 percent in Lira district indicating so. Yilma (1996) had made the same observation for coffee producers in Masaka district who also had poor access. Note that the informal avenue constitutes the largest (77.4%) source of loans for those who accessed them. This may be because these sources are easily accessible with minimum transaction costs and conditions, even though the interest rate could go as high as 50% per year.

The situation of access to extension services is equally very discouraging, with only 20% of the farmers being able to access extension services. Descriptive results indicate very few prospects for non-farm employment in the areas studied. Indeed, only 27% of the respondents indicated that they were engaged in non-farm employment activities. Pallisa district seems to have the best opportunities (31%) among the three districts followed by Tororo (28%) and Lira (18%).

59

Table 4.1b:

Other Household Characteristics in the Study Area Tororo Proportion 0.507 a (0.043) 0.275 a (0.038) 0.217 a (0.035) 0.355 a (0.041) Pallisa Proportion 0.683 bc (0.046) 0.212 a (0.040) 0.106 bc (0.030) 0.481 ac (0.049) Lira Proportion 0.600 ac (0.067) 0.236 a (0.058) 0.164 ac (0.050) 0.564 bc (0.067)

Characteristics

Primary Secondary / Tertiary Not attended Credit Access Credit from informal

0.857 a 0.98 b 0.774 a source (0.045) (0.045) (0.056) a a 0.28 0.31 0.18a Non-farm employment (0.04) (0.04) (0.06) n 138 104 55 Source: Computed from Field Survey Data. Superscripts with the different letters are significantlydifferent at the 0.10 level 1 () figures in brackets are standard errors 4.2 Testing for the Appropriateness of C-D Model The likelihood ratio, sigma-squared, gamma parameters shown in Tables 4.2 and 4.3a through 4.3d presents are results on the behavior of the error term outlined in equation 12. Basically, the statistics are designed to test for efficiency effects in the model and the appropriateness of the model to represent the data. The gamma ( ) and the log likelihood (LL) parameters are employed to test for efficiency and the appropriateness of the model, respectively. The gamma ( ) tests whether the observed variations in efficiency are simply random or systematic. The parameter is defined as the ratio of the unexplained inefficiency error term of ( ) to the total sum of errors, explained ( )
2 2 2 2 2 2 and random ( v ) or = / ( + v ). The gamma ( ) is bounded by 0 and 1,

where if

is zero inefficiency effects are not present in the model, and if it is one 60

inefficiency exists and is not random. In other words, if is not significantly different from zero, the variance of the inefficiency model in equation 12 reduces to average response function (C-D function or deterministic model) in which the inefficiency variables enter directly into the model (Battese and Coelli, 1995).

The results in Table 4.3a through 4.3d show that gamma ( ) is significantly different from zero in all the estimated samples implying that there is profit inefficiency in rice production. The observed variations in profit efficiency among the rice farmers are due mainly to differences in farm practices and characteristics of sampled rice farmers rather than random factors.

The LL test statistic is employed to further provide tests to various restrictions in the model. The restrictions are normally effected according to the researchers demands. In this study, we wanted to establish three positions, namely whether we can use a C-D model instead of any other form of the model, in this case, the translog to estimate equations 11 and 12. We also wanted to know whether farmers were operating on the frontier and whether factors included in model 13 explained the observed profit inefficiency.

The restricted frontier model as specified by the null hypothesis is defined as LL = -2{log [Likelihood (H0)]-log [Likelihood (H1)]} where LL is absolute values between LR and LU. LR (H0) is the restricted frontier (C-D model) function in the null hypothesis;

61

and LU(H1) is the unrestricted ( stochastic) frontier function in the alternate. The test
2 statistic has approximately a v distribution with equal to the number of parameters,

which are assumed to equal zero in the null hypothesis.

The first null hypothesis in Table 4.2 is set to find out whether a C-D model can be employed in estimating the two equations (11 and 12) or a translog specification. The LL test for this hypothesis is conducted using the log-likelihood function values of the estimated stochastic frontier function and the values of the corresponding C-D profit function. The rejection criterion is set at 5% such that when the product of the difference between the two (restricted and unrestricted) LL is greater than the critical value, the value obtained from Kode and Palm (1986) the null hypothesis is rejected in favor of the alternate (the translog specification). When this happens, it implies that the C-D model is not adequate, necessitating the need to use the alternative model, in this case the translog specification model. The second hypothesis tests various restrictions on joint and inefficiency effects included within the two models 12 and 13. In the first instance, we test whether farmers individually are operating at the frontier. If they are, then we reject null hypothesis at the fixed critical confidence level, got from Kode and Palm (1986), in this case 5% and accept the alternate. This would imply that farmers are profit inefficient or are not operating on the frontier.

The third hypothesis is designed to test the contribution of factors included in model 12 to observed inefficiency levels (Table 2), in case hypothesis 2 is rejected. The null hypothesis to test is that factors included in model 12 contribute significantly to observed profit inefficiency levels. Again, at the set critical point, of 5% level, the difference 62

between the LU (model 11 plus model 12 estimated simultaneously) and the restricted LR (inefficiency effects specified in model 12) is greater than the critical value got from Kode and Palm (1986). Hence the null hypothesis is rejected. Rejecting the null hypothesis implies that these factors contribute significantly to explaining the observed profit inefficiency.

The results presented in Table 4.2 indicate that the null hypothesis for adequacy of the CD model form is rejected at 5 per cent level of significance in all models. This implies that we should select the translog specification instead of the C-D specification and hence that is what was selected. The second hypothesis is also rejected indicating that the rice farmers are not operating on the frontier. Similarly, the third hypothesis is rejected implying that the variable play a significant role in explaining the observed inefficiency.

To conclude this section, the three hypotheses tested show that first, the C-D model is not an adequate representation of the data, second, that all farmers are not operating on the efficient profit frontier and third, that in general the variables included in inefficiency model adequately explain the observed variations. Therefore, the next section concentrates on estimating the translog profit function using MLE method.

63

Table 4.2: Hypotheses

Hypotheses Testing for the Models and its Inefficiency Effects Pooled Sample -80.13 -127.44 94.62 13.40 reject H0 -80.13 -186.04 212.22 10.37 reject -80.13 Tororo -19.41 -47.91 57.00 13.40 reject H0 -19.41 -81.97 125.12 10.37 reject -19.41 Pallisa 55.48 26.80 29.68 13.40 Lira 17.14 -10.34 27.48 13.40

1) H0:C-D is an adequate LLU representation of Profit frontier Function LLR LL Critical value* (5%) Decision LLU 2) H0: = 0 = d =0, d Each farm is operating on profit frontier LLR LL Critical value (5%) Decision LLU 3) H0: 0 = d =0, d Variables included in the inefficiency effect model have no effect on the level of profit inefficiency. LLR LL Critical value (5%) Decision

reject H0 reject H0 55.48 17.14 -59.07 211.10 10.37 reject -55.48 -29.77 93.82 10.37 reject 17.48

-166.23 86.10 13.40 reject

-72.15 105.48 13.40 reject

-45.74 101.2 13.40 reject

-18.15 35.29 13.40 reject

Source: Field Survey Data. *the corresponding critical values were obtained from Kodde and Palm (1986) LLR and LLU Likelihood function values of the restricted and unrestricted function, respectively. LL computed Likelihood Ratio Value which is the absolute difference between LLU and LLR multiply by 2

64

4.3 Estimation of Frontier Profit Function: Translog Model As explained in chapter 3, the estimation of the stochastic frontier profit function (objectives 2) was undertaken. The dependent variable was restricted profit from an output of one season. Estimation was done in two phases: first using the pooled sample data and then using district level samples, individually. The pooled sample included all the three district samples in the study sites, namely Tororo, Pallisa and Lira. The results for the pooled sample and the district specific samples are presented in Tables 4.3a through 4.3d. The discussion concentrates on MLE results.

All the estimated coefficients in the pooled sample (MLE) carry the theoretically expected signs in the MLE model and are statistically significant, except in the case of estimates associated with other inputs(Table 4.3a). Similarly, estimates associated with all the five variables were statistically significant in Tororo sub-sample (Table 4.3b). However, estimates on cost of hired labor carried unexpected positive sign. In Pallisa sub-sample, the estimates for all the five variables were statistically significant and carried the theoretically expected signs (Table 4.3c). In the Lira sub-sample, three variables namely other inputs, area under rice and capital were statistically significant (Table 4.3d). However, although estimates on cost of hired labor have negative sign, it is not significant. A comparative analysis of coefficient estimates for the three districts show that costs of other inputs, area under rice and capital are the most influential variables in rice production (Tables 4.3a through 4.3d). The estimates associated with these three variables were statistically significant. Costs of other inputs affect profit efficiency negatively whereas area under rice and capital has the opposite influence.

65

Table 4.3a: Frontier Profit Function among Rice Producers in selected Districts Dependent Variable = Normalized Profit in UgShs OLS Coefficients 7.04 -0.14 -0.01 -0.12 0.17 0.02 -0.01 -0.02 0.02 0.03 -0.08 0.00 0.02 0.05 0.04 -0.03 -0.01 0.02 0.02 0.02 0.03 0.24 p-v1 0.00 0.09 0.74 0.00 0.00 0.07 0.74 0.18 0.41 0.04 0.09 0.89 0.58 0.31 0.20 0.47 0.67 0.41 0.46 0.60 0.13 MLE Coefficients 8.11 -0.12 -0.01 -0.20 0.09 0.02 -0.03 0.01 0.03 0.02 -0.05 0.09 0.03 0.01 0.04 -0.07 0.01 0.01 0.00 -0.04 -0.03 0.27 0.69 p-v1 0.00 0.03 0.67 0.00 0.00 0.05 0.07 0.07 0.15 0.12 0.14 0.07 0.18 0.78 0.07 0.03 0.15 0.44 0.89 0.18 0.01 0.00 0.00 80.13

Constant Cost Hired labor ( p1 ) Other inputs ( p 2 ) Imputed cost of family labor (Ushs/ha) ( p3 ) Rice acreage ( z1 ) Capital ( z 2 ) p1 x p 2 p1 x p 3 p1 x z1 p1 x z 2 p 2 x p3 p 2 x z1 p2 x z 2 p 3 x z1 p3 x z 2 z1 x z 2 p1 x p1 p2 x p2 p3 x p3 z1 x z1 z2 x z2 Sigma-squared Gamma Log likelihood n

166.22 253

Source: Field Survey data. p-v1 are p values computed from t-ratios (source: Abramowitz and Stegan. http://www.graphpad.com) Hand book of Math functions . Imputed cost of family labor also play an important role in profit efficiency in the three districts, except in Lira in which it carries a pervasive (positive) sign. The negative sign on the estimated coefficient for this variable in two districts (Tororo and Pallisa) and the 66

pooled sample imply negative impact on profitability of the rice enterprises. Many of the studies reviewed earlier, such as Ali and Flinn (1989), Abdulai and Huffman (2000) and Rahman (2003) reported similar results.

The estimated coefficients on costs of other inputs, which included costs of fertilizers and seeds and insecticides, are statistically significant and carry a negative sign in all results. Also, the estimated coefficients on capital have the positive expected sign and are statistically significant in all districts (Tables 4.3b through 4.3d).

67

Table 4.3b:

Frontier Profit Function among Rice Producers in Tororo District

Dependent Variable = Normalized Profit in UgShs OLS Variables Constant Cost of hired labor ( p1 ) Other inputs ( p2 ) Imputed cost of family labor (Ushs/ha) ( p 3 ) Rice hectarage( z1 ) Capital ( z 2 ) p1 x p 2 p1 x p 3
p1 x z1 p1 x z 2

MLE Coefficients 7.37 0.14 -0.03 -0.22 0.03 0.08 0.00 0.00 0.03 0.00 -0.09 0.05 -0.01 0.06 0.04 -0.08 -0.02 0.09 0.03 -0.15 -0.01 0.55 1.00 p-v1 0.00 0.00 0.00 0.00 0.00 0.00 0.64 0.55 0.00 0.78 0.00 0.55 0.67 0.05 0.01 0.12 0.00 0.00 0.01 0.00 0.01 0.00 0.00 -19.41

Coefficients 7.25 0.00 -0.05 -0.06 0.14 0.00 -0.00 -0.02 0.07 0.03 -0.05 0.16 -0.01 0.13 0.07 0.17 -0.02 0.03 -0.03 -0.28 -0.62

p-v1 0.00 0.89 0.14 0.18 0.00 0.89 0.84 0.46 0.18 0.41 0.74 0.56 0.88 0.30 0.30 0.46 0.18 0.69 0.47 0.14 0.74

p2 x p 3 p 2 x z1 p2 x z 2

p 3 x z1 p3 x z2

z1 x z 2 p1 x p1 p2 x p2

p3 x p3
x

z1 x z1 z2 z2

Sigma-squared Gamma 55.48 Log likelihood -72.15 n 123 Source: Field Survey Data. p-v1 are p values computed from t-ratios

68

Table 4.3c: Frontier Profit Function among Rice Producers in Pallisa District Dependent Variable = Normalized Profit in Ug Shs OLS Variables Constant Cost of hired wage ( p1 ) Other inputs( p2 ) Imputed cost of family labor (Ushs/ha) ( p 3 ) Rice hectarage ( z1 ) Capital ( z 2 ) p1 x p 2 p1 x p 3
p1 x z1 p1 x z 2

MLE p-v1 0.00 0.46 0.85 0.89 0.00 0.34 0.46 0.47 0.07 0.07 0.78 0.41 0.85 0.86 0.65 0.47 0.03 0.89 0.68 0.03 0.85 Coefficients 6.56 -0.03 -0.01 -0.02 0.04 0.02 0.00 -0.02 -0.01 0.01 0.10 0.16 0.11 -0.08 -0.06 -0.04 0.00 -0.15 0.01 -0.03 -0.02 1.00 p-v1 0.00 0.00 0.00 0.00 0.00 0.00 0.44 0.12 0.44 0.18 0.00 0.15 0.01 0.07 0.00 0.07 0.44 0.03 0.00 0.44 0.14 0.00 55.48

Coefficients 6.51 -0.06 0.00 0.01 0.17 0.05 -0.06 -0.02 0.12 0.07 0.04 0.40 -0.07 -0.02 -0.05 -0.22 -0.04 0.05 0.02 -0.55 0.01

p2 x p 3 p 2 x z1 p2 x z 2

p 3 x z1 p3 x z2

z1 x z 2 p1 x p1 p2 x p2

p3 x p3
x

z1 x z1 z2 z2

Sigma-squared Gamma Log likelihood -45.74 n 91 Source: Field Survey Data. p-v1 are p values computed from t-ratios

69

Table 4.3d:

Frontier Profit Function among Rice producers in Lira District Dependent Variable = Normalized Profit in Ug Shs OLS MLE p-v1 0.89 0.74 0.44 0.05 0.44 0.00 0.62 0.47 0.01 0.00 0.00 0.18 0.00 0.00 0.02 0.74 0.03 0.00 0.09 0.58 0.34 Coefficients 89.25 0.00 -7.43 1.81 8.94 0.30 -0.03 0.06 0.14 -0.01 0.48 2.32 1.77 -0.99 -0.55 -1.60 0.00 -1.21 0.06 0.93 -0.31 0.14 1 p-v1 0.00 0.91 0.00 0.01 0.00 0.05 0.91 0.18 0.30 0.67 0.05 0.00 0.00 0.00 0.00 0.00 0.91 0.00 0.30 0.00 0.00 0.00 0.00 17.14

Variables Constant Cost of hired labor ( p1 ) Other inputs( p2 ) Imputed cost of family labor (Ushs/ha) ( p 3 ) Rice hectarage ( z1 ) Capital ( z 2 z 2 ) p1 x p 2 p1 x p 3
p1 x z1 p1 x z 2

Coefficients 19.64 -3.3 7.64 -9.53 8.68 0.05 -0.36 0.02 0.57 0.30 2.03 -4.19 -3.96 -1.48 -0.87 1.78 -0.15 2.17 0.06 1.60 0.94

p2 x p 3 p 2 x z1 p2 x z 2

p 3 x z1 p3 x z2

z1 x z 2 p1 x p1 p2 x p2

p3 x p3
x

z1 x z1 z2 z2

Sigma-squared Gamma Log likelihood 18.15 n 39 Source: Field Survey Data. p-v1 are p values computed from t-ratios

To determine the level of profit efficiency (objective 3), two hypotheses were examined to determine whether rice farmers were operating on the frontier or not. If not, how far was each farmer operating from the frontier? The response to this question can be 70

gleaned from the value of (). In all cases, the value of ( is close to one indicating that there is inefficiency or that farmers were not operating on the frontier. How far away a given farmer was operating from the frontier is the subject of the next section. 4.4 Profit Efficiency Score Estimates: Translog Model The frequency distribution of farm-specific efficiency scores for the rice farmers is presented in Table 4.4 and figure 9 (Appendix A). The findings show that in the pooled sample, rice farmers achieved on average 66 percent level of efficiency (Table 4.4). Taking separate samples, Pallisa district had the highest mean (75 percent) efficiency levels followed by Lira (70 percent) and Tororo district (65 percent).

Table 4.4:

Frequency Distribution of Farm- Specific Profit Efficiency Index in Studied Areas-Translog Model Pooled Tororo Freq. % Freq. % 27 12 32 31 18 34 99 253 11 5 13 12 7 13 39 100 66.4 04 95 22.6 18 14 11 12 10 13 45 123 15 11 09 10 81 11 37 100 64.7 04 1.00 27.0

Efficiency Index <30 31-40 41-50 51-60 61-70 71-80 81+ n Mean Min Max S.D.

Pallisa Freq. 14 04 00 07 08 04 53 91

% 15 04 0 08 09 04 58 100 74.5 02 1.00 27.3

Lira Freq. 06 05 02 02 00 04 20 39

% 15 13 05 05 0 10 51 100 70.3 10 1.00

30.8

Source: Computed from the Survey Data

A wide variation in the level of efficiency is observed across the three districts ranging from 02 percent to 100 percent. It is worthy noting, however, that this wide variation is 71

not unique to Uganda. Similar results have been reported by other researchers elsewhere. Abdulai and Huffmans (2000) study of rice farmers in four districts in Northern Ghana reported a wide variation in the level of efficiency for rice farmers that ranged from 16 percent to a maximum of 95.5 percent. In Asia where the rice crop has a long history of production and intensive research activities, researchers obtained similar results. Ali and Flinn (1989) obtained a minimum of 13 percent and a maximum of 95.5 percent for rice farmers of Gujranwala district, Pakistan. Other authors, including Ali and Sha (1994), and Wang et al., (1996b) for Punjab Pakistan, North-west Pakistan, and China, respectively registered similar variations. While Wang et al., (1996b) reported efficiency levels ranging from 6 per cent to 93 per cent with a mean of 62 per cent and Ali et al., (1994) registered a mean profit efficiency of 75 per cent with a range of 4 per cent to 90 per cent. Studies by Rahman (2002, 2003) that covered rice farmers in Bangladesh also reported a wide variation in profit efficiency, ranging from 3.3 percent to 93.7 percent with a mean of 60 per cent for modern Aman14 rice. These similarities may be a reflection of the low level of economic transformation of many of the third world peasant economies where rice is grown.

The distribution of the efficiency in the translog model as illustrated in figure 9(Appendix A) show skeweness to the right in all cases, implying that in all the districts, the sampled rice farmers, although operating below the frontier, over 30 percent are operating close to the frontier.

14

Aman is one of the three seasons in Pakistan which is the monsoon .The other two are Aus and Boro which are dry season. These two seasons were combined and the efficiency recorded varied from 11.5 percent to 92.2 percent (Rahman, 2003).

72

The inefficiency translated into a profit loss ranging from Ug shs 74,261 to 509,871 with a mean of Ug shs 137,741 per acre per season for the pooled sample (Table 4.5). Tororo experienced the highest loss (Ushs 489,692) followed by Pallisa (Ugsh 301,571) and Lira (Ushs 364,162). Further, the Kolmogorov-Smirnov test showed that there were significant differences between the three districts in terms of mean profit loss. The most significant difference was that between Tororo and Pallisa (Table 4.6).

Table 4.5

Comparison of mean Profit loss per hectare as a result of Profit Efficiency by Districts Minimum 469514.00 281271.00 74261.00 74261.00 Maximum 509871.00 321664.00 154347.00 509871.00

District Mean n Std. Deviation Tororo 489692.50 123 11719.3852 Pallisa 301571.00 91 11915.3694 Lira 114304.00 39 24029.4971 Total 364162.03 253 137741.79 Source: Field Survey Data. Table 4.6

Tests of Significance of Mean Profit loss P-values .000 .000 .000

District Kolmogorov-Smirnov Z Tororo- Pallisa 7.232 Tororo- Lira 5.442 Pallisa- Lira 5.225 Source: Computed from Field Survey Data. 1 used Kolmogorov-Smirnov Test

In addition to profit loss, the degree of responsiveness or price elasticity was computed and results are presented in Table 4.7. For example the results show that a 1% increase in cost of other inputs in Tororo district would result into a decrease of 0.03% in profits. A 1% increase in area under rice in the same district would result into 0.03 percent increase of profit. Table 4.7 Estimated Profit Elasticities in the studied Area 73

Prices and Fixed Inputs With respect to Cost of hired labor other inputs Imputed cost of family

Tororo Price p-v1 elasticity 0.14 -0.03 -0.22 0.00 0.00 0.00

Pallisa Price p-v1 elasticity -0.03 -0.01 -0.02 0.00 0.00 0.00 0.00 0.00

Lira Price elasticity 0.00 -7.43 1.81 8.94 0.30 0.91 0.00 0.01 0.00 0.05 p-v1

labor Area under rice 0.03 0.00 0.04 Capital 0.08 0.00 0.02 Source: Computed from profit function results p-v1 are p values computed from t-ratios

4.5 Determinants of Firm-Specific Profit Inefficiency in Rice-Translog Model In line with objective number 3, estimated results based on model 12 are presented in Table 4.8. The purpose was to determine factors that explain profit inefficiency. The variables included in the model were in line with theory as explained in chapter 3. These are: non-farm employment, education, experience in rice growing, degree of specialization, access to credit and extension services.

The presentation of the results is by variables. Results on the non-farm employment variable carry opposite signs in the three districts. This is in line with what was hypothesized in chapter 3. In two districts (Pallisa and Lira), the estimated coefficients carry a negative sign and are statistically significant. Abdulai and Huffman (2000) reported similar results for rice farmers in Northern Ghana. Ali and Flinn (1989), Wang et al., (1996b) and Rahman (2002, 2003) reported similar results for farmers in Pakistan, China and Bangladesh, respectively. The results indicate that having non-farm work provides the income to buy inputs needed to raise productivity, and hence reducing

74

inefficiency. On the other hand, in Tororo district, the estimated coefficients carry positive sign, but not statistically significant. Table 4.8: Determinants of Farm-Specific Inefficiency in Rice Production in the Sampled Districts

Dependent Variable = Inefficiency Pooled Coeff p-v1 2.09 0.0 Constant ( ) 0 Non-farm employment ( 1 ) Education ( 2 ) Extension services( ) 3 Credit access( 4 ) Experience ( ) 5 Degree of specialization( 6 ) 0.37 -0.14 -0.16 -0.25 -0.07 -0.08 0 0.0 1 0.0 0 0.0 0 0.1 5 0.6 7 0.0 0

Tororo Coeff p-v1 3.84 0.0 0.06 -0.25 -0.26 -0.55 -0.44 -0.14 0 0.8 9 0.0 0 0.0 0 0.0 5 0.1 7 0.0 0

Pallisa Coeff p-v1 3.51 0.0 -0.95 -0.16 -0.47 -0.46 -0.49 -0.15 0 0.0 3 0.0 0 0.0 0 0.0 3 0.0 0 0.0 0

Lira Coeff p-v1 2.28 0.0 -0.93 -0.30 -0.28 -0.24 0.06 -0.03 0 0.0 0 0.0 0 0.0 3 0.4 4 0.7 8 0.1 7

Source: Field Survey Data p-v1 are p values computed from t-ratios

75

The results also show that the estimated coefficient on education is negative and statistically significant in all the districts, indicating reduction in profit inefficiency. This implies that to an extent more education brings about decrease inefficiency (increase in efficiency) in rice production. These results are consistent with Lockheed et al., (1980), Ali and Byerlee (1991), Ali and Flinn (1989), Bravo-Ureta and Rieger (1991), Abdulai and Huffman (2000) for rice farmers in Ghana and Wang et al., (1996b) for China. Thus, giving education to rice farmers in particular would be very beneficial in terms of reducing inefficiency in rice production. Reduction in profit inefficiency will enhance the Governments policy on commercialization of agriculture and poverty eradication.

The estimated coefficients associated with the extension services are significant in all districts. These results show that access to extension advice by rice farmers help to reduce the profit inefficiency in rice production. The results are also consistent with findings obtained by other researchers (Bravo-Ureta and Rieger, 1991; Seyoum et al., 1998; Rahman, 2002). These results therefore serve to emphasize the role of extension services in reducing profit inefficiency in rice production.

Access to credit is expected to ease the financial constraint, enhance the acquisition of the much-needed inputs, and improve revenue and subsequently profits. Indeed, the results of the Tororo and Pallisa districts show that access to credit is a significant factor in reducing inefficiency in profits. The estimated coefficients associated with experience carry the expected negative sign and are statistically significant at 10 percent level in all the districts. The studies reviewed by Ali and Byerlee (1991) reported similar results and 76

Rahman (2002) registered similar results for Bangladesh rice farmers. Sharma et al., (1999) studying allocative and economic efficiencies in swine production in Hawaiian farmers had similar results. The results imply that those with experience will be better performers than those without. Whereas this is so, education and quality of extension services given to the farmers would supplement or in some cases substitute it.

When a farmer specializes in rice production, all his/her efforts in terms of accessing information on new technology should translate into improvement of efficiency in the production of the crop. In this study a negative and statistically significant relationship between the degree of specialization in the rice crop production and profit inefficiency was observed in districts of Tororo and Pallisa implying efficient allocation of resources in these districts. Abdulai and Huffman (2000) registered similar results for rice producers in Northern Ghana. However, in Lira the results are not significant, possibly because the crop is still new and as such specialization in the crop is low.

4.6 Key Constraints to Profit Efficiency in Rice Production The previous section examined factors contributing to profit inefficiency among rice producers. The main purpose of this section is to analyze key factors contributing to profit inefficiency in rice production loss. Section 4.4 showed that rice farmers lost on average Ug shs 302,744 per hectare due to inefficiency in rice production. However, the results did not clearly indicate which of the variables led to this profit loss. The results presented in Tables 4.9a through 4.9c shed some light on this. The discussion combines all the 3 Tables discussing variable by variable. It was noted that education plays a key 77

role in actual profit made by rice farmers. The loss in profit by farmers by not going to school ranged from Ushs131, 000 to Ushs 524, 000 per hectare in Tororo district and Lira, respectively (Table 4.9a-4.9c). These results also show that farmers with tertiary education significant gain of revenue throughout the three districts. The loss was 71,000/=, 245,000/=, 286,000/= per hectare in Lira, Pallisa and Tororo, respectively. Similarly, the revenue loss between those who have had primary education and tertiary education are significantly different. However, there were no significant differences in loss of revenue between those who are literate and those who have This implies that we require farmers to go beyond primary leaving level to be more effective in utilizing opportunities to improve revenue in rice production.

In terms of efficiency, education s translated into efficiency reduction. For example, in Tororo district, those farmers who had tertiary education were more efficient (50%) than those who had no education (46%) (Table 4.9a) and these differences are significantly different. Similar results are observed across the two remaining districts.

78

Table 4.9a Profit Loss in Rice Production in Tororo District by Key Constraints Parameters n 24 61 33 5 43 80 22 101 34 55 34 Actual Profit 000s Shs Mean 968.6 a (3.8) 962.6 a (2.3) 953.5 b (3.2) 637.0 c (10.6) 1104.0a 1069.5b 1114.4a 1118.1b 815.4a (17.6) 1081.4 b (20.4) 1281.4 c (17.6) Profit loss 000s Shs Mean 524 a (3.7) 489 b (2.3) 451 c (3.2) 286 d (8.1) 467 a 523 b 487 a 549 b 515.5 a (2.7) 514.8 a (3.5) 507.1 b (5.5) Efficiency Scores Mean 45.9 a (2.5) 49.2 b (1.6) 52.7 c (2.1) 50.1 d (5.5) 57.7 a 51.1 b 56.3 a 50.9 b 59.1 a (0.5) 57.4 a (0.9) 55.3 b (1.2)

none Education Level primary secondary tertiary Credit Access Extension services Degree of specialization Yes No Yes No 0-25% 25.1-50% 50.1+

Source: Field Survey Data. Different superscripts along columns depict significant differences at the 0.10 level. The results on access to credit help to reinforce earlier observation that those who had access to credit in all districts experienced least loss in profit as compared to those who didnt have. The farmers loss ranged from Ushs 73,000 per hectare (Lira district) to Ug shs 467,000 (Tororo district) (Table 4.9a-4.9c). These results indicate that having access to credit would improve profit efficiency in Tororo and Pallisa districts from 42% to 56% and 47% 53%) for Pallisa. Efficiency would increase from 50% to 53% in Lira. Note that these improvements are statistically significant at 10% level.

79

Table 4.9b Profit Loss in Rice Production in Pallisa District by Key Constraints Education level Nil Primary Education Secondary Tertiary Credit Access Extension access Degree of specialization Yes No Yes No 0-25% 25.1-50% 50.1%+ n 8 64 17 2 41 50 8 83 36 33 22 Actual profit 000s Ushs 672.4 a (7.4) 659.5b (2.2) 574.6 c (4.6) 523.5d (29.2) 615.6a 586.8b 711.9 a 645.8b 1091.3a (17.6) 1137.8a (20.4) 924.8b (18.5) Profit loss 000s Ushs 351 a (6.3) 337 b (2.2) 281c (4.3) 245 d (12.6) 269 a 338 b 294 a 361b 513.9a (4.1) 508.2b (4.7) 499.8c (6.9) Efficiency 47.8 a (1.5) 48.9 a (0.5) 51.1 b (1.1) 53.2 a (3.1) 56.3 a 42.4 b 58.7 a 44.1 b 57.9a (0.9) 57.1a (1.2) 55.1b (1.5)

Source: Field Survey Data. Different superscripts along columns depict significant differences at the 0.10 level. Similarly, farmers who had access to extension services were significantly better off than those without in that they experienced lower revenue (Table 4.9a-4.9c). Their loss ranged from Ushs 87,000 (Lira district) to Ushs 487,000 per hectare (Tororo district). However, unlike in education, there would be marginal improvement in increase of profit efficiency level through access to extension services. Tororo and Pallisa would experience improvement of 5points from (51% to 56% in Tororo district and 44% to 59% in Pallisa district) whereas Lira, farmers had the least increase (48% to 50%).These differences are statistically significant at 10% level. The reason could be that even though farmers had

80

access to extension services, the messages may have not been relevant, since messages were not crop specific. Until very recently, NARO, the organization charged with the responsibility of generating technologies for farmers had in the past attached low priority to rice, especially paddy. In terms of policy, there is need to realize that there is a very big percentage of farmers depending on this crop for their livelihood. The crop therefore deserves special attention in terms of budget allocation (research as well as human) to generate high yielding varieties for dissemination. Results indicate that there are statistical differences between farmers who specialize in rice in the three districts.

Table 4.9c

Profit Loss in Rice Production in Lira District by Key constraints

Educati on level Nil n 8 23 6 2 21 18 20 19 15 17 7 Actual profit 000s Ushs 199.1a (4.9) 237.3b (2.6) 191.3 a (6.1) 153.7c (20.0) 155.0 a 307.5b 174.7 a 323.8 b 1151a 1056a 975.5b Profit loss 000s Ushs 131 a (2.4) 126 b (1.4) 97c (2.8) 71 d (4.9) 73 a 155 b 87 a 169 b 510 a 514a 521a Efficiency 34.2 a (1.7) 46.9 b (1.0) 49.3 c b (2.0) 53.8 c d (3.4) 52.9 a 49.6 b 50.2 a 47.8 b 59.4a 58.2a 57.6b

Education

Primary Seconda ry Tertiary

Credit Access Extension access

Yes No Yes No 0-25%

Degree of specialization

25.150% 50.1+

Source: Field Survey Data. Different superscripts along columns depict significant differences at the 0.10 level.

81

4.6 Summary This chapter presented both descriptive and econometric results. Since the likelihood test statistic showed that the translog model was a better choice, the study concentrated on presenting and discussing results from the translog model. These results showed that area under rice and capital had positive effect on profit levels in all the three districts. On the other hand, costs of other inputs exerted negative influence on profit efficiency in the three districts under study. In addition, imputed cost of family labor played an important role in Tororo and Pallisa districts only.

The next level of analysis concentrated on computing farm-specific profit scores. The results revealed a wide variation in profit efficiency among sampled rice farmers ranging from 02 per cent to 100 percent in the three districts. Pallisa district had the highest score of profit efficiency with a mean of 74.5 percent level of efficiency and the percentage of the farmers who scored above 61 per cent was 71.4. Pallisa was followed by Lira with a mean score of 70 per cent and Tororo with a mean score of 64.7 percent. The potential for improving profit efficiency in production of rice by adopting the best frontier technology is possible in all districts. For example the rice farmers in Tororo need to improve on agronomic practices by 36 percent to move to the frontier while in Pallisa and Lira the required improvement would be 25 and 30 per cent, respectively. The next level of analysis examined factors contributing to observed level of inefficiency.

Six inefficiency effects (variables) were included in the inefficiency model. These were non-farm employment, education, and access to extension services, credit access, and 82

experience in years of production of rice and degree of specialization in rice production. The findings showed that overall; education and access to extension services were the most influential variables on profit levels. These two variables act as a catalyst in improving efficiency. The degree of specialization in rice production had a negative influence in Tororo and Pallisa districts only.

A deeper analysis of the constraints showed that farmers on average lost Ushs 302,744 per hectare due to inefficient production. However, the results indicated that education, credit and extension access contributed significantly to explaining the observed level of inefficiency. Credit access is hence important for maintaining high profits. Farmers lose most profit when they lack education. They lose up to Ushs 524,000/=, 467,000 and 487,000 per hectare for lacking education, access to education and extension services, respectively.

83

CHAPTER V SUMMARY, CONCLUSIONS AND POLICY RECOMMENDATIONS 5.0 Introduction This study set out to evaluate profit efficiency levels among rice producers in Eastern and Northern Uganda. Since rice is the number one crop in terms of returns to labor and is inceasingly becoming a major food crop, especially in institutions such as schools and hospitals, its profit efficiency levels are very crucial to the attainment of both food and income security for farmers.

Rice has faced stagnating yield levels in the past five years (1997-2004), possibly reflecting low research priority previously accorded to the sub-sector. Thus, the main questions addressed in this study were first, whether rice farmers are producing at the production frontier. Second, if not how this translates into profit levels, and exploration of the main determinants of profit levels in rice production. Third, an assessment of factors affecting profit efficiency levels among rice producers in Eastern and Northern Uganda. Fourth, an investigation of how much profit was being lost due to allocative and technical inefficiency.

Chapter one gave the background information of the whole study and stated the main objectives. In chapter two, the theoretical exposition of the economic efficiency measurements and the theoretical modeling was undertaken.

84

Generally, the literature reviewed in chapter two highlighted the fact that there are two main strands in the theoretical developments in frontier modeling to handle efficiency measurements. These are frontier and non-frontier approaches. The frontier approaches aim at locating the best practice profit function and through either deterministic or stochastic models, farmer-specific profit efficiency levels are estimated. In non-frontier approaches, location of best practice isoquant is not required and non-parametric methods such as mathematical programming or Data Envelopment Method can be used to estimate a profit function and inefficiency levels.

The literature reviewed also highlighted the fact that the efficiency levels estimated depend on the approach used. In the case of deterministic models, all the observed inefficiencies are attributed to differences in farmers practices, whereas in the stochastic model, there is an error term that is split between the observed ( ) and the unobserved ( ) components. The observed inefficiency ( ) is interpreted as inefficiencies due to technical and allocative inefficiencies of individual farmers and the unobserved ( ) is attributed to random factors, such as weather and policy changes.

In the second part of chapter two, theoretical modeling for the study is discussed. This study discussed the stochastic frontier translog profit function model by Ali and Flinn (1989) and Rahman (2002 and 2003). Briefly, the model states that the normalized profit of farm j (defined as gross revenue less variable cost divided by farm specific price for rice) is a function of prices of variable inputs, fixed inputs, dummy variables for

85

exogenous factors and an appropriate error term. This model was specified fully together with the inefficiency model and estimated simultaneously.

In chapter three we discussed the empirical model, data sources, and definitions of variables. The data used were collected from three districts namely Tororo, Pallisa and Lira from a sample of 253 rice farmers. The study adopted a translog model suggested by Ali and Flinn (1989) and Rahman (2002 and 2003). The variables used in the empirical model were; normalized profit on the jth farm as a dependent variable, imputed cost of family labor, normalized cost of hired labor and cost of other inputs as variable costs. Rice hectares and capital were modeled as fixed factors. In the inefficiency model, the variables were; profit inefficiency for jth farm as a dependent variable and the independent variables were non-farm employment, educational level, extension services, credit access, years of farming (experience), and degree of specialization.

The maximum-likelihood estimates of the parameters of the translog stochastic function and inefficiency effects were estimated by using FRONTIER 4.1; Coelli (1996a). This package estimates the two models simultaneously. To test for the functional form, the respective log likelihood tests were computed and the results compared with the critical values obtained from Kodde and Palm (1966). The tests carried out were to assess whether there was a need of using a more sophisticated model such as a translog instead of the C-D model. Also tested, were various restrictions on the role of parameters and inefficiency effects. Furthermore, profit efficiency levels were computed followed by the estimation of profit losses due to profit inefficiency. Based on the estimates of the profit 86

frontier function, the basic features of production structure were computed, namely profit elasticity with respect to price of inputs.

Summary The results of the Log likelihood tests showed that the C-D model was not the right model necessitating the adoption of a frontier translog model. Thus the profit frontier translog model was estimated in this study.

The analysis from the translog model showed that the variables rice hectarage and capital had a positive influence on the profit levels while imputed cost of family labor and costs of other inputs had a negative effect on profit efficiency levels in all studied areas.

The analysis of profit efficiency levels revealed that the sampled farmers from all the three districts (Tororo, Pallisa and Lira) were not operating at the profit frontier. They had different levels of efficiency, with a wide variation (2%-100%) and a mean of 66 percent. Pallisa district scored the highest mean level (74.5%) with about 70 per cent of the farmers scoring 61 per cent and above. These results imply that Pallisa district had the highest potential for improving efficiency levels or moving the farmers to operate on the profit frontier, in this case by 25 per cent. The efficiency levels in the remaining two districts were 65 per cent and 70 per cent in Tororo and Lira districts, respectively.

87

Further analysis of profit loss due to allocative and technical inefficiency showed that farmers in Tororo experienced the highest loss (Ushs 489,692/=) followed by Pallisa (Ushs 301,571/=) and Lira (Ushs 114,304/=). The elasticity estimates showed that, reduction on costs in other costs, expansion in area under rice, and investing in capital would have the greatest positive impact on profits.

In analyzing the sources of inefficiency of rice farmers, six factors were identified. These were non-farm employment, education, experience, access to credit, access to extension services and degree of specialization in rice growing. Non-farm employment had opposite signs in the three districts, but in two districts (Pallisa and Lira) had negative sign and statistically significant. These results implied that those who accessed non-farm employment in these districts could have used the income earned to purchase inputs to increase productivity and hence reduce inefficiency in rice production.

Lack of education was found to have an impact on profit inefficiency levels in all the three districts. Lack of it contributed to the loss of profit efficiency by as much as 10 points . The significant differences were observed between those who are illiterate and those who have at least primary education. The implication is that to improve efficiency in rice production primary level of education is absolutely necessary.

Lack of extension services was found to be statistically significant and influencing profit inefficiency negatively in all districts. These results reinforce the already acknowledged

88

view that extension access is a necessary lubricant to adoption of new technology, which has positive impact on profit efficiency. The most crucial point is to pass on the relevant messages to farmers. In the crop under study case, such information was found to be limited as the crop has in the past received very low priority in terms of budget allocation (research and human). It is recommended that NARO- an organization charged with the responsibility to carry out research on crops such as rice to refocus its efforts on rice. In addition, the organization should foster linkages with private sector to come in where they are weak, particularly in providing credit or supplying seed.

The degree of specialization in rice production was found to be an important variable in influencing profit inefficiency levels in Tororo and Pallisa districts. The results tend to suggest that specialization in rice production reduces profit inefficiency in Tororo and Pallisa districts implying efficient use of resources on specialization. In these two district it would pay for the farmers to specialize the crop.

Credit access was found to be a significant factor in reducing inefficiency in Tororo and Pallisa districts. Farmers who did not have access to credit lost as much as Uganda Shs 549, 000 per hectare in Tororo district and 338,000/= in Pallisa district. The efficiency levels were 6 points below those who accessed in both districts. Experience as a factor was important in Tororo and Pallisa districts. Farmers with experience in rice production were more efficient than those without.

89

5.2 Conclusions and Policy Recommendations The main objective of this study was to estimate a frontier profit function for rice producers in Eastern and Northern Uganda. The subsidiary objectives were to determine farm-specific profit efficiency levels and explain inefficiency levels observed. The study results from the frontier profit function showed that the major variables affecting profit efficiency were imputed wage of family labor and area under rice. Imputed cost of family labor had a negative influence on profits whereas area under rice had the opposite effect.

These results, therefore, imply that in order to improve profit levels in rice production there is need to increase area under rice and reduce family labor in rice production. Currently, rice farmers in the study area are operating an average of 0.57, 0.72 and 0.21 hectares in Tororo, Pallisa and Lira districts, respectively. But area expansion may imply increasing family labor, which negatively affects profit efficiency. Thus, this suggests that land-augmenting technologies such as improved seeds would be the most appropriate approach. This serves to re-emphasize the need for research stations to strengthen the breeding programs in order to come up with high yielding varieties for release to rice farmers. The recent release of upland varieties (NARIK 1, 2, 3) is hence very encouraging (NARO, 2003). Similar efforts are required for lowland rice.

The study results also showed that the majority of rice farmers were not operating on the profit frontier, given the technology and that there was potential to do so by eliminating the observed inefficiencies. The variables found to explain the profit inefficiency levels among rice producers were lack of extension services and low educational levels. 90

Therefore, if rice farmers have to reduce profit inefficiency, which implies moving to the profit frontier, access to extension services by rice producers must be improved. The results in this study help to reinforce the governments policy of bringing extension services nearer to the farmers.

In the context of rice production, the Ministry of Agriculture Animal Industry and Fisheries (MAAIF) and NARO should not only focus research on upland rice, but also on low land rice in terms of instituting a breeding program for new varieties of rice. Without this, extension officers will have limited knowledge to disseminate to rice farmers. Also, given the low educational levels in all the three districts, the Government policy on Universal Primary Education (UPE) requiring all school going children to attend school is in the right direction. However, this policy may not succeed unless special strategies are put in place to discourage children from dropping out from school due to the urge to make quick cash in rice gardens.

Lastly, to reduce profit inefficiency levels in all districts the issue of non-farm employment must be tackled. One way is create employment opportunities outside the farm. This would enable some of the farmers to access jobs from which they can earn income. The income would be used to purchase inputs to use on the farm and hence improve on productivity. Alternatively, they need not stay on the farm, but could combine non-farm activities effectively by employing labor-saving technologies to ease the weeding and bird-scaring burdens. This implies that the concerned organizations should come up with varieties which are less palatable to the birds. This would help to 91

reduce on the drop out of school going as these are the ones who are mainly employed to perform this ardors task.

5.3 Recommendations for Further Research This study is one of a few studies the author is aware of that has pursued a rigorous analysis of efficiency issues in the production of agriculture commodities in selected districts of Uganda. Appleton and Balihuta (1996) studied the impact of education on agriculture and Yilma (1996) studied technical efficiency issues in Masaka district. Therefore, there is need to replicate the current study in other districts for the whole country. With the country well underway in implementing the Plan for Modernization of Agriculture, which emphasizes commercialization of agriculture, it may also be necessary to conduct such studies for other crops in Uganda.

Furthermore, this study covered lowland rice areas only. Hence, the scope needs to be widened to cover the upland rice. Secondly, to know what is happening to efficiency levels, we need good panel data on the crop to trace the impact of the technology generated on productivity. This can only be possible when longitudinal studies are carried out systematically. Finally, this study limited itself to commodity production issues, yet marketing and consumption issues are equally important in improving profit efficiency of rice farmers. Thus a study to examine rice marketing issues is pertinent.

92

REFERENCES

Abdulai, A., and Huffman, W. (2000). Structural adjustment and economic efficiency of Rice Farmers in Northern Ghana. Economic Development and Cultural Change, 504-519. Abramowitz, M. and Stegan I.A, Hand book of Math Functions http://www.graphpad.com Afriat, S.N. (1972). Efficiency estimation of production function. International Economic Review, 13, 568-598. Agricultural Policy Committee APC (1997) Report on Economics of Crops and Livestock Production , Agricultural Policy Secretariat, Ministry of Planning and Economic Development. Aigner, D.J., and Chu, S.F. (1968). On estimating the industry production function. American Economic Review, 58, 826-839. Aigner, D.J., Lovell, C.A.K., and Schmidt, P. (1977). Formulation and estimation of stochastic frontier production function models. Journal of Econometrics, 6, 21-37 Akinwumi, A., and Djato, K. K. (1997). Relative efficiency of women as farm managers: Profit function analysis in Cote dIvoire. Agricultural Economics, 16, 47-53. Akinwumi, A., and Djato, K. K. (1996). Farm size, relative efficiency and agrarian policy in Cote dIvoire: Profit function analysis of rice farms. Agricultural Economics, 14, 93-102. Ali, M., and Byerlee, D. (1991). Economic efficiency of small farmers in a changing world: A survey of recent evidence. Journal of International Development, l3 (1), 1-27. Ali, M., and Flinn, J. (1989), Profit efficiency among Basimati rice producers in Pakistan Punjab. American Journal of Agricultural Economics, 71(2), 303-310. Ali, F., Parikh, A., and Sha, M.K. (1994). Measurement of profit efficiency using behavioral and stochastic frontier approaches. Applied Economics, 26, 181-188. Appleton, S., and Balihuta, A. (1996). Education and agricultural productivity: Evidence from Uganda Journal of International Development, 8, 415-445.

93

Armstrong, R.A., and Eperjesi, F. (2000). The use of data analysis methods in Optometry: basic methods. OT40; 12: 36-40. Armstrong, RA. Armstrong. R.A., and Eperjesi, F. (2002). The use of data analysis methods in optometry: comparing the difference between two groups. OT41; 2: 27-31. Bakan, D. (1966). The test of significance in psychological research. Psychological Bulletin, 66(6), 423-437. Battese, G.E.(1992). Frontier production functions and technical efficiency: A survey of empirical applications in agricultural economics. Agricultural Economics, 7, 185208. Battese, G.E., and Coelli, T.J. (1995). A model for technical inefficiency effects in a stochastic frontier production function for panel data. Empirical Economics, 20, 325-332. Battese , G.E., and Corra, G.S. (1977). Estimation of a production frontier model: With application to pastoral zone of Eastern Australia. Australian Journal of Agriculture Economics., 21, 169-179 Battese, G.E., and Safraz, H. (1998). Technical efficiency of cotton farmers in Vehari District of Punjab Pakistan. Centre for Efficiency and Productivity(CEPA Working Papers.No8/98). Department of Econometrics, University of New England , Armidale, Australia. (Working Papers no. 8/98), ArmidaleCetre for Efficiency and Productivity Analysis (CEPA). http:/www.une.edu.au/econometrics/cepawp.htm Belbase, K., and Grabwoski, R. (1985). Technical efficiency in Nepalese agriculture. The Journal of Developing Areas, 19, 515-520. Binger, B.R. and Hoffman, E. (1998). Microeconomics with Calculus. Massachusetts: Addison-Wesley. Reading,

Bravo-Ureta, B.E., and Pinheiro, A.E. (1993). Efficiency analysis of developing country agriculture: A review of the frontier function literature. Agriculture and Resource Economics Review, 22, 88-101. Bravo-Ureta, B.E. and Rieger, L. (1991). Dairy farm efficiency measurement using stochastic frontiers and neoclassical duality. American Journal of Agricultural Economics Review, 22(1), 88-101. Chinese Rice Study Team of the Peoples Republic of China (1982). The feasibility for rice irrigation in Uganda, October 1982. A Report written for the Republic of Uganda

94

Coelli, T. J. (1995). Recent development in frontier modeling and efficiency measurement. Austrarian Journal of Agricultural Economics, 39, 219-245. Coelli, T.J. (1996a). A guide to frontier version 4.1:A computer program for stochastic frontier production and cost function estimation mimeo, Department of Econometrics, University of New England, Armidale. Coelli, T.J. (1996b). Measurement and sources of technical efficiency in Australian coalfired electricity generation. Centre for Efficiency and Productivity Analysis(CEPA Working Papers No. 1/96). Department of Econometrics University of New England, Armidale, Australia. Cohen, J. (1988). Statistical power analysis for the behavioral sciences, (2nd ed.). Hillsdale, N.J.: Academic Press. Dawson, P.J., and Lingard, J. (1991). Approaches to measuring technical efficiency on Philippine rice farms, Journal International Development, l3(3), 211-228. Debrew, E. (1951). The coefficient of resource utilization. Econometrica, 19, 273-292. Fan, S. (1999). Technological change, technical and allocative efficiency in Chinese Agriculture: The case of rice production in Jiangsu. (EPTD Discussion Paper no 39, International Food Policy Research Institute IFPRI).Washington, D.C. Farrell, M. J. (1957). The measurement of productivity efficiency. Journal of Royal Statistical Society Series, 120, 253-290. Food and Agriculture Organization (1985). Women in Developing Agriculture FAO, Rome. Food and Agricultural Organization (2004). Food Supply Situation and Crop Prospects in Sub-Saharan Africa. FAO Global Information and Early Warning Systems (GIE WS) Africa report no. 1 http://faodoc.org/docrep/006. Jagwe, J. Okoboi, G., Arayo, E. and Abele, S. (2005). Market Opportunity identification study for 5 selected key crops in Kabalore district, Western Uganda. CRS-Food net Uganda. A Report to Catholic Diocese of Fort Portal Development Bureau, Uganda. Jamison, D.T., and Moock, P.R. (1984). Farmer education and farm efficiency in Nepal:The role of schooling extension services and cognition skills. World Development ,12, 67-86.

95

Jondrow, J., Lovell, K.C.A., Materov, I. S., & Schmidt, P. (1982). On the estimation of technical inefficiency in the stochastic frontier production function model. Journal of Econometrics, 19, 233-238. Junankar, P.N. (1989). The response of peasant farmers to price incentives: The use and misuse of profit functions. The Journal of Development Studies, 25(2), 162-182. Kalirajan, K.P., and Obwona, M.B. (1994). A measurement of firm-specific and inputspecific technical and allocative efficiencies. Applied Economics, 26,393-398. Kalirajan, K.P., and Shand, R.T. (1988). Firm and product-specific technical efficiencies in a multi-product cycle system. The Journal of Development Studies. 25(1), 8496. Khan, M.H. and Maki, D.R. (1979).Effects of farm size on economic efficieincy: The case of Pakistan. American Journal Agricultural Economics. 61(1), 54-69. Kmenta, J (1986). Elements of Econometrics (2nd Edition) NewYork: MacMillan Publishing Company. Kodde, D.A., and Palm, A.C. (1986). Wald Criteria for jointly testing equality and inequality restrictions. Ecometrica, 54, 1243-1248 Koopmans, T.C. (1951). An analysis of production as an efficient combination of activities. In Activity Analysis of Production and Allocation, (Ed.), Cowlles Commission for Research in Economics, Monograph No.13, Wiley, New York. Knight, F. (1933). The Economic Organization. Harper and Row: New York. Lau, I.J., and Yotopolous, P.A. (1971). A test for a relative and application to Indian agriculture. American Economic Review, 61, 94-109. Lingard, J., Castillo, L., and Jayasuriya, S. (1983) Comparative efficiency of rice farms in Central Luzon, The Philippines. Journal of Agricultural Economics, 34(2), 163173. Lockheed, M.E., Jamison, D.T., & Lau, L.J. (1980). Farmers education and farm efficiency : A survey. Economic Development and Cultural Change, 29, 37-76. Maxwell, S. E., and Delaney, H. D. (1990). Designing experiments and analyzing data: A model comparison perspective. Belmont, CA: Wadsworth Publishing. McGraw. Mbowa, S. (1996). Farm size and economic efficiency in sugar cane production in Kwazul-Natal. Un published doctoral dissertation, University of Natal, Durban.

96

Meeusen, W., and Van den Broeck, J (1977). Efficiency estimation from Cobb-Douglas production function with composed error. International Economic Review, 18, 435-44. Moock, P. (1976). The Efficiency of women as farm managers: Kenya. American Journal of Agricultural Economics, 58(5), 831-835. Ministry of Agriculture Animal Industry and Fisheries (2000). Plan for modernization of agriculture: Eradicating poverty in Uganda: Government strategy and operation framework, Kampala. Nkonya, E. , Pender, J., Jagger, P. Sserunkuma, D., Kaizzi, C. and Ssali, H. (2004) Sratagies for Sustainable land management and poverty reduction in Uganda. Reseaerch Report 133 International Food Policy Research Institute, Washington D.C. National Agricultural Research Organization, (2003). Participatory multiplication and testing of improved upland rice varieties in Uganda, A Progress Report on Cereal Research for 2002. Nsanzugwanko, M. D., Battesse, G.E., and Fleming, E.M. (1996). The technical efficiency of small agricultural producers in Central Ethiopia. Centre for Efficiency and Productivity(CEPA Working Papers.No5/96). Department of Econometrics, University of New England , Armidale, Australia. http://www.une.edu..au/econometrics/cepawp.htm Ochollah, A.R., OgengaLatigo, M.W., and. Nsubuga, E.N.B (1997). Impact of upland rice cultivation on crop choice and income of farmers in Gulu and Bundibugyo districts. African Crop Science Proceedings, 3, 1407-1411. Phillips, J.M. (1994). Farmer education and farmer efficiency: A meta-analysis. Economic Development and Cultural Change, 43(1), 150-165. Pindyck, R.S. and Rubinfield, D. (1981) Econometric models and economic forecasts. New York:McGraw Hill . Quismbing, A.R. (1994). Gender differences in agricultural productivity: A survey of empirical evidence. The World Bank: Washington D.C. (Paper Series No.36) Education and Social Policy Department. Rahman. S. (2002). Profit efficiency among Bangladesh rice farmers Manchester: University of Manchester. (School of economic studies. discussion paper series No. 0203). Rahman. S. (2003). Profit efficiency among Bangladeshi rice farmers. Food Policy, 28, 483-503.

97

Sabiiti, A.G. (1995). An economic enalysis of production of cow pea in Northern and Eastern Uganda. Unpublished masters thesis, Makerere University, Kampala. Sadoulet, E., and De Janvry, A. (1995). Quantitative development policy analysis. Baltimore: The Johns Hopkins University Press. Saleem, S. T. (1988). Relative efficiency of cotton farms in Sudanese irrigated agriculture World Development, 16(8), 975-984. Schultz, T.W. (1964). Transforming traditional agriculture. Chicago: University of Chicago Press. Schultz, T.W. (1975). The Value of Ability to deal with disequilibilia. Journal of Economic Literature, 13, 827-846. Sevilla-Siero, C. A. (1991). On the use and misuse of profit functions measuring the price responsiveness of peasant farmers: A comment. The Journal of Development Studies, 27(4), 123-136. Seyoum, E.T., Battese, G.E., and Fleming, E.M. (1998). Technical efficiency and productivity of maize producers in eastern Ethiopia: A study of farmers within and outside the Sasakawa-Global 2000 project. Agricultural Economics, 19, 341348. Sharma, K. R., PingSun, L., and Halina M. Z. (1999). Technical, allocative and economic efficiencies in swine production in Hawaii: a comparison of parametric and nonparametric approaches. Agricultural Economics, 20, 23-35. Ssenteza, J.F.G. (1993). An conomic analysis of rice industry in Uganda: The case of Kibimba rice company. Unpublished masters thesis, Makerere University, Kampala. Taylor, T.G., and Shonkwiler, J. S. (1986). Alternative stochastic specifications of the frontier production in the analysis of agricultural credit programs and technical efficiency. Journal of Development Economics, 21, 149-160. Thiam, A., Boris Bravo-Ureta, E., and Teodoro, E. R. (2001). Technical efficiency in developing country agriculture: A meta-analysis. Agricultural Economics, 25, 235-243. Timmer, C.P. (1971). Using a probabilistic frontier production function to measure technical efficiency. Journal of Political Economy, 79, 776-94. Uganda Bureau of Statistics (2004). Statistical Abstract. Ministry of Finance, Planning and Economic Development.

98

Uganda Bureau of Statistics (2005). The 2002 Uganda Population Housing Census, main Report The Republic of Uganda. Wang, J., Cramer, G. L, and Wailes, E. J. (1996a). A shadow-price frontier measurement of profit efficiency in Chinese Agriculture. American Journal of Agricultural Economics, 78, 146-156. Wang, J., Cramer, G. L, and Wailes, E. J. (1996b). Production efficiency of Chinese agriculture: Evidence from rural household survey data. Agricultural Economics, 15, 17-28. Weier, S. (1999). The effects of education on farmer productivity in rural Ethiopia (WPS99-7). Oxford: University of Oxford, Department of Economics ,Centre for the Study of African Economies. Wilson, P., Hadley D., Ramsden, S., and Kaltsas, I. (1998). Measuring and explaining technical efficiency in UK potato production. Journal of Agricultural Economics 49(3), 294-305. World Bank (1993). Growing out of Poverty: A World Bank Country Study (Uganda). Washington D.C.: The World Bank. www.stata.com, accessed 2nd April ,2004 Yilma, M. (1996). Measuring smallholder efficiency: Uganda coffee and food crop production. Unpublished dissertation, Goteborgs Universitet : Ekonomiska Studier Utgivna av Nationalekonomiska Institutionen Handelsh, KompendietGoteborg, Sweden . Yotopolous, P.A., and Lau, L.J. (1973). A test for relative economic efficiency: Some further results. American Economic Review, 63(1), 214-223.

99

APPENDIX A 1. DATA NORMALIZATION

Figure 1a Normalized Profit
200

100

Std. Dev = 1969.62 Mean = 772.6 0 N = 297.00

0 0. 00 30 0.0 00 28 0.0 00 26 0.0 00 24 0.0 00 22 0.0 00 20 0.0 00 18 0.0 00 16 0.0 00 14 0.0 00 12 0.0 00 10 .0 00 80 .0 00 60 .0 00 40 .0 00 20 0 0.

normalized profit1(profit1/farm gate price of rice)

100

Figure 1b Log of Normalized Profit

40

30

20

10 Std. Dev = 1.09 Mean = 6.10 0 N = 274.00

5 .2 10 75 9. 25 9. 75 8. 25 8. 75 7. 25 7. 75 6. 25 6. 75 5. 25 5. 75 4. 25 4. 75 3. 25 3. 75 2. 25 2.

log of normalized profit1

F ig ure 2 a Norm alized Hire d W ag e ra te

120 100

80

60

40

20 0
0.0

S td . De v = 31 3 .7 3 Me a n = 2 0 4 .9 N = 2 97 .0 0
.0 00 22 .0 00 20 .0 00 18 .0 00 16 .0 00 14 .0 00 12 .0 00 10 0.0 80 0.0 60 0.0 40 0.0 20

n o rm a lize d h ire d wa g e ra te

101

Figure 2b Log Normalized Hired Wage Rate

60 50

40

30

20 Std. Dev = 1.23 Mean = 4.95 0 -1.50 -.50 .50 1.50 2.50 3.50 4.50 5.50 6.50 7.50 -1.00 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 N = 234.00

10

log of normalized hired wage

Figure 3a Family Labour in Rice Production

120 100

80

60

40

20 0
0.0

Std. Dev = 512.44 Mean = 465.2 N = 297.00

.0 00 48 0 . 00 44 .0 00 40 0 . 00 36 0 . 00 32 0 . 00 28 0 . 00 24 0 . 00 20 .0 00 16 0 . 00 12 0.0 80 0.0 40

T otal Ma ndays Rice

102

F igure 3b Lo g of F am ily La bour

50

40

30

20

10

Std. Dev = 1.06 Mean = 5 .67 N = 2 97 .00

5 2.7 5 2.2 5 3.2 5 3.7 5 4.7 5 4.2 5 5.2 5 5.7 5 6.2 5 7.2 5 6.7 5 7.7 5 8.2

lo g o f to ta l m an d a ys in rice

F ig ure 4 a Norm alized O the r Input C o sts

120 100

80

60

40

20 0

S td . De v = 22 3 .6 9 Me a n = 2 2 2 .0 N = 2 89 .0 0
.0 00 19 0 .0 0 18 0 .0 0 17 0 .0 0 16 0 .0 0 15 0 .0 0 14 0 .0 0 13 0 .0 0 12 0 .0 0 11 0 .0 0 10 .0 0 90 .0 0 80 .0 0 70 .0 0 60 .0 0 50 .0 0 40 .0 0 30 .0 0 20 .0 0 10 0.0

NT INC O S R

103

F ig ure 4 b L o g o f No rm alized O the r Inp ut C os ts

50

40

30

20

10

S td . De v = .9 0 Me a n = 5 .0 3 N = 2 89 .0 0 2 .2 5 2 .7 5 3 .2 5 3 .7 5 4 .2 5 4 .7 5 5 .2 5 5 .7 5 6.2 5 6.7 5 7.2 5

L NT O IC O S

F ig ure 5 a R ic e A c rea g e
200

100

S td . De v = 1.7 1 Me a n = 1 .3 0 0.0 2.0 4.0 6 .0 8 .0 1 0 .0 1 2 .0 1 4 .0 1 6 .0 1 8 .0 2 0 .0 N = 2 97 .0 0

R ic e a cre a g e

104

8 8

105

Fig ure 6 a C apital

200

100

Std. Dev = 34160.9 0 Mean = 25491 .2 0 N = 297.00

.0 00 00 44 0.0 0 00 40 0.0 0 00 36 0.0 0 00 32 0.0 0 00 28 0.0 0 00 24 0.0 0 00 20 0.0 0 00 16 0.0 0 00 12 .0 0 00 80 .0 0 00 40 0.0

to ta l co st o f ca pita l

Figure 6b Log of C apital

60

50

40

30

20

10 0
0 7.0 0 7.5 00 8. 0 8.5 00 9. 0 9.5

Std. Dev = .88 Mean = 9.76 N = 292.00

.00 13 .50 12 .00 12 0 .5 11 0 .0 11 0 .5 10 0 .0 10

106

log o f cap ita l

Figure 6b Log of Capital

60

50

40

30

20

10 0
0 7.0 0 7.5 00 8. 0 8.5 00 9. 0 9.5

Std. Dev = .88 Mean = 9.76 N = 292.00

.00 13 .50 12 .00 12 0 .5 11 0 .0 11 0 .5 10 0 .0 10

lo g o f cap ita l

107

F igure 7 a Exp e rience in R ic e G ro w ing

120

100

80

60

40

20

S td . De v = 9 .1 2 Me a n = 1 1 .7 N = 2 8 1 .0 0 0 .0 5 .0 1 0 .0 1 5 .0 2 0 .0 2 5 .0 3 0 .0 3 5 .0 4 0 .0 4 5 .0 5 0 .0

ye a rs in rice g ro win g

F ig ure 7 b L o g o f E xp e rie nc e in R ic e G ro w ing

50

40

30

20

10

S td. Dev = .87 Mean = 2.14 N = 281.00 0.0 0 .25 .50 1.00 .75 1.50 2.00 2 .50 1.75 2.25 3.00 3.50 4.00 1 .25 2.75 3 .25 3.75

lo g o f Y e a r s in R ic e g ro win g

----------------------------------------------------------------------------------------------------

108

REGRESIN DIAGNOSTICS Ramsey RESET test using powers of the fitted values of ln_profit Ho: model has no omitted variables F(3, 221) = 0.0818 Prob > F = 0.761 Breusch-Pagan / Cook-Weisberg test for heteroskedasticity Ho: Constant variance Variables: fitted values of ln_profi chi2(1) = 0.0673 Prob > chi2 = 0.942 VIF Variable VIF 1/VIF ----------------+-------------------------beta15 | 7.3400 0.1363 beta14 | 7.1337 0.1608 beta12 | 6.9273 0.1853 beta10 | 6.7210 0.2098 beta_9 | 6.5147 0.2344 beta_6 | 6.3083 0.2589 beta11 | 6.1020 0.2834 beta20 | 5.8957 0.3080 beta18 | 5.6893 0.3325 beta17 | 5.4830 0.3570 beta13 | 5.2767 0.3816 beta_7 | 5.0703 0.4061 beta_1 | 4.8640 0.4306 beta16 | 4.6577 0.4551 beta_8 | 4.4513 0.4797 beta_4 | 4.2450 0.5042 beta_2 | 4.0387 0.5287 beta_5 | 3.8323 0.5533 beta19 | 3.6260 0.5778 delta_6 | 3.4197 0.6023 beta_3 | 3.2133 0.6269 delta_2 | 3.0070 0.6514 delta_4 | 2.8007 0.6759 delta_3 | 2.5943 0.7004 delta_5 | 2.3880 0.7250

109

delta_1 | 2.1817 delta_7 | 1.9753 --------------+-----------------------Mean VIF | 4.5545 . .

0.7495 0.7740

Shapiro-Wilk W test for normal data Variable | Obs W V z --------------+-----------------------------------------------------ln_profi | 253 0.99060 1.72000 1.26200 beta_1 | 253 0.99086 1.67292 1.15065 beta_2 | 253 0.99111 1.62585 1.03931 beta_3 | 253 0.99137 1.57877 0.92796 beta_4 | 253 0.99163 1.53169 0.81662 beta_5 | 253 0.99189 1.48462 0.70527 beta_6 | 253 0.99214 1.43754 0.59392 beta_7 | 253 0.99240 1.39046 0.48258 beta_8 | 253 0.99266 1.34338 0.37123 beta_9 | 253 0.99292 1.29631 0.25988 beta10 | 253 0.99317 1.24923 0.14854 beta11 | 253 0.99343 1.20215 0.03719 beta12 | 253 0.99369 1.15508 -0.07415 beta13 | 253 0.99395 1.10800 -0.18550 beta14 | 253 0.99420 1.06092 -0.29685 beta15 | 253 0.99446 1.01385 -0.40819 beta16 | 253 0.99472 0.96677 -0.51954 8beta17 | 253 0.99497 0.91969 -0.63088 beta18 | 253 0.99523 0.87262 -0.74223 beta19 | 253 0.99549 0.82554 -0.85358 beta20 | 253 0.99575 0.77846 -0.96492 delta_1 | 253 0.99600 0.73138 -1.07627 delta_2 | 253 0.99626 0.68431 -1.18762 delta_3 | 253 0.99652 0.63723 -1.29896 delta_4 | 253 0.99678 0.59015 -1.41031 delta_5 | 253 0.99703 0.54308 -1.52165 delta_6 | 253 0.99729 0.49600 -1.63300 delta_7 | 252 0.99972 0.05100 -6.93200

Prob|>z| 0.10349 0.13600 0.16851 0.20102 0.23353 0.26604 0.29855 0.33106 0.36357 0.39608 0.42859 0.46110 0.49361 0.52612 0.55862 0.59113 0.62364 0.65615 0.68866 0.72117 0.75368 0.78619 0.81870 0.85121 0.88372 0.91623 0.94874 1.00000

2.

Estimation of Frontier Profit Function: Cobb-Douglas Model

Below is a brief discussion of results from C-D model. Table A1 summarizes Ordinary OLS and MLE results for the C-D profit function model. The coefficients for all the variables (Cost of hired labor, imputed cost of family labor, area under rice capital, other

110

input costs in a pooled sample carry the theoretically expected signs and are statistically significant with exception cost of hired labor (Table A1). In Tororo district, the estimated 2coefficients associated with three variables, namely other inputs, imputed wage of family labor and rice hectares are statistically significant and carry the theoretically expected signs for both OLS and MLE results (Table A2).

The Pallisa district results are shown in Table A2. Of the five variables included in the model, four variables were significant and carried the expected signs in MLE estimation techniques. The estimates associated with cost of hired labor, other inputs costs and imputed family labor costs carried negative signs. Area under rice and capital had positive signs. Note that all the estimates associated with the preceding discussed variables are statistically significant. In Lira district, only three variables of the five included in the model were significant and carried the expected sign. These were other inputs, area under rice and capital. Table A1: Coefficient Estimates of C-D Frontier Profit Function in all three

studied Districts Dependent variable = Normalized Profit in Ug Shs Variables OLS Coefficients 6.96 Constant ( pi ) -0.03 Cost of hired labor (Ushs /Ha) ( p1 ) -0.03 Other input costs (Ushs/Ha) ( p 2 ) Imputed cost of family labor -0.11 (Ushs/Ha) ( p 3 ) 8Area under rice (ha) ( z1 ) Capital ( z 2 ) Sigma squared 0.20 0.02 0.00 0.01 111 0.13 0.02 0.31 0.00 0.07 0.00 MLE coefficients 7.43 -0.01 -0.02 -0.12

p-v 0.00 0.17 0.00 0.00

p-v1 0.00 0.17 0.07 0.00

Gamma Log likelihood -186.65 Source: Field Survey Data. p-v1 are p values computed from t-ratios

0.63 -127.44

0.00

112

Table A2:

Coefficient Estimates of C-D Frontier Profit Function-Tororo District Dependent Variable = Normalized Profit in Ug Shs

Variables

OLS Coefficients pi ) 7.20 Constant ( -0.02 Cost of hired labor (Ushs /Ha) ( p1 ) -0.04 Other input costs (Ushs/Ha) ( p 2 ) Imputed cost of family labor -0.06 (Ushs/Ha) ( p 3 ) Area under rice (Ha) ( z1 ) 0.16

p-v 0.00 0.47 0.06 0.00 0.00 0.89

MLE coefficients 7.89 0.00 -0.05 -0.08 0.10 0.00 0.89 0.92 -47.91

p-v1 0.00 0.91 0.00 0.00 0.00 0.91 0.03 0.00

0.00 Capital ( z 2 ) Sigma squared 0.23 Gamma Log likelihood -81.99 Source : Field Survey Data p-v1 are p values computed from t-ratios

113

Table A3:

Coefficient Estimates of C-D Frontier Profit Function-Pallisa District

Dependent Variable = Normalized Profit in Ug Shs Variables OLS Coefficients pi ) 5.94 Constant ( 0.04 Cost of hired labor (Ushs /Ha) ( p1 ) -0.01 Other input costs (Ushs/Ha) ( p 2 ) Imputed cost of family labor -0.01 (Ushs/Ha) ( p 3 ) Area under rice (Ha) ( z1 ) 0.18 0.00 0.63 0.05 0.01 0.40 0.99 26.80 0.00 0.17 0.00 0.00 0.02 Capital ( z 2 ) Sigma squared 0.23 Gamma Log likelihood -55.08 Source: Field Survey Data. p-v1 are 2p values computed from t-ratios p-v 0.00 0.09 0.67 0.65
1

MLE coefficients 6.78 -0.03 -0.02 -0.02

p-v1 0.00 0.00 0.00 0.03

114

Table A4:

Coefficient Estimates of C-D Frontier Profit Function-Lira District Dependent Variable = Normalized Profit in Ug Shs

Variables Constant ( p 0 ) Cost of hired labor (Ushs /Ha) ( p1 ) Other input costs (Ushs/Ha) ( p 2 ) Imputed cost of family labor (Ushs/Ha) ( p 3 ) Area under rice (Ha) ( z1 )

OLS Coefficients p-v1 20.51 -0.06 -1.28 0.06 1.41 0.01 0.15 0.00 0.60 0.00 0.00

MLE coefficients p-v1 20.02 -0.01 -1.03 0.08 0.96 0.12 0.38 0.95 -10.32 0.00 0.67 0.00 0.52 0.00 0.15 0.05 0.00

0.66 Capital ( z 2 ) Sigma squared 0.32 Gamma Log likelihood -29.77 Source: Field Survey Data. p-v1 are p values computed from t-ratios8 2.2 Profit Efficiency Score Estimates

Table A5 and figure 8 show the distribution of farm-specific profit efficiency model results of the C-D model. The efficiency levels were obtained by getting the exponential of scores from the estimates of inefficiency model 12. The results show a wide variation of efficiency levels in the districts, with 24 farmers in the sample getting less than 40 per cent level of efficiency and 66 per cent getting over 81 %( Table A5). This is in contrast to translog model results, which registered only 30%. Within the districts, Tororo has the highest mean score of 79 per cent followed by Pallisa at 75 per cent and Lira at 65 per cent. Again, these results demonstrate that the translog results are superior. Figure 8 shows this clearly by the fact that the scores are skewed positively. What contributes to the observed levels of efficiency is the subject of the next discussion. Table A5 Farm Specific Efficiency Scores in a Profit Frontier C-D model

115

Efficiency Scores

Pooled

Tororo n 8 5 2 0 22 86 123 % 6.5 4.07 1.62 0 17.89 69.92 100 79

Pallisa n 13 5 5 9 2 57 91 % 14.29 5.49 5.49 9.89 2.20 62.63 100 75 02 98

Lira % 23.1 10.3 5.1 10.3 7.7 43.6 100 65 04 96

n % <40 24 9.4 41-50 6 2.3 51-60 17 6.7 61-70 20 7.9 71-80 20 7.9 81+ 166 65.6 n 253 100 Mean 77.2 Min 9.1 Max 95.0 Source: Field Survey Data. 2.3

9 4 2 4 3 17 39

Determinants of Farm-Specific Profit Inefficiency in Rice-C-D Model

The purpose of this section is to explain the variation in performance by farmers of the above observed profit efficiencies. The results presented in Table A6 suggest that the results are at variant with those reported in translog model. Generally, the same variables have not performed well. Whereas education and extension service estimates are significant in all the three districts in translog model, they are significant in Tororo and Pallisa (education) and Pallisa (extension services). Degree of specialization has major influence in Tororo and Pallisa in translog model whereas it was only so in Pallisa in C-D model. Non- farm employment carries expected negative sign in Pallisa and Lira in translog model but had little influence in all districts in C-D model. Similarly, experience had influence in Pallisa district in translog model but only so in Tororo district.in C-D model.

Table A6: C-D model

Determinants of Firm-Specific Profit Inefficiency in Rice Production-

Dependent: Variable = Inefficiency effects () 116

Pooled coefficients p-v1 Constant ( 4.10 0.001

Tororo coefficients p-v1 2.84 0.1 6

Pallisa coefficients p-v1 3.14 0.0 0

Lira Coefficients p-v1 1.07 0.1 8

) 0 Non-farm
employment ( ) 1 Education

0.14 ( -0.24

0.46 0.00

-1.55 -0.34

0.9 1 0.0 3

-0.78 -0.16

0.1 2 0.0 0

0.46 -0.17

0.4 6 0.1 7

2 ) Extension
services ( 3 ) Credit Access ( 4 ) Experience

-0.14 -0.27

0.00 0.17

0.19 -1.40

0.7 7 0.0 3 0.1 4 0.1 2

-0.50 -0.32 -0.27

0.0 0 0.0 0 0.1 7 0.0 0

-0.25 -0.02 -0.07

0.1 8 0.9 1 0.9 5 0.1 7

(yrs) ( ) 5 -0.19 0.14 -0.46 Degree of specialization ( ) 6 -0.21 0.03 -0.50 Source: Field Survey Data. p-v1 are p values computed from t-ratios

-0.16

-0.14

117

Figure 8:
100 90 80 70 60 percentage 50 40 30 20 10 0

Profit Efficiency Distribution C-D Model

Pooled Tororo Pallisa Lira

<40

41-50

51-60

61-70

71-80

81+

M ean

Efficiency Scores

118

Figure 9:

Profit Efficiency Distribution Translog Model

100 90 80 70 60 Pooled Tororo Pallisa Lira

percentage 50 40 30 20 10 0 <30 31-40 41-50 51-60 61-70 71-80 81+ Efficiency Scores

119

APPENDIX B QUESTIONNAIRE PROFIT EFFICIENCY AMONG RICE PRODUCERS IN EASTERN AND NORTHERN UGANDA Introduction:

Rice is slowly gaining prominence both as a means of livelihood and the diets of Ugandans. It is more prominent in the Eastern Uganda. However, there exists limited knowledge on the crop.

The purpose of this study is to evaluate profit efficiency in rice production Please note that your responses will remain confidential.

Name of Enumerator---------------------------------------------------------------------------------Date of Interview-------------------------------------------------------------------------------------District--------------------------------------------------------------------------------------------------Sub country--------------------------------------------------------------------------------------------Village/parish-------------------------------------------------------------------------------------------

SECTION A SOCIO-DEMOGRAPHIC

Respondent name and number-----------------------------------------------------------------------

120

1. Respondent sex. Male/Female 2. Head of household Yes No

3. How many are you in the household including non-biological? No------------------------No. of adults Males Females No. Children <18 Males Females

4. Please, fill the following table with the information regarding the household.

Member household

of Age (years)

Education level

Main

No.

of Secondary

No.

of

occupation years

occupation years

121

SECTION B: RICE PRODUCTION 5. How much land (total) do you have? No. of acres--------------------------------------------6. How much of this land was occupied by agricultural crops last season?------------------7. When did you start growing rice? Year---------------------------------------------------------8. How many plots of rice do you have? No.-----------------------------------------------------9. What is the area of the plots and who owns them? Plot 1. 2 3 4 Area/acres Type of ownership

10. How did you/they get the plots? Plot 1 2 3 4 Source of ownership

11. How many of these did you cultivate this year? No------------------------------------------

12. Who in the family carried out the following agricultural activities in the plots mentioned above in the last season? Activity
Hsl Head Spouce1 Spouce2 Spouce 3

Children <18 years School going Not going to school

male female male female

Opening land Seed bed

122

prep Planting Weeding Scaring bird Harvesting Threshing Marketing

123

13. How many days did the member of the household take in each activity? Number of working days in the week Member of the family Spouce1 Spouce2 Spouce3 Children <18years School going Not going Male Female Male Female Opening land Seed of bed

Activity

Hsld head

preparation Planting Weeding Scaring birds Harvesting Threshing

14. Did you hire labor in rice activities last season? Yes/No. If yes, for what activities and how many days and if paid in cash, at what rate?

No. of days Activity 8 8

Hired labor Plot 1

Plot 2

Plot 3

Plot 4

Wage rate (shs)

124

15. Did you use inputs (fertilizers, manure, pesticides) in your rice last seasons? Yes/No 16. If yes which ones did you use and how much in each plot? Plot 1 2 3 4 Type (Quantity Units and Price)

17. If no, why?--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------18.What farm implements did you use in rice production? Type of equipment Hoes Pangas Ox plough Basket Wheel barrow Others (specify) Number owned Year bought Value when bought shs.

19. Are there any crops you grow that compete directly with rice directly? Yes/No 20. If yes, in which way? (be specific or explain).

Crops

Land

Time

Labour

125

SECTION C: HARVESTING AND STORAGE 21. How many bags of rice did you harvest last season? No. of bags-------------------------22. How many bags of these did you sell last season? Unmilled -----------------------------------------------------------------------------------------------------------Milled ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------23. Did you store any last season after harvest? Yes/No 24. If yes, in which form did you store. 1) Unmilled 2) milled ..3) both 25. Which containers did you sue to store (Tick) 1) Gunny bags 2) Floor 3) Traditional granary 4) others (specify)

26. How long did you store last season? (Tick) 1) Sold straight away 20 1 month after 3) 1-2 months 4)>2 months

27. What are the main problems you face in storage (mention them in order of importance)? 1)-------------------------------------------------------------------------------------------------2)-------------------------------------------------------------------------------------------------3)-------------------------------------------------------------------------------------------------4)------------------------------------------------------------------------------------------------28. How did you dry your un-milled rice last season? 1) on bare grounds 2) on mats 30 cemented area 40 others specify

29. If cemented, how much did it cost you to do so and when did you do it?

126

Ushsyear.. 30. If mats, how many and how much did you pay and when did you buy them? Ushs.year. 31. Did you get any credit (formal or formal) to use in rice production last season? Yes/No. 32. If yes, fill the following table

Type of credit From where/whom Amount Interest rate

Last season Formal

Informal

33. If no, why not ------------------------------------------------------------------------------------------------------34. Did an extension officer visit you about rice production last season? Yes/No. 35.If yes, how many times last year? 1) Once a month 2) 3 times a month 3)Once in 6 months 4) Not at all. 36. If visited, what message did they carry? Message -------------------------------------------------------------------------------------------------37. If they did not come, did you try to look for advice from extension agents? Yes/No

127

38. If yes, what type of information did you look for and from whom? Type of information Media (source)

39. Apart from extension agents how else do you get information on production of rice? 1) Radio 2) neighbor 3) newspapers 4) family

40. What type of information did you get? Type of information Source

SECTION D: MARKETING 41.In which form did you market your rice last season? 1) un milled 2)milled 3) both depending on need of funds

42. Last season where do you market it? 1) Traders came to my home market 4) Used any of the methods depending on convenience. 128 2) took to the mill 3) took to the local

43. If traders came to your home, what price did you get per kilo on average? 1) Un milled Ush/kg..2) Milled rice ..Ushs/kg 44. How far is the market from your home? No of miles---------------------------------------45. If you took to the market, what price did you get per kg? 1) Before milling Ush/kg------------------------------------------------------------------2) After milling Ushs./kg------------------------------------------------------------------46.. Are there times when you fail to market your milled rice? Yes/No. 47. If 2yes, what do you think are the reasons? ------------------------------------------------------------------------------------------------------

129

APPENDIC C: MAP SHOWING STUDY DISTRICTS

Lira Pallisa Tororo

130