r... 0..
3.1.5:: .6

 

.9
ct.

. I. I. .1 . 9.1.

:

.3»... 5.5.

2.

.53.. e

3
A. 1

car.

_.H. ,J

u.

“HIT.

as:

 

a; .6.

N2.

3
:

3,.

5! .3. t

4.1.»...

 

 

 

 

 

 

 

 

r n A 3
4... .15

|‘. J

VI:

 

 

 

{(1.71 ‘

an.

 

3
wt

4..
3,an 5.5.: . .,

- .axwmfi

.. . .. m. ,. .. ., ms Rxfiuh
561.“...zfimfi4s . H. : . :5 .fifiwtv. 3a..
z e- ‘ ;.........u..b..3.9n. . . Wm. . .J....ku . a

. . , .. . 3...

 

. . .
25:. 3......) .. .0

 

 

a»?
a...“

x
. .11

$412.35.... ..
u...o.u.31?3

iv... .91) knit. .1

RX“.

«35 _
.tl‘lu‘nyl‘ .
:9}:

(Frf...
n); .X...

.! .mwvmflhfi ..
. til 3 ‘1
: {zfifl h;
a.) ‘9' A .AO.‘ ‘- 31., M

 

a}!!! :1 .I’.i..c
:71} . I.
. {it

3...? .19.
t: 9...; A

n .
. .\‘-‘T
(J ‘u

.u'no

 

 

1......3 .1... .3

lIBRARY
Michigan State
University

This is to certify that the

dissertation entitled'

EFFICIENCY AND PRODUCTIVITY OF

NEPALESE AGRICULTURE

presented by

NAZMUL CHAUDHURY

has been accepted towards fulfillment
of the requirements for

PhD d . AGRICULTURAL ECONOMICS
egree m
m

@4254

Major prof or

 

 

Date éflfl/OZJ/Jflo 0

MS U is an Affirmative Action/Equal Opportunity Institution 0-12771

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJC|RCIDateDue.p65-p.15

EFFICIENCY AND PRODUCTIVITY or NEPALESE AGRICULTURE
By

Nazmul Chaudhury

A DISSERTATION

Submitted to
Michigan State University
in partial fulfilment of the requirements
for the degree of

DOCTOR OF PHILOSOPY

Department of Agricultural Economics
Department of Economics

2001

ABSTRACT
EFFICIENCY AND PRODUCTIVITY OF NEPALESE AGRICULTURE
By

Nazmul Chaudhury

In the first chapter, a stochastic production fi'ontier fiamework is used to examine the
technical efficiency of rice production for a sample of irrigated farmers in the Rupandehi
district of Nepal. Coefficient estimates from the production frontier indicate that: (a) source
of irrigation and varietal choice are the two most important factors which enhance rice yields;
(b) mid-season water stress and long term non—use of organic fertilizer, are the two primary
factors which adversely efl‘ect rice yields. Farm level technical efficiency measures derived
from the production frontier model, suggests that on average, rice yields could have
potentially been increased by slightly over half a metric ton per hectare, corresponding to a
18% average increase in output, via a more efficient utilization of available resources at the
current state of technology. I then explore for the relationship between technical efficiency
and two sets of variables: (1) farmers’s grasp of agronomic principles and knowledge; (2)
socio-economic environment in which the farmer operates. I am particularly interested in
examining how education and land ownership size is related to efficiency, two specific issues
which have received considerable attention in the literature. I find a significant relationship
between secondary education and efficiency. Average technical inefiiciency in rice production
is reduced by 16% in plots farmed by households in which the primary farm manager has
completed more than five years of schooling. I also find a significant inverted U-shaped

relation between technical inefiiciency and land ownership size (i.e., a significant U-shaped

efficiency-size relationship). However, the significance of this efficiency-size relationship
appears to be sensitive to model specification/endogeneity bias.

In the second chapter, a stochastic production frontier approach embedded in a meta-
production function framework is used to : (a) estimate the district level rates of technical
change in Nepalese agriculture based upon econometric estimation of the underlying
production technology; (b) derive point estimates of district level technical efficiency from
parametric estimation of the production technology; and (c) given that I am interested in
comparing efficiency levels across districts, I also construct confidence intervals around the
time varying technical efficiency estimates, using non-parametric bootstrap methods.

When literacy is included in the analysis, there appears to be no significant growth
in TFP in the Terai region, while there appears to be a severe decline in TFP growth in the
Hill region. Significant negative TFP grth rates in the Hill region might reflect a pernicious
decline in the quantity/quality of the natural resource base, however, it is not possible to
explicitly examine the interplay between natural resource degradation and agricultural
productivity given the data available for this study. This study also highlights the fact that
there exists substantial scope for increasing output via a better utilization of existing inputs
and technologies in the Terai region. For example, agricultural output in 1991 could have
been increased by 40 % in the Terai region via a more eflicient utilization of existing inputs
and technologies. There also exists tremendous potential for increasing output in the Hill
region, however, this potential is perhaps being squandered due to natural resource

degradation.

ACKNOWLEDGMENTS

I would first and foremost like to express my gratitude and sincere appreciation to my
major advisors, Professors John Strauss and Thomas Reardon, for their unwavering support,
guidance and encouragement. I would also like to thank the other members of my
committee, Professors Peter Schmidt, Scott Swinton and Richard Harwood. I remain
indebted to Professor Schmidt for having the patience and generosity to guide me through
a challenging tour of the frontiers of ‘stochastic production-frontier’ theory. I am also
indebted to Professor Harwood given that I have benefitted enormously from his erudition
of the field of sustainable agriculture and extensive knowledge of Nepalese agriculture. Also,
Professor Harwood made it financially viable for me to pursue graduate studies at Michigan
State University, through his generous support via the MC. Mott Fellowship in Sustainable
Agriculture.

1 am grateful to Dr. Peter Hobbs who allowed me to gain full access to the
CIMMYT/NARC data set used in the first chapter of this dissertation; I am also indebted to
Dr. Hobs for sharing with me his insightfirl reflections on the complex agronomical,
economical, political and informal institutional factors which shape agricultural performance
in South Asia. I am also grateful for the financial and institutional support provided to me
by the International Maize and Wheat Research Institute (CIMMYT), with particular thanks
to Dr. Larry Harrington and Dr. Prabhu Pingali. I would also like to thank Mr. Dongol of the
Central Bureau of Statistics (Nepal) and Dr. Champak Pokharel of the National Planning

Commission (Nepal) for allowing me full access to their data archives - the second chapter

iv

of this dissertation would not have been possible without their Herculean effort to help me
locate rather elusive data. In particular, Mr. Dongol spent an enormous amount of his
valuable time and effort to help me - I will forever be grateful for his magnanimity.

I am grateful to my dear friends, Janet Owens, Adnan Ozair, Frederik Schlingemann,
Jouni Paavola, Jean Bumshin, Jim Myers, Kei Kajisa, Veronique Massanet, Hassen
Abdulkerim and Sohail Maklai, for their moral and intellectual support. I am especially
grateful to Janet Owens who was always there, particularly in times of extreme duress, to
help me retain my humanity and to enjoy the simple joys of life which makes this ephemeral
existence an eternal joy.

I would also like to especially thank my sister, Akhi, her compassion and concern
for the non-Benthamite welfare of both human and non-human life forms, and her vigilance
against the hegemony of status quo dogma, has allowed to gain a more holistic appreciation
of sustainable development. I thank my mother for sheltering me under her aegis of love and
tolerance, and for instilling in me the belief that the essence of religion is love, and the holy
book is life itself, with the unfolding chapters of human experiences to help guide the soul.
Finally, I would like to thank my father for his intellectual and moral probity which has

served as an ideal towards which I shall always strive for.

TABLE OF CONTENTS

LIST OF TABLES ................................................... vii
LIST OF FIGURES ................................................... ix
CHAPTER 1

DETERMINANTS OF PRODUCTION EFFICIENCY IN A DYNAMIC
AGRICULTURAL REGION OF NEPAL

I. Introduction .................................................. 1
II. Agricultural Issues in the Terai ................................... 3
III. Data Section ................................................ 6
IV. Analytical Framework, Empirical Procedures, and Findings ............ 12
EMPIRICAL PROCEDURES AND FINDINGS : ................ 15
DETERMINANTS OF TECHNICAL INEFFICIENCY : ........... 18
V. Conclusion ................................................. 27
BIBLIOGRAPHY .................................................... 42
CHAPTER 2
PRODUCTIVITY GROWTH OF NEPALESE AGRICULTURE
I. Introduction ................................................. 49
II. Analytical Framework ........................................ 55
III. Data Section ............................................... 69
IV. Empirical Findings .......................................... 73
Results Without Literacy : Terai Region ........................ 73
Results Without Literacy : Hill Region ......................... 77
Results With Literacy : Terai Region .......................... 80
Results With Literacy : Hill Region ............................ 82
V. Conclusion ................................................. 83
BIBLIOGRAPHY ................................................... 117

vi

CHAPTER 1

Table 81

Table S2

Table 1

Table S3

Table 2

CHAPTER 2

Table 81

Table 82

Table 1

Table 2

Table 2.2 :

Table 3

Table 4

Table 5

Table 6

Table 7

Table 7.2 :

Table 8
Table 9

Table] 0

Table l 1

Table 12

: Within-Estimates of Production Function w/o Literacy - Terai

: Within-Estimates of Production Function w/o Literacy - Hill

LIST OF TABLES

: Changes in Plot level Macro and Micro Nutrients ............

: Summary Statistics of Production Function Variables ..........
: OLS and ML estimation results of Production Function .........
: Summary Statistics of Variables used to explore variation in TE . .

: OLS estimation results of determinants of variation in TE .......

: Grth Rate of Production Function Variables - Terai .........

: Growth Rate of Production Function Variables - Hill ..........

: District level RTC w/o Literacy - Terai .....................

Comparison of RTC to Growth Rates w/o Literacy - Terai ......

: Bootstrap CI for RTC estimates w/o Literacy - Terai ..........
: District level TE 1981 w/o Literacy - Terai ..................

: District level TE 1991 w/o Literacy - Terai ..................

: District level RTC w/o Literacy - Terai .....................

Comparison of RTC to Growth Rates w/o Literacy - Hill .......

: Bootstrap CI for RTC estimates w/o Literacy - Hill ..........
: District level TE 1981 w/o Literacy - Hill ..................

: District level TE 1991 w/o Literacy - Hill ...................
: Within-Estimates of Production Function with Literacy - Terai . .

: District level RTC with Literacy - Terai ...................

vii

3O

31

32

34

35

86

87

.88

89

9O

91

92

93

.94

95

96

97
98

99

100

101

Table 12.2: Comparison of RTC to Growth Rates with Literacy - Terai . . . . 102

Table 13 : Bootstrap CI for RTC estimates with Literacy - Terai ......... 103
Table 14 : District level TE 1981 with Literacy - Terai ................ 104
Table 15 : District level TE 1991 with Literacy - Terai ................. 105

Table 16 : Within-Estimates of Production Function with Literacy - Hill . . 106

Table 17 : District level RTC with Literacy - Terai .................... 107
Table 17.2: Comparison of RTC to Grth Rates with Literacy - Hill ...... 108
Table 18 : Bootstrap CI for RTC estimates with Literacy - Hill .......... 109
Table 19 : District level TE 1981 with Literacy - Hill ................. 110
Table 20 : District level TE 1991 with Literacy - Hill .................. 111
Table ARI :Districts with Local Agricultural Research Centers - Terai ...... 112
Table AR2zDistricts with Local Agricultural Research Centers - Hill ....... 113
Table 21 : Summary Results .................................... 114

viii

CHAPTER 1
Figure 1 :
Figure 2 :
Figure 3 :
Figure 4 :
Figure 5 :

CHAPTER 2
Figure l :

Figure 2 :

LIST OF FIGURES

Long-Run Average Rice and Wheat Yield for Rupandehi District . . 37

District vs. Sample Rice and Wheat Averages ................. 38
Changes in Rice-Yield by Irrigation status of farmer ............ 39
Changes in Rice fields by FYM application of farmer ......... 40
Changes in Rice Yield by education category of farmer ......... 41
TB and Agricultural Research Centers - Terai ................. 115
TE and Agricultural Research Centers - Hills ................. 116

ix

Chapter 1

DETERMINANTS OF PRODUCTION EFFICIENCY IN A DYNAMIC
AGRICULTURAL REGION OF NEPAL

I. Introduction

Nepal, which used to be a food-grain surplus country in the 1970's, has changed
to a food grain-deficit country in the 1990's (Banskota 1992). Domestic cereal production
and food availability per capita is on a decline in Nepal (Ali, Hobbs and Velasco 1993).
Export earnings from both the agricultural and non-agricultural sectors are insufficient to
allow Nepal to pursue a food security policy which primarily depends upon cereal imports
(ADB 1992). Labor absorbing industrial development has yet to emerge as a significant
factor in the Nepalese economy. Thus, Nepal cannot afford to spend its reserves of hard
currency on procurement of cereals from the international market. Nepal must rather rely
upon strategies which enhance domestic cereal production in an arable land-constrained
environment.

Currently there is a dearth of empirical studies on the productivity of Nepalese
agriculture at either the national, regional or the commodity specific level. The
importance of having credible productivity measures of Nepalese agriculture is firrther
highlighted by the troubling fact that there is growing concern and evidence that
productivity growth in various intensive cropping systems of South Asia is either slowing
down or even declining (Hobbs and Morris 1995; Cassman and Pingali 1995; Byerlee
1992). This intensification induced decline in productivity grth could be associated
with various factors such as long term changes in soil physical characteristics/decline in
soil fertility, ground water depletion and water quality degradation (Pinagli and Rosegrant

1993).

Most agricultural productivity studies in South Asia have been at the national level
and have focused on macro-level determinants (e. g., investment in agricultural research
and extension) of productivity change. It is difficult in national level studies to explicitly
control for the effects of micro-level factors (e. g., specific farm level management
practices) on agricultural productivity. Thus, regional and cropping system specific
studies can provide a platform for more “fine-tuned” productivity measures and allow for
exploration of micro-level detemrinants of productivity.

Economic efficiency in production is detemrined by the technique of applying
inputs and levels of application of inputs. Technical efficiency (TE) reflects the firm’s
ability to obtain the maximum possible output fiom a given set of inputs. Allocative
efficiency (AE) reflects the ability of the firm to maximize profits, by equating the marginal
revenue product with the marginal costs of inputs. These efficiency measures depend
upon the factors which determine the firm manager’s technical knowledge, and the socio-
economical environment in which the firm manager operates (Kalirajan 1990). While TE
measures can be estimated from farm level input-output data, estimation of AE measures
require cost/price information. It is not surprising then that TE is the most widely used
measure of farm efficiency in the developing country literature (Bravo-Ureta and Evenson
1994), reflecting the fact that farm level cost of production data is relatively more scarce
compared to mere physical input-output data. Analytically, estimation of farm level AB is
firrther complicated by a host of market failures which plagues the agricultural sector of
many developing countries (e. g., AB in production will fail to hold if the farm household
faces either a credit/liquidity constraint or an input supply constraint). Thus, while farm

level estimation of TE does not (explicitly) require assumptions of market efficiency, most

farm level AE measures are obtained . under the restrictive (and probably not realistic)
assumption of a perfectly competitive market environment.

The focus of this study will be to estimate the technical efficiency of rice
production for a sample of farmers in the rice-wheat cropping-system of the Terai region
of Nepal. I will also explore for the relationship between farm level technical efficiency
and farmer’s technical knowledge and the socio-economic environment in which the
farmer operates. This study is thus, both an effort to fill the gap in the productivity
literature on Nepalese agriculture, and to add to the growing literature on farm level
efficiency of South Asian agriculture‘.

Section II of this paper provides brief background information of agricultural
issues pertinent to the Terai region of Nepal; Section III briefly discusses the nature of the
project which generated the data set used in study; Section IV lays the analytic framework
and empirical procedures used in this study, and presents the findings from the analysis ;
and Section V highlights the main conclusions of this study and suggests relevant policy
implications.

II. Agricultural Issues in the Terai

Nepal can be divided into three agro-ecological zones: The Terai (lowlands); The
Hills (middle mountains); and The Mountains (the High Mountains, and the High
Himalayas). The Terai comprises 54% of the cultivated land area and 45% of the
population (Banskota 1992). The importance of the Terai stems fi'om the fact that it
remains as the only food grain-surplus region in Nepal and is the most favorable area for

intensified agriculture.

 

‘ Estimates of AE could not be conducted due to limited infomation in the data set.

3

Small-holders constitute the majority of farmers in the Terai. Agricultural land is
the primary productive resource in rural Nepal, and accounts for more than 88% of the
value of farm assets in the Terai (Banskota 1992). However the distribution of this
primary means of production is highly skewed, with 16.1 % of the farmers owning 62.8%
of the land (Banskota 1992). Land—holding is also the major determinant of access to
production credit in the Terai (Yadav, Otsuka and David 1992). Existing government
credit programs in Nepal do not appear to be benefitting small farmers, but rather large
farmers have been the predominant beneficiaries (Banskota 1992). It is estimated that
77% of formal lending goes to large-scale farmers (Nelson 1987). In a comprehensive
analysis of the government’s agricultural credit policy in Nepal, Banskota (1992, pp.70)
concludes that “small and marginal farmers have been left out, and access to modern
inputs to enhance productivity is denied as they do not have the power to purchase the
new technology which is embodied in inputs.”

There is a tremendous potential for irrigation in most of the Terai which is
endowed with a relatively high water table (combined with monsoon rains which provide a
good source for recharge). Groundwater sources of irrigation water can be easily tapped
into through small-scale, low-cost irrigation schemes. However the overall irrigation
system in Nepal remains underdeveloped and underutilized. Upadhyaya and Thapa (1994)
find that irrigation is the most significant determinant of MV seed adoption, intensity of
fertilizer use and cropping intensity in Terai (and throughout all major agricultural regions
in Nepal).

The introduction of wheat in the rice mono-culture has allowed farmers to fit in

winter wheat within the traditional rice-based cropping pattern of the Terai. However,

delays in rice harvest interfere with the optimal planting date of wheat. The short time
available for turnaround for planting wheat often leads to sub-optimal land preparation and
use of other inputs. Also, raising two major crops may affect soil quality and increase
biotic stress (Hossain 1994).

Diagnostic surveys in the rice-wheat cropping system in the Terai have identified
soil nutrient deficiencies (associated with nutrient mining) as one of the long term
(sustainability) problems common to both rice and wheat, which if unaddressed will
increasingly limit rice and wheat yields (Harrington et al. 1993). Plot-level data from
long-term experimental stations in the Terai indicates decline in yields of both continuous
rice and wheat (even in plots which were treated with recommended applications of
inputs) rotations. The fact that both cereals, and different varieties of these cereals
“experienced significant declines in yield despite constant levels of management suggests
that soil fertility or other as yet unidentified factors were depressing yields” (Hobbs and
Morris 1995, pp.31).

Ali (1996) examines the efficiency of wheat production drawing upon the same
data source used in this study. The author finds that cropping practices such as
continuous rice-wheat rotation and discontinuous organic fertilizer application on plots
which exhibited both poor soil quality and drainage conditions, resulted in a negative
impact on wheat yield. However, both the estimated yield firnction and associated
efficiency measures presented in his study could be biased due to the exclusion of labor

input in the analysis.

III. Data Section

A collaborative project has been formed by researchers from the International
Maize and Wheat Improvement Center (CIMMYT), the International Rice Research
Institute (IRRI), and the national agricultural research systems (NARSs) of Bangladesh,
India, Nepal and Pakistan, to examine the issue of sustainablility of the irrigated rice-wheat
systems of South Asia. The first country selected for micro-level monitoring was Nepal.
The data set used in this study stems from CIMMY T’s monitoring data of plot level
practices and resource use in the major rice-wheat regions of Nepal. This study draws
upon the survey data of 170 farmers in the Bhairahawa (a major rice-wheat region) study
area located in the Rupandehi District of the Terai region of Nepal.

The primary objective of the project was to examine if there are signs of an inter-
temporal decline in productivity in actual farmer fields. To this end, given budgetary
considerations, emphasis was placed on fine-tuning data collection (perhaps too fine-tuned
as we shall see) on plots which would most likely exhibit adverse consequences of
intensification. Also, “interference” due to jumbling of issues were avoided (e.g., sample
consists of farmers who are predominantly owner-operators, even at the plot level,
therefore avoiding incorporation of incentive issues associated with various forms of
contractual arrangements in land which would complicate the productivity analysis).

Farmers in the Rupandehi district are some of the most intensive farmers (in terms
of irrigation water use, chemical fertilizer use, adoption of Modern Varieties, etc) in
Nepal, benefitting from the marketing channels of the neighboring Indian state of Utter
Pradesh. Farmers in the sample were pre-stratified according to whether they were

participating in the Bhairahawa Lumbini Groundwater Project, a tubewell irrigation

development project funded by the World Bank. The sample included 82 farmers from the
groundwater project and 88 farmers fi'om outside the project (referred to as non-project
farmers)? Sampling was designed so that all farmers within a stratum would have the
same probability of being selected. In both strata, farmers were selected through a
multiple-stage sampling.

Similar to patterns observed in this region, land holdings are extremely fragmented
(while the average land holding size is 2 ha in this sample, the average number of plots is
10). After a particular farm household was selected, the principal farm manager was then
asked to identify “an important” (in terms of being a productive plot fi'om the view point
of the farmer) rice-wheat plot owned by the household. Most of the plots (90%) were
farmed on a continuous rice-wheat rotation - again, plots which most likely would exhibit
productivity decline due to soil fatigue or other intensification factors.

Plot level data was then collected on farmer’s practices and levels of input use, rice
and wheat yields, resource quality, and other related variables. Data collection began in
1990/91 by scientists from the Bhairahawa Wheat Research Farm, CIMMYT, IRRI, and
extension workers from the district Agricultural Development Office. The same plots
have continued to be monitored since then. Figure 1 shows the long-run trend in both rice
and wheat yield at the district level. Figure 2 compares the sample and district averages
for both crops. We see that our sample farmers somewhat follow district patterns except
for the fact that average yields for both crops are consistently higher in our sample

(particularly for rice).

 

2 Project farmers and non-project farmers do not come from the same village.

7

Our sample is biased towards farmers who use irrigation, and continuously farm on
Danda landtype. In this sample, two primary land types are identified, Khala (lower
terraces characterized by heavier soils and poor drainage) vs. Danda (upper terraces
characterized by lighter soil and few drainage problems, however drought prone). While
constructing categories for land type (Danda or Khala) or land ownership holding was
fairly straight forward (these categories were static during the sample period),
categorizing farmers as “continuous” (plot under a continuous rice-wheat rotation is
planted with rice followed by winter wheat during each crop year, and the same rotation is
followed every subsequent crop year, and that plot is never allowed to remain fallow) v.s.
“non-continuous” (which in our study area is primarily either rice-wheat-mustard or rice-
wheat-pulse) ; and ‘irrigated’ v.s. ‘non-irrigated”, was a bit more complicated :
Continuous v.3. Non-Continuous : Under ‘ideal’ circumstances a continuous farmer should
enter every year under both the rice and wheat sub-sections of the panel, while a ‘non-
continuous’ farmer should enter every year in the rice sub-section and every other year in
the wheat subsection. However, certain farmers had to keep their plot idle in particular
years due to various non-agronomic factors (primarily pre-planting financial constraints).
Thus, ‘continuous’ farmers are labeled as those who farm under a continuous rice-wheat
regime, leaving the plot fallow only due to non-agronomic reasons;
Irrigated v. s. Non-Irrigated : In this study, we categorized a household as “non-irrigated”
if the household did not use any source of irrigation water throughout the sample period
(1991-1996). On the other hand, we classified a household as “irrigated” if the household
irrigated its plot at least once during the sample period. Incidence of irrigation is more

prevalent in wheat (which is grown during the dry winter season)

There were hardly any farmers in this panel who did not use chemical fertilizers
and modern varieties. In the rice sub-section twenty nine percent of the farmers applied
nitrogen fertilizer in a “discontinuous” manner (i.e, not every year they appeared in the
panel), while 71% of farmers applied nitrogen fertilizer every year that they appeared in
the panel (only one farmer never applied any nitrogen fertilizer) ; fifiy three percent of the
farmers applied non-organic phosphorus in a discontinuous manner, while 47% of farmers
applied phosphorus fertilizer every round of the panel (only five farmers never applied any
phosphorus fertilizer) ; Ninety percent of the farmers in the sample used modern rice
varieties every year they appeared in the panel, while only 5% never used any modern
varieties. In the wheat sub-section thirty one percent of the farmers applied nitrogen
fertilizer in a discontinuous manner, while 69% of farmers applied nitrogen fertilizer every
year that they appeared in the panel (only two farmers never applied any nitrogen
fertilizer) ; thirty three percent of the farmers applied non-organic phosphorus in a
discontinuous manner, while 67% of farmers applied phosphorus fertilizer every round of
the panel (only one farmer never applied any phosphorus fertilizer) ; unlike the rice sub-
section where farmers had intimate detail about every single rice variety that they used, the
vast majority of the farmers claimed not to know what type of wheat variety they were
using. We do not have sufficient information on other macro nutrient application (e.g.,
potassium) nor on any micro nutrient application (e. g., zinc). We do have information in
certain years regarding use of farm yard manure at the plot level. However, organic

fertilizer of such nature makes it difficult to quantify its “composition”.

Farm households in this sample region hold two key natural resource assets in their
portfolio which they draw upon for agricultural production: soil, and groundwater’ (only
irrigated farmers utilize this resource of course). There is negligible soil erosion in this
particular region, and it seems unlikely that soil erosion will arise as a major problem in the
near future, thus, leaving soil fertility as the key soil related issue. Soil samples were
collected fi'om each plot and analyzed during 1991 and 1995. In Table 81 we notice an
alarming trend of both macro and especially, micro nutrient depletion (e.g., phosphorus
has declined in 92% of the plots, while magnesium has declined in 84% of the plots).
Macro nutrient decline in such a short time period comes as a surprise, given that most.
farmers use chemical fertilizers. While the implications of macro nutrients deficiency on
plant grth has been throughly studied in various agro-ecological zones around the
world, we yet do not have a clear understanding between micro nutrient and cereal yield
interaction (particularly in this region).

In 1991, the average rice yields for irrigated farmers in the sample was 3.3 mt/ha.
The average rice yields for the same irrigated farmers in 1996 was close to 4.5 mt/ha.
Thus, in a period of five years, average rice yields for irrigated rice cultivated on the same
plots, have increased by one mt/ha. In this sample, even non-irrigated farmers, farmers
who never apply organic fertilizer (F YM), and farmers without formal education, enjoyed
average increases in rice yields over the sample period (see Figure 3 - Figure 5). These
yields have occurred without any significant increases of variable inputs (1 do not

however, have information on changes in labor input over time), without noticeable

 

3 The data set does not contain information on actual quantity of water use. Irrigation information appears
as number of irrigations given to the plot. Source of irrigations turns out to be a better measure of
irrigation use.

10

changes in plant variety adoption, without new types of crop rotations, without adoption
of new types of planting/land preparation techniques, nor without adoption of new
farming technologies. Average rice yield increase in the same plot without any major
changes in inputs/technology, does suggest that it is plausible that farmers in our sample
have made efficiency gains in rice production (albeit, we cannot extrapolate from plot level
performance to total farm level performance).

There are several limitations to this data set. First, it would have been better to
have data on each plot owned by the household. This would have allowed us to explicitly
examine household level productivity strategies over time, instead of merely examining
plot level efliciency tied to certain household characteristics. To examine plot level
technical efficiency in this study, I rely upon primal representations of the production
technology (e.g, yield firnction). However, we are still left to deal with the problem of
endogenity. Inputs used at the farm level are endogenous to the household decision
framework, which for rural agricultural households involve simultaneously both
consumption and production choices. Then, even if we were lucky enough to have
“seperability” (Strauss, Singh and Squire 1986) between household consumption and
production decisions, production inputs will still be governed by the matrix of relative
prices, degree of risk aversion and other constraints faced by the farm household pursuing
various production strategies, making the model nonseperable. I do not have sufficient
information to construct robust instruments to address the problem of endogeneity in the
primal estimation. On a final note, given that (limited) information on labor use in rice
production is available only for the first round of the panel, I estimate TE measures of rice

production using data fi'om only the first round of the panel.

11

IV. Analytical Framework, Empirical Procedures, and Findings

Working with cross-sectional data limits the possible estimation strategies available
to calculate plot level TE (for a thorough review of the issues involved with both cross-
section and panel data based estimates of TE via econometric and non-parametric
methods, see Fried, Lovell and Schmidt 1993). I will adopt an econometric approach
towards estimating plot level TE, and within that framework, I will explore the “stochastic
production frontier” approach (Aigner, Lovell and Schmidt 1977; Meeusen and Van' de
Broeck 1977; Jondrow et al. 1982). The general idea behind this framework is that each
firm faces its own production frontier, and that frontier is randomly determined by a host
of stochastic factors outside the control of the firm (Green 1993). However, once the
firm-specific production fi'ontier is randomly placed, any deviation from that fi'ontier is due
to firm-specific technical inefficiency. A “generic” characterization of our yield fianction
within the stochastic frontier framework is as follows:

Yr: f(xi»B)+ei = f(xiaB)+vi'ui (El)
where, y: crop yield ; x: vector of inputs (including variable inputs and fixed factors) ; [5:
vector of unknown parameters; e: error structure; i: firm index.

The error term, ei, can be decomposed into two parts: (1) vi, the “familiar”
independently and identically distributed (across plots) two sided N(0,o,,z) random variable
representing model this-specification, measurement error and random shocks (taking on
values which can either be, negative, zero, or positive); and (2) u,, a random variable
associated with firm-specific factors which influence whether firm/farmer i attains
maximum efficiency of production (taking on values which can either be positive or zero) .

This firm-specific time-invariant technical inefficiency parameter is known to the farmer,

12

but not to the analyst. A value of 0 for ui indicates that the farm is operating on its
frontier. Any value greater than 0 indicates that the farm is below the fi'ontier, implying
that the farm’s practices conditioned by its environment leads it to produce less than the
maximum possible output. The random variables v, and ui are assumed to be independent.

The compound disturbance in this model (ei = vi + 1.1,), while asymmetrically
distributed, can still be estimated from maximum likelihood (ML) estimation of Eq (1).
However, before we can proceed with ML estimation, certain assumptions regarding the
distributional properties of ui have to be made. One of the most common characterization
of ui is to assign it a truncated (halt) normal distribution N(0, of). While TE estimates
have been found to be fairly consistent under various specifications of the firnctional form
of the production technology, TE estimates are sensitive to the distributional specification
of u (Bravo-Ureta and Evenson 1994). Unfortunately, currently there is no satisfactory
method for choosing nor testing the ‘validity’ of any distribution (e.g., half normal or
gamma distribution). This remains as the primary drawback of estimating TE measures in
a stochastic frontier framework using cross-sectional data. Because we do not have
plausible instruments in our data set, I also assume that firm-specific level of inefficiency is
uncorrelated with the level of inputs (x). Recent developments in this field utilizes panel
data to avoid some of the troublesome aspects related with estimating TE measures based
upon cross-section data (e. g., Comwell, Schmidt and Sickles 1990). Since I do not have
labor data over time, I cannot estimate TE measures using panel data methods.

For the half-normally distributed inefficiency term, the log-likelihood firnction is

(for details please see Aigner, Lovell and Schmidt 1977) :

l3

-&‘1

 

Kan/3,1,0): -N*ln0' — c+ i [ln<D( o

1 £1
_ —# _ 2 2
) 2 (a) 1 (E >
where, N : number of observations ; o’ = of + of ; l. = o,, / o, ; and, (.) : (I) is the cdf of
the standard normal distribution.
I obtain the estimates for o, and o, from the ML estimation of the yield function.
To then obtain the farm specific measures of technical inefficiencies (TIE), I use the

following formula (Jondrow, Lovell, Materov, and Schmidt 1982):

0’11 ¢(£r*/l/a') 8:51
(1+12)[¢(-£r*/l/0’)- 0' 1 (EB)

 

E[url£i] =

where, the pdf and cdf, (p(.) and <D(.) respectively, is evaluated at 85*1/ o .

After obtaining the farm/plot specific TIE measures, 1 then employ a multi-variate
regression setting to explore for some of the possible ‘determinants’ of farm-specific
efficiency.

Empirical Procedures and Findings

Data on labor input use is available for 101 plots during the 1991 rice season. Out
of the 101 farmers, I segregate farmers into 87 “irrigated” and 14 “non-irrigated” farmers
(i.e., farmers who do not have access to any type of irrigation source). Since it would be
unwise to include these two categories in the same estimation model, I restrict the analysis
to only irrigated farmers, and then try to control for other pertinent factors within the
multi-variate regression setting. It is hypothesized that plot level rice‘yield (kg/ha) for

irrigated farmers is a firnction of (x):

14

Variable inputs (besides irrigation) : (1) labor - pre-harvest family and hired labor
(hrs/ha) (2) nitrogen - chemical nitrogen application (kg/ha) ; (3) phosphorus - chemical
phosphorus application (kg/ha) ; (4) animal-hrs - hours of animal traction used during
plowing and planking operations (hrs/ha) ; tractor-hrs - hours of tractor use during
plowing and planking operation (hrs/ha) ;

Irrigation - we only have information on the number of times the plot was irrigated,
however, we can control for the source of irrigation which is viewed to be a better
measure of the impact of irrigation : (5) number-irrg - number of times the plot was
irrigated; (6) tubewell - binary indicator variable taking on the value of 1 if the source of
irrigation for the plot was tubewell ; (7) canal - binary indicator variable taking on the
value of 1 if the source of the source of irrigation of the plot was canal ; otherirrg - binary
indicator variable taking on the value of 1 if the source of irrigation was neither tubewell
nor canal (e. g., pond), not included in the regression;

Varietal Choice of rice plant (all binary indicators) : (8) Saryu 49; (9) Savitri ; (10)
Masuli ; (l 1) Janaki ; othervar - other types of modern varieties, not included in the
regression (all 87 farmers in our data set planted traditional varieties) ;

Timing of Rice Transplant : (12) transplant-date - timing of crop establishment, binary
indicator variable which takes on the value of 1 if transplant date was later than optimal
date (all farmers in our data set transplanted rice from seed-bed, as opposed to employing
direct seeding method);

Plot specific weather eflects : (13) drought - binary indicator taking on the value of 1 if

the plot suffered fi'om severe drought (captures mid-season water-stress);

15

Management Practices with long-term carry-over efi’ects on soil fertility : (14) years-cm
- number of years the plot has been cultivated under the continuous rice-wheat rotation (a
value of 0 indicates plots farmed under non-continuous rotations); (15) never-manure -
binary indicator variable which takes on a value of 1 if field yard manure (FYM), or
organic fertilizer, has never been applied to the plot.

Fixed Characteristics : (16) plot-size - size of plot (ha) ; (17) land-type - binary indicator
which takes on the value of 1 if the plot is of land-type danda (upper terraces with
generally good drainage) , 0 if the land-type is khala (lower terraces with generally poor
drainage conditions) ; (18) soil-medium - binary indicator taking on the value of 1 if the
soil-type of the plot is classified as medium ; (19) soil-heavy - binary indicator taking on
the value of 1 if the soil-type of the plot is classified as heavy ; soillight - binary indicator
variable taking on the value of 1 if the soil-type of the plot is classified as light, not
included in the regression (see Table 82 for descriptive statistics of variables used in the
estimation) .

I first introduced (right-hand side) variables linearly; then I added squared terms
for labor, nitrogen, phosphorus, animal-hrs, tractor-hrs, and years-crw : squared terms
which did not “add” to the estimation were removed from the final analysis (only the
squared terms for years-crw added to the specification) ; finally, I interacted only certain
variables to keep the model both plausible and parsimonious - variable inputs were
interacted with each other, irrigation, variety, land-type and soil-type ; years-crw was
interacted with variable inputs, irrigation, variety, land-type, soil-type, and never-manure.

Since none of the variable interactions proved to be significant (cut-off point for

16

significance was generous, set at the 30% level), they were subsequently not included in
the final analysis.

Table 1 shows both the OLS and ML parameter estimates of the stochastic
production frontier. As we can see in Table 1, both the magnitude and sign of the OLS
and ML estimates are quite similar. However, the standard errors of the production
firnction parameters obtained via OLS are incorrect. As long as the firm-specific level of
inefficiency, u,, is not equal to zero, the standard errors of the parameters under OLS will
be incorrect even if: a) all the right-hand side variables are exogenous; b) 11,, is
uncorrelated with the level of inputs. Given that the ML estimation indicates that on #0,
the stand errors of the production firnction parameters obtained via ML are more
appropriate. Thus, the ML estimates of o, and o, are used (along with the half-normality
assumption) to obtain the firm-specific technical inefficiency measures - which shall be
addressed in the next section.

While my primary objective in this paper is to discuss issues related to technical
efficiency, I shall briefly highlight some results of the production function estimate in
Table 1 and discuss them briefly:

1. Labor input has a positive and significant (at the 5% level) impact on rice yield.
Exclusion of labor did not significantly alter the coefficient estimates of the other
explanatory variables, suggesting that labor could be treated as orthogonal in this
estimation (results not reported). However, this does not imply that I could have dropped
labor from the analysis and used the full panel (recall that labor data is collected only in the

first year of the panel), given that even if this orthogonal relationship held inter-

l7

temporally, omitted labor would of course still be masked in the residuals, and thus, bias
any attempt to derive efliciency measures which rely upon residuals from the estimation.
2. The two most important factors which increase rice yield are : (1) tubewell irrigation
significantly (at the 1% level) boosts average rice yield by almost one ton/ha; (2) planting
plots with Savitri variety significantly (at the 1% level) boosts average rice yields also by
almost one ton/ha;
3. The two most significant factors which adversely effect rice yields are : (1) severe mid-
season water stress drastically reduces average rice yields by 1.3 ton/ha (at the 1%
significance level), even among this group of irrigated farmers; (2) plots in which organic
fertilizers have never been used, on average have lower rice yields of 825 kg/ha
(significant at the 1% level);
4.There is an inverted-U shape relationship between average rice yields and the number of
years the plot has been under the continuous rice-wheat rotation. Holding everything else
constant, on average, an additional year the plot is planted under the continuous regime
increases rice yields by 86 kg/ha ; yields start to decline after the plot has been farmed
under the continuous rice-wheat regime for more than 11 years. However, the quadratic
term is not significant in the ML estimation. Also, while we do not have actual data on
rice yields prior to 1991, most farmers report a general upward trend in rice yields over
time. I will address this issue later on in the paper.
‘Determinants’ of Technical Inefliciency

Using parameter estimates presented in Table l and, using equation (E3), I derive
the farm/plot level technical inefliciency measures (TIE). The average TIE for this group

of irrigated rice farmers in 1991, was found to be 594 kg/ha (with a standard deviation of

18

105). Average TIE was (statistically) similar across various categories of farmers (e. g.,
project vs. non-project; continuous vs. non-continuous). This suggests, that on average,
rice yields in 1991 could have potentially been increased by slightly over half a ton/ha,
corresponding to a 18% average increase in output, via a more efficient utilization of
current input levels and technology.

In this section of the paper I will explore for factors which might help to account
for the variation in TIE across farms/plots. Before I do so, I should mention that this
“two-step” procedure (i.e., first estimating efficiency measures, and then estimating a
regression model exploring for factors which influence variation in efficiency) is one of the
two general approaches in this literature (Bravo-Ureta and Rieger 1990; Kalirajan 1991).
The other approach includes socio-economic variables which are thought to influence
efficiency directly in the production frontier estimation (e.g., Battese, Coelli and Colby
1989; Battese and Coelli 1995). For example, Battese, Heshmati and Hjalmarsson (1998),
employ maximum likelihood methods to simultaneously estimate: (1) the parameters of the
production frontier model; (2) and the parameters of a second model which examines the
determinants of the variation in individual level mean technical inefficiency. Whether to
directly include socio-economic variables in the production frontier (or simultaneously
estimate technical (in)efliciency and the determinants of technical (in)efficiency), or to
examine the determinants of technical (in)efficiency in a separate analysis, tends to be left
to the discretion of the researcher given that this debate has yet to be meaningfully
resolved. In this section I am particularly interested in exploring the impact of two socio-
economic factors on the variation of technical inefficiency, education and ownership of

land, via a “two-step” approach.

19

The private and social returns to schooling have been a prominent research issues
in the field of development economics. The impact of education has wide implications,
ranging fiom it’s effects on agricultural productivity (e.g., Chaudhury 1979; Lockheed,
Jamison and Lau 1980, Jamison and Lau 1982; T.P. Schultz 1988; Singh 1990 ) to child
health outcomes (e.g., Caldwell 1979; Strauss and Thomas 1995). The specific evidence
on the relationship between education and farm level technical efficiency, however,
suggests that education, particularly, only elementary education (4-5 years of schooling)
might not have a significant effect on the technical efficiency of traditional farmers
(Cotlear 1986; Azhar 1991; Bravo-Ureta and Evenson 1994). Farmers operating in more
or less “static” low-technology, low-growth areas, might not have the opportunity to
adequately utilize skills acquired through low levels of education to begin with. However,
as Chaudhury (1979) and Singh (1990) pointed out, the impact of education (more
specifically returns to secondary education) on agricultural productivity increases with the
diffusion of new technologies and changes in the incentive structure. However, on the
other hand, there is evidence that primary schooling plays a significant role in shaping
allocative efficiency (e. g., Foster and Rosenzweig 1996). There is no a priori reason why
empirical studies of technical efficiency should consistently establish that secondary
schooling effects dominate primary schooling effects, while empirical studies of
profit/allocative efficiency tend to find the contrary. Given that in this particular study I am
only examining factors that influence technical efficiency, I refrain fiom any further
discussion of this issue - I do surmise that secondary schooling should have a significant

impact on technical efficiency in this district.

20

Ever since Sen (1962) sparked off the modern debate between farm size and
productivity, primarily in the context of yield and other partial-productivity-ratios (PRRs),
several studies in different countries have affirmed the inverse relationship between farm
size and productivity (Berry and Cline 1979; Cornia 1985). Synthesis of the findings on
the size-productivity relationship by Rudra and Sen (1980), show that the inverse size-
productivity relationship attenuates or even turns positive with the introduction of
irrigation. This debate has been firrther invigorated by evidence from various studies which
show that this once strong negative relationship between farm size and yield is either being
weakened or even reversed as a consequence of capital-led agricultural intensification
(Griffin 1974; Rao 1975; Berry and Cline 1979; Ghose 1979; Roy 1981;Deola1ikar 1981).
A U-shaped relationship between farm size and productivity could arise if large farmers
have a lower credit cost advantage while small farmers have a lower labor cost advantage,
which is associated with their lower labor supervision cost advantage (Binswanger and
Rosenzweig 1986). Two major criticisms levied against this general line of argument are
that: (1) PPRs do not provide a comprehensive measure of productivity (TE, AE or TFP
measures are more appropriate); (2) differences in quality of inputs, particularly land
quality, was not factored in the analysis. The ‘second wave’ of empirical studies
suggested that while the initial productivity of large farms might have been higher during
the early stages of the Green Revolution, small farmers did manage to catch up (Barker
and Herdt 1985; Hazel] and Ramasamy 1991). The initially higher productivity of large
farms “was soon mitigated by government credit programs and by induced institutional
innovations, such as the interlinking of credit with product and input markets, which

lowered the cost of credit to small-scale and tenant farmers” (David and Otsuka 1994, pp.

21

4). We certainly have yet to observe this phenomenon in Nepal. Reigniting the debate,
Binswagner, Deiniger and Feder (1995) find the persistence of a negative size-productivity
relationship in LDCs even after adjusting for quality of inputs, distinguishing between
ownership and operational holding size, accounting for the number of family members able
to act as supervisors, and using comprehensive measures of productivity. Van Zyle,
Binswanger, and Thitle (1995) also find an inverse size-productivity relationship in South
Afiican agriculture (using both TE and TFP measures). Thus, in lieu of a resurgence of
sound empirical validation for the size-productivity relationship and the particular setting
of this study, I hypothesize that efficiency and farm-size should be negatively related (or a
positive association between inefficiency and farm-size).

Technical inefficiency is assumed to be a firnction of two sets of variables: (A)
farrners’s grasp of agronomic principles and technical knowledge ; (B) broader socio-
economic environment. The first set consists of : (1) years-variety - number of years the
farmer has used the variety (recall variety was controlled for in the production frontier
estimate), as a measure of variety specific knowledge; (2) age - age of the principle farm
manager, as proxy for experience; (3) elderly - the number of elderly people in the
household, as a proxy for stock of experience; - the number of elderly people in the
household; (4) primary - binary dummy variable which takes on a value of 1 if the primary
farm manager has completed between 1 to 5 years of schooling (31% of farmers in our
sample); (5) secondary‘ - binary dummy variable which takes on a value of 1 if the primary
farm manager has completed between 6 or more years of schooling (31% of farmers in our

sample).

 

‘There were only 3 farm managers who had completed two years of post-secondary schooling (i.e., beyond
grade 10), however, they were also included in the secondary category.

22

The second set of variables consists of: (6) land-holding - amount of land (in
hectares) owned’ by the household (in this particular sample, none of the farmers rented
in/out any land) ; (7) buffalo - number of buffalos owned by the household; (8) cattle -
number of cattle owned by the household; (9) tractor - binary indicator variable which
takes on the value of 1 if the household owns a tractor; (10) farmworkers — number of
family members working on the farm and also acting to supervise hired labor; (11) non-
farmworkers - number of family members engaged in non-farm employment. Kalirajan
(1990) finds that technical efficiency was positively related to non-farm income among a
sample of farmers from the Philippines, however, the effect could go either way,
particularly if household time and effort fetches a higher return for non-farm activities
compared to farming.

However, most of these asset and labor variables are potentially endogenous.
Empirical studies by Rosenzweig and Wolpin (1989) and Binswanger and Rosenzweig
(1993), have shown that Indian farmers choose their asset portfolio in response to their
ability to cope with risk. In a credit/contingency constrained environment, buying and
selling of assets provide a way to smooth consumption in the face of income fluctuations.
Thus, buffalo, cattle, and tractor, are all potentially endogenous. However, land-holding
can still be treated as exogenous, given that farm land transactions are still uncommon in
this region (most land transactions in South Asia are in tenancy - selling of farm land
occurs only in exceptional situations, such as during a severe famine). Similarly,
household labor supply (on-farm and off-farm) decisions are potentially endogenous (e.g.,

Benjamin 1992), particularly in this environment where it is quite likely that farm

 

5 Unweighted gini coefficient of landholding in the sample was 0.74.

23

households fail to ‘separate’ consumption and production decisions. Given that I do not
have robust instruments, I will present results both with and without including bufl’alo,
cattle, tractor, fannworkers, and non-farmworkers.

The third set of variables consists of: (12) project - binary indicator variable which
takes on a value of 1 if the plot is farmed by a project farmer. Since project farmers are
formally involved in an externally financed group irrigation project, project participation
could result in group ‘synergies’ which might increase efficiency. The remaining non-
project farmers are covered by local extension services; (13) migrant -binary indicator
variable which takes on a value of 1 for migrant households. Migrants from the Hill
region of Nepal are perceived to be more productive than the local indigenous farmers on
the region. While this might seem surprising, given that ‘local’ farmers would possess
more specialized knowledge associated with farming in this region, agronomists with long-
term experience in this study area stress that migrants in general are more willing to adopt
new farming technologies and are more likely to use organic fertilizer (manure) compared
to indigenous farmers; (14) distance-road : distance (in km) of village from nearest main
(paved) road; (15) electricity - binary dummy variable which takes a value of 1 if the
village has electricity. Besides these two sets of variables, I also control for : (16) plot-
accessability - binary indicator variable taking on a value of 1 if the plot was near the
homestead, it is assumed that if the household faces labor/supervision limitations, then
plots far away from the homestead might suffer from inefficiencies in production; (17)
farm-parcels - number of plots the total operational landholding size is divided into,

similarly it is assumed that if the household faces labor limitations/limitations , then more

24

fragmented holdings might reduce technical efficiency. Table S3 provides summary
statistics of these variables.

Besides the variables mentioned above, I also include Village Development
Committee (VDC) dummies in the estimation - VDC is an important level of
administration in rural Nepal (villages are aggregated to wards, wards are aggregated to
VDCs, and VDCs are aggregated to form the district). Ignoring the issue regarding VDC
dummies for now, the coefficient estimate of primary and secondary schooling might be
biased negatively due to a host of factors. The rich wage-earnings literature has
thoroughly examined the issue regarding the overestimation of the private returns to
schooling due to unobserved ability (and other omitted variables such as school quality).
Unobserved individual specific ability which is certainly masked in the error term of the
regression, is certainly correlated with the dependent variable, plot specific technical
inefficiency, and also correlated with explanatory variables such as schooling. Bias in any
one coefficient (such as schooling) would bias all the other coefficients (such as
landholding). I do not have sufficient data to construct instruments necessary to test the
robustness of our cross-section analysis, nor can we draw upon familiar panel data
techniques to address this heterogeneity factor (e. g., fixed effects regression should
remove time-invariant heterogeneity influences). While I cannot control for individual
fixed effects, I can at least control for VDC fixed effects. The 87 fanners in our working
sample fall into 26 VDCs, and there is an average of 3.8 farmers per VDC (in our sample
there are five households which uniquely fall into a VDC category, thus, for those cases
VDC dummies are equivalent to individual fixed effects). Controlling for VDC fixed

effects (via VDC dummies) would control for potential variations in school quality and

25

other infi'astructure/market characteristics which might have direct and indirect effects on
production efficiency.

The impact of these factors on the TIE of farmers in the sample are explored via a
multivariate OLS regression framework. Explanatory variables first enter linearly; then I
added squared terms for years-variety, age, elderly, land-holding, buffalo, cattle,
farmworkers and non-farmworkers. Besides the squared term for land-holding, none of
the other squared terms added to the linear specification, and were not included in the final
estimation. Results‘5 presented in Column (1) of Table 2 show that the only significant
factor influencing technical efficiency appears to be secondary schooling (significant at the
2% level). Average technical inefficiency is reduced by 16% in plots farmed by
households in which the primary farm manager has completed more than five years of
schooling. I do observe an inverted U-shaped relation between technical inefficiency and
land ownership size (i.e., an U-shaped efficiency-size relationship), however, the
relationship is statistically weak (land-holding and land-holdingz, are individually and
jointly insignificant at the 10% level) 7.

It should be noted that the standard errors computed fiom this OLS regression are
incorrect. However, given that our dependent variable is rather ‘unorthodox’, there exists
no clear-cut correction procedures for this ‘two step’ approach. Another issue of concern

which I addressed earlier is that several of the right-hand side variables presented in

 

8 Subset regressor test strongly suggests that the VDC fixed effects belong in the regression
model (at the 1% level of significance ) - coefficient estimates of the 25 VDC dummies are not
reported.

9I also replaced the landholding variable with binary dummies for land quartile, however,
none of the three quartile dummies included in the regression were significant. Results not
reported.

26

Column (1) of Table 5 are potentially endogenous - I am particularly concerned with the
asset and labor variables. Given that I do not have robust instruments, I re-run the OLS
regression, dropping buffalo, cattle, tractor, farmworkers, and non-farmworkers. The new
set of results are presented in Column (2) of Table 2. We now observe the emergence of a
significant inverted U-shaped relation between technical inefficiency and land ownership
size, or a significant U-shaped efficiency-size relationship: land-holding and (land-
holding)2 are individually and jointly significant at the 5% level. The coefficient on
secondary-education becomes slightly more negative and significant. Thus, while the
(statistical) significance of the relationship between landholding size and efficiency in this
sample is quite sensitive to model specification/endogeneity bias, the effect of secondary
education appears to be quite consistent.
V. Conclusion

Using a stochastic production frontier framework, I estimated plot level technical
inefficiency for a sample of irrigated farmers in the Rupandehi district of Nepal. The
results suggests that rice yields among this group of farmers could on average be increased
by slightly over half a metric ton per hectare (mt/ha), through better utilization of available
resources at the current level of technology. How realistic is it for farmers in our sample
to achieve such efficiency gains ? I had previously mentioned, in a period of five years,
average rice yields for irrigated rice cultivated on the same plots, have increased by one
mt/ha. Thus, the efficiency estimates obtained from this study are plausible.

Despite the potential for greater efficiency of rice production, we should remind
ourselves of the grim picture stemming from the soil sample analysis. At this moment,

there is no evidence of declining productivity in rice production (coefficient estimates from

27

the production frontier indicate that we are on an upward trend), however, macro and
particularly, micro nutrient depletion might arise as serious limiting factors in the near
future. It would rather be more prudent to begin a serious collaboration between
agronomists, soil scientists, economists and farmers to explore ways to mitigate nutrient
mining. Economists can help to design more efficient institutions which are capable of
delivering macro and miro nutrients to farmers in a timely and adequate manner. To that
end, we need further exploration of farm level demand characteristic, and marketing and
infrastructure conditioners which shape supply characteristics. If we can apply any lessons
learned from the high-end Green Revolution areas (e. g., Indian Punjab), economists must
ensure that the incentive structure does not lead to an over-utilization of chemical inputs
and associated environmental damage.

With the rapid shrinking of the last arable land frontier of Nepal, increases in
efficiency will have to play a major role in increasing agricultural output. Policy makers
have various instruments at their disposal to stimulate both greater levels of technical
efficiency (e.g., facilitating a more efficient coordination between agricultural research,
extension services and farmer feedback) and allocative efficiency (e. g., foster institutional
innovations to deliver credit to collateral poor farmers). In Schultz’s influential 1975
paper, “Value of the Ability to Deal with Disequilibria”, he stressed the point that a
farmer’s ability to realign production activities in response to a rapidly changing
environment is affected by his/her “allocative ability”, which in turn is conditioned by
his/her level of education and experience. Investment in education and extension services
are critical instruments by which policy makers can increase agricultural efficiency,

particularly appropriate in a dynamic environment. In this paper, I have highlighted the

28

relationship between education and technical efficiency in rice production in our sample of
farmers, albeit in a rather restrictive and incomplete manner. I only had information on the
primary farm manager, while ideally one should do an analysis that incorporates
information on educational attainment for all family farm workers (and education level of
hired farm workers). Also, the analysis would be reinforced if one had information on
quality of schooling and information on broader family background factors. Given the
limited data available to me for this analysis, and the potential biases in the estimates, it
still can be reasonably argued that results from my analysis suggest that secondary school
education has payoffs in increasing technical efficiency in farm production. Further
detailed exploration of the relationship between education and agricultural productivity in
this dynamic region of Nepal would have immediate relevance to help guide policy makers
in designing appropriate strategies to enhance agricultural efficiency, which will translate

to higher farm incomes.

29

Table 81

Changes in Plot level Macro and Micro Nutrients

 

Element % of Plots undergoing decfine Average % change in
in element between 1991 and 1995 plot level element

Nitrogen 41 5.5
Phosphorus 92 -42.5
Potassium 59 -34.1
Organic matter 43.8 1.82
Sulfur 93 -32
Magnesium 84 ’ -17.74
Manganese 92 -43
Aluminum 77 -55
Copper 89 -38.3

Iron 92 -35.2

30

Table 82

 

Van'ble Mean Standard Deviation
rice yield 3281 .01 1 130.40
labor 51.73 31.91
nitrogen 40.72 24.96
phosphorus 16.15 10.91
animal—hrs 28.35 42.47
tractor-hrs 1 .32 7.74
number-irrg 1 1 .23 26.84
plot-size 0.58 0.77
years-cm 6.99 6.37
tubewell 0.31 0.47
canal 0.12 0.32
Saryu49 0.28 0.45
Savitri 0.24 0.43
Masuli 0.15 0.36
Janaki 0.07 0.25
transplant-date 0.54 0.50
drought 0.06 0.23
land-type 0.76 0.43
soil-medium 0.55 0.50
soil-heavy 0.36 0.48
never-manure 0.29 0.46

31

Dependent Variable (Y) : Rice Yield( Yp = 3281.01 kg/ha )

Table l : OLS and ML Estimation Results of Production Function

( Number of Observations : 87 )

 

OLS ML
(1) (2)
R2 0.62
Adjusted R2 0.53
Variable m Barc 5141
Labor 8.750 8.744
(2.030)‘ (1.912)‘
Nitrogen 11.836 11.835
(1.955)‘ (1.430)
Phosphorus -27.340 -27.940
(-2.040)‘l (-1.463)
Animal-Hrs 6.640 6.710
(0.649) (0.611)
Tractor-Hrs 16.579 16.601
(0.831) (0.801)
Number-hrg 1 1.027 1 1.024
(1.982)‘ (1.330)
Tubewell 813.58 814.67
(2.820)" (3.080)”
Canal 253.71 253.66
(0.460) (0.353)
Saryu49 -367.69 -367.51
(-1.254) (-1.690)
Savitri 865.23 865.23
(2.705)” (2.957)”
Masuli 226.40 226.40
(0.631) (0.747)
Janaki 371.42 371.42
(0.750) (0.601)
Late-Transplant -l 70.01 -l70. 14
(-0.678) (-0.568)
Drought -1312.0 -1 312.0
(-2.720)" (-2.845)“'
Never-Manure -825. 17 -825. 17
(4.281)" (-3.284)"

32

Table 1 Continued

 

OLS ML
(1) (2)
Variable [x2 Bare Bur -
Years-CRW 85.874 85.874
(1.755)‘ (1.701)’
(Years-CRW)2 .4005 .4012
(-l.671)‘ (-1.506)
Plot-Size 131.06 131.014
(0.640) (0.351)
Land-Type 215.55 216.81
(0.774) (0.800)
Soil-Medium -76.326 -76.533
(—0.177) (-0.141)
Soil-Heavy -34.801 «34.81 1
(-0.077) (0.061)
Intercept 4121.3 4436.0
(4.298)" (2.611)”
7. 0.69
(0.391)
0’ 772.82
(1.714)
Log likelihood -694.108
Joint F test 3.21 2.86
{Hoz SYN“, = BWMW,’ = 0} [0.09] [0.23]

 

Number in parenthesis represent 1 statistics ([9 / s. e. ) ; Number in brackets represent p-values

‘ represents significance at the 5% level ; ” represents significance at the 1% level

33

Table ss

 

Variable Mean Standard Deviation
Technical lnefticiency 593.70 105.33
age 42.67 12.07
year-variety 3.25 2.47
elderly 0.36 0.61
farmworkers 3.84 2.07
oftan'nworkers 0.62 1 .37
primary 0.20 0.41
secondary 0.42 0.50
landholing 2.07 2.24
buffalo 1.22 1.71
cattlem 1.92 1.40
distance-road 0.34 0.77
migrant 0.37 0.49
project 0.49 0.50
farmparcles 10.08 12.50
tractor 0.02 0.15
electricity 0.25 0.44
plot-accessability 0.59 0.50

Note: average number of years of schooling completed in the sample was 3.9 years

34

Table 2 : OLS Estimation Results of Determinants of Variation in TE

Dependent Variable : TIE ( TIErr = 593.703 kg/ha )

( Number of Observations : 87 )

 

R2 0.687 0.534
Adjusted R2 0.488 0.491
@ble Bore Bars
(1) (2)
Years-Variety -5.830 -5.947
(-0.977) (-1. 122)
Age -1 .923 -2.10
(-l .350) (-l .465)
Elderly -8.945 -9.03
(0.311) (-0.921)
Primary-Education -15.670 -16.31 1
(-0.397) (-0.794)
Secondary-Education -96.784 -99.391
(-2.60)" (-2.87)"
Land-Holding 49.446 57.397
(1.453) (1 .983)‘
(Land-Holding): -9.326 .101 17
(-l.501) (-2.192)‘
Buffalo -1 1.977
(—0.813)
Cattle -5.107
(-0.390)
Tractor -352.473
(-0.980)
Fannworkers -5.460
(-0.352)
Non-Farmworkers 14.860
(1.287)
Project 122.89 119.381
(1.361) (1.173)
Migrant -66.516 -70.872
(-0.984) (-1.351)

35

Table 2 Continued

 

Variable Bars Bars—
(1) (1)
Distance-Road -17.438 47.532
(—0.582) (0591)
Electricity -10.771 -11.421
(-0.471) (-0.716)
Plot-Accessability ~19.810 -22.361
(-0.671) (-1 .07)
Farm-Parcels -19.213 -l9.927
(—0.953) (-1.105)
Intercept 587. 82 561.91
(7.271)“I (6.829)"
JointFtest 1.54 3.31
{Ho3 Bind-Holding = Valid-Holding): = 0} 10-671 [0-06]
Joint F test 3.75 3.83
(Ho: All village dummies = 0) [0.01] [0.01]

 

Note : The 25 VDC Dmnmy coefficient estimates are not reported ; Null Hypothesis that VDC dummies
are not different from zero, was rejected at the 1% significance level for both specifications

Number in parenthesis represent t statistics (flou- / s. e. ) ,° Number is brackets represent p-values

‘ represents significance at the 5% level; ‘ ‘ represents significance at the 1% level

36

Figure 1

 

 

3
Z

 

 

 

 

 

 

g V
o
E
“
g.
a /\/\ Rf j
a. i'.”
g l h i X I \
5 t \ " ll! 1
3 A f \ I
3 ‘5. \t I' H
5' q 4’ y
g. i I!
1000 ’
‘ I
5 t f
1 .r
l I
if
i
500
o I I I I I I I f I I I I I I I I
It ‘3 O N “L ‘0 ’\ Cb '\
eeea‘a 9“ <5” 9" a”e"e"’e"~¢"e"e“e‘befpe‘be°

 

L—o—rue. - «Ir “M0811

 

37

Figure 2

 

 

 

 

8

Yield (kg/ha)

 

3

_ .«X

 

 

 

 

 

I, R
x.. . ... .. I, "
“x
1991 1992 1993 1994 1995
Year

 

—O-—Riee:SampleAverage - - I - -Rlce:DistrictAverage
'A Wheat:Sample Average *- - Meat: DistrictAverage

 

 

 

38

Average Rice Yield (11W)

3

8

E

1500

1000

Figure 3

 

 

 

 

 

 

 

 

 

 

 

 

.’ \ / .
x ’ 5, \ / .
if ' '_ xv ’1 s
J ,l' ‘ \\ /
If I " \ \ I! I
, ., ‘ ‘lr .
I’ ’ i .
.‘1 " I,
3
‘1
1991 1992 1998 1994 1995 1996
Year

 

—O—-Project-lrrigated I-u- Non-Project- Irrigated - - a- - -Non.lnlgared]

39

Figure 4

 

 

 

 

 

 

§

 

8

Average Rice Yield (kg/ha)

 

3

 

 

 

 

1992 1993 1994 1995 1996
Year

 

L— e - Never Applies-FYM +Applles-FYMj

 

40

mam—

 

cozuoauméuucooem 4.. cozaoaumémetmllfll cosmoaumbz. . 0 . .

 

mmmw

vmmv

eee>

mam?

«mar

ram?

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

e. 2:2“.

com

000?

oomw

ooow

oomN

ooom

comm

ooov

oomv

ooom

(err/6).) maul earn eBeraAv

41

BIBLIOGRAPHY

Aigner, D., Lovell, C.K., and Schmidt, P., 1977; Formulation and Estimation of Stochastic
Frontier Production Function Models; Journal of Econometrics 6:21-38.

Ali, Mubarik, 1996; Quantifying the socio-economic determinants of sustainable crop
production: an application to wheat cultivation in the Terai of Nepal; Agricultural
Economics 14: 45-60.

Antle, John M., 1983; Sequential Decision Making in Production Models; American
Journal of Agricultural Economics 65(2): 282-90.

Azhar, R., 1991; Education and technical efficiency during the green revolution in
Pakistan; Economic Development and Cultural Change; (39) : 651-665.

Banskota, K., 1992; Agriculture and Sustainable Development. In Nepal: Economic
policies for sustainable development. Asian Development Bank. Philippines. pp. 51-71.

Barker, R., and Herdt, R.W., 1985; The rice economy of Asia. Washington, DC:
Resources for the Future.

Battese, G.E., TJ. Coelli and TC. Colby, 1989; Estimation of frontier production
firnctions and the efficiencies of Indian farms using panel data from ICRISAT’s village
level studies; Journal of Quantitative Economics; (5) : 327-348.

----, A. Heshmati and L. Hjalmarsson, 1998; Efficiency of Labor Use in the Swedish
Banking Industry : A Stochastic Frontier Approach; Unpublished Paper.

Benjamin, D., 1992; Household Composition, Labor Markets, and Labor Demand:
Testing for Separation in Agricultural Household Models; Econometrica; Vol(60).

Berry, RA, and Cline, W.R., 1979; Agrarian Structure and Productivity in Developing
Countries. Johns Hopkins University Press. Baltimore MD.

Binswanger, H.P., H. Deininger, G. Feder, 1995; Power, Distortions, Revolt and Reform
in Agricultural Land Relations; In: Handbook of Development Economics Vol III; Edited
by T.N. Srinivasan and J. Behnnan.

~---, and Sillers, D., 1983; ‘Risk aversion and Credit Constraints in Farmer’s Decision-
making: A Reinterpretation’; Journal of Development Studies 20: 5-21.

----, MR. Rosenzweig, 1986; Behavioral and Material Determinants of Production
Relations in Agriculture; Journal of Development Studies (22), pp. 503-30.

----, M.R. Rosenzweig, 1993; Wealth, Weather Risk and the Composition and Profitability
of Agricultural Investments; Economic Journal; Vol(103).

42

----, Khandker, S., and Rosenzweig, MR, 1993; How Infiastructure and financial
institutions affect agricultural output and investments in India; Journal of Development
Economics 41: 337-366.

Bravo-Ureta, BE. and Evenson, RE, 1994; Efficiency in agricultural production: the
case of the peasant farmers in eastern Paraguay; Agricultural Economics 10: 27 - 38.

Byerlee, Derek, 1990; ‘Technical Changes, Productivity, and Sustainability in Irrigated
Cropping Systems of South Asia: Emerging Issues in the Post-Green Revolution Era’;
CW T Working Paper 90/06; Mexico, D.F.: CIMMYT.

---, 1992;’Technica1 Change, Productivity, and Sustainability in irrigated cropping
systems of South Asia: Emerging issues in the post-green revolution era’; Journal of
International Development, Vo. 1(4), No.5, 477-496.

Cadwell, J.C., 1979; Education as a Factor in Mortality Decline: An Examination of
Nigerian Data; Population Studies (33).

Comwell, C., Schmidt, P., Sickles, RC, 1990; Production Frontiers with Cross-Sectional
and Time-Series Variation in Efficiency Levies; Journal of Econometrics, Vol(46), 185-
200.

Cotlear, D., 1986; Farmer Education and farm efficiency in Peru : the role of schooling,
extension services and migration. Discussion Paper EDT49; Education and Training
Department, World Bank, Washington, DC.

Chaudhury, M.K., and L.W. Harrington, "The Rice-Wheat System in Haryana: Input-
Output Trends and Sources of Future Productivity Growth", Mexico, D.F.: C.C.S.
Haryana Agricultural University Regional Station (Kamal) and CIMMYT, 1993.

Chaudhuri, DP, 1979; Education, Innovation and Agricultural Development. Croom
Helm; London.

Cassman, K.G., and Pingali, PL, 1995; Extrapolating Trends fi'om Long-Term
Experiments to Farmers’ Fields: The Case of Irrigated Rice Systems in Asia; In:
Agricultural Sustainability: Economic, Environmental and Statistical Considerations.
Edited by Vic Barnett, Roger Payne and Roy Steiner; John Wiley & Sons Ltd., England.
pp. 63-84.

David C. Cristian, and Otsuka, K., 1994; Modern Rice Technology and Income
Distribution in Asia. Lynne Rienner Publishers, Inc. USA.

CGIAR, 1989; ‘Sustainable agricultural production: imlications for international
agricultural research; FAO Res. Technol. Paper 4.

43

Deolalikar, A., 1981; The Inverse Relationship between Productivity and Farm Size: A
test using regional data from India; American Journal of Agricultural Economics 63: 275-
279. .

Dhawan, B.D., 1977; Tubewell Irrigation in the Gangetic Plains; Economic and Political
Weekly 12(39) : A91-A104.

Evenson, R., and Rosegrant, M., "Determinants of Productiviy Growth in Asian
Agriculuture Past and Future", Paper presented at the 1993 AAEA International Pre-
Conference on "Post-Green Revolution Agricultural Development Strategies in the Thirld
World: What Next?", 1993.

Economic Policies for Sustainable Development, 1992; Asian Development Bank.
Philippines.

Ehui, SK, and Spencer, D.S.C, 1992; Measuring the sustainability and economic viability
of tropical farming systems: a model from sub-Saharan Afiica; Agricultural Economics
9(2): 279-296.

Evenson, Robert E., and Rosegrant, Mark W., 1993; ‘Determinants of productivity
grth in Asian agriculture past and future’; Paper presented at the 1993 AAEA
International Pre-Conference on “Post-Green Revolution Agricultural Development
Strategies in the Third World: What Next?

Faeth, P., "Evaluating Agricultural Policy And the Sustainabiltiy of Production Systems:
An Economic Framework", Journal of Soil and Water Conservation, March-April: 94-99,
1993.

The Measurement of Productive Efficiency : Techniques and Applications’ ; edited by:
Fried, O.H., Lovell, Knox CA, and Schmidt, Shelton S. ; Oxford University Press.

Foster, AD, and Rosenzweig, R., 1996; Technical Change and Human Capital Returns
and Investments : Evidence from the Green Revolution; American Economic Review
86(4): 931-953.

Fujisaka, S., Harrington, L., and Hobbs, P., 1994; "Rice-Wheat in South Asia: Systems
and Long-Tenn Priorities Established Through Diagnostic Research", Agricultural '
Systems, Vol(46), 169-187.

Ghose, AK, 1979; Farm Size and Land Productivity in Indian Agriculture: A
Reappraisal; Journal of Development Studies 16(1): 27-49.

Green, W.H., 1993 ; in ‘The Measurement of Productive Efficiency : Techniques and

Applications’ ; edited by: Fried, O.H., Lovell, Knox CA, and Schmidt, Shelton S. ;
Oxford University Press.

44

Griffin, K., 1974; The political economy of agrarian change: an essay on the Green
Revolution.
Cambridge: Harvard University Press.

Harrington, L.W., Fujisaka S., Hobbs P.R., Adhikary C., Giri 6.8., and Cassaday K.,
1993; ‘Rice-Wheat Cropping Systems in Rupandehi District of the Nepal Terai:
Diagnostic Surveys of Farmers’ Practices and Problems, and Needs for Further Research;
Mexico, D.F.: CM/IYT, NARC and IRRI.

---, "Indicators and Adoption: economic issues in research on the sustainability of
agriculture", Invited Symposim Full Paper, Symposium on Soil Productivity and Nutrient
Cycling in Relation to low input Sustainable Agriculture, 15th World Congress of Soil
Science, Acapulco, Mexico, 1994

Harwood, RR, 1987; Low Input technologies for sustainable agricultural system. In:
V.W. Ruttan and CE. Pray (Eds). Sustainable Agricultural Systems. Westview Press,
Boulder, CO/London.

Hazel], Y., and Ramasamy, C., 1991. Green Revolution reconsidered: The impact of high
yielding rice varieties in South India. Baltimore and London: John Hopkins University
Press.

Hobbs, P., and Morris, M., 1995; Meeting South Asia’s Future Food Requirements from
Rice-Wheat Cropping Systems: Priority Issues Facing Researchers in the Post-Green
Revolution era. CIMMYT.

Hossain, Mahabub., 1994; ‘Rice-Wheat Production System in Eastern India and
Bangladesh: Recent Development and Economic Constraints’; In Sustainability of the
Rice-Wheat Production Systems in Asia; Food and Agriculture Organization of the United
Nations. '

Hsieh, SC, and Ruttan, V.W., 1967; Environmental, Technological, and Institutional
Factors in the Growth of Rice Production: Philipinnes, Thailand, and Taiwan; Food
Research Institute Studies 7(3): 307-341.

Ishikawa, S., 1967. Economic Development in Asian Perspective. Tokoyo: Kinokuniya.

Jamison, D.T., and L]. Lau , 1982; Farmer Education and Farm Efliciency; Johns
Hopkins Press, Baltimore.

Jondrow, J., Lovell, C.A.K., Materov, 1.8. and Schmidt, P., 1982; On the estimation of
technical inefficiency in the stocastic frontier production function model; Journal of
Econometrics 19: 233-238.

Kalirajan, KR, 1990; On Measuring Economic Efficiency; Journal of Applied
Econometrics 5: 75-85.

45

Lockheed, M., D. Jamison and L. Lau, 1981; Farmer education and farm efficiency : a
survey; Economic Development and Cultural Change; (29) : 37-76.

Lynam, John K. and Herdt, Robert W., 1989; Sense and Sustainability: Sustainability as an
objective in International Agricultural Reserach; Agricultural Economics, (3), 381-398.

Meeusen, W. And J. Van den Broeck, 1977; Efficiency estimation from Cobb-Douglas
production functions with composed error; International Economic Review 18: 43 5-444.

Nelson, CC, 1987; Agricultural Price Policy in Nepal; Economic Staff Paper No.35;
Asian Development Bank.

Nepal: Economic Policies For Sustainable Development, 1992; Asian Development Bank;
Philippines.

Ostrom, E., 1995; In: Annual World Bank Conference on Development Economics; (Eds)
Bruno, M., and Pleskovic, B.; The World Bank; Washington, DC.

Pingali, PL, and Binswanger, HP, 1988. Population Density and Farming Systems: The
Changing Locus of Innovations and Technical Change. World Population Trends and their
Impact on Economic Development pp. 165-186.

----., and Rosegrant, Mark W., 1993; ‘Confronting the environmental consequences of the
Green-Revolution in Asia’; Paper presented at the 1993 AAEA International Pre-
Conference on “Post-Green Revolution Agricultural Development Strategies in the Third
World: What Next?

Pudasini, S., 1984; ‘Production and Price Responsiveness of Crops in Nepal’;
WGN/ADC Project Research Paper Series No.25, Agricultural Projects Services Center,
Nepal.

Rice-Wheat Atlas of Nepal; Edited by Huke R., Huke E., and Woodhead T., 1993;1RRI,
CIA/HWY T, NARC.

Rao, C.H.H., 1975; Technological change and distribution in Indian agriculture. New
Delhi: MacMillan Company.

Rosenzweig, M., and Wolpin, KL, 1993; Credit Market Constraints, Consumption
Smoothing and the Accumulation of Durable Production Assets in Low-Income

Countries; Journal of Political Economy; Vol(101).

Ruttan, V.W., 1977; The Green Revolution: Seven Generalizations; International
Development Review 19(1): 16-23.

Roy, R, 1980; Farm Size and Labour Use: A Comment; Economic and Political Weekly
15(39): 1631-1632.

46

---, 1981; Transitions in Agriculture: Empirical Indicators and Results (evidence from
Punjub, India); Journal of Peasant Studies; 8(2): 212-241.

Rudra, Ashok, and Sen, AK, 1980; Farm Size and Labor Use : Analysis and Policy;
Economic and Political Weekly; Vol(XV), No. 5, 6, 7.

Sadoulet, E., and de Janvry, A., 1995. Quantitative Development Policy Analysis. The
Johns Hopkins University Press. Baltimore.

Schultz, T.P, 1988; Education Investments and Returns; in Chenery and Srinivasan.

Schultz, T.W., 1975; Value of the Ability to deal with Disequilibria; Journal of Economic
Literature; (13) 827-846.

Singh, 1., Squire, L., and Strauss, J., 1986. Agricultural household models: extensions,
applications and policy. Baltimore: John Hopkins University Press.

---, 1990; The Great Ascent : The Rural Poor in South Asia; Johns Hopkins Press,
Baltimore.

Strauss, J., 1990; Households, Communities, and Preschool Children’s Nutritional
Outcomes : Evidence from Rural Cote d’Ivoire; Economic Development and Cultural
Change; (38).

---, and Thomas, D., 1995; Human Resources : Empirical Modeling of Household and
Family Decisions; In: Handbook of Development Economics, vol. 3A; (Eds) Behrman, J .,
and Srinivasan, T.N.; North Holland Press, Amsterdam.

Timmer, CR, 1988; The Agricultural Transformation; In: Handbook of Development
Economics, vol. 1; (Eds) Chenery, H., and Srinivasan, T.N.; North Holland Press,
Amsterdam.

Upadhyaya, H.K., and Thapa, GB, 1994; Modern Variety Adoption, Wage Differentials,
and Income Distribution in Nepal; In: Modern Rice Technology and Income Distribution
in Asia. (Eds) Cristina, CD, and Otsuka, K.; Lynne Rienner Publishers, Inc.

Van Zyle, Johan, Binswanger, H., and Colin,T., 1995; The Relationship bewtween Farm

Size and Efficiency in South African Agriculture; Policy Research Working Paper 1548;
The World Bank.

47

Whitaker, M. and S. Lalitha. 1993. Quantifying the relative productivity and sustainability
of alternative cropping systems; Resource Management Program Economics Group
Progress Report 115, ICRISAT.

48

Chapter 2
PRODUCTIVITY GROWTH OF NEPALESE AGRICULTURE
I. Introduction

The neoclassical grth models pioneered by Abramovitz (1956), Swan (1956),
Solow (1957), Fabricant (1959), and Kendrick (1961), highlighted the role of (exogenous)
technological change in driving macro economic growth. The “Solow residuals” - the
residual grth in output not accounted for by the grth in factor inputs, were supposed
to measure the contribution of technological progress (also often referred to as “Total
Factor Productivity” growth). However, long before the emergence of “new grth
theory” in macro economies which began to emphasize the role of investments in human
capital and research-and-development (R&D) as endogenous drivers behind technological
change (e. g., Romer 1986, 1987, 1990; Grossman and Helpman 1991; Aghion and Howitt
1992), Griliches’ seminal empirical analysis on the measurement and explanation of
productivity grth in United States agriculture, highlighted the role of education and
public expenditures on agricultural research and extension (R&E), as the principle drivers
of grth in the agricultural sector (e.g., Griliches 1963; 1964;]967). Building upon
Griliches’ work, various researchers have explored the determinants of productivity
grth in the context of international agricultural development (e.g., Hayami and Ruttan
197], 1985; Evenson and Kislev 1975, Boyce and Evenson 1975, Evenson and McKinsey
1991; Binswanger and Ruttan 1978; Mundlak and Hellinghausen 1982; Lau and
Yotopoulos 1989).

Discourse on how to measure productivity and exploration of the determinants of

productivity growth, continues to be an important agenda in international agricultural

49

research. Particularly within a developing country context, productivity increasing
technological change in agriculture has been recognized as one of the principal catalysts of
growth. Ideally, the positive impacts of increasing productivity comes about through an
interplay of boosting output, increasing demand for agricultural labor, lowering food
prices, improving rural incomes, stimulating rural non-farm employment, and increasing
purchasing power of poor consumers. Thus, the linkages between increasing agricultural
productivity and its multipliers, helps to initiate the economic transformation fi'om a
predominantly agrarian to a primarily industrial and service-oriented economy (Tirnmer
1988;Me11or 1995).

The agenda of this study is to examine the productivity of Nepalese agriculture at
the aggregate district (agro-ecological) level. Nation level analysis is useful in the sense
that it provides a general picture of overall agricultural performance. However, given that
in this study, I am relying upon estimates based upon primal representation of the
underlying production technology, it is often difficult to estimate “individual country
production firnctions from individual country data The first difficulty is insufficient
variation of the quantities of inputs, due to multicollinearity ..., or due to restricted range
of variations ..., or due to approximate constancy of factor ratios resulting fiom
approximate constancy of relative factor prices Insufficient variation in the data due to
any one of the above-mentioned reasons results in imprecision, unreliability, and possible
under identification of the estimated parameters of the production function .. The second
difficulty is the general inability of separate identification of the level of technological
change of the production function of an individual country and its biases or the degree of

returns of scale from time-series data on output and inputs of that country alone (Lau and

50

Yotopoulous 1989 , p.242). Thus, not only can district level analysis differentiated by
agro-ecological zones provide a sharper picture of agricultural performance, it also avoids
some of the pitfalls mentioned above.

Currently there is a dearth of empirical studies on the productivity of Nepalese
agriculture at either the national, regional or commodity specific level. For example, an
unpublished M.S. thesis in 1973 was the first and subsequently for 26 years, has been the
only study in which someone attempted to estimate the Total Factor Productivity (TFP)
grth of Nepalese agriculture (Shah 1973). The Nepalese economy is predominantly
agrarian with more than 90% of the population living in rural areas (and about 80% of the
active labor force employed in agriculture). Unfortunately the performance of the
agricultural sector over the last few decades has been dismal, and “can generally be
summarized as stagnant, with increasing population pressure leading towards
fragmentation of land, lower labor productivity and further poverty” (Pokharel 1993,
p.43). Nepal, which used to be a food-grain surplus country in the 1970's, has changed to
a food grain-deficit country in the 1990's (Banskota 1992). Domestic cereal production
and food availability per capita is on a decline in Nepal (Ali, Hobbs and Velasco 1993).
Export earnings from both the agricultural and non-agricultural sectors are insufficient to
allow Nepal to pursue a food security policy which primarily depends upon cereal imports
(ADB 1992). Labor absorbing industrial development has yet to emerge as a significant
factor in the Nepalese economy. However, currently only the agricultural sector has the
size and the multipliers necessary to stimulate broad based economic growth. Thus,

Nepal cannot afford to spend its reserves of hard currency on procurement of cereals from

51

the international market. Nepal must rather rely upon strategies which enhance domestic
cereal production in an arable land-constrained environment.

Thus, it is imperative that we analyze prospects for productivity grth in
Nepalese agriculture. 1 am aware of the apprehensions Griliches expressed more than
three decades ago regarding ‘mere’ estimation of technical change - “Identification of
measured grth in total factor productivity .. provides methods for measuring technical
change, but provides no genuine explanation of the underlying changes in real output and
input. Simply relabeling these changes as Technical Progress or Advances of Knowledge
leaves the problem of explaining growth in total output unsolved.” (Griliches 1967,
p.309). However, given the paucity of sound empirical diagnosis on the state of Nepalese
agriculture, this study provides a necessary starting point from which to launch a
systematic investigation of the productivity of Nepalese agriculture. 1

Changes in productivity grth reflect changes in scale, technology, human
capital, quantity and quality of the natural resource base, and efficiency. Economic
efliciency in production is determined by the technique of applying inputs and levels of
application of inputs. Technical efficiency (TE) reflects the ability to obtain the maximum
possible output from a given set of inputs. This is the most widely used measure of the
efliciency of a production unit. Allocative efficiency (AE) reflects the ability to maximize
profits, by equating the marginal revenue product with the marginal costs of inputs. While
TE measures can be estimated from (primal) input-output data, estimation of AE measures
requires (dual) cost/price information. It is not surprising then that TB is the most widely
used measure of efficiency in the developing country literature (Bravo-Ureta and Evenson

1994), reflecting the fact that cost of production data is relatively more scarce compared

52

to physical input-output data. Analytically, estimation of AB is further complicated by a
host of market failures which plagues the agricultural sector of many developing countries
(e. g., AB in production will fail to hold if farmers faces either a credit/liquidity constraint
or an input supply constraint). Thus, while estimation of TE does not (explicitly) require
assumptions of market efficiency, most AE measures are obtained under the restrictive
(and probably unrealistic) assumption of a perfectly competitive market environment.
Most estimates of technical change or TFP are based upon non-parametric growth
accounting methods embedded in the neoclassical framework, characterized by
competitive equilibrium and constant return to scales (which imply that payments to .
factors should exhaust total product). The grth accounting approach is as follows: (a)
detail accounts of all pertinent outputs and inputs of the production process are compiled;
(b) these outputs and inputs are aggregated using various types of indexing procedures,
and these indexes are used to calculate a TFP index. The various indexing procedures
reflect economic assumptions of the underlying production technology and production
environment. The basic idea is that if ‘technological’ change occurs, then payments to
factors would not exhaust total product, and there would remain a residual output not
accounted for by increases in total factor input (Capalbo and Antle 1988). This has been
by far the most common representation of TFP. However, the principal drawback of the
grth accounting method is that it is relatively data-intensive requiring extensive factor
price information. As Block (1993) points out, the data-intensity of grth accounting
methods makes it an impractical tool for examining the productivity in the agricultural
sector of most Afiican countries. Similarly most productivity studies of South Asian

agriculture have been carried out using Indian and Pakistani data - countries like Nepal

53

lack the institutional capacity to collect pertinent detailed data on a systematic basis in
order to carry out most growth accounting exercises. Thus, in such data-constrained
cases, parametric approaches which primarily draw upon physical input-output data, are
more appropriate. While the growth accounting procedure makes strong assumptions
about the underlying production technology (e.g., constant returns to scale) and
production environment (e. g., perfect competition), the econometric approach makes an
equally strong assumption - that the nature of technological change can be represented as
a function of time'.

Specifically, in this study, I will: (a) estimate the district level rates of technical
change in Nepalese agriculture based upon parametric estimation of the underlying
production technology within the “meta-production function” framework; (b) derive point
estimates of district level technical efficiency from econometric estimation of the
production technology within the ‘stochastic production frontier framework’; and (c)
given that I am interested in comparing efficiency levels across districts, I also construct
confidence intervals around the time varying technical efficiency estimates, using non-
parametric bootstrap methods. This study, is thus, both an effort to fill the gap in the
productivity literature on Nepalese agriculture, and to add to the growing literature on the
efficiency of South Asian agriculture.

Section II of this paper lays the analytical framework and empirical procedures
used to estimate district level technical change and efficiency; Section 1H briefly discusses

the district level data sources; Section IV presents the empirical findings from the analysis;

 

'° Technological change is synonymous to productivity change under the assumption that production is
efficient or that the degree of inefficiency is constant.

54

and Section V highlights the main conclusions (and shortcomings) of this study, suggests
direction of further research, and discusses relevant policy implications.
II. Analytical Framework

To estimate district level rates of technical change (RTC), I employ a “Meta-
Production Function” approach as originally forwarded by Hayami (1969) and Hayami
and Ruttan (1970, 1985), and extended by Lau and Yotopoulos (1989) and Lau et al.
(1993). Under a meta-production function framework, it is assumed that all production
units have access to the same underlying technology. As I will show, there is
‘convergence’ between the stochastic production frontier approach (Aigner, Lovell and
Schmidt 1977; Meeusen and Van de Broeck 1977; Jondrow et al. 1982) and the resulting
estimation framework of our production technology. District level time-varying TE
estimates will be estimated following the approach of Schmidt and Sickles (1984) and
Comwell, Schmidt and Sickles (1990); bootstrap confidence intervals for the district level
TE estimates will be constructed along the lines of Schmidt and Kim (1999).

Following Lau et al. (1993), I assume that all districts within a given agro-
ecological region have access to the same technology, i.e., an underlying aggregate
production function F(.), a meta-production function. However, different districts operate
may operate on different parts of the meta-production filnction. These differences arise
fi'om possible differences in efliciencies of production, differences in quality of inputs
(man-made and natural resources), or due to various measurement errors. Despite these
differences, the measured outputs and inputs of the different states may be converted into
standardized “efficiency-equivalent” units of outputs and inputs:

w, = F(X*

ii“

13,) v j=1,...,J;i=1,...,N;t=1,..T (1)

55

where, Y is output; Xs are “conventional” inputs indexed by j (J inputs); E is education;
district index i (N districts); and time index t (T time periods). The implicit assumption is
that the meta-production function itself does not depend on i but may depend on t.

The “efficiency-equivalent” quantities of outputs and inputs are off course are not
directly observable. They are however, assumed to be linked to the measured quantities of
outputs and inputs, through possibly time-varying and district-and-commodity-specific

augmentation factors:

Y2: = Ad(t)Ya (2)
X2“ = A J-,,(t)X,-,, V j = 1,...,J (3)
B‘a. = E... + AN) (4)

There are many reasons why these commodity augmenting factors are not likely to
be identical across districts. Examples include differences in climate, topography, natural
resources and infi'astructure; differences in definitions and measurements; and differences
in the efficiencies of production. For empirical implementation, the commodity
augmentation factors for output and all inputs besides education are assumed to have a
constant exponential form with respect to time. The augmentation factor for education is

assumed to have the linear form with respect to time. Thus,

Y2: = Aoi(t)Yit = Ana 3X13 (Coi’ 0 Yr: (5)
X1}, = A J.,,(t)x,, = Ai exp (Cl; 0 x,, v j = 1,..,J (6)
Eli = Am“) '1' E11 = An. '1' cm 1% +151: (7)

where, A’s are the augmentation level parameters and C’s are the augmentation rate

parameters (assumed to be constants).

56

For this study, I assume that the meta-production function (1), takes on a Cobb-

Douglas functional form”:

J
lnY.lr= lnYo+ZajlnX',-n+a£.u (8)

2-1

By substituting equations (5) through (7) into the Cobb-Douglas form, and rearranging

terms, we get:

J J J
1n Y. = 1n Y0+ Z arm/13 +aEEn + {- ln A.+ 2 ajlnAj} + {—C..- + Z 0.} *r (9)
121

j: l j: 1

We can then rewrite (9) as:

J
1111,11: 1nY0+Zaj1anit+aEEit+A.i+C.i*t (10)

i=1

In order to estimate equation (10) within a statistical framework, I add an independently
and identically distributed (lid) two sided N(0,o,’) stochastic disturbance term 85,, having
identical variance and assume that it is uncorrelated across districts:

J
llerr= ll‘lYo-l' Z a/lanil+ aEEit‘l' A.l+ C.i*l+ 6‘11 (11)

i=1

 

" Given the limits of the panel data and the number of inputs I include in the estimation, I cannot use a
more flexible functional form such as a transcendental logarithmic function (Christensen et a1. 1973). 1
also do not use a Cobb-Douglas plus (log) interactions of selected inputs due to computational limitations
and generally weak results. I shall bring up this issue again in section V.

57

Thus, using the meta production function framework with a Cobb-Douglas
specification, district level time-invariant heterogeneity, A,- ,and district level time-varying
heterogeneity, Ci , enter the estimation in a linearly separable fashion. District level rate of
technical progress/RTC, is captured by the district-specific heterogeneity term interacted
with the time trend, Ci . The specification of equation (11) does not allow us to estimate
separate commodity-specific rates of technical change (Cji V j = 1,..,J). However, given
that in this study I am primarily interested in estimating overall rates of technical change at
the district level, the specification of equation (11) will suffice.

The specification of equation (11) is similar to the Fixed-Effects Panel data
framework regarding the estimation of technical (in)efflciency of production. A Cobb-
Douglas specification within a stochastic frontier fiamework is as follows:

J
lnYl. = 1n Y.+ Z a,lnX,-.. + aeEn . e. - TIE” (12)

I"

where, TIE, (20) represents time-varying inefficiency (i.e., firm-district technical

inefficiency is allowed to change over time). We can rewrite equation (12) as:

J
- 1n I," = an '1' Z a; In int '1' aEErr + 8t! (13)
j=1

where, 0,, = (lnYo - TIE“); We then represent the time-varying firm-district technical

inefficiency effect as an explicit firnction of time: 0,, = 0li + 02ft ; we can then rewrite

equation (13) as:

58

J
In Yr: = z a; 1n X311 + aEErr + 9,, + 62, * l + 8a (14)

i=1

The core idea behind the stochastic frontier framework is the same, regardless of
cross-section or panel data applications Let us look at the last three terms of equation
(14). The last term, 8,, is the “familiar” iid two sided N(0,o,’) random variable
representing model this-specification, and random shocks (taking on values which can
either be, negative, zero, or positive), while the TIE parameters, 0,, and 02,. represent firm-
district specific factors which influence whether firm-district i attains maximum efficiency
of production (taking on values which can either be positive or zero). At any given time
period, a value of 0 for TIE indicates that the firm-district is operating on its fiontier, and
any value greater than 0 indicates that the firm-district is below the frontier (under the
assumption that all districts are bounded by the same frontier - if that assumption does not
hold, then we cannot separate differences across districts from TIE within a district).

By construction of the model, the TIE parameters, 0,, and 02,. are correlated with
the X5, thus, suggesting a Fixed-Effects (FE) specification as advanced by Schmidt and
Sickles (1984), and extended by Comwell, Schmidt and Sickles (1990), henceforth
referred to as C38 1990. Unlike the Random-Effects characterization of this problem
which ofien requires strong distributional assumptions about the nature of the TIE
parameters, the only further assumption I need to make is that of ‘Strict
Exogeneity’(Wooldridge 1996):

E(8II|91I , 92i , lnXlil,.e.,lnX1iT,eee, lnXJil,.ee,lnx1iT’ Eil,...,EiT) : O (15)

59

When the strict exogeneity condition holds, X,,, V t = 1,..,T (and V j = 1,..,J) are
strictly exogenous conditional upon the two unobserved (or latent) effects. This follows
from the conditional mean specification:

E(1nY,,|01, , 02,, lnX,,,,...,lnX,,,,..., lnXm,...,lnX,,T, E,,,...,E,r ) =
qunxi+01,+02,*t (15.1)

Given the assumption of strict exogeneity”, ignoring issues of efficiency for the
moment, the parameter estimates of the production function (a, V j = 1,..,J) in equation
(1 1)/(14) can be obtained through a number of econometric techniques (Green 1990;
Wooldridge 1996). The specification of equation (1])/(14) is no more than that of the
standard unobserved fixed-effects model augmented by a district-specific trend as an
additional source of heterogeneity - a ‘random grth model’ (Heckman and Hotz 1988).
Whenever the dependent variable is expressed in natural logs, the 02, parameter can be
viewed as the average growth rate over a period (holding the explanatory variables fixed).
Possible estimation strategies include: simple least squares including district dummies and
district dummies interacted with a time trend; Second-Differencing and then applying least
squares; First-Differencing and then standard ‘within’ FE estimation (for the nuances
pertaining to the consistency of the estimates and asymptotic properties implied by these

different techniques, please see Wooldridge 1996).

 

'2 If the strict exogeneity assumption fails to hold, then this problem can still be estimated using non-
linear instrumental variable techniques, under the assumption of ‘Weak Exogeneity’or ‘Sequential
Moment Restrictions’:

E(e,,|01, , 02, , lnX,,,,lnX,,T,,'...,1nX,,1 ,..., 1nX,,,,lnX,,T_,,...,1nX,,,, E,T,E,T_,,...,E,, ) =
0 , t = 1,..,T

For this study I will use the efficient instrumental variables approach advanced by
C88 1990 as a comprehensive approach towards estimating panel data models with
heterogeneity in slopes as well as in intercepts. C88 1990 specifies a systematic
fiamework in which to obtain consistent estimates of both the production function
parameters, and the time-varying district level productivity/(in)efficiency". Following
CCS 1990, I rewrite the data-intensive representation of equation (1 l) in matrix notation
as follows:

Y = lnYo + X13 + Qu + a, (16)
where: Y is the (NT x 1) output vector of stacked 1n Yit (stacked by each N which is
observed T times); lnY0 is (NT x 1) vector of Is; X is the corresponding (NT x K) input
matrix following the stacking order of Y - now including E,,; B is a (K x 1) vector of
parameters to be estimated; Q is the (NT x NL) block-diagonal matrix (with e on the off-
diagonals) representing district level time-invariant and time-varying heterogeneity”; and,
u is the (N *L x 1) parameter vector of 05 (L = 2 in our setting) - for example, the first 2
rows of u would be 011 and 02,, and the last two rows would be 0m and 021., , respectively.

Given that I will be dealing with the case in which L sT, Q will have full column
rank, and thus, 0 will be identified. Let PQ = Q(Q’Q)"Q’ be the projection onto the
column space of Q (NT x NT). Let MQ = ( IN.T - PQ ) be the projection onto the null

space of Q (NT x NT). The consistent “within” estimator is given by:

,0», = (X’MQxy' X’MQY (17)

 

'3 While instead of specifying the firm-district level effect as 01,, = 01i + 022,t + 03,t2 as in
CSS 1990, in this study we end up with only a linear time effect.

"' For example, the “first block” would be (there would be N such blocks): T rows of Is in
the first column; T rows of the time-trend (1 to T) in the second column.

61

The within estimator is an instrumental variable (IV) estimator, with instruments
MQ; note, that since ManYo = 0, the constant intercept term of the production firnction
cannot be consistently estimated. Similar to the standard FE within model (i.e., when 02,
= 0), equation (16) can be transformed by MQ and the parameter estimates of the
production firnction can be obtained via least square regression of MQY on MQX.

I can then estimate district level technical (in)efficiency measures using the within

residuals (éw = Y - Xflw) following CSS 1990; Schmidt and Sickles (1984). For each

district, we regress the T district residuals on a constant and a time-trend (i.e., a least
square regression with T-L degrees of freedom), to get 0 1:62: (which are consistent V i
and t, as T - co). Once the @1552: estimates have been obtained (e.g., either through

‘two-step’ CSS 1990 method, or through ‘one-step’ OLS regression of output on inputs

and district level dummies and district level dummies interacted with time), for each time

period, we can evaluate (HAL-z + 632:: * t) V i = 1,..,N. Then we can define:

éit = max,=, ..... N(élir+é2ir*l) (18)

For each time period, we can obtain the technical inefficiency estimates for each district as

follows:

I‘ll-l = 0.. - (91,, 02.117) (19)

For any given time period, the district with the lowest value of T11: can be

thought of as the best district in the sample (with a value of 0 indicating that the

62

production in the district is occurring on it’s frontier). Thus, T10 is an estimate of

relative rather than absolute inefficiency. Given that the production technology follows a
logarithmic specification, the technical efficiency estimate, TE,” for each time period (and

for each district) can be expressed as:

TE" = exp‘ T1”) . (20)

Thus, technical efficiency estimates are also expressed “relative”to the most

efficient (best) district (with a value of 1 indicating that the production in the district is

occurring on it’s frontier). With N fixed, as T - co, 91,62 , are consistent estimates of

01 and 02 (V i and t), and similarly I7" is a consistent estimate of 0., (V i and t).

However, given that in this study 1 am using a sample with a relatively small T(=]1), flu
may be biased upwards - the “max” operator in equation (18) induces upward bias, since

the largest (6111+ 02n*l)is more likely to contain positive estimation error than

negative error. This bias is large when N is large (relative to T), and when

(510+ éZit*t)IS measured imprecisely. Upward bias in (9111+ éer*l)induces an

upward bias in T70 , and thus a downward bias in TE: - thus, efficiency would be

underestimated given that the level of the fiontier has been overestimated. For a rigorous
exposition of the asymptotic properties of these type of models, please see Park and Simar

(1994)

63

Given that I have not made any distributional assumptions regarding the nature of

the time-invariant and time-varying heterogeneity (or technical inefficiency), at this
juncture I only have ‘point estimates’ of flit and T Err . I cannot rank (in)efficiency levels

across firms-districts with statistical precision. The overwhelming majority of past

empirical studies of efficiency have tended to overlook this issue. We can use (non-
parametric) bootstrapping to construct confidence intervals both I72: and TE". (Simar

1992; Hall, Hardle and Simar 1993, 1995; Kim and Schmidt 1999). The bootstrap method

was introduced in 1979 as a simulation-based method for estimating the standard error of

A

any given estimator, 6 , regardless of the mathematic complication of the estimator (the

following section closely follows Efron and Tibshirani 1993). The core of this technique

rests upon the notion of a bootstrap sample. For example, let X = (Xl,...,xn) be an
observed random sample from an unknown probability distribution F, with a

corresponding estimator 19 = S (X ) , and unknown standard error 56r(é). Letfi be
the empirical distribution, which places a probability of 1/ n on each of the observed values

. A bootstrap sample, X . , is defined to be a random sample of size n drawn from F .
Thus, while X represents the actual data set, X I , represents a resampled version of X -

the bootstrap data points(x‘l,...,x‘n) are a random sample of size n drawn with
replacement from the population of n objects (Xl,...,xn) . Therefore, the bootstrap data

‘ . . .
set (x 1,...,xtn)C0n51SlS of the same members of the onglnal data set (xl,...,xn) , some

(,4

appearing zero times, some appearing once, and some appearing more than once. The

corresponding bootstrap replication of the estimator is, 9. = S (X t). The bootstrap

estimate of the standard error of the estimator/statistic ’ 9 , is a plug-in estimate that uses

the empirical distribution function}? instead of the unknown distribution F . The

following is the algorithm for estimating non-parametric (since the estimate is based on
the non-parametric estimate of the population) standard errors of any estimator from an

observed random sample X :

(1) Select B independent bootstrap samples X ”,...,X .8 , each of size n with values

drawn with replacement from X .

(2) Evaluate the bootstrap replication corresponding to each bootstrap sample,

é'(b)= S(X"’)Vb =1,...,B

(3) Approximate ser(9) by the sample standard deviation of the B replications:

 

B

M = 1): Iran). é(.)]2/(B- 1) ,

B
where, 9(.) = Z 9.(b)/ B ; and asymptotically, lim 33' = sci .
b=1 B-) 00

65

Now turning to the specifics of this study, I will outline the bootstrap procedure on

how to simulate the standard errors and construct confidence intervals for the estimate13
for which I am primarily interested in, T E: . Given that the within estimation described

above is for all practical purposes analogous to doing least squares with district dummies
and district dummies interacted with time - “the elaborate matrix results in their (CSS
1990) paper notwithstanding, for a moderately sized data set, the most expeditious way to

handle this model is brute force, OLS” (Green 1990, p. 113), I will follow Green’s advice

when running the bootstrap simulation. Thus, 911,921, is estimated in a ‘one-step’

procedure from the least square regression of output on inputs plus district dummies and
district dummies interacted with time, i.e., OLS regression of equation (14). The
bootstrap samples will then be drawn by resampling the OLS residuals. For the first

iteration: (1) I resample the original OLS residuals; (2) generate a corresponding pseudo

data set Y (I) (or a resampled Y) using the parameter estimates of the inputs and

911,921, Y(1)=X,90LS +91i+92i*t+é(1)0LS ; (3) redo the OLS estimation using
original X and resampled Y (I), to get 9(1)0Ls,91r(1) ,921“) ;(4) use 91r(l),921(1) to

estimate TI“’..,TE“’.-.(Vi,r), I repeat this procedure B(=2000 in this study) times to obtain

 

'5 Confidence intervals for 01 and 02 can of course be constructed via standard parametric techniques
given that we already have the standard errors for these estimates, however, in this study I will also use
bootstrapping methods to construct confidence intervals for these estimates.

66

bootstrap flammfllr‘wflzw),TIu‘b),TEu‘b). Proof of the validity of applying

bootstrapping techniques for this problem is given by Hall, Hardle and Simar (1995).

The percentile bootstrap is the simplest bootstrap technique for constructing

confidence intervals. Let G be the cumulative distribution firnction (cdf) of the B

bootstrap replicates, 9”). The 1-2a percentile interval is defined by the a and 1-0.

A

percentiles of G , with the lower and upper bound of the estimator given by:

[9%ro,9°/.up] = [G'l(a),G—l(l - (1)]. Since by definition:

G"(a) = 9"“),

which represents the 100rrth percentile of the bootstrap distribution, we can write the

percentile interval as:

[é%10,é%up] = [é‘(a)’é*(l-a)]

For example, for a 90% confidence interval (CI), with B = 2000 and o. = 0.05, the
percentile interval is the interval ranging from the 100'" to the 1900* ordered values of the
2000 bootstrap replicates. While the percentile technique is relatively simple, it might be
an inaccurate method for relatively small time periods. The bias-corrected and accelerated
(BCA) technique, is considered to be a substantial improvement over the percentile
method in both theory and practice (for details please see Efron and Tibshirani 1993). The
plug-in estimator tends to introduce downward bias in the bootstrap replicator. The BCA

method automatically corrects for this bias. The BCA 1-2a CI is given by:

67

[9%10, 9%up] = [9am I) ,9‘(a2)], where:

20 + 2‘“) 20 + 20'“)

1— am 299)} i “2 = ‘1’ {2” 1- 0(2.+ Iva-9)}

 

 

a]: ¢{20+

where, <1>(.) Is the standard normal cdf and 2‘“) is the 100ath percentile point of <1>(.). For

example, 2‘09”: 1.645, and d>(l.645) = 0.95. Note that if 9 = 20 = O , then a] = a ,

and 0.2 = (l-a), thus, I revert back to the percentile CI specification. The value of the bias

correction 20 is obtained fiom evaluating inverse standard normal cdf at the proportion of

A

bootstrap replications which are less than the original point estimate 9:
20 = <D"(# {9.(b) < 9} / B); 20 measures the median bias of 9.011619 , in normal

units.

There are various methods that can be used to compute the acceleration factor, a .

A widely used method is to express it in terms of jacknifed values of a statistics.
Jacknifing is similar to bootstrapping, and was used before the days of powerfill PCs to

save on computing time. Let X(i) be the original sample with the ith data point deleted;

let the corresponding estimator be defined as 9(i) = S(X (1')), and define

NT
9(.) = Z 9(i) / NT . An expression for the acceleration factor in terms of the jacknifed

i=1

values is as follows:

68

M A A
Z (00- 60')?

 

9:

6* l; «6‘0- émfl”

The acceleration factor refers to the rate of change of the standard error of 9 with

respect to the true parameter value 9 , measured on a normalized scale.

III. Data Section

I draw upon three data sources for this study: (1) Revised Crop Area Statistics ;
National Planning Commission (NPC) 1994; (2) published and unpublished data from the
Central Bureau of Statistics (CBS); (3) unpublished data from the Agricultural Input
Corporation (AIC). Information on area planted to and production of major agricultural
crops for the year 1974 - 1991 stems fi'om the NPC 1994 data set. The NPC 1994 data is
a revision/readjustment of area and production data collected by the Department of Food
and Agricultural Marketing Services (DFAMS). This revision has brought about a drastic
reappraisal of the performance of Nepalese agriculture (Nepal: Agricultural Perspective
Plan 1995). Using unrevised DFAMS data, the growth rate of foodgrain production over
the decade 1981-1991, was 5.03% per annum (compared to the population grth rate
of 2.5% per annum). However, the area planted had been systematically under-measured
in the DFAMS data set. The NPC 1994 data set corrects for this error (among other
revisions). Using the NPC 1994 data, the growth rate of foodgrain production turns out

to be 2.3% per annum, thus, less than the population growth rate during that decade.

69

Information of labor, livestock, literacy rates and rainfall, were obtained fi'om the CBS.
Fertilizer consumption data was obtained fi'om the AIC.

Nepal can be divided into three agro-ecological zones (CBS 1997) : The Terai
(lowlands); The Hills (middle mountains); and The Mountains (the High Mountains, and
the High Himalayas). The Terai region is an extension of the Indo-Gangetic plains,
comprises 23% of the land mass (and most of it arable) and 47% of the population; The
Hill region comprises 42% of the land mass (however, only one tenth of its area is suitable
for cultivation) and 46% of the population. The Mountain region comprises 35% of the
land mass (less than 2% of the land in this region is suitable for cultivation) and 7% of the
population. The importance of the Terai stems from the fact that it remains as the only
food grain-surplus region in Nepal and is the most favorable area for intensified
agriculture.

The three agro-ecological zones are divided into 75 districts: 20 districts fall in the
Terai region; 39 districts fall into the Hill region; while, 16 districts fall into the Mountain
region. In this study, I will exclude districts which fall in the Mountain region from my
analysis. This is primarily due to the fact that data from the Mountain region tends to be
erratic and in many instances may be unreliable. In terms of agricultural production, the
Mountain region is a very minor player in Nepal, with most of the action taking place on
the flood plains of the Terai or on the slopes and valleys of the Hill region. I also restrict
the time frame of my analysis to cover the period 1981-1991, the period for which more
complete data are available. The variables used in this study are as follows:

(1) Output: Aggregated value of Rice, Maize, Millet, Wheat, Barley, Oilseed, Potato,

Tobacco and Sugarcane output, valued at 1991 prices (in Nepalese rupees). Time-series

70

data on area and production (dissaggregated to the district level) is only available for these
major food and cash crops.

(2) Land: area planted to Rice, Maize, Millet, Wheat, Barley, Oilseed, Potato, Tobacco
and Sugarcane. Unfortunately I do not have time-series district level information on area
planted under modern varieties (MV), nor area under irrigation. This is problematic, since
changes in the quality of conventional inputs will now be meshed into changes in TFP
(Alston, J ., G. Norton and P. G. Pardey 1995). This source of bias will be more acute in
the Terai region where there has been more than a 50% increase in the ratio of irrigated
land-to-cropped land between 1981 and 1991 (70% of the land under irrigation in Nepal is
in the Terai)“. In this study, 1 do, however, control for major agro-climatic differences by

conducting separate regional analyses.

(3) Labor-Male; (4) Labor-Female: Number of economically active males and females
(age 10 and above) engaged in agriculture. Again, there is the issue of labor quality. We

will address this issue later on in this section.

(5) Fertilizer: Quantity of chemical fertilizer (NPK) sold by the AIC. Nepal imports it’s
chemical fertilizers, and the AIC, until very recently had a monopoly on fertilizer imports
and nation-wide wholesale distribution (at a very subsidized rate - the subsidy burden of
fertilizer alone consumes 50% of the development budget under the Ministry of
Agriculture). Thus, official estimates of fertilizer use are quite reliable (even afier

accounting for ‘leakages’ - smuggling of fertilizer from Nepal to India).

 

" District level information on irrigated land can be obtained only for two years, 1981 and 1991, from the
National Sample Census of Agriculture 1981and National Sample Census of Agriculture 1991. However,
there is no systematic district level irrigation for years between 1981-1991, thus, could not be included in
this study.

71

(6) Livestock: number of draft animals. Animal traction for field operations is widespread
while tractor use is still rare. Tractor use is catching on in some districts, particularly in
the Terai, however, it is a reasonable assumption that exclusion of this input will not
significantly affect results (I do not have information on tractor hours or number of

tractors per district for all districts for the period 1981-1991).

(7) Literacy-Male; (8) Literacy-Female: ratio of literate males (age 10 and above) to all
males (age 10 and above); ratio of literate females (age 10 and above) to all females (age
10 and above). Ideally I want a reasonable proxy for the average education level (not to
even mention anything about the quality of education) of the labor force engaged in
agriculture. Unfortunately, I only have access to district level literacy rates (a proxy for
education which is widely used in these type of studies). Besides reflecting the literacy
level of the labor force engaged in agriculture, this measure is confounded by the literacy
level of the non-agricultural workforce, and more problematically, by non-working
students. Districts with higher productivity levels, might be more wealthy, and thus, have
higher demand for child schooling, resulting in higher district level literacy rates - thus, the
causality could even run in the oppositive direction. Ideally, we should also difl’erentiate
the effect of education by primary and secondary levels, given that have may have
differential effects on agricultural productivity. Therefore, given that literacy rates
constitute a questionable proxy for labor quality, I will present results both with and

without literacy rates.

Besides the eight explanatory variables, in the estimations I also explicitly control
for, Rainfall: amount of annual rainfall as recorded in 2] weather stations. This is the only

variable which is not expressed at the district level since usually more than one district falls

72

under the coverage of a particular weather station. Classification of districts by weather
stations (and relevant adjustments) was provided by the government of Nepal’s
Department of Hydrology and Meteorology. I present descriptive statistics of the
variables used in the study in Table S] and Table SZ,for the Terai and Hill region,
respectively. The grth rates of these variables in the Terai and Hill region are presented

in Table Sl.1 and Table 82.2.
IV. Empirical Findings

As was previously mentioned, since the available measure of education is an
imperfect proxy, 1 present results both with and without male/female literacy rates We'
can then see how our district level RTC and technical (in)efficiency estimates are altered
due to inclusion of our imperfect measure of education, and to see if the resulting changes

are plausible.
Results Without Literacy : T erai Region

Table 1 presents the within district estimates for the Terai region. The standard
errors of the variable inputs are robust errors based on Huber-White “sandwich”
estimation procedure. These standard errors should be robust to heteroskedasticity (since
I do explicitly control for heterogeneity in this model) and more importantly, serial-
correlation. I tested for serial-correlation using pooled cross-section time-series data
within a simple OLS framework (Wooldridge 1996), i.e., without including the time-
invariant and time-varying heterogeneity terms. Dropping the first three time periods, I

regressed agricultural output on: (1) all inputs and rainfall ; (2) first, second and third lags

of the pooled OLS residuals, fill—l,fiil—2,fill-3. The coefficients on

73

iiu-lfiu -2,iiu-3, were 0.85, 0.6], and 0.05 respectively, with t-statistics of 57.63,

10.12, 1.08, respectively. Thus, serial-correlations appear to follow a AR(2) process and
most importantly, the process appears to be stationary (if the process had been non-
stationary, then the Huber-White procedure would have been insufficient/inappropriate to
obtain the correct standard errors of the population parameters). Also using pooled data,
I carried out a Lagrange Multiplier (LR) test for group-wise heteroschedasticity (Green
1993); as expected, the null hypothesis that the pooled OLS residuals were

homoschedastic was soundly rejected at the 1% level of significance".

All of the variable inputs, except for land, are statistically insignificant (regardless
of robust or non-robust standard errors). I had expected fertilizer to have a significant
positive elasticity, given that there has been a ‘take-off in chemical fertilizer use in the
Terai during this period (with an average per annum growth rate of 21.8% - see Table
$12). The estimates of the production elasticities generally correspond to estimates
obtained fi'om cross-country studies involving deve10ping country data (e.g., Lau and
Yotopoulos 1989)“, except for the production elasticity of male labor and land. The
coefficient on agricultural male labor force participation is negative - which is a very rare
finding, but, male labor is insignificant. As we can see in Table 81.2 and $2.2, males have
been deserting the agricultural sector in both regions (female labor force participation
rates in agriculture have also been dropping in the Terai, while they have been going up in

the Hills). Despite negligible land expansion in the Terai during this period, we find that

 

'7 I conducted these test for both Terai and Hill regions, with and without literacy; tests indicate serial
correlation in all models, thus, it is appropriate to include robust standard errors (test results not presents).

" For example, most production elasticities of labor tend to be around 0.40; livestock elasticities tend to
vary considerably anywhere between the range of 0.30 - 0.09; so does fertilizer elasticities, 0.08 - 0.05.

74

land is highly significant (with a t-statistics of 7.8) and has the largest magnitude efl‘ect on
agricultural output. A 1% increase in agricultural land during the 80's in the Terai, on
average would have boosted output by more than 2% per annum. Unfortunately I do not
have information on land quality, area under modern varieties (MV), nor do I have
information on irrigation for this study. It is likely that the large elasticity of land reflects
increases in productivity due to expansion of irrigation and area planted to MV (and
possible changes in cropping intensities). The amount of rainfall also significantly effects

agricultural output as expected (with a t-statistics of 4.6).

Point estimates of district level RTC (02,) for the Terai region are presented in
Table 2". All of the 0,'s are highly significant, as well as most of the rates of technical
change (02's). Thus, district level time-varying and time-invariant heterogeneity
significantly affects agricultural productivity. The average rate of technical change, or
TFP growth for the 20 Terai districts during the 80's was 3%, with Saptari registering the
highest grth rate at 7.4% (and Bardiya coming in last with a rate of 0.5%). A 3% per
annum rate of technical change or TFP growth seems too high, particularly for Nepal
(relative to the per annum output growth rate of 3.37% in the Terai). TFP growth rates
for other South Asian countries usually have been found to hover around 1% per annum
(e.g., Evenson and Rosegrant 1993a; Rosegrant and Evenson 1993; 1995). Looking at
Table 2.2, we see that district level RTC estimates are generally large relative to actual
district level grth rates in output. Thus, it seems as if upward bias in the RTC estimates

is very strong. As seen in Table 3, the confidence interval for most of the RTC estimates

 

'9 The null hypothesis that the 02's are jointly equal to zero, is soundly rejected at the 1% level of
significance. '

75

are too wide to rank the districts distinctly. For example, while Saptari, Mahotari and
Jhapa have point estimates of RTC above 5%, there are 10 other districts with upper
confidence band value above 5% (districts can however, be grouped based on ranges of

rates of technical change.

Table 4 presents district level technical efficiencies for the first time period, 1981.
The average technical efficiency of production is 60%, thus, implying that on average,
agricultural production in the Terai region in 1981 could have been increased by 40% via a
more efficient utilization of available resources. Based on point estimates, it appears as if
Parsa was operating on its frontier in 1981, and was the most efficient district (with Jhapa
coming in as the least efficient district, operating at only 36% of its potential). However,
when we look beyond mere point estimates, we see that at least two other districts,
Bardiya and Nawalparasi, have upper bound values of 1 (based on the BCA approach). It
is again difficult to rank districts distinctly; however, I can again group districts based on
ranges of efficiency levels. Table 5 shows district level technical efficiencies for the last
time period, 1991. The average technical efficiency of production was 70%, thus,
implying that on average, agricultural production in the Terai region in 1991 could have
been increased by 30% via a more efficient utilization of available resources (average
technical efficiency levels have risen by 10% over 1981-1991, or by 1% per annum). Both -
the 198] and 199] TE estimates seem rather high compared to results from the farm-plot
sample analysis in Part I, where I found that farmers could have increased average output
by 18%, via a more efficient utilization of available resources. As we see in Table 5,
district rankings based solely on point estimates have changed. While Parsa still remains

as the most efficient district, 10 districts experienced increases in their efiiciency of

76

production, while there was a fall in efficiency levels in 9 districts. Factoring in
information fi'om the confidence bands, it appears as if there is some weak sign of
‘convergence’, while only 3 districts could have been operating on the frontier in 198], 6

districts might well have been operating on the frontier in 1991.

Table ARC] shows the year in which a local agricultural research center had been
established in that district (9 out of the 20 districts in the Terai have a local agricultural
research center). It is interesting to note that, Parsa which in 1981 was ranked the most
efficient district in the Terai, has the oldest agricultural research center in the Terai (and is
considered by many agronomists to be one of the best local agricultural research centers).
However, if we simply plot district level technical efficiency against the number of years
(up till 1981) there has been an agricultural research center based in that district, we
realize that the relationship is rather week (see Figure 1). For example, Bardiya which has
the second-highest point estimate for the Terai, has never had a local agricultural research
center based in that district. Of course, districts without local agricultural research centers
could still benefit from research/extension linkages with research centers of other districts.
Also, we need a richer understanding of the quality of research of the local agricultural
research centers, and the effectiveness of those centers in working with extension
networks to disseminate agronomic information to farmers. I however, do not have

access to such institutional information for this study.
Results Without Literacy .° Hill Region

Table 6 shows the within estimates for the Hill Region. The elasticities are in line
with previous studies, albeit on the low side for fertilizer elasticity. Unlike results from the
Terai region, land does not appear to be a significant determinant of agricultural output.

77

Also, unlike results fiom the Terai region, male labor and fertilizer are significant (at the
5% level of significance). A 1% increase in the male labor force engaged in agriculture
increases agricultural output by 0.36%; while a 1% increase in the use of chemical
fertilizers results on average in a 0.02% increase in output. Rainfall is also an important

factor in the Hill region, albeit not as significant as in the Terai.

Similar to results obtained from the Terai region, all of the 01's are highly
significant, as well as most of the rates of technical change (please see Table 7)". The
average rate of technical change, or TFP growth for the 39 Hill districts during the 80's
was 2.5% (the growth rate of output during this period in the Hill region was 2% per
annum). Kavre registered the highest growth rate at 7.4%, while the capital, Kathmandu,
come in last with a negative TFP grth rate of -l .85% (Bhaktapur was the only other
district with a negative TFP grth rate). Negative TFP grth rates may reflect decline
in the quantity and quality of the natural resource base (Lynam and Herdt 1989; Cassman
and Pingali 1994; Ali and Byerlee 1998). However, since I do not have any pertinent
measures such as soil nor water quality, I cannot surmise about the impact of resource
degradation on productivity. Given widespread deforestation and associated loss of top-
soil in the Hill region, I am surprised that more districts did not register a negative TFP
growth rates. Again the RTC estimates appear to be too high relative to actual output
growth rates (see Table 7.2). Even after factoring in confidence intervals (please see

Table 8), I can still say that Kavre exhibited the highest rate of technical change in the Hill

 

3° The null hypothesis that the 02's are jointly equal to zero, is soundly rejected at the 1% level of
significance.

78

region during the 80's. However, the remaining Hill districts cannot be ranked in such a

clear-cut fashion, but I can group districts based on ranges of rates of technical change.

Table 9 presents district level technical efficiencies for the first time period, 1981.
The average technical efficiency of production was 45%, thus, implying that on average,
agricultural production in the Hill region in 1981 could have been increased by 55% via a
more efficient utilization of available resources. Thus, compared to results from the Terai
region, agricultural production is less efficient in the Hill region. Based on point
estimates, it appears as if Kathmandu was operating on its frontier in 1981, and was the
most efficient district (Myagdi was the least efficient district, operating at only 16% of its
potential). Factoring in confidence intervals, only one other district, Syangja, has an upper
bound value of 1. It is again difficult to distinctly rank districts - what we can do is to use
the confidence bands to group districts based on efficiency ranges. Table 10 presents
district level technical efficiencies for the last time period, 1991. The average technical
efficiency of production was 41%, thus, implying that on average, agricultural production
in the Hill region in 1991 could have been increased by 59% via a more efficient utilization
of available resources. It appears as if there was no eficiency growth during the 80's in
the Hill region; rather efficiency fell by 4% during 1981-1991. Thus, average TFP grth
in the Hill region must have taken place without increases in the level of technical
efficiency. While Kavre emerges definitively as the most efficient district in 1991, there is
no indication of ‘convergence’ in the Hill region (based on point estimates, 4 districts
maintained their rankings, 17 districts experienced increases in their efficiency of

production, while 18 districts experienced decreases in their level of efficiency).

79

Table ARC2 shows the year in which a local agricultural research center had been
established in that district (11 out of the 39 districts in the Hill region have a local research
center). We see that Kathmandu, the nation’s capital, which in 1981 was ranked the most
efficient district in the Hill region, has the oldest agricultural research center in the Hill
region. However, similar to what we see in the Terai, there does not appear to be a
coherent relationship between district level technical efficiency and how long a local

agricultural research center has been based in that district (see Figure 2).
Results With Literacy .' T erai Region

Unlike in Part I where I found that secondary education had a significant effect in
lowering production inefficiencies among a sample of farmers in the Rupandehi district, I
do not find a significant effect of education when drawing upon aggregate district data in
the Terai region. While including male and female literacy rates does not significantly alter
the production elasticities, RTC and efficiency estimates are however, altered in a
significant manner. Table 11 presents within estimates for the Terai region, controlling for
male and female literacy rates. As shown, this increases the elasticities of male and female
labor, however, they still remain insignificant. Neither male nor female literacy rate is
significant. Also, while the elasticity of literacy tends to vary between 0.10 - 0.20 in most
developing country studies, the literacy elasticity in this study is very low (for both males
and females). .Table 12 shows that while most of the 01 parameter estimates are still
strongly significant, most of the 02, or RTC parameter estimates are now insignificant".
Thus, I cannot make a strong case that there is significant variation in technical change

across districts. Ignoring the issue of statistical significance for now, the average annual

 

2' The null hypothesis that the 02's are jointly equal to zero, cannot be rejected.

80

rate of technical change, or TFP grth for the 20 Terai districts during the 80's (after
controlling for literacy rates), was -0.47%. Sixty five percent of the districts in the Terai
region now appear to experienced negative TFP growth. The RTC estimates are more
plausible in comparison to actual grth rates (please see Table 12.2), however, now they
appear to be biased downwards. However, too much cannot be made of these estimates
given their standard errors are too large (see Table 13 for the confidence intervals of the

RTC point estimates).

Table 14 presents district level technical efficiencies for the first time period, 1981.
The average technical efficiency of production was 47% (compared to 60% when I do not
include literacy). Rankings based upon point estimates are quite different than the case in
which we do not include literacy. Based on point estimates, it appears now appears as if it
is Bardiya which was operating on its frontier in 1981 (with Jhapa again coming in as the
least efficient district, operating at only 19% of its potential). However, when considering
the CI intervals, we see that at least three other districts, Parsa, Nawalparasi, and Sunsari,
have upper bound values of 1 (based on the BCA approach). It is again difficult to rank
districts distinctly, however, I can group districts based on ranges of efficiency levels.
Table 15 presents district level technical efficiencies for the last time period, 1991. The
average technical efficiency of production was 57%( compared to 70% when I do not
include literacy). Thus, average technical efficiency levels had also risen by 10% over p
1981-1991, or by 1% per annum. Also, as reported in Table 15, district rankings based
solely on point estimates have changed : 10 districts experienced increases in their
efficiency of production, while there was a fall in efficiency levels in 6 districts. Again,

there does not seem to be any signs of ‘convergence’.

8]

Results With Literacy : Hill Region

Similar to the Terai case, while including male and female literacy rates does not
significantly alter the production elasticities, RTC and efficiency estimates are altered in a
significant manner. However, unlike the Terai case (where neither male nor female
literacy was significant), female literacy rate is mildly significant (t-statistics of 1.7), but,
with a low elasticity (see Table 16). A 1% increase in female literacy rate in the Hill
region appears to (on average) increase output by 0.04% (please see Table 16). From
results presented in Table 17, opposite to the Terai case, the 02/RTC parameter are mostly
significant”, while most of the 01 parameter estimates are now statistically insignificant.
Thus, I can make the case that significant variation in district level rates of technical
change exists in the Hill region. Controlling for literacy in the estimation, results in
negative rates of technical change, or decreasing TFP grth in all Hill regions. The
average decline in TFP grth is -8.4% per annum. As shown in Table 18, the
confidence intervals for the RTC estimates are quite wide. For example, the rate of
technical change in Kavre could either be negative, zero, or positive. Even though the
RTC estimates exaggerate the extent of decline in factor productivity (e. g., compare
district level grth rates and RTC in Table 17.2), given the statistical significance of
most of the RTC estimates, it is possible that these result might very well stem from the
ravages of resource degradation in the Hill region of Nepal. Using comprehensive soil and
water quality data, Ali and Byerlee (1998) find that widespread resource degradation in

the wheat-rice system of Pakistan resulted in negative TFP growth. Similarly, Cassman

 

2’ The null hypothesis that the 02's are jointly equal to zero, is soundly rejected at the 0% level of

significance.
82

and Pingali (1994) associate the decline in TFP in the lice-wheat system of the Indian
Punjab to intensification induced soil degradation. In the Hills, it is not unsustainable
agricultural intensification which is the culprit; rather it is whole scale deforestation and
associated loss of top soil which poses the greatest threat to agricultural productivity.
Unfortunately, I do not have the data with which to examine this critical issue within the

Nepalese context.

Table 19 presents district level technical efficiencies for the first time period, 1981.
The average technical efficiency of production was 46% (these average results are almost
, identical to those obtained in the case when literacy was not include). Based on point
estimates, it appears as if Dhading was operating on its frontier in 1981 (with Parbat
coming in as the least efficient district, operating at only 16% of its potential). Factoring
in confidence intervals, two other districts, Nuwakot and Salyan, have upper bound
values of 1. Table 20 presents district level technical efficiencies for the last time period,
1991. The average technical efficiency of production was 40%. It appears that average
efficiency levels fell slightly during the 80's in the Hill region. Again no indication of

convergence is evident in the Hill region.
V. Conclusion

Using a stochastic fiontier approach embedded within a meta-production function
framework, I explore for district level technical change and efficiency of agricultural
production for the Terai and Hill regions of Nepa during the period 1981-1991. I present
a summary of the results obtained from the district level analysis in Table 21. As shown,
the ‘story’ we wish to tell about technical change and efficiency in the Terai and Hill
regions during this period, hinges upon whether we include literacy in the estimation.

83

Certainly, education belongs in the estimation of a meta-production filnction. Ignoring
education would result in over estimating the rate of technical chanage. When literacy, an
imperfect measure of the education level of the agricultural labor force, is included, there
is indeed a drop in the district level rates of technical change. Alas, the reductions seems
rather drastic, particularly for the Hill region. On the other hand, inclusion of literacy only
bring about a -21% drop in average technical efficiency levels in the Terai region, while it
does not appear to alter the average efficiency level in the Hill region. A drop in average
efficiency levels in the Hill region (with and without inclusion of literacy) is notable
between 1981 and 1991. Efficiency levels in general stay the same or increase over time
as farmers within a district ‘learn’ to better use existing resources. Declining efficiency
levels can however, be indicative of the changing production environment in the Hill
region. Technical efficiency reflects the ability to maximize production, given a fixed set
of inputs, which includes natural resources. If the quantity and the quality of the natural
resource base is declining, and there exits limited substitutability between manmade and
non-manmade inputs, then we could very well experience a decline in efficiency levels.
The district level analysis was severely curtailed due to a lack of better proxies for
education, from not having irrigation information for the Terai region, and from not
having any measures of the quality and quantity of natural resources used in the
production process, particularly in the Hill region. I however, able to control for irrigation
when exploring for factors which might help to explain the variation in technical efficiency

across districts for the first and last period of the panel.

Despite these deficiencies, the district level analysis provides an empirical ‘starting-

point’ for examining the productivity of Nepalese agriculture at a aggregate (meso) level.

84

In spite of potential downward bias in the technical efficiency estimates, this study shows
that there exits substantial scope for increasing output via a better utilization of existing
inputs and technologies in the Terai region. There also exists tremendous potential for
increasing output in the Hill region. However, this potential is perhaps being squandered
due to natural resource degradation in the Hills. The significantly negative rates of
technical change for the Hill districts (after inclusion of literacy) could very well be written
off as nothing more than a statistical anomaly. However, if they do reflect an underlying
decline in TFP grth due to degradation of the natural resource base, then we do have
cause for concern, and therefore, this issue remains as an unanswered empirical question.
To properly address that question requires a strong institutional commitment to help

strengthen Nepal’s capacity to collect such complex data in a systematic fashion.

85

 

omod odd oood- Nvod word- odd- oood. ..ood «when
oood mhod wood oo pd ovod- oNod- owod- mood 33233.
oood omod mood on Pd odd- hNod- oood. o ..od oxcmm
mood oNod wood 9. ...o oood god- vood- omod Ecocaaam
oood mood mood- omod oood o rod. Nood. NNod 53¢ch
mood mmod wood vFNd NmNd hNod- n rod oNod demo
PNwd mood hvod- mo Pd :od «mod- o rod oNod tam—Sm
ohod god oood ovwd oNod o_.o.o- oood onod 3233262
Nmod oood hmwd on...o otdl omod- mood. oood 355w”.
vood omod ”Pod NhNd ode- m ..od- oood ..nod caucuses.
omod ovod sood. or ...o mvod- wood- mood oood 93.8.2
wood wood oood hood muod m Pod- o _.o.o omod __m=m¥
omod ovod swod- om vd mood- oood- Nood- oood «:25
vsod obod o w Pd hNPd ovod- wood- _.Nod Nvod «32mm
shod Sod wood mood ovod- god- Food wvod 33:20
ovod oNod o rod No «d o Pod- vvod. hood- hvod tmflww
oood omod «sod. oood vood- ovod- o_.od vmod Sam
omod mood wood oo_.d ovod- ovod- Nood hmod mound
Nbod mvod Nvod- vm Pd oNod- omod- god omod Eatww
omod mvod wood de oowdl ohod- mood omod :30sz
“#5823 2-980;... xoofigj donate“. “7.003 25.33 96.. 59:0 8:65

32.33 8:8. 2.. .8 .839. 3:: e5 5 85. 5326

..m 03¢...

86

Table 82

Growth Rates in the Hill Region for the period 1981-1991

 

District Output Land Labor- Labor-F Fertilizer Livestock Literacy-M LiteractF
Kavre 0.056 0.015 -0.032 0.004 -0.031 0.090 0.045 0.065
Achham 0.038 0.017 -0.023 0.027 -0.012 0.030 0.063 -0.030
Rukum 0.033 0.017 -0.015 0.046 0.008 0.020 0.069 0.099
Lalitpur 0.032 0.004 -0.076 0.004 0.037 0.014 0.046 0.066
Jajarkot 0.031 0.012 -0.015 0.033 0.039 0.024 0.074 0.066
Lamjung 0.030 0.008 -0.039 0.054 -0.003 0.013 0.024 0.059
Khotang 0.030 0.011 -0.026 0.026 -0.045 0.025 0.049 0.082
Gorkha 0.028 0.003 -0.040 -0.014 0.191 0.011 0.072 0.117
Baitadi 0.027 0.012 -0.035 0.030 0.014 0.027 0.059 0.049
Makwanpur 0.025 0.008 -0.010 0.022 0.008 0.013 0.041 0.059
Myagdi 0.025 0.011 -0.046 0.034 0.024 0.041 0.054 0.084
Parbat 0.025 0.009 -0.034 0.008 0.033 0.010 0.030 0.096
Dadeldhura 0.025 0.002 -0.034 0.065 0.095 0.066 0.064 0.020
Dhading 0.024 0.007 -0.022 0.019 0.018 0.011 0.070 0.098
Baglung 0.023 0.017 -0.045 0.015 0.013 0.011 0.044 0.099
Palpa 0.023 -0.006 -0.041 0.027 -0.014 0.013 0.045 0.074
Tanahu 0.023 0.003 -0.043 0.005 0.030 0.014 0.057 0.099
Rolpa 0.023 0.013 -0.018 0.027 -0.012 0.086 0.069 0.028
Surithet 0.022 0.006 -0.005 -0.001 0.020 0.024 0.056 0.094
Sindhuli 0.021 0.002 -0.017 0.029 -0.038 0.017 0.052 0.061
Dhankuta 0.021 -0.001 -0.018 0.037 0.036 0.013 0.033 0.083
Udayapur 0.020 0.000 -0.009 0.011 0.057 0.014 0.061 0.088
Ramechhap 0.019 0.012 0.024 0.098 0.026 0.096 0.049 0.063
Bhojpur 0.018 0.004 -0.033 0.013 -0.012 0.012 0.049 0.063
Terhathum 0.016 0.001 -0.031 -0.003 0.017 0.014 0.035 0.089
Kaski 0.016 0.003 -0.062 -0.006 0.018 0.012 0.037 0.074
Arghakhanchi 0.016 0.013 -0.055 0.062 0.011 0.130 0.050 0.088
Nuwakot 0.015 0.008 -0.016 0.013 0.101 -0.017 0.048 0.079
Gulmi 0.015 0.004 -0.051 0.015 -0.007 0.098 0.027 0.070
Salyan 0.014 0.010 -0.015 -0.004 0.013 0.019 0.076 0.090
Okhaldhunga 0.014 0.011 -0.029 0.001 0.006 0.020 0.063 0.119
Panchthar 0.01 1 0.005 -0.024 -0.014 0.009 0.017 0.045 0.106
Pyuthan 0.010 0.014 -0.030 0.003 0.023 0.018 0.062 0.097
Syangja 0.010 0.005 -0.043 0.013 -0.019 0.011 0.036 0.106
Dailekh 0.009 0.011 -0.017 0.012 0.008 0.016 0.048 0.046
Doti 0.008 0.006 -0.034 0.034 0.002 0.034 0.087 0.051
llam 0.007 0.007 -0.013 -0.002 0.032 0.013 0.035 0.071
Bhaktapur -0.004 -0.009 -0.055 0.004 -0.060 0.014 0.048 0.086
Kathmandu -0.021 -0.038 -0.123 -0.051 0.010 0.015 0.031 0.040

87

Table 1
Parameter Estimates of the Production Function (Without Literacy) - Terai
Dependent Valrable : Natural Log of Agricultural Output (Mean Value : 6.63)

Explanatory Variables (besides net school enrolment rates) are in Natural Logs

 

Independent Variables Win-Estimates
land 2.234
(0.285)‘
labor-male -0.400
(0.260)
labor-female ' 0.130
(0.080)
fertilizer 0.030
(0.023)
livestock 0.052
(0.040)
rainfall 0.170
(0.037)*
Observations 220
# of Districts 20
it of Time Periods 11

robudstandarderrorsinparentheees;'signiticantat1%level

Note : robust errors based on White's "Sandwich Method" are heteroskedasticity/serial—correhtion
consistent estimates

88

Table 2

Parameter Estimates (Without Literacy) of time-invariant and time-varying heterogeneity :

District

Saptari
Mahotari
Jhapa
Kapilbastu
Rautahat
Bara

Banke
Siraha
Dhanusa
Rupandehi
Sadahi
Chitawan
Kailall
Morang
Kanchanpur
Parsa
Nawalparasi
Sunsari
Dang
Bardiya

" : Standard-Error

 

 

Terai
Thrre-lnvarient heterogernlty
Estimate SE' t-stat
4.978 0.926 -5.377
-4.621 0.884 -5.230
-5.127 0.979 -5.236
-4.988 0.970 -5.141
4.915 0.931 -5.281
-4.560 0.939 -4.855
4.388 0.851 -5.157
-4.781 0.930 -5.143
-4.628 0.913 -5.067
-4.837 0.957 -5.057
-4.668 0.898 -5.199
4.623 0.917 -5.043
4.862 0.930 -5.225
-4.846 0.968 -5.005
4.459 0.883 -5.053
-4.067 0.894 -4.550
-4.197 0.859 -4.888
-4.324 0.892 -4.850
-4.654 0.917 -5.077
-4.116 0.829 -4.967

" : Rate of Technical Change (time-varying helerogeniety)

89

che
Estimate SE * t-stat
0.074 0.010 7.067
0.053 0.012 4.433
0.051 0.010 5.010
0.044 0.011 4.179
0.044 0.010 4.366
0.042 0.012 3.542
0.042 0.010 4.049
0.041 0.010 4.030
0.038 0.010 3.770
0.033 0.010 3.277
0.029 0.011 2.724
0.028 0.010 2.820
0.025 0.011 2.252
0.022 0.010 2.198
0.016 0.010 1.519
0.016 0.010 1.535
0.014 0.010 1.369
0.009 0.010 0.894
0.007 0.011 0.626
0.005 0.011 0.425

Table 2.2

Terai - Without Literacy

 

District GRYA RTC
Mahotari 0.058 0.053
Sanahi 0.058 0.029
Jhapa 0.057 0.051
Bara 0.054 0.042
Saptari 0.047 0.074
Dhanusa 0.042 0.038
Bardiya 0.042 0.005
Siraha 0.039 0.041
Kailali 0.038 0.025
Morang 0.036 0.022
Kanchanpur 0.031 0.016
Rautahat 0.030 0.044
Nawalparasi 0.030 0.014
Sunsari 0.026 0.009
Dang 0.026 0.007
Chitawan 0.022 0.028
Rupandehi 0.020 0.033
Banke 0.013 0.042
Kapiibastu 0.005 0.044
Parsa 0.001 0.016

" GRY : growth rate of output

90

Table 3

Bootstrap Confidence Intervals tor RTC Esimates (Without Literacy) :

90% Level - Terai

 

Percentile BCA

District RTC Estimate LB UB LB UB

Saptari 0.0735 0.0568 0.0892 0.0672 0.0985
Mahotari 0.0532 0.0358 0.0706 0.0470 0.0845
Jhapa 0.0506 0.0361 0.0648 0.0448 0.0735
Kapiibastu 0.0443 0.0286 0.0596 0.0381 0.0687
Rautahat 0.0441 0.0293 0.0587 0.0381 0.0662
Bara 0.0418 0.0243 0.0594 0.0361 0.0702
Banke 0.0417 0.0249 0.0573 0.0355 0.0647
Siraha 0.0407 0.0258 0.0553 0.0348 0.0635
Dhanusa 0.0377 0.0233 0.0532 0.0322 0.0648
Rupandehi 0.0331 0.0182 0.049 0.0276 0.0611
Sarlahi 0.0286 0.0138 0.0436 0.0226 0.0525
Chitawan 0.0282 0.0132 0.0429 0.0237 0.0544
Kailali 0.0250 0.0087 0.0416 0.0192 0.0519
Morang 0.0222 0.0079 0.0371 0.0163 0.0439
Kanchanpur 0.0158 0.0004 0.0313 0.0104 0.0413
Parsa 0.0155 0.0004 0.0303 0.0087 0.0345
Nawalparasi 0.0141 -0.0010 0.0293 0.0079 0.0379
Sunsari 0.0093 -0.0055 0.0242 0.0027 0.0308
Dang 0.0067 -0.0091 0.0219 0.0012 0.0313
Bardiya 0.0048 -0.0108 0.0212 -0.0013 0.0306

91

Table 4

Bootstrap Confidence Intervals for Technical Efficiency Esimates (Without Literacy) :
90% Level - Terai

Technical Efficiency Estimates for the First Period (1981)

 

Percentile BCA

District Efficiency (T: 1981) L8 UB LB UB

Parsa 1.000 0.915 1.000 0.962 1.000
Bardiya 0.943 0.816 1.000 0.891 1.000
Nawalparasi 0.877 0.765 0.967 0.839 1.000
Sunsari 0.768 0.673 0.846 0.733 0.931
Banke 0.745 0.637 0.831 0.71 1 0.890
Kanchanpur 0.676 0.589 0.755 0.642 0.826
Bara 0.627 0.529 0.724 0.593 0.818
Mahotari 0.597 0.513 0.674 0.565 0.751
Dhanusa 0.583 0.505 0.651 0.556 0.708
Chitawan 0.581 0.505 0.652 0.556 0.701
Sanahi 0.555 0.487 0.621 0.531 0.703
Dang 0.551 0.474 0.620 0.526 0.682
Siraha 0.502 0.431 0.574 0.476 0.644
Rupandehi 0.471 0.393 0.549 0.443 0.626
Morang 0.462 0.381 0.545 0.432 0.649
Kailali 0.456 0.389 0.528 0.430 0.595
Rautahat 0.441 0.378 0.499 0.418 0.554
Saptari 0.426 0.368 0.481 0.405 0.530
Kapiibastu 0.410 0.341 0.486 0.386 0.548
Jhapa 0.359 0.292 0.428 0.333 0.523

Note: EMciency estimates are expressed 'relative' to the most ancient dstrict in the sample

(Which has a value of 1)

92

Tabie 5

Bootstrap Confidence Intervals for Technical Efficiency Esimates (Without Literacy) :

90% Level - Terai

Technical Efficiency Estimates for the Last Period (1991)

 

Percentile BCA

District Efficiency (T =1 99 1) LB UB LB UB

Parsa 1.000 0.891 1.000 0.946 1.000
Banke 0.968 0.841 1.000 0.908 1.000
Mahotari 0.871 0.756 0.954 0.822 1.000
Nawalparasi 0.865 0.757 0.951 0.821 1.000
Bardiya 0.847 0.733 0.933 0.806 1.000
Bara 0.816 0.664 0.957 0.753 1.000
Saptari 0.762 0.654 0.847 0.712 0.890
Dhanusa 0.728 0.617 0.813 0.681 0.875
Sunsari 0.722 0.618 0.801 0.681 0.863
Kanchanpur 0.678 0.587 0.755 0.643 0.813
Chitawan 0.659 0.564 0.730 0.621 0.786
Siraha 0.646 0.544 0.728 0.604 0.781
Sanahi 0.633 0.534 0.714 0.594 0.764
Rautahat 0.587 0.501 0.661 0.552 0.694
Rupandehi 0.562 0.461 0.648 0.523 0.705
Kapilbastu 0.547 0.457 0.620 0.509 0.668
Jhapa 0.510 0.402 0.610 0.464 0.662
Dang 0.505 0.412 0.579 0.467 0.624
Kailali 0.502 0.411 0.584 0.463 0.646
Morang 0.494 0.395 0.590 0.448 0.630

Note: Elliciency estimates are expressed 'Ielative' to the most elllcient dstrict in the sample
(which has a value of 1)

93

Table 6
Parameter Estimates of the Production Function (Without Literacy) - Hill

Dependent Vslrable : Natural Log oIAgricuA‘urd Output (Mean Value : 5.507)

 

Explanatory Variables are in Natural Logs
Independent Veirables Win-Estimates
land 0.1 16
(0.23)
labor-male 0.36
(0.1 8)“
labor-female 0.017
(0.087)
fertilizer 0.022
(0.01 O)“
livestock 0.056
(0.047)
rainfall 0.066
(0.030)“
Observations 429
# of Districts 39

# of Time Periods

11

robflstandarderrorsinparentheses;“slgniflcentatthe5%level

Note : robud errors based on White's ”Sandwich Method” are heteroskedasticity/serial—correlation
consisted estimates

94

Table 7

Parameter Estimates (Without Literacy) of time-invariant and time-varying heterogeneity :Hill

District

Kavre
Achham
Rukum
Jajarkot
Myagdi
Khotang
Baitadi
Lamjung
Lalitpur
Gorkha
Surkhet
Makwanpur
Palpa
Dhading
Dadeldhura
Tanahu
Parbat
Sindhuli
Baglung
Rolpa
Dhankuta
Udayapur
Bhojpur
Ramechhap
Salyan

Doti
Terhathum
Kaski
Okhaldhunga
Arghakhanchi
Gulmi
Panchthar
Pyuthan
Dailekh
Nuwakot
Syangja
Ilam
Bhaktapur
Kathmandu

" : Standard-Error

 

 

Tine-invarient heterogeneity
Estimate SE " t-stat
5.267 0.651 8.093
4.191 0.512 8.183
4.814 0.599 8.036
4.095 0.526 7.791
4.050 0.497 8.142
5.003 0.633 7.905
4.302 0.527 8.168
4.897 0.596 8.217
4.867 0.497 9.798
5.329 0.676 7.887
5.137 0.647 7.934
5.535 0.649 8.531
5.040 0.634 7.949
5.262 0.625 8.426
4.412 0.545 8.092
5.261 0.657 8.008
4.798 0.599 8.007
5.239 0.638 8.206
4.795 0.581 8.249
4.700 0.557 8.432
5.183 0.627 8.262
4.969 0.593 8.385
5.201 0.621 8.382
4.826 0.578 8.343
5.103 0.648 7.874
4.588 0.550 8.341
4.919 0.567 8.681
5.427 0.672 8.080
4.673 0.522 8.955
4.694 0.552 8.508
5.077 0.626 8.111
5.295 0.626 8.454
4.714 0.565 8.340
4.565 0.548 8.331
5.621 0.646 8.699
5.770 0.734 7.866
5.465 0.617 8.863
5.270 0.475 11.103
5.915 0.610 9.692

" : Rate of Technical Change (time-varying heterogeniety)

95

RTC“

Estimate SE * t-stat
0.088 0.011 7.813
0.050 0.010 5.295
0.040 0.009 4.333
0.039 0.009 4.286
0.038 0.009 4.074
0.037 0.009 4.033
0.037 0.009 3.967
0.034 0.009 3.775
0.032 0.009 3.636
0.032 0.009 3.625
0.030 0.009 3.322
0.029 0.009 3.225
0.028 0.009 3.169
0.028 0.009 3.124
0.028 0.009 3.148
0.028 0.009 3.090
0.028 0.009 3.090
0.027 0.009 3.045
0.026 0.009 2.783
0.025 0.009 2.767
0.024 0.009 2.742
0.024 0.009 2.761
0.022 0.009 2.443
0.021 0.009 2.356
0.021 0.009 2.289
0.020 0.009 2.207
0.020 0.009 2.247
0.020 0.009 2.239
0.020 0.009 2.154
0.019 0.009 2.144
0.018 0.009 2.000
0.016 0.009 1.753
0.015 0.009 1.681
0.014 0.009 1.589
0.014 0.009 1.551
0.010 0.009 1.120
0.010 0.009 1.101
-0.002 0.009 -0.202
-0.018 0.010 -1.854

Table 7.2

Hill - Without Literacy

 

District GRYA RTC

Kavre 0.056 0.088
Achham 0.038 0.050
Rukum 0.033 0.040
Lalitpur 0.032 0.032
Jajarkot 0.031 0.039
Lamjung 0.030 0.034
Khotang 0.030 0.037
Gorkha 0.028 0.032
Baitadi 0.027 0.037
Makwanpur 0.025 0.029
Myagdi 0.025 0.038
Dadeldhura 0.025 0.028
Parbat 0.025 0.028
Dhading 0.024 0.028
Baglung 0.023 0.026
Palpa 0.023 0.028
Tanahu 0.023 0.028
Rolpa 0.023 0.025
Surkhet 0.022 0.030
Sindhuli 0.021 0.027
Dhankuta 0.021 0.024
Udayapur 0.020 0.024
Ramechhap 0.019 0.021

Bhojpur 0.018 0.022
Terhathum 0.016 0.020
Arghakhanchi 0.016 0.019
Kaski 0.016 0.020
Nuwakot 0.015 0.014
Gulmi 0.015 0.018
Salyan 0.014 0.021

Okhaidhunga 0.014 0.020
Panchthar 0.011 0.016
Pyuthan 0.010 0.015
Syangja 0.010 0.010
Dailekh 0.009 0.014
Doti 0.008 0.020
Ilam 0.007 0.010
Bhaktapur -0.004 -0.002
Kathmandu -0.021 -0.018

" GRY : growth rate of ouptut

96

Table 8

Bootstrap Confidence Intervals for RTC Esimates (Without Literacy) : 90% Level - Hill

 

Percentile BCA

District RTC Estimate LB UB LB UB

Kavre 0.088 0.070 0.105 0.081 0.115
Achham 0.050 0.035 0.065 0.044 0.073
Rukum 0.040 0.027 0.054 0.035 0.062
Jajarkot 0.039 0.025 0.052 0.034 0.064
Myagdi 0.038 0.025 0.052 0.033 0.058
Khotang 0.037 0.022 0.050 0.031 0.058
Baitadi 0.037 0.023 0.049 0.031 0.057
Lamjung 0.034 0.020 0.047 0.029 0.055
Lalitpur 0.032 0.018 0.045 0.027 0.062
Gorkha 0.032 0.019 0.045 0.028 0.055
Surkhet 0.030 0.016 0.042 0.026 0.052
Makwanpur 0.029 0.015 0.042 0.024 0.053
Palpa 0.028 0.016 0.041 0.023 0.047
Dhading 0.028 0.014 0.041 0.023 0.051
Dadeldhura 0.028 0.014 0.041 0.023 0.049
Tanahu 0.028 0.013 0.040 0.023 0.048
Parbat 0.028 0.013 0.040 0.023 0.052
Sindhuli 0.027 0.014 0.041 0.022 0.046
Baglung 0.026 0.011 0.039 0.021 0.051
Rolpa 0.025 0.012 0.038 0.021 0.047
Dhankuta 0.024 0.012 0.038 0.020 0.045
Udayapur 0.024 0.012 0.039 0.020 0.048
Bhojpur 0.022 0.008 0.035 0.017 0.043
Ramechhap 0.021 0.008 0.034 0.016 0.044
Salyan 0.021 0.007 0.035 0.016 0.046
Doti 0.020 0.006 0.034 0.015 0.040
Terhathum 0.020 0.007 0.034 0.015 0.042
Kaski 0.020 0.006 0.032 0.015 0.041
Okhaldhunga 0.020 0.006 0.033 0.015 0.043
Arghakhanchi 0.019 0.006 0.032 0.015 0.040
Gulmi 0.018 0.004 0.031 0.012 0.039
Panchthar 0.016 0.002 0.028 0.010 0.034
Pyuthan 0.015 0.001 0.029 0.011 0.040
Dailekh 0.014 0.001 0.028 0.009 0.036
Nuwakot 0.014 0.001 0.026 0.009 0.036
Syangja 0.010 -0.004 0.024 0.005 0.032
Ilam 0.010 -0.004 0.022 0.005 0.035
Bhaktapur -0.002 -0.015 0.011 -0.006 0.021
Kathmandu -0.018 -0.032 -0.004 -0.023 0.005

Note : Theta2 represents time-varying district level heterogeneity

97

Table 9

Bootstrap Confidence Intervals for Technical Efficiency Esimates (Wlthout Literacy) :

90% Level - Hill

Technical Efliciency Estimates for the First Period (1981)

 

Percentile BCA

District Efficiency (T=1981) LB UB LB UB

Kathmandu 1.000 0.864 1.000 0.931 1.000
Syangja 0.889 0.704 1.000 0.800 1 .000
Nuwakot 0.769 0.669 0.845 0.729 0.912
Makwanpur 0.716 0.619 0.789 0.677 0.855
Ilam 0.655 0.566 0.730 0.622 0.773
Kaski 0.637 0.541 0.712 0.602 0.763
Gorkha 0.584 0.495 0.651 0.551 0.705
Kavre 0.581 0.494 0.661 0.548 0.730
Panchthar 0.556 0.482 0.624 0.525 0.670
Dhading 0.545 0.473 0.606 0.520 0.651
Tanahu 0.544 0.469 0.606 0.515 0.663
Bhaktapur 0.533 0.371 0.681 0.477 0.760
Sindhuli 0.532 0.461 0.592 0.504 0.651
Bhojpur 0.509 0.438 0.564 0.486 0.615
Dhankuta 0.502 0.431 0.556 0.475 0.607
Surkhet 0.481 0.414 0.535 0.457 0.575
Salyan 0.461 0.393 0.511 0.438 0.549
Gulmi 0.448 0.387 0.496 0.425 0.534
Palpa 0.436 0.378 0.484 0.413 0.518
Khotang 0.424 0.367 0.476 0.401 0.511
Udayapur 0.405 0.341 0.458 0.383 0.495
Terhathum 0.384 0.316 0.437 0.359 0.471
Lamjung 0.380 0.320 0.431 0.357 0.468
Lalitpur 0.368 0.269 0.457 0.334 0.498
Rukum 0.352 0.302 0.393 0.333 0.430
Ramechhap 0.350 0.293 0.396 0.331 0.429
Parbat 0.342 0.290 0.385 0.322 0.413
Baglung 0.341 0.281 0.389 0.320 0.431
Pyuthan 0.31 1 0.255 0.360 0.292 0.392
Rolpa 0.310 0.249 0.360 0.291 0.384
Arghakhanchi 0.306 0.245 0.352 0.283 0.385
Okhaidhunga 0.300 0.228 0.357 0.276 0.394
Doti 0.276 0.222 0.321 0.257 0.349
Dailekh 0.268 0.211 0.314 0.247 0.346
Dadeldhura 0.233 0.184 0.273 0.217 0.297
Baitadi 0.210 0.164 0.251 0.193 0.273
Achham 0.191 0.146 0.233 0.174 0.258
Jajarkot 0.171 0.130 0.208 0.157 0.227
Myagdi 0.164 0.118 0.208 0.145 0.228

Note: Eiliciency estimates are expressed 'relative' to the most elliclent dstrict in the sample

(which has a value of 1)

98

Table 10

Bootstrap Confidence intervals for Technical Efficiency Eshnates (Wlthout Literacy)

90% Level - Hil
Technical Elliciency Estirnetes for the Last Period (1991)

 

Percentile

District analog (T81991) L8 as LB ua

Kavre 1 .000 1 .000 1 .000 1 .000 1.000
Syenas 0.708 0.602 0.844 0.661 0.892
W 0.685 0.61 1 0.767 0.653 0.816
Nuwakot 0.633 0.565 0.717 0.606 0.748
Kathrnmdu 0.601 0.500 0.730 0.554 0.813
Gorkha 0.577 0.508 0.650 0.549 0.685
Kaeld 0.557 0.494 0.628 0.532 0.669
llarn 0.519 0.447 0.594 0.487 0.633
Dhading 0.516 0.456 0.589 0.489 0.629
Tanahu 0.514 0.452 0.583 0.488 0.626
Sindiuli 0.500 0.442 0.575 0.473 0.614
Panchthar 0.466 0.405 0.533 0.440 0.566
Surirhet 0.466 0.41 1 0.529 0.442 0.555
Dhanlnna 0.460 0.401 0.527 0.435 0.572
Bhopur 0.453 0.400 0.520 0.431 0.567
Khotmg 0.439 0.384 0.499 0.414 0.538
Palpa 0.415 0.362 0.479 0.393 0.520
Salyan 0.407 0.358 0.462 0.386 0.489
Gum 0.383 0.334 0.443 0.364 0.469
Lamjung 0.382 0.321 0.453 0.356 0.486
Rukum 0.378 0.327 0.437 0.357 0.472
Bhaktapur 0.375 0.261 0.534 0.330 0.639
Udayapur 0.371 0.312 0.439 0.345 0.485
Lalitpur 0.364 0.271 0.487 0.323 0.555
Terhathum 0.336 0.280 0.408 0.310 0.455
Parbat 0.323 0.274 0.382 0.299 0.407
Baglung 0.316 0.266 0.376 0.297 0.431
Ramechhap 0.310 0.262 0.369 0.291 0.415
Rolpa 0.285 0.235 0.350 0.262 0.385
Arghekmnchi 0.266 0.221 0.324 0.245 0.352
Okhaidhmga 0.262 0.204 0.335 0.238 0.376
Pymhan 0.260 0.219 0.310 0.241 0.350
Doti 0.242 0.195 0.303 0.221 0.334
Achham 0.227 0.179 0.288 0.205 0.321
Dailekh 0.222 0.180 0.274 0.204 0.302
Dadeldhura 0.220 0.175 0.280 0.202 0.319
Ballad 0.218 0.173 0.276 0.197 0.31 1
Jaiarkot 0.182 0.141 0.231 0.165 0.268
Myagdi 0.172 0.128 0.230 0.151 0.264

Note: Efficiency estimates are expressed 'relative' to the most efficient dshict in the sample

(which has a value of 1)

Table 1 1
Parameter Estimates of the Production Function (With Literacy) - Terai
Dependent Vairable :Netural Log of Agriculturd Oumuf (Mean Value : 6.63)

Explanatory Variables (besides net school enrollment rates) are in Natural Logs

 

Independent Vairables Wlthin-Estimates
land 2.300
(0.295)‘
labor-male -0.523
(0.45)
labor-female 0.420
(0.31)
fertilizer 0.0130
(0.01 5)
livestock 0.042
(0.040)
rainfall 0.190
(0.034)‘
literacy-male 0.01 1
(0.017)
literacy-female 0.024
(0.016)
Observations 220
# of Districts 20
# of Time Periods 11

robust standard errors in parentheses;'significantatthe 196de

Note : robust errors based on White's ”Sandwich Method“ are heteroskedasticity/sedat-correlation
consistent estimates

100

Table 12

Paraneter Estimates (With Literacy) of tine-invariant and tine-varying heterogeneity

 

 

 

Terai
Tine-invariant heterogeneity RTC“

District Estimate SE " t-stet Estimate SE " t-stet
Saptari 5.084 1.179 5.160 0.047 0.021 2.206
Mahotari 5.466 1.002 5.457 0.025 0.025 1.012
Ram 5.699 1.0080 5.654 0.016 0.028 0.579
Bare 5.454 1 .0930 -4.990 0.01 1 0.024 0.469
Siraha 5.665 1.070 5.295 0.010 0.028 0.371
Kapibastu 5.761 1.050 5.489 0.004 0.034 0.122
Dhanusa 5.518 1.0740 5.137 0.004 0.024 0.165
Sarhhi 5.503 1 .0250 5.368 -0.002 0.024 -0.072
Banke 5.2590 0.989 5.318 -0.003 0.031 '0.065
Rupandehi -6.004 1 .215 -4.940 -0.005 0.022 -0.242
Kailali 5.792 1 .1 27 5.1 40 «0.007 0.022 5.309
Jhapa 5.574 1 .2900 5.096 -0.010 0.032 -0.330
Parsa 5.0900 1.065 -4.781 -0.0130 0.027 -0.487
Chitawan 5.943 1 .234 -4.818 -0.033 omo -1 .093
Morang 5.106 1.21 1 5.044 -0.033 0.034 -0.997
Kanchanpur 5.4650 1 .069 5.1 13 -0.034 0.039 -0.872
Nawalparasi 5.194 1.069 -4.861 -0.034 0.030 -1 .155
Bardiya -4.870 0.913 5.334 -0.038 0.035 -1.065
Dang 5.616 1.095 5.130 -0.047 0.036 -1.301
Sunsari 5.453 1.192 -4.573 -0.089 0.041 -2.177

' : Standard-Error

" : Rate of Technical Change (time-varying heterogeniety)

101

Terai - With Literacy

Table 12.2

 

District GRYA RTC
Mahotari 0.058 0.025
Sarlahi 0.058 -0.002
Jhapa 0.057 -0.010
Bara 0.054 0.01 1
Saptari 0.047 0.047
Dhanusa 0.042 0.004
Bardiya 0.042 -0.038
Siraha 0.039 0.010
Kailali 0.038 -0.007
Morang 0.036 -0.033
Kanchanpur 0.031 -0.034
Rautahat 0.030 0.016
Nawalparasi 0.030 -0.034
Sunsari 0.026 -0.089
Dang 0.026 -0.047
Chitawan 0.022 -0.0330
Rupandehi 0.020 -0.005
Banke 0.013 -0.003
Kapiibastu 0.005 0.004
Parsa 0.001 -0.0130

" GRY : growth rate of output

102

Table 13

Bootstrap Confidence Intervals for RTC Eshnates (With Literacy) : 90% Level - Terai

 

Percentile

District RTC Estimate LB UB LB UB

Saptari 0.0472 0.0166 0.0791 0.0349 0.0950
Mahotari 0.0251 5.0119 0.0578 0.0115 0.0779
Rm 0.0161 5.0250 0.0550 0.0006 0.0779
Bara 0.0114 5.0244 0.0454 5.0024 0.0699
Siraha 0.0102 5.0284 0.0498 5.0053 0.0687
Kapibestu 0.0041 5.0457 0.0502 5.0163 0.0624
Dhenma 0.0039 5.0303 0.0362 5.0091 0.0563
Sarlahi 5.0017 5.0354 0.0296 5.0144 0.0513
Banke 5.0026 5.0480 0.0389 5.0205 0.06%
Rwandehi 5.0053 5.0382 0.0247 5.0181 0.0436
Kailali 5.0069 5.0417 0.0227 5.0200 0.0445
Jhapa 5.0104 5.0576 0.0327 5.0287 0.0585
Parsa 5.0130 5.0525 0.0246 5.0291 0.0507
Chitawan 5.0330 5.0787 0.0084 5.0505 0.0328
Morang 5.0334 5.0833 0.0133 5.0527 0.0401
Kanchanpur 5.0342 5.0918 0.0195 5.0565 0.0601
Nawalparasi 5.0343 5.0798 0.0060 5.0517 0.0335
Bardiya 5.0375 5.0889 0.0105 5.0578 03393
Dang 5.0471 5.1023 0.0008 5.0687 0.0344
Sunsari 5.0886 5.1462 5.0330 5.1087 0.0188

103

Table 14

Bootstrap Confidence htervals for Technical Efficiency Eshnates (With Lleracy) :
90% Level - Terai

Technical Eiliclency Estimates for the First Period (1981)

 

Percentile BCA

District Efficiency (T =1981) LB UB LB UB

Bardiya 1 0.8890 1 0.9330 1

Parsa 0.8220 0.6022 1 0.7212 1

Nawalparasi 0.7249 0.4934 0.9769 0.6259 1

Banks 0.7015 0.5602 0.8192 0.6457 0.8764
Mahotari 0.5866 0.4786 0.6776 0.5493 0.7878
Bare 0.5853 0.4135 0.7458 0.5207 0.8813
Kanchanpur 0.5531 0.3799 0.7167 0.4775 0.7945
Sarlahi 0.5503 0.4368 0.6523 0.5077 0.7192
Dhanusa 0.5453 0.3924 0.6957 0.4844 0.8305
Sunsari 0.5304 0.2917 0.8543 0.4334 1

Siriia 0.4736 0.3516 0.5920 0.4283 0.7188
Dang 0.4693 0.3210 0.6132 0.4034 0.6772
Rautahat 0.4601 0.3734 0.5435 0.4264 0.6194
Kaplhestu 0.4274 0.3336 0.5332 0.3898 0.6316
Kaila! 0.4099 0.2663 0.5735 0.3526 0.71 15
Ohm 0.3434 0.1753 0.5949 0.2756 0.7826
Rupandehi 0.3321 0.1860 0.5219 0.2708 0.6179
Saptul 0.3231 0.1912 0.4830 0.2694 0.5914
Morang 0.2915 0.1641 0.4426 0.2397 0.5568
Jhapa 0.1869 0.0899 0.3272 0.1455 0.4333

Note: Efficiency estimates are expressed 'relative' to the most elllcient dstn‘ct in the sample

(which has a value of 1)

104

Table 15

Bootstrap Confidence Intervals for Technical Efficiency Esimates (With Literacy)
: 90% Level - Terai

Technical Efficiency Estimates for the Last Period (1991)

 

Percentile BCA

District Efficiency (T =1 991) LB UB LB UB
Mahotari 1 0.8174 1 0.8958 1
Parsa 0.9568 0.754 1 0.8809 1
Bardiya 0.9114 0.7584 0.9842 0.8606 1
Banke 0.9066 0.7217 1 0.8347 1
Bara 0.8701 0.6999 0.9907 0.8039 1
Dhanusa 0.7522 0.6172 0.8223 0.7043 0.8779
Sarlahi 0.7178 0.5891 0.7935 0.6758 0.8749
Rautahat 0.7165 0.5523 0.8134 0.6628 0.8894
Siraha 0.6957 0.5614 0.7753 0.6503 0.8882
Saptari 0.6866 0.4808 0.8617 0.6054 0.9803
Nawalparasi 0.6820 0.4825 0.8361 0.6053 0.9486
Kapiibastu 0.5906 0.4886 0.6483 0.5545 0.6876
Kanchanpur 0.5209 0.3254 0.7144 0.4460 0.8876
Kailali 0.5073 0.3939 0.5851 0.4620 0.6503
Rupandehi 0.4176 0.2822 0.5354 0.3633 0.6520
Dang 0.3885 0.2595 0.5137 0.3345 0.6183
Chitawan 0.3274 0.1726 0.5257 0.2681 0.7193
Sunsari 0.2900 0.1519 0.4895 0.2426 0.8768
Morang 0.2769 0.1554 0.4200 0.2268 0.5681
Jhapa 0.2233 0.1097 0.3925 0.1761 0.5859

Note: Efficiency estimates are expressed ”relative" to the most efficient dstn’ct in the sample

(Which has a value of 1)

105

Table 16
Parameter Estimates of the Production Function (With Literacy) - Hill 9
Dependent Vairable : Natural Log of Agricultural quaut (Mean Value : 5.507)

ExtpfanatoryVarlablesmesidesliteracyrdesnrehNaturalLogs

 

Independent Vairables Wlthin-Estimates
land 0.180
(0.16)
Iabonmale 0.46
(0.21 0)“
labor-female ' 0.335
(0.234)
fertilizer 0.024
(0.01 1)”
livestock 0.069
(0.063)
rainfall 0.063
(0.029)“
literacy-male 0.049
(0.032)
literacy-female 0.039
(0.023)“
Observations 429
it of Districts 39
# of Time Periods 11

robust standard errors in parentheses ; “ significant at the 10% level

Note : robust errors based on White's “Sandwich Method" are hetemskedasticitylsend—conelation
consistent estimates

106

Table 17

Parameter Estimates (With Literacy) of time-invariant and time-varying heterogeneity

District

Kavre
Lamjung
Achham
Makwanpur
Gulmi
Jajarkot
Parbat
Sindhuli
Dhankuta
Nuwakot
Ramechhap
Dailekh
Rukum
Khotang
Baglung
Myagdi
Palpa
Dhading
Rolpa
Bhojpur
Terhathum
Ilam

Kaski
Baitadi
Panchthar
Udayapur
Syangja
Surkhet
Arghakhanchi
Salyan
Pyuthan
Kathmandu
Lalitpur
Dadeldhura
Gorkha
Okhaidhunga
Doti
Tanahu
Bhaktapur

' : Standard-Error

 

 

Terai
Tine-lnvarient heterogeneity
Estimate SE " t-stat
1.4665 0.9414 1.558
0.5016 0.9143 0.549
1 .4593 0.7083 2.060
1.7015 0.9325 1.825
0.5797 0.9664 0.600
1.6405 0.7069 2.321
0.3390 0.9370 0.362
1.7857 0.8983 1.988
0.8099 0.9658 0.839
2.1278 0.9140 2.328
1.4965 0.8235 1.817
1.3717 0.7801 1.758
1.8425 0.8274 2.227
1.2824 0.9169 1.399
1.1304 0.8482 1.333
0.9868 0.7051 1 .400
1.0057 0.9338 1.077
2.1769 0.8608 2.529
1 .8440 0.7660 2.407
1 .4651 0.9033 1 .622
0.5114 0.9098 0.562
1.1814 0.9336 1.265
0.8343 1.0091 0.827
1.1048 0.7591 1.455
1.3999 0.9269 1.510
1.6884 0.8410 2.008
1.1205 1.1024 1.016
1.4472 0.9301 1.556
1.1987 0.8057 1.488
2.0394 0.8906 2.290
1 .6426 0.7929 2.072
0.8265 0.9776 0.845
0.8795 0.7728 1.138
1.1854 0.7787 1.522
1 .8974 0.9444 2.009
1.6453 0.7436 2.213
1 .9936 0.7463 2.671
1 .3845 0.9453 1 .465
1.4960 0.7426 2.015

A : Rate of Technical Change (time-varying heterogeniety)

107

che
Estimate SE ' t-stat
5.0352 0.0189 -1 .862
5.0424 0.0126 5.365
5.0642- 0.0235 -2.732
5.0664 0.0144 4.611
5.0674 0.0133 5.068
5.0675 0.0185 5.649
5.0730 0.0160 4.563
5.0755 0.0164 4.604
5.0769 0.0148 5.196
5.0778 0.0143 5.441
5.0811 0.0171 4.743
5.0839 0.0173 4.850
5.0863 0.0206 4.189
5.0888 0.0180 4.933
5.090 0.0164 5.488
5.0901 0.0182 4.951
5.0906 0.0165 5.491
5.0931 0.0181 5.144
5.0952 0.0219 4.347
5.0985 0.0170 5.794
5.0998 0.0167 5.976
5.1001 0.0157 5.376
5.1009 0.0169 5.970
5.1042 0.0239 4.360
5.1052 0.0167 5.299
5.1052 0.0190 5.537
5.1057 0.0173 5.110
5.1066 0.0192 5.552
5.1087 0.0176 5.176
5.1108 0.0213 5.202
5.1131 0.0195 5.800
5.1137 0.0164 5.933
5.1149 0.0194 5.923
5.1169 0.0267 4.378
5.1205 0.0203 5.936
5.1220 0.0197 5.193
5.1240 0.0254 4.882
5.1252 0.0201 5.229
5.1452 0.0191 -7.602

Table 17.2

Hill - With Literacy

 

District GRY“ RTC
Kavre 0.056 5.035
Achham 0.038 5.064
Rukum 0.033 5.086
Lalitpur 0.032 5.115
Jajarkot 0.031 5.068
Lamjung 0.030 5.042
Khotang 0.030 5.089
Gorkha 0.028 5.121
Baitadi 0.027 5.104
Makwanpur 0.025 5.066
Myagdi 0.025 5.090
Dadeldhura 0.025 5.117
Parbat 0.025 5.0730
Dhading 0.024 5.093 .
Baglung 0.023 5.0900
Palpa 0.023 5.091
Tanahu 0.023 5.125
Rolpa 0.023 5.095
Surkhet 0.022 5.107
Sindhuli 0.021 5.076
Dhankuta 0.021 5.077
Udayapur 0.020 5.105
Ramechhap 0.019 5.081
Bhojpur 0.018 5.099
Terhathum 0.016 5.100
Arghakhanchi 0.016 5.109
Kaski 0.016 5.101
Nuwakot 0.015 5.078
Gulmi 0.015 5.067
Salyan 0.014 5.111
Okhaidhunga 0.014 5.1220
Panchthar 0.011 5.105
Pyuthan 0.010 5.113
Syangja 0.010 5.106
Dailekh 0.009 5.084
Doti 0.008 5.1240
Ilam 0.007 5.100
Bhaktapur 5.004 5.145
Kathmandu 5.021 5.114

GRY“ : growth rate of output

108

Table 18

Bootstrap Confidence intervals for RTC Esimates (With Literacy) :
90% Level - Hill

 

Percentile BCA

District RTC Estimate LB UB LB UB

Kavre 5.0352 5.0638 5.0061 5.0470 0.0137
Lamjung 5.0424 5.0618 5.0244 5.0496 5.0132
Achham 5.0642 5.0991 5.0282 5.0797 5.0124
Makwanpur 5.0664 5.0869 5.0439 5.0755 5.0323
Gulmi 5.0674 5.0886 5.0476 5.0752 5.0351
Jajarkot 5.0675 5.0946 5.0388 5.0784 5.021 1
Parbat 5.0730 5.0977 5.0505 5.0822 5.0350
Sindhuli 5.0755 5.0997 5.0506 5.0858 5.0361
Dhankuta 5.0769 5.0993 5.0553 5.0855 5.0364
Nuwakot 5.0778 5.0995 5.0580 5.0869 5.0445
Ramechhap 5.0811 5.1055 5.0552 5.0921 5.0396
Dailekh 5.0839 5.1 101 5.0577 5.0955 5.0411
Rukum 5.0863 5.1161 5.0559 5.1002 5.0358
Khotang 5.0888 5.1151 5.0625 5.1003 5.0437
Baglung 5.0900 5.1151 5.0672 5.0997 5.0515
Myagdi 5.0901 5.1169 5.0629 5.1015 5.0473
Palpa 5.0906 5.1 156 5.0669 5.1008 5.0488
Dhading 5.0931 5.1199 5.0666 5.1053 5.0484
Rolpa 5.0952 5.1288 5.0620 5.1098 5.0449
Bhojpur 5.0985 5.1233 5.0726 5.1083 5.0557
Terhathum 5.0998 5.1246 5.0746 5.1102 5.0613
Ilam 5.1001 5.1225 5.0758 5.1099 5.0657
Kaski 5.1009 5.1260 5.0755 5.1114 5.0593
Baitadi 5.1042 5.1394 5.0683 5.1207 5.0491
Panchthar 5.1052 5.1299 5.0806 5.1156 5.0631
Udayapur 5.1052 5.1331 5.0758 5.1177 5.0603
Syangja 5.1057 5.1325 5.0806 5.1158 5.0635
Surkhet 5.1066 5.1338 5.0770 5.1178 5.0597
Arghakhanchi 5.1087 5.1345 5.0818 5.1199 5.0674
Salyan 5.1108 5.1414 5.0784 5.1236 5.0655
Pyuthan 5.1131 5.1418 5.0837 5.1262 5.0658
Kathmandu 5.1137 5.1381 5.0890 5.1228 5.0732
Lalitpur 5.1149 5.1431 5.0855 5.1261 5.0664
Dadeldhura 5.1169 5.1555 5.0780 5.1336 5.0556
Gorkha 5.1205 5.1507 5.0889 5.1327 5.0684
Okhaidhunga 5.1220 5.1505 5.0915 5.1337 5.0735
Doti 5.1240 5.1606 5.0858 5.1385 5.0616
Tanahu 5.1252 5.1547 5.0955 5.1366 5.0755
Bhaktapur 5.1452 5.1736 5.1170 5.1572 5.0959

Note: Theta2 represents time-varying district level heterogeneity

109

Table 19

Bootstrap Confidence intervals for Technical Efficiency Esimates (With Literacy) :
90% Level - Hill

Technical Efficiency Estimates for the First Period (1981)

 

Percentile BCA
District Efficiency (T =1 981) LB UB LB UB
Dhading 1 0.8857 1 0.9501 1
Nuwakot 0.9667 0.8295 1 0.9107 1
Salyan 0.8563 0.7463 0.9307 0.81 10 0.9842
Doti 0.8072 0.6226 0.9915 0.7322 1
Gorkha 0.7357 0.61 15 0.8220 0.6922 0.8809
Rukum 0.7206 0.6212 0.7993 0.6821 0.8415
Rolpa 0.7154 0.5643 0.8409 0.6577 0.9247
Sindhuli 0.6883 0.5997 0.7454 0.6569 0.7898
Makwanpur 0.6385 0.5294 0.7156 0.6001 0.7753
Udayapur 0.6062 0.5239 0.6703 0.5716 0.7186
Jajarkot 0.6001 0.4413 0.7696 0.5355 0.8738 -
Pyuthan 0.5745 0.4728 0.6678 0.5379 0.7209
Okhaidhunga 0.5709 0.4395 0.7102 0.5169 0.8205
Kavre 0.5208 0.421 5 0.6015 0.4861 0.6629
Ramechhap 0.5126 0.4262 0.5801 0.4802 0.6577
Achham 0.5022 0.3722 0.6409 0.4472 0.7273
Bhojpur 0.4881 0.4067 0.5534 0.4559 0.5893
Bhaktapur 0.4805 0.3190 0.6836 0.4155 0.8588
Surkhet 0.4756 0.3903 0.5346 0.4437 0.5808
Panchthar 0.4543 0.3564 0.5450 0.4142 0.5970
Dailekh 0.4511 0.3598 0.5406 0.4147 0.5940
Tanahu 0.4385 0.3504 0.5082 0.4062 0.5568
Khotang 0.4106 0.3385 0.4662 0.3832 0.5009
Arghakhanchi 0.3702 0.2963 0.4385 0.3407 0.4855
Ilam 0.3670 0.2833 0.4453 0.3306 0.5262
Dadeldhura 0.3623 0.2899 0.4305 0.3328 0.4880
Baglung 0.3523 0.2806 0.4167 0.3253 0.4816
Syangja 0.3434 0.2167 0.4853 0.2941 0.5964
Baitadi 0.3385 0.2596 0.4198 0.3035 0.4597
Palpa 0.3108 0.2492 0.3613 0.2850 0.4039
Myagdi 0.3051 0.2173 0.4143 0.2656 0.4756
Lalitpur 0.2674 0.1869 0.3636 0.2309 0.4518
Dhankuta 0.2591 0.1901 0.3323 0.2296 0.3934
Kaski 0.2591 0.1873 0.3315 0.2272 0.3922
Kathmandu 0.2539 0.1739 0.3448 0.2189 0.4339
Gulmi 0.2078 0.1506 0.2706 0.1841 0.3210
Lamjung 0.1970 0.1466 0.2499 0.1769 0.3020
Terhathum 0.1878 0.1320 0.2533 0.1619 0.3181
Parbat 0.1624 0.1137 0.2169 0.1430 0.2835

Note: Eliiciency estimates are expressed 'relative' to the most eliiclent dstrict in the sample

(which has a value of 1)

110

Table 20

Bootstrap Confidence Intervals for Technical Efficiency Esimates (With Literacy) :
90% Level - Hill

Technical Efficiency Estimates for the Last Period (1991)

 

Percentile BCA

District Efficiency (T =1 99 1) LB UB LB UB
Nuwakot 1 0.9678 1 0.9669 1
Dhading 0.8876 0.7897 0.9928 0.8428 1
Kavre 0.8252 0.6911 0.9655 0.7676 1
Makwanpur 0.7406 0.6501 0.8321 0.7018 0.8809
Sindhuli 0.7287 0.6450 0.8150 0.6907 0.8559
Jajarkot 0.6887 0.5069 0.9565 0.6057 1
Rukum 0.6846 0.5809 0.7910 0.6449 0.8410
Salyan 0.6372 0.5618 0.7176 0.6067 0.7653
Rolpa 0.6219 0.4945 0.7783 0.5625 0.8450
Achham 0.5953 0.4306 0.8256 0.5062 0.9356
Doti 0.5263 0.4006 0.6849 0.4654 0.7479
Ramechhap 0.5132 0.4210 0.6240 0.4772 0.6739
Gorkha 0.4967 0.4225 0.5743 0.4624 0.6064
Udayapur 0.4769 0.3999 0.5648 0.4402 0.6068
Dailekh 0.4390 0.3358 0.5638 0.3904 0.6119
Pyuthan 0.4177 0.3424 0.5050 0.3819 0.5461
Bhojpur 0.4107 0.3509 0.4775 0.3852 0.5160
Khotang 0.3805 0.3260 0.4423 0.3546 0.4723
Okhaidhunga 0.3796 0.2867 0.4986 0.3348 0.5522
Surkhet 0.369 0.3109 0.4308 0.3428 0.4581
Panchthar 0.3575 0.2961 0.4261 0.3275 0.4563
Baglung 0.3226 0.2664 0.3833 0.2970 0.4193
Ilam 0.3038 0.2459 0.3748 0.2764 0.4152
Lamjung 0.2904 0.2350 0.3541 0.2662 0.3977
Palpa 0.2830 0.2340 0.3403 0.2606 0.3731
Tanahu 0.2824 0.2253 0.3473 0.2583 0.3719
Arghakhanchi 0.2811 0.2218 0.3473 0.2514 0.3809
Myagdi 0.2793 0.1927 0.3968 0.2379 0.4565
Dhankuta 0.2705 0.2182 0.3327 0.2476 0.3694
Baitadi 0.2691 0.1915 0.3711 0.2324 0.4280
Syangja 0.2688 0.1885 0.3700 0.2337 0.4183
Dadeldhura 0.2536 0.1786 0.3543 0.2210 0.4249
Bhaktapur 0.2535 0.1573 0.3978 0.2062 0.4917
Gulmi 0.2387 0.1953 0.2881 0.2206 0.3153
Kaski 0.2128 0.1578 0.2765 0.1869 0.3125
Lalitpur 0.1908 0.1285 0.2775 0.1592 0.3314
Kathmandu 0.1835 0.1226 0.2680 0.1556 0.3284
Parbat 0.1763 0.1369 0.2217 0.1598 0.2484
Terhathum 0.1560 0.1 157 0.2066 0.1372 0.2395

Note: Efficiency estimates are expressed 'relative' to the most efficient (9st in the sample

(which has a value of 1)

111

Table ARC1

Districts with local agricultural research center
and year of establishment (Terai)

Year Agricultural Research Center Efficiency (T=1981)

 

District was established (without literacy)
Parsa 1959 1.000
Bardiya " 0.943
Nawalparasi " 0.877
Sunsari 1960 0.768
Banke 1960 0.745
Kanchanpur ' 0.676
Bara 1972 0.627
Mahotari 1965 0.597
Dhanusa 1961 0.583
Chitawan 1972 0.581
Sariahi 1976 0.555
Dang * 0.551
Siraha " 0.502
Rupandehi 1972 0.471
Morang * 0.462
Kailali " 0.456
Rautahat "' 0.441
Saptari * 0.426
Kapiibastu * 0.410
Jhapa ‘ 0.359

Note: " Indicates that district does not have a local agricultural research center

112

Table ARC2

Districts with local agricultural research center
and year of establishment (Hill)

Year Agricultural Research Center Efficiency (T =1981)

 

District was established (without literacy)
Kathmandu 1955 1.000
Syangja * 0.889
Nuwakot 1971 0.769
Makwanpur * 0.716
Ilam * 0.655
Kaski 1960 0.637
Gorkha ' 0.584
Kavre 1961 0.581
Panchthar * 0.556
Dhading * 0.545
Tanahu 1977 0.544
Bhaktapur * 0.533
Sindhuli ' 0.532
Bhojpur " 0.509
Dhankuta 1961 0.502
Surkhet ' 0.481
Salyan 1991 0.461
Gulmi * 0.448
Palpa " 0.436
Khotang " 0.424
Udayapur * 0.405
Terhathum * 0.384
Lamjung " 0.380
Lalitpur 1965 0.368
Rukum * 0.352
Ramechhap ' 0.350
Parbat * 0.342
Baglung " 0.341
Pyuthan * 0.31 1
Rolpa ‘ 0.310
Arghakhanchi * 0.306
Okhaidhunga * 0.300
Doti 1962 0.276
Dailekh 1968 0.268
Dadeldhura " 0.233
Baitadi " 0.210
Achham ’ 0.191
Jajarkot “ 0.171
Myagdi * 0.164

Note: " Indicates that district does not have a local agricultural research center

113

Table 21

 

 

Without Literacy With Literacy
GRY ARTC ATE81 ATE91 ARTC ATE81 ATE91
TERAI 3.37% 3% 60% 70% 5.47% 47% 57%
HILL 2% 2.50% 45% 41% 5.40% 46% 40%

where.

GRY : growth rate of output

ARTC : average rate of district level technical change

ATE81 : average level of distirct level technical efficiency in 1981

ATE81 : average level of distirct level technical efficiency in 1991

114

Technical Efficiency 1981 (Terai)

Figure 1

 

1.200

 

11!!)

 

0.800

O.

 

0.600

 

 

...,?

 

0.200

 

 

0.000

f I f

5 10 15 20

Number of years since an agricultural research center has been

established in that district

115

Technical Efficiency 1981 (Hill)

Figure 2

 

1.200

 

1.000

 

 

 

 

 

 

 

0.300
e
it
e
0.600
e
e
0
0.400
e
. e
0.200 :
0.000 . . r . T
o 5 1o 15 20 25

Number of years since an agricultural research center has been
established in that district

116

BIBLIOGRAPHY

Agricultural Statistics of Nepal, Revised Cropped Area Series, 1974/75-1991/92;
National Planning Commission 1994.

Abramovitz, M.; Resources and Output Trends in the United States since 1870; American
Economic Review; Papers and Proceedings, May 1956; pp. 5-23.

Aghion, P., and Howitt, P., 1992; A Model of Growth through Creative Destruction;
Econometrica (March, 56): 323-351.

Aigner, D., Lovell, OK, and Schmidt, P., 1977; Formulation and Estimation of Stochastic
Frontier Production Function Models; Journal of Econometrics 6121-38.

All, M., and Byerlee, D., 1988; Productivity Growth and Resource Degradation in
Pakistan’s Punjab: A Decomposition Analysis; unpublished manuscript.

Azhar, R., 1991; Education and technical efficiency during the green revolution in
Pakistan; Economic Development and Cultural Change; (3 9) : 651-665.

Binswanger, H.P., Khandker, S., and Rosenzweig, M.R., 1993; How Infrastructure and
financial institutions affect agricultural output and investments in India; Journal of
Development Economics 41: 337-366.

Bravo-Ureta, BE. and Evenson, RE, 1994; Efficiency in agricultural production: the
case of the peasant farmers in eastern Paraguay; Agricultural Economics 10: 27 - 38.

Central Bureau of Statistics, Nepal; 1981.

Central Bureau of Statistics, Nepal; 1996.

Capalbo, Susan, and Antle, J.H., 1988; Agricultural Productivity: Measurement and
Explanation; Resources for the Future; Washington , D.C.

Cassman, K.G., and Pingali, PL, 1995; Extrapolating Trends from Long-Term
Experiments to Farmers’ Fields: The Case of Irrigated Rice Systems in Asia; In:
Agricultural Sustainability: Economic, Environmental and Statistical Considerations.

117

Edited by Vic Barnett, Roger Payne and Roy Steiner; John Wiley & Sons Ltd., England.
pp. 63-84.

Comwell, C., Schmidt, P., Sickles, RC, 1990; Production Frontiers with Cross-Sectional
and Time-Series Variation in Efficiency Levles; Journal of Econometrics, Vol(46), 185-
200.

Cotlear, D., 1986; Farmer Education and farm efficiency in Peru : the role of schooling,
extension services and migration. Discussion Paper EDT49; Education and Training
Department, World Bank, Washington, DC.

Champak, P. Pokharel, I993; Nepalese Agriculture: Performance and Future Direction; in
Future of Nepalese Economy, Eds,. Madan K. Dahal; NEFAS/FES.

Datt, G., and M. Ravallion, 1998; “Why Have Some Indian States Done Better than
Others at Reducing Rural Poverty?”; Economica (65), 17—3 8.

Denny, M., and Fuss, M., "A general approach for intertemporal and interspatial
productivity comparisions", Journal of Econometrics, Vol(23), 315-330, 1983.

Efron, B., and RI. Tibshirani, 1993; An Introduction to the Bootstrap; Chapman and Hall,
New York.

Evenson, R., and Kislev, 1971; Agricultural Research and Productivity; Yale University
Press; New Haven.

----, and Rosegrant, Mark W., 1993; ‘Determinants of productivity growth in Asian
agriculture past and firture’; Paper presented at the 1993 AAEA International Pre-
Conference on “Post-Green Revolution Agricultural Development Strategies in the Third
World: What Next? ”.

Grossman, G. M., and Helpman, E., 1991; Innovation and Grth in the Global
Economy; Cambridge; MIT Press.

Green, W.H., 1993 ; in ‘Ihe Measurement of Productive Efficiency .' Techniques and
Applications’ ; edited by: Fried, O.H., Lovell, Knox CA, and Schmidt, Shelton S. ;
Oxford University Press.

118

---, 1993b; Econometric Analysis, Second Edition; Macmillan Publishing Company, New
York.

Grilliches, Zvi, 1963; The Sources of Measured Productivity Growth: United States
Agriculture, 1940-1960; Journal of Political Economy; (August, vol. LXXI): 331-346.

---, 1964; Research Expenditures, Education, and the Aggregate Production Function;
American Economic Review; (December, no.6).

---, Explanation of Productivity Change; Review of Economic Studies; (July, no.99).

Hall, P., W. Hardle, and L. Simar, 1993; On the Inconsistency of Bootstrap Distribution
Estimators; Computational Statistics and Data Analysis (16): 11-18.

----, W. Hardle, and L. Simar, 1993; Iterated Bootstrap with Applications to Frontier
Models; Journal of Productivity Analysis (7): 257-282.

Hsieh, SC, and Ruttan, V.W., 1967; Environmental, Technological, and Institutional
Factors in the Growth of Rice Production: Philipinnes, Thailand, and Taiwan; Food
Research Institute Studies 7(3): 307-341.

Hyami, Y., 1969; Sources of agricultural productivity gap among selected countries;
American Journal of Agricultural Econorrrics (51): 564-567;

----, and Vernon W. Ruttan, 1971; Agricultural Development, An International
Perspective; The John Hopkins Press, Baltimore.

----, and Vernon W. Ruttan, 1985; Agricultural Development, An International
Perspective, Revised and Expanded Edition; The John Hopkins Press, Baltimore.

Jamison, D.T., and L.J. Lau , 1982; Farmer Education and Farm Efficiency; Johns
Hopkins Press, Baltimore.

Jorgenson, D., and Z. Griliches, 1967; The Explanation of Productivity Change; Review
of Economics and Statistics, (34), 249-283.

Jondrow, J., Lovell, C.A.K., Materov, LS. and Schmidt, P., 1982; On the estimation of
technical inefficiency in the stocastic frontier production function model; Journal of
Econometrics 19: 233-238.

119

 

Kalirajan, K.P., 1990; On Measuring Economic Efficiency; Journal of Applied
Econometrics 5: 75-85.

Lau, L., and Pan Yotopoulos, 1989; The Meta-Production Approach to Technological
Change in World Agriculture; Journal of Development Economics (31): 241-269.

----, D. T. Jamison, Shu-Cheng Liu, and Steve Rivkirr, 1993; Education and Economic ~
Growth, Some Cross-Sectional Evidence from Brazil; Journal of Development Economics
(41): 45-70.

Lockheed, M., D. Jamison and L. Lau, 1981; Farmer education and farm efficiency : a
survey; Economic Development and Cultural Change; (29) : 37-76.

Lovell, 1993; The Measurement of Productive Efficiency.

Meeusen, W. And J. Van den Broeck, 1977; Efficiency estimation from Cobb-Douglas
production firnctions with composed error; International Economic Review 18: 435-444.

Mellor, John W., 1995; Agriculture on the Road to Industrialization; Johns Hopkins
University Press, Baltimore.

Mundlak, Y., and Rene Hellinghousen, 1982; The Intercountry Agricultural Production
Function: Another View; American Journal of Agricultural Economics (64), November:
664-672.

Nelson, GO, 1987; Agricultural Price Policy in Nepal; Economic Staff Paper No.35;
Asian Development Bank.

Nepal: Economic Policies For Sustainable Development, 1992; Asian Development
Bank; Philippines.

Nepal, Agricultural Perspective Plan, 1995; Prepared for the National Planning
Commission and Asian Development Bank by Agricultural Projects Services Center and
John Mellor Associates.

Ostrom, E., 1995; In: Annual World Bank Conference on Development Economics; (Eds)
Bruno, M., and Pleskovic, B.; The World Bank; Washington, DC.

120

Park, B.U., and Simar, 1994; Efficient Serrriparametric Estimation in a Stochastic Frontier
Model; Journal of the American Statistical Association (89): 929-936.

Pingali, PL, and Binswanger, HP, 1988. Population Density and Farming Systems: The
Changing Locus of Innovations and Technical Change. World Population Trends and their
Impact on Economic Development pp. 165-186.

Pudasini, S., 1984; ‘Production and Price Responsiveness of Crops in Nepal’;
HMGN/ADC Project Research Paper Series No.25, Agricultural Projects Services Center,
Nepal.

Ruttan, V.W., 1977; The Green Revolution: Seven Generalizations; International
Development Review 19(1): 16-23.

Rice-Wheat Atlas of Nepal; Edited by Huke R., Huke E., and Woodhead T., 1993;1RRI,
CIWY T, NARC.

Rao, C.H.H., 1975; Technological change and distribution in Indian agriculture. New
Delhi: MacMillan Company.

Schmidt, P., and RC. Sickles, 1984; Production Frontier and Panel Data; Journal of
Business and Economic Statistics (2): 367-374;

----, and Y. Kim, 1999; Bootstrap Confidence Intervals for Efficiency Levels in Stochastic
Frontier Models with Panel Data; unpublished manuscript, Department of Econorrrics,
Michigan State University.

Simar, L., 1992; Estimating Efficiencies from Frontier Models with Panel Data: A
Comparison of Parametric, Non-Parametric and Semi-Parametric Methods with
Bootstrapping; Journal of Productivity Analysis (3): 171-203.

Rosegrant, M., and Robert E. Evenson, 1993; Agricultural Productivity Growth in
Pakistan and India: A Comparative Analysis; The Pakistan Development Review (32:4
Winter, Part I): 433-451;

----, and Robert E. Evenson, 1995; Total Factor Productivity and Sources of Long-Term
Grth in Indian Agriculture; EPTD Discussion Paper No.7; IFPRI; Washington, DC.

121

 

Romer, P. M., 1986; Increasing Returns and Long Run Growth; Journal of Political
Economy; (94) Ocober; pp. 1002-1037.

---, 1990; Endogenous Technological Change; Journal of Political Economy; (October,
Part2): 871-8102.

Solow, R. M.; Technical Change and Aggregate Production Function; Review of
Economics and Statistics; August 1957; pp. 312-320.

Shah, B; Total Factor Productivity Growth of Nepalese Agriculture; Unpublished MS.
Thesis; University of New South Wales, Australia; 1973.

Timmer, CR, 1988; The Agricultural Transformation; In: Handbook of Development
Economics, vol. 1; (Eds) Chenery, H., and Srinivasan, T.N.; North Holland Press,
Amsterdam.

Wooldridge, J; 1996; Econometric Analysis of Cross-Section and Panel Data; Unpublished
Manuscript; Department of Economics, Michigan State University; 1996.

122

   

llljllllzllflllflllllllflfllfllflfllJl