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 Agricultural Policy Committee

CRS The Chinese Rice Study Team

FAO Food and Agriculture Organization

GDP Gross Domestic Product

LDC Less Developed Countries

MAAIF Ministry of Agriculture Animal Industry and Fisheries

NAARI 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.

x

TABLE OF CONTENTS

DECLARATION.................................................................................................................ii

DEDICATION....................................................................................................................iv

ACRONYMS...................................................................................................................viii

ABSTRACT.......................................................................................................................ix

LIST OF TABLES............................................................................................................xiv

LIST OF FIGURES..........................................................................................................xvi

CHAPTER I 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 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

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 METHODOLOGY................................................................................28

3.0..........Introduction.........................................................................................................28

3.1..........Approaches to Measuring Efficiency.................................................................28

3.2..........Deterministic Versus Stochastic Frontier Models..............................................29

xi

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

3.6..........Data Analysis/Model Implementation................................................................54

CHAPTER IV 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

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 SUMMARY, CONCLUSIONS AND POLICY

RECOMMENDATIONS.................................................................................84

5.0 .........Introduction.........................................................................................................84

5.1..........Summary.............................................................................................................87

5.2.......... Conclusions and Policy Recommendations.......................................................90

5.3.......... Recommendations for Further Research...........................................................92

xii

REFERENCES..................................................................................................................93

APPENDIX A..................................................................................................................100

APPENDIX B..................................................................................................................119

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 Model52

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 Districts66

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

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..................................74

xiv

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

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 2: Frontier MLE and OLS Stochastic Profit Function.................................13

xvi

CHAPTER I

INTRODUCTION

1.1 Background

Agriculture plays an important role in Uganda’s 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 country’s 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.

3

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.

4

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.

5

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 d’Ivoire

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 (LDC’s) 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 = x1……xn, 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 practice’1production

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.

1 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.

8

Figure 1: Stochastic Production Frontier

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

x 2/y

R >

Q

P

Q'

D'

D

x1/y

B

A

O

9

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 = x (p, w, z) 2

Q = q (p, w, z) 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

π = 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:

π 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:

πi = normalized profit on firm i defined as gross revenue minus variable cost

divided by the output price.

Pij = prices of variable input j on firm i divided by the output price.

Zik = level of fixed input on firm i where k are a number of fixed inputs.

i = 1,………………………….., n number of farms in the sample.

11

℮ i = 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

E O

Figure 2: Frontier MLE and OLS Stochastic Profit Function

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

M

Normalized input price given fixed resources Pί/Zj

F

D

D

M

E

.·

$ Normalized Profit

P

·

E

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 farmer’s inability to make optimal decisions may be due to external factors,

which lie outside his/her prevue. These include untimely input supply, bad weather, non-

conducive 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 farmer’s 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 non-

participants, 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 μ. The major weakness with this model is that any

measurement error and any other source of variations in the dependent variable is

embedded in μ and can’t 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 μ. 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

2 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 non-

conventional 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 Cobb-

Douglas (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 μ. 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 Farrell’s 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 large-

scale 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.

5 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 and 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 don’t 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 d’Ivoire. 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 λ7 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 γ was close to one. The average profit efficiency

7 λ is the likelihood- ratio test statistic which is = -2{log [Likelihood (H0)]-Log [Likelihood (H1)]} and has approximately χ2

ν distribution with ν equal to the number of parameters assumed to be zero in the null hypothesis (Rahman, 2002). λ 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

d’Ivoire 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 d’Ivoire.

The authors concluded that there was no economic rational for biasing rice development

strategies towards male farmers in Cote d’lvoire 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 Farrell’s (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.

3.2 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 = ƒ (х, ß) е-μ...............................................................................................................(6)

where ƒ (х, ß) denotes the frontier production function and μ is a one-sided non-negative

distribution term. This model imposes constraint of μ≥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 ν

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:

ƒ (х, ß) е ν-μ........................................................................................................................(7)

where ƒ ( х, ß), is as defined in (6) and ν-μ is error term, ν represents factors external to

the farmer and are assumed to be independently and identically distributed (iid) as

Ν(0,σν2); μ 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 farmer’s 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 μ (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 Junanker’s 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 firm’s 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 Flinn’s model

specified in equation 8:

π ј = ƒ (Pιј, Zκј, Dіј).exp e ј....................................................................................................(8)

31

Where

π ј = normalized profit of јth farm defined as gross revenue less variable cost,

divided by commodity prices from farm j.

Pιј = prices of the variable inputs on jth farm,

Zκј = kth fixed factors on jth farm and

Dіј = exogenous variables on jth farm,

e ј = an error term, and ј = 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 ј = νј- µ ј..............................................................................................................................................(9)

where νј and µј are random error terms and inefficiency effects of the farm ј, respectively.

When µј = 0, the firm lies on the frontier but if µ ј>0 the farm is profit inefficient and

incurring losses.

The inefficiency effects (µј) in equation (9) which are non-negative random variables are

assumed to be identically and independently distributed such that µј is defined by the

truncation (at zero) of the normal distribution with a mean of

and variance where are the variable representing socio-economic characteristics

of farm j to explain inefficiency and δ0 and δd are the unknown parameters to be

32

estimated. The profit efficiency of the farm in the context of stochastic frontier is given by:

ζj = E [ exp(-µ ј)| e ј] = E[exp(- ..............................................................(10)

where ζj 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 µј upon observed value of ζj.

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 don’t 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

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

34

.............................................................................................(11)

where:

.........................................................................................................(12)

for all

restricted normalized profit computed for farm defined as gross revenue less

variable costs divided by farm specific rice price

ln = natural log

= price of variable inputs normalized by price of output where (for i =1,

2, and 3) so that:

= the cost of hired labor normalized by price of rice ( )

= the cost of “other inputs” normalized by price of rice ( )

= Imputed cost of family labor normalized by the price of rice ( )

= the quantity of fixed input (

= 1, 2)

where :

= land under rice (hectares under rice) for each farm j

35

= capital used in farm j (sum of total cost of hoes and pangas)10.

μ = 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

α0, αi , rik ø , β , , φ , δ0 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 ( ) and imputed cost of family labor ( ) 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 10 The items were assumed to be used up in one production year therefore no depreciation is necessary

36

Djato, 1996 and 1997; Sharma et al., 1999). Ali and Flinn, (1989) treat family labor as

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 female’s hour of work is

not equivalent to a man’s hour. The practice has therefore been to value women’s hour

differently. In a number of cases, it has been put at 75% of men’s. Basing on the

prevailing average wage rate, Abdulai and Huffman (2000) treated females’ and

children’s labor as equivalent to half of the man’s. 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; children’s labor was 50% of the adult

labor. Variable (fertilizer and insecticide) use in production enhance productivity

37

Table 3.1: Variables Included in the Frontier Profit Function Models and their Descriptions

Variable Descriptions Expected

sign

Normalized profit of the th farm defined as

gross revenue less variable cost divided by farm

specific price (dependent). Note the j is

suppressed.

Variables

Normalized cost of hired labor divided by price

of rice.-ve

Normalized cost of “other inputs” (fertilizer and

pesticides) divided by price of rice. -ve

Normalized imputed cost of family labor on the

farm x the prevailing wage rate divided by price

of rice.

-ve

Fixed factors

Land under rice in hectares on farm +ve

Cost of capital (hoes and pangas) used in farm -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.

38

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

multiplying the quantity of each input by their respective prices and treated as ( )

(Table 3.1). It is hypothesized that the cost of inputs affects profit efficiency negatively.

Land ( ) 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 ( ) 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 d’Ivoire.

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.

List of Variables Descriptions Expected sign

μ Inefficiency effects

Intercept term

Non farm employment 1= have non-farm employment = 0 other wise

Education level of a respondent in years -ve

Extension service visits to farm 1= received extension visits 0=otherwise

-ve

Credit access by farmer 1=access 0 = otherwise -ve

Experience measured by years in rice production by farmer

-ve

Degree of specialization in rice (acreage in rice/total crop acreage) in farm

-ve

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 ( ), 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 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 ( ) 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

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

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

and Butaleja. River Mpologoma, whose source lies on the slopes of Mt Elgon, drains the

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

44

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 (1993-

2000) 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 scheme’s office that served as a

sampling frame. In case of Tororo district, large-scale farmers, who have opted to divide

up their farms in ¼ 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/secretary’s 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/Cook-

Weisberg 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 Komogrov-

Smirnov (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: Skewness and Normality Variables (Unstandardized)- Translog Model

Statistics Profit Hired wage rate

Family wage

Other inputs

Rice Hectares

Capital Experience (years)

Mean 772.6 204.9 465.2 222.0 1.3 25491.2 11.7

S.E. 18.2 29.7 13.2 9.9 1982.2 .54

Std 114.3 313.7 512.4 223.7 1.7 34160.9 9.1

Variance in coefficient

3879420 98427 262597 50038 2.91166967247

83.3

Skewness 12.7 3.4 3.9 3.4 6.1 7.6 1.4

Kurtosis 193.5 14.9 25.4 16.9 56.7 88.7 2.3

Minimum -311.5 .00 10.0 10.0 .05 00 1.0

Maximum31187.0 2239.3 4860.0 1929.4 20.5 46000 51.0

Source: computed from Field survey

51

Table 3. 4: Skewness and Normality Variables (Standardized)- Translog Model

Statistics Profit Hired

wage

rate

Family

wage

Other

inputs

Rice

Hectares

Capital Experience

(years)

Mean 6.09 4.95 5.67 5.02 -0.17 9.75 2.14

S.E. .06 0.08 0.06 0.05 0.05 0.05 0.05

S.D 1.09 1.22 1.06 0.89 1.02 0.87 0.86

Variance in

Coefficient1.19 1.51 1.14 0.81 1.04 0.77 0.75

Skewness -0.16 -1.01 -0.68 -0.26 -0.46 0.06 -0.44

Kurtosis 1.17 3.81 0.85 0.07 0.45 0.48 -0.24

Minimum 2.3 -1.29 2.3 2.3 -3.0 6.91 0

Maximum 10.4 7.71 8.49 7.56 3.02 13.04 3.93

Source: computed from Field survey

The results of the computed presented Cohen’s 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 statisticConstant 7.65 15.2 30.4Cost of hired labor 0.12 1.71 23.42“Other inputs” -0.02 -3.27 -6.54Imputed cost of family labor -0.28 -25.5 -0.51Rice hectares 0.05 7.26 14.52Capital 0.13 4.22 8.44Cost of hired labor x “other inputs” -0.03 -1.62 -3.24Cost of hired labor x Imputed cost of family labor

-0.01 -1.71 -3.42

Cost of hired labor x rice hectarage 0.02 1.19 2.38Cost of hired labor x capital 0 0.43 0.86“Other inputs” x Imputed cost of family labor

0 0.04 0.09

“Other inputs” x rice hecatrage 0.13 3.24 6.48“Other inputs” x capital 0.03 1.74 3.48Imputed cost of family labor x rice hectarage -0.04 -1.52 -3.04Cost of family labor x capital 0.04 2.72 5.44Rice hecatarage x Capital -0.05 -2.68 -5.36Cost of hired labor2 0.01 1.84 3.68“Other inputs”2 -0.02 -1.23 -2.46Imputed cost family labor2 -0.01 -1.12 -2.24Rice hectarage2 -0.01 -4.33 -8.66Capital2 -0.03 -4.03 -8.06Constant -2.75 -7.91 -15.82Non-farm employment -0.32 -1.82 -3.64Education 0.16 2.67 5.34Experience 0.32 7.52 15.04Credit access 0.37 2.37 4.74Extension services 0.24 1.86 3.72Degree of specialization 0.23 3.67 7.34Source: 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

3.6 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 γ is conducted, with the β parameters (excepting β0) set the

OLS values and the β0 and σ2 parameters adjusted according to the corrected ordinary least

squares formula as set in Coelli (1995). Any other parameters (μ,η or δ’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).

The likelihood function is expressed in terms of the variance parameters, σ2 = +

and γ = / (Rahman, 2002, 2003).

54

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

distribution of inefficiency effects (error terms μ and ν) 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 Cobb-

Douglas (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, Tororo’s 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 Government’s

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

Characteristics Tororo Pallisa LiraMean Mean Mean

Age of household head (years)

41.09a

(1.22)140.02 a

(1.39)41.22a

(1.43 )

Household size 9.37 a

(0.47)8.41 a

(0.49)8.36 a

(0.66)Number of children 6.42a

(0.35)5.36 bc

(0.31)5.40 ac

(0.38)Size of land (Ha) 1.77 a

(0.12)2.87 b

(0.34)2.02 a

(0.20)Total (Ha) 1.25 a

(0.08)2.14 b

(0.20)1.13 a

(0.10)Rice (Ha) 0.57 a

(0.04)0.72 b

(0.09)0.18 c

(0.02)Imputed family labour cost Ha (Ush)

137475.47a

(44.57)101518.50b

(94.59)117682.50c

(22.86)Hired Labour cost per Ha (Ush)

57280.66 a

(6035.95)46994.06 a

(9010.90)10374.81 b

(1575.41)“other input cost” per Ha (Ushs)

4897.57 a

(1255.92)2557.73 a

(926.12)2895.13 a

(852.01)Experience (years) 14.71 a

(0.81)12.16 b

(0.96)4.75 c

(0.30 )Degree specialization 3.23a

(0.09)3.90b (0.11)

3.56c

(0.15)Number of plots 1.72 a

(0.09)1.21 bd

(0.05)1.12 cd

(0.05)n 138 104 55

Source: Computed from Field Survey Data.Superscripts with the different letters are significantly different at the 0.10 level1 () figures in brackets are standard errors

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

Characteristics Tororo Pallisa Lira

Proportion Proportion Proportion

Primary

0.507 a

(0.043)

0.683 bc

(0.046)

0.600 ac

(0.067)

Secondary / Tertiary

0.275 a

(0.038)

0.212 a

(0.040)

0.236 a

(0.058)

Not attended

0.217 a

(0.035)

0.106 bc

(0.030)

0.164 ac

(0.050)

Credit Access0.355 a

(0.041)0.481 ac

(0.049)0.564 bc

(0.067)Credit from informal

source0.857 a

(0.045)0.98 b

(0.045)0.774 a

(0.056)

Non-farm employment0.28a

(0.04)0.31a

(0.04)0.18a

(0.06)n

138 104 55Source: Computed from Field Survey Data.Superscripts with the different letters are significantlydifferent at the 0.10 level1 () 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 ( ) and random ( )

or = / ( + ). The gamma () is bounded by 0 and 1, where if is zero

60

inefficiency effects are not present in the model, and if it is one 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 researcher’s 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

statistic has approximately a 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 C-

D 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 Testing for the Models and its Inefficiency Effects

Hypotheses Pooled Sample Tororo Pallisa Lira

1) H0:C-D is an adequate representation of Profit frontier Function

LLU -80.13 -19.41 55.48 17.14

LLR -127.44 -47.91 26.80 -10.34LL 94.62 57.00 29.68 27.48

Critical value*(5%)

13.40 13.40 13.40 13.40

Decision reject H0 reject H0

reject H0 reject H0

2) H0: = 0 = d =0, d

Each farm is operating on profit frontier

LLU -80.13 -19.41 55.48 17.14

LLR -186.04 -81.97 -59.07 -29.77LL 212.22 125.12 211.10 93.82Critical value(5%)

10.37 10.37 10.37 10.37

Decision reject reject reject reject

3) H0: 0 = d =0, d

Variables included in the inefficiency effect model have no effect on the level of profit inefficiency.

LLU -80.13 -19.41 -55.48 17.48

LLR -166.23 -72.15 -45.74 -18.15LL 86.10 105.48 101.2 35.29Critical value(5%)

13.40 13.40 13.40 13.40

Decision reject reject reject rejectSource: 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 MLE

Coefficients p-v1 Coefficients p-v1

Constant 7.04 0.00 8.11 0.00Cost Hired labor ( ) -0.14 0.09 -0.12 0.03

“Other inputs” ( ) -0.01 0.74 -0.01 0.67Imputed cost of family labor (Ushs/ha) ( )

-0.12 0.00 -0.20 0.00

Rice acreage ( ) 0.17 0.00 0.09 0.00

Capital ( ) 0.02 0.07 0.02 0.05

x -0.01 0.74 -0.03 0.07

x -0.02 0.18 0.01 0.07

x 0.02 0.41 0.03 0.15

x 0.03 0.04 0.02 0.12

x -0.08 0.09 -0.05 0.14

x 0.00 0.89 0.09 0.07

x 0.02 0.58 0.03 0.18

x 0.05 0.31 0.01 0.78

x 0.04 0.20 0.04 0.07

x -0.03 0.47 -0.07 0.03

x -0.01 0.67 0.01 0.15

x 0.02 0.41 0.01 0.44

x 0.02 0.46 0.00 0.89

x 0.02 0.60 -0.04 0.18

x 0.03 0.13 -0.03 0.01Sigma-squared 0.24 0.27 0.00Gamma 0.69 0.00Log likelihood -

166.2280.13

n 253Source: 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

66

on the estimated coefficient for this variable in two districts (Tororo and Pallisa) and the

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 UgShsOLS MLE

Variables Coefficients p-v1 Coefficients p-v1

Constant 7.25 0.00 7.37 0.00Cost of hired labor ( ) 0.00 0.89 0.14 0.00

“Other inputs” ( ) -0.05 0.14 -0.03 0.00Imputed cost of family labor (Ushs/ha) ( )

-0.06 0.18 -0.22 0.00

Rice hectarage( ) 0.14 0.00 0.03 0.00

Capital ( ) 0.00 0.89 0.08 0.00

x -0.00 0.84 0.00 0.64

x -0.02 0.46 0.00 0.55

x 0.07 0.18 0.03 0.00

x 0.03 0.41 0.00 0.78

x -0.05 0.74 -0.09 0.00

x 0.16 0.56 0.05 0.55

x -0.01 0.88 -0.01 0.67

x 0.13 0.30 0.06 0.05

x 0.07 0.30 0.04 0.01

x 0.17 0.46 -0.08 0.12

x -0.02 0.18 -0.02 0.00

x 0.03 0.69 0.09 0.00 x -0.03 0.47 0.03 0.01

x -0.28 0.14 -0.15 0.00

x -0.62 0.74 -0.01 0.01

Sigma-squared 0.55 0.00Gamma 55.48 1.00 0.00Log likelihood -72.15 -19.41n 123Source: 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 MLE

Variables Coefficients p-v1 Coefficients p-v1

Constant 6.51 0.00 6.56 0.00Cost of hired wage ( ) -0.06 0.46 -0.03 0.00

“Other inputs”( ) 0.00 0.85 -0.01 0.00Imputed cost of family labor (Ushs/ha) ( )

0.01 0.89 -0.02 0.00

Rice hectarage ( ) 0.17 0.00 0.04 0.00

Capital ( ) 0.05 0.34 0.02 0.00

x -0.06 0.46 0.00 0.44

x -0.02 0.47 -0.02 0.12

x 0.12 0.07 -0.01 0.44

x 0.07 0.07 0.01 0.18

x 0.04 0.78 0.10 0.00

x 0.40 0.41 0.16 0.15

x -0.07 0.85 0.11 0.01

x -0.02 0.86 -0.08 0.07

x -0.05 0.65 -0.06 0.00

x -0.22 0.47 -0.04 0.07

x -0.04 0.03 0.00 0.44

x 0.05 0.89 -0.15 0.03 x 0.02 0.68 0.01 0.00

x -0.55 0.03 -0.03 0.44

x 0.01 0.85 -0.02 0.14

Sigma-squaredGamma 1.00 0.00Log likelihood -45.74 55.48n 91Source: 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

Variables Coefficients p-v1 Coefficients p-v1

Constant 19.64 0.89 89.25 0.00Cost of hired labor ( ) -3.3 0.74 0.00 0.91

“Other inputs”( ) 7.64 0.44 -7.43 0.00Imputed cost of family labor (Ushs/ha) ( )

-9.53 0.05 1.81 0.01

Rice hectarage ( ) 8.68 0.44 8.94 0.00

Capital ( ) 0.05 0.00 0.30 0.05

x -0.36 0.62 -0.03 0.91

x 0.02 0.47 0.06 0.18

x 0.57 0.01 0.14 0.30

x 0.30 0.00 -0.01 0.67

x 2.03 0.00 0.48 0.05

x -4.19 0.18 2.32 0.00

x -3.96 0.00 1.77 0.00

x -1.48 0.00 -0.99 0.00

x -0.87 0.02 -0.55 0.00

x 1.78 0.74 -1.60 0.00

x -0.15 0.03 0.00 0.91

x 2.17 0.00 -1.21 0.00 x 0.06 0.09 0.06 0.30

x 1.60 0.58 0.93 0.00

x 0.94 0.34 -0.31 0.00

Sigma-squared 0.14 0.00Gamma 1 0.00Log likelihood 18.15 17.14n 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 (γ). 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 Pallisa LiraEfficiency Index

Freq. % Freq. % Freq. % Freq. %

<30 27 11 18 15 14 15 06 1531-40 12 5 14 11 04 04 05 1341-50 32 13 11 09 00 0 02 0551-60 31 12 12 10 07 08 02 0561-70 18 7 10 81 08 09 00 071-80 34 13 13 11 04 04 04 1081+ 99 39 45 37 53 58 20 51n 253 100 123 100 91 100 39 100

Mean 66.4 64.7 74.5 70.3Min 04 04 02 10Max 95 1.00 1.00 1.00S.D. 22.6 27.0 27.3 30.8Source: Computed from the Survey Data

71

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

not unique to Uganda. Similar results have been reported by other researchers elsewhere.

Abdulai and Huffman’s (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

District Mean n Std. Deviation Minimum Maximum

Tororo 489692.50 123 11719.3852 469514.00 509871.00

Pallisa 301571.00 91 11915.3694 281271.00 321664.00

Lira 114304.00 39 24029.4971 74261.00 154347.00

Total 364162.03 253 137741.79 74261.00 509871.00

Source: Field Survey Data.

Table 4.6 Tests of Significance of Mean Profit loss

District Kolmogorov-Smirnov Z P-values

Tororo- Pallisa 7.232 .000

Tororo- Lira 5.442 .000

Pallisa- Lira 5.225 .000

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.

73

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

Tororo Pallisa Lira

Prices and Fixed Inputs Price

elasticity

p-v1 Price

elasticity

p-v1 Price

elasticity

p-v1

With respect to

Cost of hired labor 0.14 0.00 -0.03 0.00 0.00 0.91

“other inputs” -0.03 0.00 -0.01 0.00 -7.43 0.00

Imputed cost of family

labor

-0.22 0.00 -0.02 0.00 1.81 0.01

Area under rice 0.03 0.00 0.04 0.00 8.94 0.00

Capital 0.08 0.00 0.02 0.00 0.30 0.05

Source: Computed from profit function resultsp-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

74

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

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 μ

Pooled Tororo Pallisa Lira

Coeff p-v1 Coeff p-v1 Coeff p-v1 Coeff p-v1

Constant ( ) 2.09 0.00 3.84 0.00 3.51 0.00 2.28 0.00

Non-farm employment ( ) 0.37 0.01 0.06 0.89 -0.95 0.03 -0.93 0.00

Education ( ) -0.14 0.00 -0.25 0.00 -0.16 0.00 -0.30 0.00

Extension services( ) -0.16 0.00 -0.26 0.00 -0.47 0.00 -0.28 0.03

Credit access( ) -0.25 0.15 -0.55 0.05 -0.46 0.03 -0.24 0.44

Experience ( ) -0.07 0.67 -0.44 0.17 -0.49 0.00 0.06 0.78

Degree of specialization( ) -0.08 0.00 -0.14 0.00 -0.15 0.00 -0.03 0.17

Source: Field Survey Datap-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

Government’s 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 Actual Profit 000’s Shs

Profit loss 000’s Shs

Efficiency Scores

Education Level

n Mean Mean Meannone 24 968.6 a

(3.8)524 a

(3.7)45.9 a

(2.5)primary 61 962.6 a

(2.3)489 b

(2.3)49.2 b

(1.6)secondary 33 953.5 b

(3.2)451 c

(3.2)52.7 c

(2.1)tertiary 5 637.0 c

(10.6)286 d

(8.1)50.1 d

(5.5)

Credit AccessYes 43 1104.0a 467 a 57.7 a

No 80 1069.5b 523 b 51.1 b

Extension services

Yes 22 1114.4a 487 a 56.3 aNo 101 1118.1b 549 b 50.9 b

Degree of specialization

0-25% 34 815.4a

(17.6)515.5 a

(2.7)59.1 a

(0.5)25.1-50% 55 1081.4 b

(20.4)514.8 a

(3.5)57.4 a

(0.9)50.1+ 34 1281.4 c

(17.6)507.1 b

(5.5)55.3 b

(1.2)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

didn’t 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

n Actual profit000’s Ushs

Profit loss000’s Ushs

Efficiency

Education

Nil 8 672.4 a

(7.4)351 a (6.3)

47.8 a (1.5)

Primary 64 659.5b

(2.2)337 b (2.2)

48.9 a (0.5)

Secondary 17 574.6 c

(4.6)281c

(4.3)51.1 b (1.1)

Tertiary 2 523.5d

(29.2)245 d

(12.6)53.2 a (3.1)

Credit AccessYes 41 615.6a 269 a 56.3 a

No 50 586.8b 338 b 42.4 b

Extension access

Yes 8 711.9 a 294 a 58.7 a No 83 645.8b 361b 44.1 b

Degree of specialization

0-25% 36 1091.3a

(17.6)513.9a

(4.1)57.9a

(0.9)25.1-50% 33 1137.8a

(20.4)508.2b

(4.7)57.1a

(1.2)50.1%+ 22 924.8b

(18.5)499.8c

(6.9)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

Education level

n Actual profit000’s Ushs

Profit loss000’s Ushs

Efficiency

Education

Nil 8 199.1a

(4.9)131 a (2.4)

34.2 a (1.7)

Primary 23 237.3b

(2.6)126 b (1.4)

46.9 b (1.0)

Secondary

6 191.3 a (6.1)

97c

(2.8)49.3 c b

(2.0)Tertiary 2 153.7c

(20.0)71 d (4.9)

53.8 c d

(3.4)

Credit AccessYes 21 155.0 a 73 a 52.9 a

No 18 307.5b 155 b 49.6 b

Extension accessYes 20 174.7 a 87 a 50.2 a

No 19 323.8 b 169 b 47.8 b

Degree of specialization

0-25% 15 1151a 510 a 59.4a

25.1-50%

17 1056a 514a 58.2a

50.1+ 7 975.5b 521a 57.6b

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.

5.1 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 government’s 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 d’Ivoire. Agricultural Economics, 16, 47-53.

Akinwumi, A., and Djato, K. K. (1996). Farm size, relative efficiency and agrarian policy in Cote d’Ivoire: 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 , 185-208.

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. Reading, Massachusetts: Addison-Wesley.

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 coal-fired 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 input-specific 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), 84-96.

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), 163-173.

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., Ogenga–Latigo, 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 master’s 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, 341-348.

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 master’s 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, Kompendiet-Goteborg, 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

100

101

102

103

104

8

8

105

106

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

107

REGRESIÓN 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 0.1363beta14 | 7.1337 0.1608beta12 | 6.9273 0.1853beta10 | 6.7210 0.2098beta_9 | 6.5147 0.2344beta_6 | 6.3083 0.2589beta11 | 6.1020 0.2834beta20 | 5.8957 0.3080beta18 | 5.6893 0.3325beta17 | 5.4830 0.3570beta13 | 5.2767 0.3816beta_7 | 5.0703 0.4061beta_1 | 4.8640 0.4306beta16 | 4.6577 0.4551beta_8 | 4.4513 0.4797beta_4 | 4.2450 0.5042beta_2 | 4.0387 0.5287beta_5 | 3.8323 0.5533beta19 | 3.6260 0.5778delta_6 | 3.4197 0.6023beta_3 | 3.2133 0.6269delta_2 | 3.0070 0.6514delta_4 | 2.8007 0.6759delta_3 | 2.5943 0.7004delta_5 | 2.3880 0.7250

108

delta_1 | 2.1817 0.7495delta_7 | 1.9753 0.7740--------------+------------------------Mean VIF | 4.5545

.

. Shapiro-Wilk W test for normal dataVariable | Obs W V z Prob|>z|--------------+------------------------------------------------------ln_profi | 253 0.99060 1.72000 1.26200 0.10349beta_1 | 253 0.99086 1.67292 1.15065 0.13600beta_2 | 253 0.99111 1.62585 1.03931 0.16851beta_3 | 253 0.99137 1.57877 0.92796 0.20102beta_4 | 253 0.99163 1.53169 0.81662 0.23353beta_5 | 253 0.99189 1.48462 0.70527 0.26604beta_6 | 253 0.99214 1.43754 0.59392 0.29855beta_7 | 253 0.99240 1.39046 0.48258 0.33106beta_8 | 253 0.99266 1.34338 0.37123 0.36357beta_9 | 253 0.99292 1.29631 0.25988 0.39608beta10 | 253 0.99317 1.24923 0.14854 0.42859beta11 | 253 0.99343 1.20215 0.03719 0.46110beta12 | 253 0.99369 1.15508 -0.07415 0.49361beta13 | 253 0.99395 1.10800 -0.18550 0.52612beta14 | 253 0.99420 1.06092 -0.29685 0.55862beta15 | 253 0.99446 1.01385 -0.40819 0.59113beta16 | 253 0.99472 0.96677 -0.51954 0.623648beta17 | 253 0.99497 0.91969 -0.63088 0.65615beta18 | 253 0.99523 0.87262 -0.74223 0.68866beta19 | 253 0.99549 0.82554 -0.85358 0.72117beta20 | 253 0.99575 0.77846 -0.96492 0.75368delta_1 | 253 0.99600 0.73138 -1.07627 0.78619delta_2 | 253 0.99626 0.68431 -1.18762 0.81870delta_3 | 253 0.99652 0.63723 -1.29896 0.85121delta_4 | 253 0.99678 0.59015 -1.41031 0.88372delta_5 | 253 0.99703 0.54308 -1.52165 0.91623delta_6 | 253 0.99729 0.49600 -1.63300 0.94874delta_7 | 252 0.99972 0.05100 -6.93200 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

109

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 ShsVariables OLS MLE

Coefficients p-v coefficients p-v1

Constant ( ) 6.96 0.00 7.43 0.00

Cost of hired labor (Ushs /Ha) ( ) -0.03 0.17 -0.01 0.17

“Other input costs” (Ushs/Ha) ( ) -0.03 0.00 -0.02 0.07

Imputed cost of family labor -0.11 0.00 -0.12 0.00

110

(Ushs/Ha) ( )

8Area under rice (ha) ( ) 0.20 0.00 0.13 0.00

Capital ( ) 0.02 0.01 0.02 0.07

Sigma –squared - 0.31 0.00

Gamma - 0.63 0.00

Log likelihood -186.65 -127.44

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

111

Table A2: Coefficient Estimates of C-D Frontier Profit Function-Tororo District

Dependent Variable = Normalized Profit in Ug Shs

Variables OLS MLE

Coefficients p-v1 coefficients p-v1

Constant ( ) 7.20 0.00 7.89 0.00

Cost of hired labor (Ushs /Ha) ( ) -0.02 0.47 0.00 0.91

“Other input costs” (Ushs/Ha) ( ) -0.04 0.06 -0.05 0.00

Imputed cost of family labor

(Ushs/Ha) ( )

-0.06 0.00 -0.08 0.00

Area under rice (Ha) ( ) 0.16 0.00 0.10 0.00

Capital ( ) 0.00 0.89 0.00 0.91

Sigma –squared 0.23 0.89 0.03

Gamma 0.92 0.00

Log likelihood -81.99 -47.91

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

112

Table A3: Coefficient Estimates of C-D Frontier Profit Function-Pallisa District

Dependent Variable = Normalized Profit in Ug Shs

Variables OLS MLE

Coefficients p-v1 coefficients p-v1

Constant ( ) 5.94 0.00 6.78 0.00

Cost of hired labor (Ushs /Ha) ( ) 0.04 0.09 -0.03 0.00

“Other input costs” (Ushs/Ha) ( ) -0.01 0.67 -0.02 0.00

Imputed cost of family labor

(Ushs/Ha) ( )

-0.01 0.65 -0.02 0.03

Area under rice (Ha) ( ) 0.18 0.00 0.05 0.00

Capital ( ) 0.02 0.63 0.01 0.17

Sigma –squared 0.23 0.40 0.00

Gamma 0.99 0.00

Log likelihood -55.08 26.80

Source: Field Survey Data.p-v1 are 2p values computed from t-ratios

113

Table A4: Coefficient Estimates of C-D Frontier Profit Function-Lira District

Dependent Variable = Normalized Profit in Ug Shs

Variables OLS MLE

Coefficients p-v1 coefficients p-v1

Constant ( ) 20.51 0.01 20.02 0.00

Cost of hired labor (Ushs /Ha) ( ) -0.06 0.15 -0.01 0.67

“Other input costs” (Ushs/Ha) ( ) -1.28 0.00 -1.03 0.00

Imputed cost of family labor

(Ushs/Ha) ( )

0.06 0.60 0.08 0.52

Area under rice (Ha) ( ) 1.41 0.00 0.96 0.00

Capital ( ) 0.66 0.00 0.12 0.15

Sigma –squared 0.32 0.38 0.05

Gamma 0.95 0.00

Log likelihood -29.77 -10.32

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

114

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

Efficiency Scores

Pooled Tororo Pallisa Lira

n % n % n % %<40 24 9.4 8 6.5 13 14.29 9 23.141-50 6 2.3 5 4.07 5 5.49 4 10.351-60 17 6.7 2 1.62 5 5.49 2 5.161-70 20 7.9 0 0 9 9.89 4 10.371-80 20 7.9 22 17.89 2 2.20 3 7.781+ 166 65.6 86 69.92 57 62.63 17 43.6n 253 100 123 100 91 100 39 100Mean 77.2 79 75 65Min 9.1 02 04Max 95.0 98 96Source: Field Survey Data.

2.3 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

115

had influence in Pallisa district in translog model but only so in Tororo district.in C-D

model.

Table A6: Determinants of Firm-Specific Profit Inefficiency in Rice Production-

C-D model

Dependent: Variable = Inefficiency effects (μ)

Pooled Tororo Pallisa Liracoefficients p-v1 coefficients p-v1 coefficients p-v1 Coefficients p-v1

Constant (

)4.10 0.001 2.84 0.16 3.14 0.00 1.07 0.18

Non-farm

employment

( ) 0.14 0.46 -1.55 0.91 -0.78 0.12 0.46 0.46

Education (

)-0.24 0.00 -0.34 0.03 -0.16 0.00 -0.17 0.17

Extension

services ( )-0.14 0.00 0.19 0.77 -0.50 0.00 -0.25 0.18

Credit

Access ( )-0.27 0.17 -1.40 0.03 -0.32 0.00 -0.02 0.91

Experience

(yrs) ( )-0.19 0.14 -0.46 0.14 -0.27 0.17 -0.07 0.95

Degree of specialization ( ) -0.21 0.03 -0.50 0.12 -0.16 0.00 -0.14 0.17Source: Field Survey Data.p-v1 are p values computed from t-ratios

116

Figure 8: Profit Efficiency Distribution C-D Model

117

Figure 9: Profit Efficiency Distribution Translog Model

118

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

119

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 No. Children <18

Males Females Males Females

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

Member of

household

Age

(years)

Education

level

Main

occupation

No. of

years

Secondary

occupation

No. of

years

120

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 Area/acres Type of ownership

1.

2

3

4

10. How did you/they get the plots?

Plot Source of ownership

1

2

3

4

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

121

12. Who in the family carried out the following agricultural activities in the plots

mentioned above in the last season?

Children <18 years

Activity School going Not going to

school

Hsl Head Spouce1 Spouce2 Spouce 3 male female male female

Opening land

Seed bed

prep

Planting

Weeding

Scaring bird

Harvesting

Threshing

Marketing

122

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

Activity Hsld

head

Spouce1 Spouce2 Spouce3 Children <18years

School going Not going

Male Female Male Female

Opening of

land

Seed bed

preparation

Planting

Weeding

Scaring birds

Harvesting

Threshing

123

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 Hired labor

Activity Plot 1 Plot 2 Plot 3 Plot 4 Wage rate (shs)

8

8

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 Type (Quantity Units and Price)

1

2

3

4

17. If no, why?-----------------------------------------------------------------------------------------

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

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

124

18.What farm implements did you use in rice production?

Type of equipment Number owned Year bought Value when bought shs.

Hoes

Pangas

Ox plough

Basket

Wheel barrow

Others (specify)

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?

Ushs…………………year……………………………………………..

126

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

Last season

Type of credit Formal Informal

From where/whom

Amount

Interest rate

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

128

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 2) took to the mill 3) took to the local

market

4) Used any of the methods depending on convenience.

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