WOMEN’S LABOUR CONTRIBUTION TO PRODUCTIVITY AND EFFICIENCY IN

AGRICULTURE: EMPIRICAL EVIDENCE FROM BANGLADESH

Sanzidur Rahman1

[Original submitted March 2009, Revision received October 2009, Accepted November 2009]

ABSTRACT

This paper examines the contribution of women’s labour input to productivity and efficiency in
crop farming using a large survey dataset of 1,839 households from 16 villages in two agro-
ecological regions of Bangladesh. Results reveal that female labour accounts for a substantial
28% of total labour use (mainly supplied from the family) and contributes significantly to
productivity as well as technical efficiency. Contrary to expectation, the cost share of female
labour input is significantly higher than the male share, and has a substitution relationship with
all other inputs, including male labour. The estimated mean level of technical efficiency is 0.90,
implying that crop output might be increased by 10% by eliminating technical inefficiency. Both
male and female education have a significant impact on improving technical efficiency. Other
significant technical efficiency shifters are farming experience, family size and crop
diversification. Owner operators are found to be technically inefficient relative to the tenants.
Policy implications include creation of a hired labour market for female labour so that more
women can be involved in the production process, and can contribute to towards improving
productivity and efficiency. Also, investment in education for both men and women, strategies to
promote crop diversification, and effective regulation/modification of the tenancy market will
significantly improve technical efficiency in this case.
JEL classification: O33; Q18; C21

Key words: Women’s labour contribution, Multiple crop farming, Stochastic frontier, Input

distance function, Technical efficiency, Bangladesh.

1
Sanzidur Rahman is a Senior Lecturer with the School of Geography, Earth and Environmental Sciences,
Faculty of Science and Technology, University of Plymouth, UK, Phone: +44-1752-585911, Fax: +44-1752-585998
E-mail: srahman@plymouth.ac.uk. The author gratefully acknowledges the contributions of two anonymous referees
and the chief editor which have improved the paper substantially. However, all caveats remain with the author.

1
1. Introduction

There is a widespread agreement that rural women in Asia play an important role in

agriculture (Kaur and Sharma, 1991; Unnevehr and Stanford, 1985) though its reflection is yet to

be seen in the formulation of agricultural development policies (Agarwal, 1998). For example,

about 84% of all economically active women are involved in agriculture in India with a positive

correlation between agricultural growth rates and employment of female agricultural labour

(World Bank, 1991). A commonly held view on women's involvement in agricultural production

in Bangladesh is that they are involved only in the post-harvest processing of crops, limiting their

contribution to national economy (Rahman, 2000). In fact, a simple change in the definition of

women’s work increased the estimate of women in the labour force from 3.2 million in the

Labour Force Survey (LFS) in 1985/86 to 21 million in the LFS in 1989 and the increase was

largely in rural regions (Rahman and Routray, 1998). The recent estimate of women in the labour

force stands at 12.2 million2 in the LFS in 2005/06 (BBS, 2007). It is believed that women’s

labour accounts for at least 25% of the value added from sowing to post-harvest operation in rice

production (Scott and Carr, 1985).

Women in Asian societies, particularly from poorer households, balance the multiplicity

of demands on their labour time between economic (wage earning and income-replacing work

like fuelwood and water collection, care of livestock) and domestic activities (cooking, cleaning

and child care) by working longer hours (Kabeer, 1994). For example, the average working day

for women is estimated at 13.2 hours in India (Kaur and Sharma, 1991) and 11.1 hours (domestic

and agriculture work only) in Bangladesh (Zaman, 1995). Although it is widely held that the
2
There has been an increase in the eligible age from population aged 10+ years to 15+ years to be included in the

pool of “economically active population” from 1999/2000. Therefore, if we take into account the number of women

aged between 10–14 years, the number of women in labour force will increase by another 3-4 million.

2
gender division of labour in Bangladesh is strictly demarcated, with women being responsible for

agricultural work within the household and not allowed to undertake field work (Begum, 1985;

Abdullah, 1985), contrasting evidence is also available (Zaman, 1995). Bangladeshi women

spent an average of 3.1 hours per day on agricultural work (4.4 hours per day during busy

season) while men spent 5.1 hours (Zaman, 1995) which is not substantially lower from an

average of 4.4 hours for rural women in India (Kaur and Sharma, 1991). Actually, the share of

women in labour use ranges between 11–18% in foodgrain (rice and wheat) production and 14–

48% in non-cereal (highest for vegetables) production in Bangladesh (Rahman, 2000).

Given that Bangladeshi women do undertake field work in agriculture, a further question

is whether they are as productive and efficient as men. An argument often used against women

farmers is that they are less efficient when compared to their male counterparts (FAO, 1985).

Whether women are more or less efficient than men in farming is a hotly debated issue and

results vary among the few studies that were undertaken in Africa during the 1990s (Adesina and

Djato, 1997), while none is yet available for Bangladesh. Also, there has been criticism of the

methods employed to examine gender differences in productivity in the literature (see

Quisumbing, 1996 for a review). The dominant use of a dummy variable approach for headship

on the input side of the production function as the stratifying variable (as seen in the literature)

disguises the nature of the household structure and the intra-household decision making process,

as it does not provide information on the decisions taken by family members (Quisumbing, 1996;

Aly and Shields, 2010). Also, most of these studies used deterministic models which assume

farmers to be fully efficient in their production technologies. With the development of stochastic

frontier analysis by Aigner, Lovell and Schmidt (1977), a large number of studies followed

which typically place farming efficiency of developing country farmers in the range of 60% to

3
82% (e.g., Bravo-Ureta et al., 2007; Rahman, 2003; Coelli et al., 2002; Ali and Flinn, 1989;

Wang et al., 1996). Failure to recognise that farmers are generally and inherently inefficient will

bias estimates of gender differences in productivity. Furthermore, most of these studies do not

take into account any relative disadvantage faced by women (e.g., educational opportunities)

when examining gender differences in productivity (Quisumbing, 1996).

The aims of this study are: (i) to determine the level of women’s labour contribution to

productivity in multiple-crop farming in Bangladesh, while allowing for inefficiency amongst

producers; (ii) to determine the influence of women’s labour input on production efficiency,

while simultaneously controlling for women’s educational level as a technical efficiency shifter,

thereby addressing both major shortcomings of the existing literature. We undertake this task by

employing a stochastic input distance function approach.

The paper is organised as follows. Section 2 reviews relevant literature examining gender

differences in agricultural productivity. Section 3 presents the analytical framework, the model,

and various performance measures developed to address the research objectives. Section 4

presents the results. The final section concludes and draws policy implications.

2. Gender differences in agricultural productivity

Quisumbing (1996), based on a comprehensive review of a number of studies undertaken during

the 1970s, 1980s and 1990s, noted that the data used to measure gender differences in

agricultural productivity is flawed. Most studies applied pooled regression with dummy variables

to account for male and female farm managers and found insignificant coefficients on the

dummies, implying that both men and women are equally efficient. Quisumbing (1996) argued

that using such dummies exclude information on decisions by family members who are not

household heads, and recommended the use of a more disaggregated measure of gender

4
difference. She also argued that in all these studies information did not exist to identify whether

the processes of allocation of human and physical capital to men and women had a bearing on

any observed gender differences in productivity. For example, underinvestment in girls’

education by parents in their resource allocation decisions could lead to lower probabilities of

female farmers adopting new technologies and, therefore, being less efficient (Quisumbing,

1996). Jacoby (1992) provided a very detailed analysis of productivity differences between men

and women in peasant agriculture of the Peruvian Sierra and identified a gender division of

labour, implying that male and female labour are not perfectly substitutable. He further

concluded that the use of animal traction and land affect the marginal productivity of male and

female labour differently and, therefore, these two types of labour cannot be aggregated. Adesina

and Djato (1997), using a deterministic profit function analysis, concluded that the relative

degree of efficiency of women is similar to that of men in Cote d’Ivoire, although their measure

of gender difference using a dummy variable has been criticised by Quisumbing (1996). More

recently, Aly and Shields (2010) examine productivity differences of female and male labourers

in Nepalese agriculture using two approaches: a Cobb-Douglas production function and a ray-

homothetic function. They conclude that, although there is a gender gap in productivity, once

differences in irrigation and type of seeds used by male and female farmers are included in the

model, the magnitude of the difference is reduced and the estimated coefficient becomes

insignificant. However, their study, although an improvement over those available in the

literature, still assumes farmers to be fully efficient in their production technologies, which may

bias the results.

Recently, two studies analyse the influence of women’s input on technical efficiency with

mixed results. Bozoglu and Ceyhan (2007) use a stochastic production frontier to analyze the

5
technical efficiency of vegetables production in Samsun province, Turkey with a small sample of

75 farmers. They conclude that women’s participation in farm decision-making significantly

improves technical efficiency. However, Hasnah et al., (2004) do not find any significant

influence of the share of female labour input as a technical efficiency shifter in the oil palm

sector in West Sumatra, Indonesia.

The contribution of our study to the existing literature is two fold: (a) we provide an

explicit examination of women’s labour contribution to agricultural productivity while

recognising that farmers may not be perfectly efficient in their production process (i.e., an

improvement over Jacoby’s (1992) and Aly and Shield’s (2010) work); (b) we provide a

simultaneous examination of women’s labour contribution to production efficiency, while

explicitly controlling for women’s educational achievement as a technical efficiency shifter (i.e.,

addressing Quisumbing’s (1996) major criticism of ignoring women’s relative disadvantage with

respect to human and physical capital). In addition, our analysis is conducted on a large dataset of

1,839 households (details in Section 3.2).

3. Research Methodology

3.1 Analytical framework

In order to examine women’s labour contribution in the farming sector characterised by

multiple crop production, a multi-output, multi-input production technology specification is

required. A distance function approach (either output-orientated or input-orientated) is

appropriate here, and can be analyzed using either parametric or non-parametric methods. In

addition, the distance function approach allows the production frontier to be estimated without

assuming separability of inputs and outputs (Kumbhakar, et al., 2007). An output oriented

approach to measure technical efficiency is appropriate when output is endogenous (e.g., revenue

6
maximization case) and inputs are exogenous, whereas an input oriented approach is appropriate

when inputs are endogenous (e.g., cost minimization case) and output is exogenous (Kumbhakar

et al., 2007). There is a criticism that parameter estimates of the distance functions may be

affected by simultaneous equations bias (e.g., Sickles et al., 1996; Atkinson et al., 1999; and

Alvarez, 2000). These authors went to correct this criticism by use of instrumental variables,

although they did not clearly specify the source of suspected simultaneous equations bias (Coelli,

2000). Some also simply argue that because the ratios of inputs appear on the right-hand-side

of the estimating equation (in the case of an input distance function) there must be a

simultaneous feedback problem because these input variables are assumed to be “endogenous”

variables. However, Coelli (2000) clearly demonstrated that OLS (Ordinary Least Squares)

provides consistent estimates of the parameters of the input distance function under an

assumption of cost minimizing behaviour. We assume that this conclusion can be generalized to

MLE (Maximum Likelihood Estimation) procedure as well. In fact as Coelli (2000) concludes,

“distance functions are no more subject to possible endogeneity criticisms than production

functions … when cost minimising behaviour is a reasonable assumption, the input distance

function has a clear advantage over the production function, because the distance function has an

endogenous dependent variable and exogenous regressors, while the production function has

the converse …. (and) distance functions release us from the shackles of the single-output

assumption, (making) the case for the use of distance functions further strengthened” (p.20-21).

We use an input-orientated stochastic distance function to address our research questions. This is

because, in an economy like Bangladesh, on the one hand, inputs are scarce, particularly the land

input, and on the other hand, farmers are often constrained by cash/credit (Rahman, 2009).

Therefore, it is logical to assume that cost minimization is the prime concern. Also, the choice of

7
a stochastic distance function approach instead of a deterministic approach (i.e., Data

Envelopment Analysis) has been adopted because of its ability to separate the random noise (e.g.,

weather variation, measurement errors, etc.) from technical inefficiency effects. For example,

weather variation could be a major issue when examining farming performance over a crop year

cycle, particularly, in a country like Bangladesh, implying that choice of a stochastic approach is

more appropriate.

Specifically, the following series of questions is addressed: (a) whether female labour

input affects productivity; (b) whether the productivity of male and female labour differs; (c)

what is the relationship between female and male labour inputs (d) whether female labour input

affects technical efficiency; (e) whether women’s educational achievement affects technical

efficiency. We specify the actual amount of female labour used in farming as an independent

variable in the stochastic input distance function and the share of female labour to total labour as

one of the technical efficiency shifters. Also, the highest educational level of male and female

members in the household, as well as other indicators representing farm characteristics, are

included in the inefficiency effects model to explain the underlying causes of deviation from the

frontier.

We begin by defining the production technology of the farm using the input set, L(y),

which represents the set of all input vectors, x ∈ R+K , which can produce the output vector

y ∈ R+M . That is,

L( y ) = {x ∈ R+K : x can produce y} (1)

The input-distance function is then defined on the input set, L(y), as

DI ( x, y ) = max{ρ : ( x / ρ ) ∈ L( y )} (2)

8
DI(x,y) is non-decreasing, positively linearly homogenous and concave in x, and increasing in y.

The distance function, DI(x,y), takes a value which is greater than or equal to one if the input

vector, x, is an element of the feasible input set, L(y) [DI(x,y) ≥ 1 if x ∈ L(y)]. Furthermore, the

distance function is unity if x is located on the inner boundary of the input set. Thus, the input

distance function can be interpreted as the multi-input input-requirement function allowing for

deviations (distance) from the frontier, which are interpreted in terms of technical efficiency

(Morrison-Paul and Nehring, 2005).

3.2 The model

For empirical implementation of the distance function, a functional form must be

specified. Also, it is particularly important in a multi-output and multi-input context to

minimize a priori restrictions on the relationships among inputs and outputs, and, therefore, a

flexible technological representation, allowing for substitution effects within the function, is

desirable for the empirical implementation of the model (Morrison-Paul et al., 2000). We select

the translog (TL) functional form used by many (e.g., Rahman, 2009, Morrison-Paul and

Nehring, 2005; Irz and Thirtle, 2004; Morrison-Paul et al., 2000; Coelli and Perelman, 1999).

The translog input distance function with M outputs and K inputs for the I farms (denoted

i) is given as:

1
ln Di = α 0 + ∑ α k ln X ki + ∑∑ α kl ln X ki ln X li +
k 2 k l
1
∑β
m
m ln Ymi + ∑ ∑ β mn ln Ymi ln Yni +∑∑
2 m n k m
τ km ln X k ln Ym (1)

The summation sign over k, l implies summation over all K inputs Xk, and similarly over m, n for

the M outputs Ym.

9
 Certain regularity conditions must hold for this function. These are: homogeneity of

degree one in inputs and symmetry of the cross effects. Therefore, the following constraints are

required:

∑ α = 1, ∑ α
k kl
kl = 0, ∑ τ km = 0(k = 1,....K ), and
km

α kl = α lk (k , l , = 1,....K ), and ( 2a )
β mn = β nm (m, m = 1,.....M ), respectively. (2b)

Following Lovell et al., (1994), we impose these constraints by normalizing the function by one

of the inputs. In this case, equation (1) becomes

1
ln Di / X 1i = α 0 + ∑ α k ln X ki* + ∑∑ α kl ln X ki* ln X li* +
k 2 k l
1
∑β
m
m ln Ymi + ∑ ∑ β mn ln Ymi ln Yni +∑∑
2 m n k m
τ km ln X k* ln Ym

= TL ( X * , Y ) (3)

Rewriting this function with – lnDi = – ui as a one sided error term, and including “white

noise” error terms vi representing random factors such as measurement error or unobserved

inputs, provides the following estimating equation:

− ln X 1,i = TL( X * , Y) − u i + vi (4)

Coefficient estimates for this equation have the opposite signs from those of a standard input

requirement function (Morrison-Paul and Nehring, 2005). However, to interpret various

performance measures derivable from equation (4) more similarly to those from more familiar

functions, we reverse their signs following Morrison-Paul et al., (2000) and Morrison-Paul and

Nehring, (2005):

ln X 1,i = −TL( X * , Y) − ui + vi (5)

This equation is now written as a standard stochastic production frontier form (with a

two-part error term representing deviations from the frontier and random error). This model can

10
be estimated econometrically using the maximum likelihood method, assuming that vi are

2
independently and identically distributed with mean zero and variance, σ v ; and the ui are non-

negative random variables independently distributed as truncations at zero of the normal

distribution with unknown variance, σ u2 , and unknown mean, µ, defined by:

µ = δ 0 + ∑ δ d Z di ( 6)
d1

Estimates of the parameters of the equations (5) and (6) were obtained using maximum

likelihood procedures in a single stage as detailed by Coelli and Perelman (1999). STATA

Software Version 8 was used for the analyses (Stata Corp, 2003).

3.3 The performance measures

Various performance measures of the production process can be derived as elasticities

from this estimated model. The combined first-order input elasticities represent scale economies

showing the extent to which productivity increases with input growth. The second-order

elasticities reflect production complementarities that reflect economic impacts from output

jointness (Morrison-Paul and Nehring, 2005). Specifically, for the input distance function, the X-

Y scale economy relationship is represented by the sum of individual input elasticities and

reflects how much overall input use must increase to support a 1% increase in all outputs.

Formally, the individual input elasticity summarizing the input expansion required for a 1%

increase in Ym is − ε D ,Ym = −∂ ln D / ∂ ln Ym = ∂ ln X 1 / ∂ ln Ym = ε Ym . Such a measure can be

thought of as an “input share” of Ym (relative to X1). In combination, these elasticties represent

scale economies: − ε D ,Y = −∑ ∂ ln D / ∂ ln Ym = ∑ ∂ ln X 1 / ∂ ln Ym = ∑ ε Ym = ε Y . The extent of

m m m

scale economies (for proportional changes in all inputs) is implied by the short-fall of εY from 1

(Morrison-Paul and Nehring, 2005).

11
 The first-order elasticities ε Ym and ε Y can also be decomposed into second-order effects

reflecting output compositions as scale expands. This information is implied by technological

bias measures indicating how the Ym input elasticity or share ( ε Ym ) reflects a change in another

output. Such measures provide insights about the output jointness of the production system.

Specifically, ε Ym,Yn = ∂ε Ym / ∂ ln Yn represents the increase in the Ym input share as Yn increases. If

ε Ym,Yn < 0 , output jointness or complementarity is implied; that is - input use does not have to

increase as much to expand Ym if the Yn level is greater. This elasticity is represented by the

cross-output coefficient estimate β mn : ε Ym,Yn = β mn = ε Yn,Ym (Morrison-Paul and Nehring, 2005).

In addition to information about output patterns, some insight into input contributions can

be obtained from the input distance function using the duality between the input distance function

and the cost function. Since the input distance function completely describes the production

technology, one may use it to describe the characteristics of the frontier or surface technology,

including curvature, i.e., the degree of substitutability along the surface technology, (Grosskopf

et al., 1995). Therefore, the indirect Morishima elasticity of substitution (MES) from an input

distance function can be computed as (Blackorby and Russell, 1989):

d ln[ Dk / Dl ] D D
MES X ,kl = − = X k kl − X k kk (7 )
d ln[ X k / X l ] Dl Dk

where the subscripts on the distance functions refer to partial derivatives with respect to inputs.

Because of the duality between the input distance function and the cost function, the first

derivatives of the distance function with respect to inputs yield the normalized shadow price of

that input, and therefore, the first component of the definition may be thought of as the ratio of

the percentage change in the shadow prices brought about by a 1% change in the ratio of inputs

(Kumar, 2006). This represents the change in relative marginal products and input prices required

12
to affect substitution under cost minimisation. High values reflect low substitutability and low

values reflect relative ease of substitution between the inputs (Kumar, 2006; Morrison-Paul et al.,

2000). The MES can be simplified as follows (Kumar, 2006):

MES X ,kl = ε X ,kl − ε X ,kk (8)

where ε X ,kl and ε X ,kk are the constant output cross and own elasticity of shadow prices with

respect to input quantities. The first term provides information on whether pairs of inputs are net

substitutes or net complements, and the second term is the own price elasticity of demand for the

inputs (Kumar, 2006). It should be noted that these elasticities are indirect elasticities. Therefore,

for εX,kl>0 net complements are indicated, and for εX,kl<0 net substitutes are indicated (Kumar,

2006). Also, the MES may not be symmetric, i.e., MES X ,kl ≠ MES X ,lk .

The Allen elasticity of substitution (AES) may also be derived from the input distance

function as follows (Kumar, 2006):

D * Dkl
AES X ,kl = = ε X ,kl / S k (9)
Dk * Dl

where Sk is the first order derivative of the translog input distance function with respect to input

Xk, i.e., S k = ∂ ln D / ∂ ln X k = −∂ ln X 1 / ∂ ln X k* (10)

The shadow price elasticities with respect to input quantities (to compute MES and AES

presented above) are given by (Kumar, 2006):

ε X ,kl = [α kl + S k S l ] / S k if k ≠ l; and
ε X ,kk = [α kk + S k ( S k − 1)] / S k if k = k (11)

Stern (2008) developed a new derivation of the Hicks elasticity of substitution (HES)

from an input distance function. The HES is defined for movement along a constant distance line

similar to shadow elasticity of substitution derived from the cost function that is defined for

13
movement along an isocost line (Stern, 2008). The HES (also known as the direct elasticity of

substitution) is defined as:

Dkk D D
− + 2 kl − ll
D k Dk Dk Dl Dl Dl
HES X ,kl = (10)
1 1
+
D k X k Dl X l

where D is the input distance function. The elasticity is symmetric and as this elasticity is the

inverse of the traditional Hick’s elasticity of substitution, greater values indicate less

substitutability (Stern, 2008). Also, for HES>0 net substitutes are indicated.

Finally, from the one-sided error term, ui, we can quantify the level of technical

efficiency, TE = exp(ui) (for details, see Coelli et al., 1998).

3.4 Data and the study area

Primary data for the study pertains to an intensive farm-survey from two agro-ecological

regions3 conducted during 1990. A complete household census of eight villages from the Jamalpur

Sadar Thana (central sub-district) of the Jamalpur region representing wet agroecology and six

villages from the Manirampur Thana (sub-district) of the Jessore region representing dry

agroecology were conducted. The survey/census covered a total of 1,839 households (923 in

Jamalpur and 916 in Jessore, respectively). Selection of these villages was conducted by BRAC. A

multistage random sampling technique was employed to locate the districts, the thana (sub-

district), and then the villages in each of the two sub-districts. Finally, all households from these

3
These data were collected by BRAC (one of the largest national non-governmental organisations in the world) to

serve as a baseline information for a longitudinal study project, called the Village Study Project (VSP). The baseline

data collection took about 6 months engaging 16 field researchers who were stationed in the core village of each

thana. The author of this paper was a member of the core research team and contributed significantly in the design of

the baseline survey and was responsible for co-ordinating the data collection team from the head office at that time.

14
16 villages were surveyed. Details of labour input data classified by gender4 for each individual

crop5 produced over a one year crop-cycle were collected in addition to information on other inputs

and the socio-economic circumstances of the surveyed households.

Although the data collected for this study are 20 years old, little has changed with regard to

the farming practices, operating institutions and the relationship between male and female labour

use in Bangladesh over this period, except for an increase in the level of modern rice technology

adoption from 30% of gross cropped area in 1990 to 51% in 2005 (MoA, 2008). For example, a

recent farm survey in 2006 of the Gher farming system (shrimp-prawn-modern rice joint culture)

from the Southwest region of Bangladesh showed that the female labour ratio is only 11% and

most of this is supplied from the family. Therefore, we argue that our results are capable of

providing valuable information of relevance to policy makers and development practitioners alike.

3.5 The empirical model

The production structure of crop farming in Bangladesh is specified using a multi-output

multi-input stochastic input distance function. The general form of the flexible translog stochastic

input distance function for the ith farm is defined as:

4
Male and female labour input was measured separately by the amount of labour used for each of the seven specific

agricultural operations (e.g., seedbed and/or land preparation, sowing and/or transplanting, weeding, irrigation,

fertiliser and pesticide application, harvesting, and threshing and/or winnowing operations) for each of the crops

produced.
5
The crop groups are: 1) traditional rice varieties (Aus – pre-monsoon, Aman – monsoon, and Boro – dry seasons);

2) modern/high yielding rice varieties (Aus, Aman, and Boro seasons); 3) wheats; 4) jutes; 5) potatoes; 6) pulses

(include lentil, mungbean, and gram); 7) spices (include onion, garlic, chilly, ginger, and turmeric); 8) oilseeds

(include sesame, mustard, and groundnut); 9) vegetables (eggplant, cauliflower, cabbage, arum, beans, gourds,

radish, and leafy vegetables); and 10) cotton.

15
 7
1 7 7
ln X 1i = α 0 + ∑ α k ln X ki* + ∑∑ α kl ln X ki* ln X li* +
k =2 2 k =2 l =2
4
1 4 4 7 4

∑β
m =1
m ln Ymi + ∑ ∑
2 m =1 n =1
β mn ln Ymi ln Yni + ∑∑ τ km ln X k* ln Ym − u i + vi
k = 2 m =1
(8)

and

9
u i = δ 0 + ∑ δ d Z id + ζ i* (8a )
d =1

where the dependent variable X1 is the land cultivated per farm in one crop year; X* are the other

inputs normalized by the land variable (X1); v is the two sided random error and u is the one sided

error in eq. (8); ln is the natural logarithm; Z in eq. (8a) are the variables representing farm

specific characteristics to explain inefficiency; ζ is the truncated random variable6; α0, αk,, αkl,

βm, βmn, τkm,δ0, and δd are the parameters to be estimated.

The model consists of seven production inputs (X); four outputs (Y); and nine variables

representing socio-economic characteristics of the farm (Z) included in the inefficiency effects

model as predictors of technical inefficiency. The seven inputs used in the analyses are: X1 = land

under all crops (ha), X2 = total female (family supplied + hired) labour (woman-days), X3 = total

male (family supplied + hired) labour (man-days), X4 = fertilizers (kg); X5 = animal power

services (animal-pair days); X6 = irrigation (taka); X7 = pesticides (taka). The four outputs are: Y1

= traditional rice (kg); Y2 = modern rice (kg); Y3 = wheat (kg); Y4 = cash crops7 (including jute,

cotton, oilseeds, spices, pulses, potatoes, and vegetables) (Bangladeshi taka). The nine variables

6
We actually conducted a Likelihood Ratio test regarding the choice of the distribution of the inefficiency term:

half-normal versus truncated normal and the result is presented in Table 2.

7
The gross value of each output is used to construct this compound (aggregate) variable, and is expressed as

Bangladeshi Taka per farm.

16
representing socio-economic characteristics of the farm are: Z1 = female headed household

dummy; Z2 = age of the farmer; Z3 = family size; Z4 = highest educational level of the male

member in the household; Z5 = highest educational level of the female member in the household;

Z6 = share of female labour in total labour; Z7 = tenurial status dummy; Z8 = regional dummy for

farmers located in Jessore region; Z9 = Herfindahl index of crop diversification (HI)8. Table 1

presents the definitions, units of measurement, and summary statistics for all variables.

Table 1. Summary statistics of the variables per farm.

Name Description Measurement Mean Standard

deviation
Output variables
Y1 Traditional rice kg 357.5 890.9
Y2 Modern rice kg 1212.9 2106.0
Y3 Wheat kg 38.2 200.7
Y4 Cash cropsa taka 10824.9 16956.2
Input variables
X1 Land area cultivated ha 0.8 1.1
X2 Female labour woman-days 12.1 16.2
X3 Male labour man-days 94.9 127.7
X4 Fertilizer kg 148.8 283.0
X5 Irrigation taka 665.9 1502.9
X6 Pesticides taka 99.1 258.9
X7 Animal power services animal pair-days 20.1 28.8
Farm-specific variables
Z1 Female headed households 1 if head, 0 otherwise 0.1 --
Z2 Age of the farmer years 42.0 13.3
Z3 Family size persons per household 5.4 2.5
Z4 Highest level of male education completed years of schooling 4.0 4.5
Z5 Highest level of female education completed years of schooling 2.2 3.2
Z6 Share of female labour proportion of total labour 0.3 0.3
Z7 Tenurial status 1 if owner-operator 0 0.2 --
otherwise

8
The Herfindahl index (HI) is represented as HI = ∑a i
2
, 0 ≤ HI ≤ 1 , where ai represents the area share occupied

by the ith crop in total area A. A zero value denotes perfect diversification and a value of 1 denotes perfect

specialization.

17
Z8 Jessore region 1 if Jessore 0 otherwise 0.5 --
Z9 Herfindahl index of crop number 0.7 0.3
diversification
Number of observations 1839
Note: a = Exchange rate of USD 1.00 = Taka 32.9 in 1990 (BBS, 1992)

The justification for inclusion of these variables is as follows. We have specified a

dummy for female headed households to test whether women as farm managers have any

influence on technical efficiency, as the literature suggests that women as farm managers are

equally productive with men (Quisumbing, 1996).

Farmers’ age is used to account for his/her experience in farming and its consequent

influence on technical efficiency, where the results in the literature are mixed. Although Rahman

(2003) and Asadullah and Rahman (2009) conclude that older farmers tend to be technically

inefficient compared with their younger peers in Bangladesh, Llewelyn and Williams (1996) and

Battese et al., (1996) conclude otherwise for Indonesian and Pakistani farmers, respectively.

Use of the education level of the farmer as a technical efficiency shifter is fairly common

(e.g., Asadullah and Rahman, 2009; Wang et al., 1996; Wadud and White, 2000). The education

variable is also used as a surrogate for a number of factors. At the technical level, access to

information as well as the capacity to understand the technical aspects related to crop production

is expected to improve with education, thereby, influencing technical efficiency. Surprisingly, the

majority of studies on to the effects of education on farm production in Bangladesh fail to find

any significant impact. For instance Deb (1995), Wadud and White (2000), Coelli et al. (2002),

and Rahman and Rahman (2009) did not find any significant effect of education on production

efficiency. However, Asadullah and Rahman (2009), using a large dataset of 2,357 households

from 141 villages in the Matlab district in Bangladesh, conclude that education of the household

matters in raising productivity, boosting potential output and improving efficiency. In this study,

we move a step further and aim to determine whether womens’ education has an independent

18
influence on technical efficiency, while controlling for the influence of mens’ education in the

household, addressing the criticism of Quisumbing (1996).

In this study, we are specifically interested in determining whether women’s labour input

has an influence on technical efficiency. Therefore, the share of female labour in total labour

used in farming is included to account for its influence in the model, as in Hasnah et al., (2004).

Tenurial status is also seen as an important technical efficiency shifter in Bangladesh

agriculture. For example, Rahman (2003), Asadulah and Rahman (2009), and Rahman and

Rahman (2009) all note that owner operators are relatively more technically efficient than

tenants.

Another key question of interest is whether farming inefficiencies are related to the

degree of diversification (or specialization), since the literature on this issue is mixed (e.g., Coelli

and Fleming, 2004; Llewelyn and Williams, 1996; Haji, 2007 and Rahman, 2009). Specialization

of farming activity may lead to greater efficiency or vice versa. The expectation is that

specialization in production leads to efficiency gains in the division of labour and management of

resources (Coelli and Fleming, 2004). A Herfindahl index is used to represent the specialization

variable. Although, this index is mainly used in the marketing industry to analyze market

concentration, it has also been used to represent crop diversification and/or concentration (e.g.,

Llewelyn and Williams, 1996; Rahman, 2009). Finally, a regional dummy for Jessore is included

in order to examine whether geography matters for production performance.

4. Results

From the information provided in Table 1, we see that the average farm size is small (0.81

ha). Share of female labour input in total labour is 28%. Average highest educational level of

both male and female members is low. The level of female education is about half of the level of

19
male education. The family size is 5.4 persons per household, which is very close to the national

average of 5.6 according to the 1991 population census (BBS, 2000). Dominant crops are modern

varieties of rice and cash crops.

Maximum Likelihood Estimation (MLE) is used to estimate the parameters of the

stochastic input distance function and the inefficiency effects model jointly in a single stage

using STATA Version 8 (Stata Corp, 2003). Prior to discussing the results of the input distance

function and the inefficiency effects model, we report the series of hypothesis tests conducted to

select the functional form and to decide whether the frontier model is an appropriate choice

rather than a standard mean-response or average production function. We also test regarding the

choice of the distribution of the inefficiency term: truncated normal versus half-normal

distribution.

We also test for the monotonicity condition requiring that the distance function be non-

decreasing in inputs (i.e., {i.e., (∂ ln D / ∂X k ) ≥ 0} and non-increasing in outputs

{i.e., (∂ ln D / ∂Ym ) ≤ 0} (Hailu and Veeman, 2000). We then check the curvature conditions of the

input distance function. The input distance function should be concave in inputs and quasi-

concave in outputs (Hailu and Veeman, 2000). Also tests were conducted to check input-output

separability, the presence of inefficiency and returns to scale in crop farming. The results are

reported in Table 2.

The first test was conducted to determine the appropriate functional form, i.e., the choice

between a Cobb-Douglas or a translog functional form (H0: αkl = 0 for all kl). A generalised

Likelihood Ratio (LR) test confirms that the choice of a translog production function is a better

representation of the production structure.

Given the functional form, we next check the sign of the third moment and the skewness

20
of the OLS residuals of the data in order to justify the use of the stochastic frontier framework

(and hence the Maximum Likelihood Estimation procedure)9. The computed value of Coelli’s

(1995) standard normal skewness statistic (M3T) based on the third moment of the OLS

residuals is estimated at 12.49 (Table 2) which is tested against H0: M3T = 0. The null

hypothesis of ‘no inefficiency component’ is strongly rejected in all cases and, therefore, the use

of the stochastic frontier framework is justified. The coefficient of γ reported at the mid-panel of

Appendix Table A1 also confirms the strong presence of technical inefficiency. The value of γ

ranges between 0 to 1, with 0 denoting no inefficiency and 1 being perfectly inefficient.

Next, we test for the preferred distribution of the inefficiency term. The generalized LR

test result presented in Table 2 confirms that the truncated normal distribution of the inefficiency

term is a preferred choice.

Next, we check the monotonicity condition of the input distance function. The results

presented in the mid-panel of Table 2 clearly demonstrate that the required conditions hold for

all inputs and outputs. Next, we conduct the curvature conditions check by examining the

Hessian matrix of the second order partial differentials of the input distance function with respect

to inputs and outputs as described by Chiang (1984), The checks reveal that the estimated

distance function was found to be concave in inputs and quasi-concave in outputs at all data

points.

Next, we test for the separability of the inputs and outputs in the input distance function.

This hypothesis is defined mathematically by equating all cross-terms between inputs and

9
In the stochastic frontier framework, the third moment is also the third sample moment of the ui. Therefore, if it is

negative, it implies that the OLS residuals are negatively skewed and technical inefficiency is present (Rahman and

Hasan, 2008).

21
outputs (τkm) to zero (Irz and Thirtle, 2004). These restrictions are strongly rejected, which

implies that it is not possible to aggregate consistently all the outputs into a single index. This is

why the distance function is more appropriate in our context, compared with a stochastic frontier

production function, which requires output aggregation prior to estimation.

Next we determine whether the variables introduced as inefficiency effects improve the

explanatory power of the model. The null hypothesis is rejected at the 1% level, implying that

the distributions of inefficiencies are not identical across individual observations (Irz and

Thirtle, 2004).

Table 2. Hypothesis tests

Name of Parameter LR test Degrees of χ2 Critical value Outcome

test restrictions statistic freedom 5%
Functional H0: αkl = βmn = τkm = 0 3155.62 55 73.31 Cobb-
form test for all k, l ,m, and n Douglas
(Translog model is
vs. Cobb- inadequate
Douglas)
Frontier vs. H0: M3T = 0 12.49 -- 0.000 Frontier not
OLSa (i.e., no inefficiency (z-statistic) (p value of z) OLS
component)
Distribution H0: No difference 162.75 10 18.31 Truncated
of the between truncated normal
inefficiency normal versus half- distribution is
term normal distribution of a preferred
the inefficiency term choice
Monotonicit {i.e., (∂ ln D / ∂X k ) ≥ 0} for every {i.e., (∂ ln D / ∂Ym ) ≤ 0} for every
y condition input output
check
Inputs Value Outcome Outputs Value Outcome
Female 0.016 Fulfilled Traditional -0.002 Fulfilled
labour rice
Male labour 0.002 Fulfilled Modern rice -0.006 Fulfilled
Fertilizer 0.002 Fulfilled Wheat -0.002 Fulfilled
Irrigation 0.003 Fulfilled Cash crops -0.011 Fulfilled
Pesticides 0.006 Fulfilled
Animal 0.012 Fulfilled
power
Input- H0: all τkm = 0 for all k 424.38 24 36.42 Aggregating

22
output and m output into a
separability single index
will provide
inconsistent
result
Returns to H0: (Σβm) = 1 for all m 553.50 1 2.71 Considerable
scale (Scale scale
economy if economy
εY<1) exists
No H0: δd = 0 for all d 97.67 9 16.92 Inefficiencies
inefficiency are jointly
effects explained by
these
variables
a
Note: = The test is the Coelli’s (1995) standard normal skewness statistic (M3T) based on the third moment of
the residual.

Measures of economic performance

The parameter estimates of the stochastic input distance function and the inefficiency

effects model for crop farming estimated jointly in a single stage is presented in Appendix Table

A1. Three quarters of the coefficients in the input distance function are significantly different

from zero at the 10% level at least. All the variables are mean-differenced prior to estimation so

that the elasticities of the distance function with respect to input quantities and output

quantities at the sample mean correspond simply to the first order coefficients. All the signs

on the first order coefficients of inputs and outputs are consistent with a priori expectations.

The primary overall measure, representing output/input patterns and performance

incentive to increase the scale and diversity of farm operations, is the scale elasticity εY (Table 3)

The presented measure suggests significant scale economies, (εY<1 indicate scale economies).

The result of the formal test for constant returns to scale (i.e., εY=1) is presented in Table 2,

which is strongly rejected in favour of increasing returns to scale. This finding is interesting

because other studies tend to report decreasing returns to scale, particularly in cereal production,

23
in developing economies. For example, Appleton and Balihuta (1996), Weir and Knight (2004),

and Asadullah and Rahman (2009) report decreasing returns to scale in cereal/rice production for

Ugandan, Ethiopian and Bangladeshi farmers, respectively.

Table 3. Output jointness, input, output, and efficiency elasticities

Variables Symbol Value t-ratio

Output elasticities
Scale economy εY 0.45
Traditional rice εY1 0.06*** 10.11
Modern rice εY2 0.32*** 48.81
Wheat εY3 0.020 0.97
Cash crops εY4 0.05*** 15.26
Input elasticities
Female labour εX2 -0.19*** -10.85
Male labour εX3 -0.04 -1.38
Fertilizer εX4 -0.05*** -3.36
Irrigation εX5 -0.02** -2.49
Pesticides εX6 -0.02*** -3.16
Animal power εX7 -0.28*** -10.31
Land εX1 -0.41 --
Output jointness or complementarity
Traditional rice and Modern rice εY12 -0.02*** -14.35
Traditional rice and Wheat εY13 -0.00** -1.96
Traditional rice and Cash crops εY14 -0.00*** -2.74
Modern rice and Wheat εY23 -0.01*** -5.91
Modern rice and Cash crops εY24 -0.01*** -5.16
Wheat and Cash crops εY34 0.00 0.89
Efficiency elasticities
Female headed households εZ1 0.00 0.01
Age of the farmer εZ2 0.07*** 3.40
Family size εZ3 0.08*** 4.03
Highest level of male education εZ4 0.03*** 3.75
Highest level of female education εZ5 0.01* 1.90
Share of women labour εZ6 0.07*** 3.37
Tenurial status εZ7 -0.02*** -5.87
Jessore region εZ8 0.01 0.87
Herfindahl index of crop diversification εZ9 -0.04** -2.31
Note: *** significant at 1 percent level (p<0.01)
** significant at 5 percent level (p<0.05)
* significant at 10 percent level (p<0.10).

24
 The individual output contribution underlying the scale elasticity is also presented in

Table 3. These elasticities with respect to output in a distance function also represent the cost

elasticity of that particular output (Irz and Thirtle, 2004). Table 3 shows that (except wheat) all

output elasticities are significantly different from zero, implying that increasing the production of

any of these inputs will increase costs substantially. The estimate also shows that the cost

elasticity of modern rice is 0.32 whereas the estimates for cash crops as well as traditional rice

are only 0.05 and 0.06, respectively. This means that a 1% increase in modern rice output will

increase cost by 0.32%, while the corresponding figure for traditional rice is only 0.06%.

Similarly, the elasticities of the distance function with respect to input quantities are equal

to the cost shares and, therefore, reflect the relative importance of inputs in the production

process. Table 3 reveals that all seven elasticities are negative, as expected, with only one input

(male labour) being not significantly different from zero. The elasticity with respect to land is the

largest with a value of -0.41, implying that the cost of land represents 41% of total cost at the

sample mean10. This is not an unexpected result in a land scarce economy like Bangladesh. The

next highest cost input is animal power services (-0.28) which is also expected because land

preparation is dominated by the use of animal power services, particularly for rice cultivation. It

is somewhat surprising to see that the input elasticity of female labour is almost four times

greater than that of male labour. A test of the equality of these two production elasticities failed

to reject the null hypothesis (H0: α2 – α3 = 0) at 1% level of significance (χ2 = 16.40, p<0.01). The

implication is that the female labour input plays a significant role in the production process and

the contribution is significantly higher than that of male labour.

10
Elasticity of the land variable is computed using the restrictions in equation (2a), and therefore, its significance

cannot be determined

25
 To further evaluate the implications of our estimates of output complementarities and

their contribution to scale economies, we focus on the (second order) cross-effects. These

estimates are represented by the cross-parameters of the estimated functions (βmn), reproduced in

the mid-panel of Table 3. Except for wheat and cash crops, all crop combinations are negative

and significantly different from zero at 5% level at least, implying complementarities and/or

output jointness (Morrison-Paul and Nehring, 2005). As expected, the complementarities are

highest for the traditional and modern rice combination followed by the modern rice and wheat

combinations. Overall, these results suggest that significant scope economies exist in

Bangladeshi farming, consistent with the results found by Rahman (2009).

4.1 Substitutability of female labour with other inputs

Table 4 presents the results of the three types of indirect substitution elasticities, namely,

Morishima (MES), Allen (AES), and Hick’s (HES) elasticity of substitution. Approximately 60%

of the estimated substitution elasticities are significantly different from zero at the 10% level at

least. With respect to MES and HES, we see that most of the inputs are substitutes although the

degree of substitutability varies amongst inputs. We focus here on the substitutability between

female labour and all other inputs. The MES results show that female labour is a substitute for

male labour, fertilizers, irrigation and pesticides when row variables were considered and for

irrigation and pesticides when column variables are considered. The AES results show that

female labour is a substitute for male labour but a complement for animal power services. The

HES results show that the female labour input is a substitute for all other inputs except male

labour (not significant). The substitutability between female labour and other inputs implies that

as the shadow price (or cost share) of female labour increases, the farmer employs more of other

inputs. Overall, the estimated elasticity values show that male labour can be relatively easily

26
substituted for female labour compared to all other inputs. This may partly explain lower use of

female labour in farm production activities, particularly when hiring-in labour input.

Table 4. Indirect substitution elasticities

Variables Female Male labour Fertilizer Irrigation Pesticides Animal

labour power
Morishima elasticitiy of substitution (MESX,kl)
Female labour -- -0.76*** -0.89*** -0.99*** -1.01*** -0.18
Male labour 6.38 -- 4.74 4.98 3.47 -0.93
Fertilizer -0.41 -1.23** -- -0.85*** -0.58** -0.18
Irrigation -0.81** -1.49** -0.89** -- -0.58** 1.11
Pesticides -3.04*** -5.63*** -2.38*** -2.72*** -- -0.54
Animal power 0.22 -1.15*** -0.21 -0.21 -0.11 -
Allen elasticity of substitution (AESX,kl)
Female labour --
Male labour -0.04** --
Fertilizer 0.02 -0.02 --
Irrigation -0.00 -0.01 -0.00 --
Pesticides -0.01 -0.07*** 0.01* 0.00 --
Animal power 0.15*** -0.24*** 0.03 0.03*** 0.06*** --
Hick’s elasticity of substitution (HESX,kl)
Female labour --
Male labour -3.89 --
Fertilizer 0.98*** -3.14 --
Irrigation 0.70*** -1.54 0.63** --
Pesticides 2.52*** -2.63 2.39*** 1.60*** --
Animal power 1.35*** -6.13 0.89*** 0.87*** 2.95*** --
Note: *** significant at 1 percent level (p<0.01)
** significant at 5 percent level (p<0.05)
* significant at 10 percent level (p<0.10).

4.2 Efficiency effects of female labour input

Prior to the discussion of the determinants of technical inefficiency, we report summary

statistics of technical efficiency scores. The mean technical efficiency is estimated at 0.90

implying that the average farm could increase production by 11% by optimising technical

efficiency. Farmers exhibit a wide range of production inefficiency ranging from 53% to 99% in

multiple crop farming. Observation of wide variation in production efficiency is not surprising

27
and is similar to the results of Rahman, (2003), Ali and Flinn, (1989), and Wang et al., (1996) for

Bangladesh, Pakistan, and China, respectively.

The lower panel of Appendix Table A1 provides the results of the inefficiency effects

model. As mentioned above, the null hypothesis of ‘no efficiency effects’ (i.e., H0: δd = 0 for all

d) is rejected at the 1% level of significance (Table 2), implying that all these variables jointly

have an influence on the technical efficiency scores of individual farmers. The coefficients on

these inefficiency predictors show only the direction of influence and do not provide information

on the magnitude of influence. Therefore, we compute technical efficiency elasticities for these

predictors using the Frame and Coelli (2001) framework. Table 3 presents the specific measure

of responsiveness of each predictor on technical efficiency, not commonly reported in the

existing literature.

Table 5. Technical efficiency in farming

Variables Efficiency scores

Efficiency levels
Upto 60% 0.9
61 – 70% 3.8
71 – 80% 13.9
81 – 90% 22.1
90 and above 59.3
Mean efficiency level 0.9
Standard deviation 0.1
Minimum 0.5
Maximum 1.0
Number of observations 1839

Older or experienced farmers are relatively technically more efficient compared with their

younger peers, consistent with the findings of Llewelyn and Williams, (1996) and Battese et al.,

(1996) but not with Asadullah and Rahman, (2009). The elasticity estimate suggests that a 1%

increase in the age of the farmer improves technical efficiency by 0.07%. Family size also

28
significantly improves technical efficiency, perhaps through more timely supply of family labour.

Tenants are relatively technically more efficient than the owner operators, which contradicts with

the findings of other studies on Bangladesh (e.g., Rahman, 2003; Asadullah and Rahman, 2009).

The contribution of female labour input significantly improves technical efficiency. The elasticity

estimate indicates that a 1% increase in the share of female labour in total labour improves

technical efficiency by 0.07%. Both male and female education have a significant influence on

improving technical efficiency, although the level of influence of male education is three times

that of female education. The significant role of education in improving technical efficiency in

Bangladesh is also reported by Asadullah and Rahman (2009) and Sharif and Dar (1996). Crop

diversification significantly improves technical efficiency, albeit with a relatively small effect.

The elasticity estimate shows that a 1% increase in the Herfindahl index of crop diversification

improves technical efficiency by 0.04%. The direction of the association is consistent with the

findings of Rahman (2009) and Coelli and Fleming (2004) but not with Haji (2007) or Llewelyn

and Williams (1996).

5. Discussion and policy implications

Rural women in Bangladesh, as elsewhere in Asia, play an important role in agriculture.

The results of the present study confirm that female labour contributes significantly to

productivity as well as technical efficiency. However, the remunerative employment of labour

remains skewed in favour of men as they are mostly hired to meet the demand. The estimated

28% of female labour used in crop farming in our study is mainly supplied from the family,

although 12% of all households reported hiring-in female labour in addition to male labour. The

surveyed farmers are operating at a high level of technical efficiency of 90% implying that about

29
11% of the loss in potential output might be recovered if technical inefficiency could be

completely eliminated.

The deprivation of women in gainful employment, reflected by weaker participation in

the hired agricultural labour market, is largely due to cultural constructs in farming societies in

Bangladesh. For example, Kabeer (1994) noted that though men can use their labour in a variety

of ways with more ability to orient it towards income earning activities, women’s ability to

dispose their labour power is constrained by imposition of purdah as well as domestic

obligations. Furthermore, Zaman (1995) claimed the existence of male preference in the

agricultural labour market where females are hired for field agriculture only when the male

labour supply is exhausted. Our results clearly show that female labour inputs can be substituted

for male labour relative easily as compared to all other inputs. Our results also clearly

demonstrate that female education has a significant influence on improving technical efficiency,

as with the case of male education. Rahman (2000) notes that one of the major vehicles for

creating awareness of gender discrimination is investment in human capital through gender

sensitive literacy programs, as there is a positive relationship between the highest level of

education of the household members and the demand for hired labour (both male and female).

Balanced development implies that both men and women are provided with equal opportunities

in all spheres of life. The dominance of the agricultural sector in the Bangladesh economy

indicates that attempts to bridge the gap in employment opportunities between men and women

has to be sought in the agricultural sector itself, since it engages the majority of the rural

population, half of which are women (Rahman, 2000).

Apart from the information on the contribution of the female labour input, the results of

our study also reveal that considerable scale economies exist in Bangladeshi farming systems.

30
The implication is that Bangladeshi farmers could gain by increasing their farm sizes.

Conventionally, either constant or decreasing returns to scale in Bangladesh are usually reported

in the literature (e.g., Wadud and White, 2000; Coelli et al., 2002; Rahman, 2003; Asadullah and

Rahman, 2008). In addition, our results also suggest considerable scope economies in farming as

a consequence of diversification of the cropping system, which also has a significant impact on

technical efficiency.

It is encouraging to note that tenants are relatively technically more efficient than the

owner operators. Tenancy (both crop-share and/or cash-rent tenancy) is a common feature of the

agricultural sector in Bangladesh. The latest available Agricultural Census of 1996 reported that

37.5% of the farmers operated as tenants (either part or pure tenants) and 10.2% of farm holdings

were landless (BBS, 2000) and that rice is the main crop grown by tenants (Akanda et al., 2008).

In our sample, 32.5% of the farmers were part-tenants and 19.6% were pure tenants. Although a

legal system of input and output sharing exists in the tenancy market, Akanda et al., (2008) argue

that the existing economic structure does not fairly balance the returns from production between

tenants and landowners (who gain relatively more). Therefore, proper regulation and/or

modification of this important market is essential to safeguard the interests of the tenants, who

would in turn boost potential output, at least according to these results.

The policy implications of this study are clear. Creation of a hired labour market for

female labour is desirable so that more women can be involved in the production process and

contribute positively towards improving productivity and efficiency and be remunerated for their

contribution. Also, policies to promote crop diversification will improve overall productivity

through scope economies and technical efficiency. The recent thrust at the planning level to

promote crop diversification and allocating 8.9% of the total agricultural budget to this during the

31
Fifth Five Year Plan (1997–2002) appears to be a step in the right direction (PC, 1998).

Furthermore, investment in education for both male and female members of the household and

effective regulation/modification of the existing tenancy markets could significantly improve

technical efficiency of the Bangladeshi farming system.

32
 REFERENCES

Abdullah, T. 1985. Women in rice farming in Bangladesh and how technology programs can reach

them. In Women in Rice Farming. Proceedings of a Conference on Women in Rice Farming

Systems. 26-30 September 1983, International Rice Research Institute (IRRI), Philippines.

Aldershot: Gower Publishing Co. Ltd.

Adesina, A. A., Djato, K.K. 1997. Relative efficiency of women as farm managers: profit function

analysis in Cote d’Ivoire. Agricultural Economics, 16, 47-53.

Agarwal, B. 1998. “Disinherited peasants, disadvantages workers: a gender perspective on land and

livelihood”. Economic and Political Weekly, 33, A2 – A14.

Aigner, D.J., Lovell, C.A.K., Schmidt, P. 1977. Formulation and estimation of stochastic frontier

production function models. Journal of Econometrics, 6, 21–37.

Akanda, M.A.I., Isoda, H., Ito, S. 2008. Problem of share-cropping system in rice farming n

Bangladesh: a case study on Alinapara village in Sherpur district. Journal of International

Farm Management, 4, 1-13.

Ali, M., Flinn, J.C. 1989. Profit efficiency among Basmati rice producers in Pakistan Punjab.

American Journal of Agricultural Economics, 71, 303 – 310.

Alvarez, A.R. 2000. Measuring allocative efficiency in a bureaucracy: the Spanish public

hospitals sector. Unpublished PhD Thesis, Department of Economics, University of

Oviedo.

Aly, Y.H., Shields, M.P. 2010. Gender and agricultural productivity in a surplus labour

traditional economy: empirical evidence from Nepal. Journal of Developing Areas, 43,

111-124.

33
Appleton, S., Balihuta, A., 1996, Education and agricultural productivity: evidence from Uganda.

Journal of International Development, 8, 415-444.

Asadullah, M.N., Rahman, S. 2009. Farm productivity and efficiency in rural Bangladesh: the role

of education revisited. Applied Economics, 41, 17-33.

Atkinson, S.E., Cornwell, C., Honerkamp, O. 1999. Measuring and decomposing productivity

change: stochastic distance function estimation vs DEA. Working Paper 99-478,

Department of Economics, University of Georgia.

Battese, G.E., Malik, S.J., Gill, M.A. 1996. An investigation of technical inefficiencies of

production of wheat farmers in four districts of Pakistan. Journal of Agricultural

Economics, 47, 37-49.

BBS, 2007. Bangladesh Labor Force Survey, 2005/06. Bangladesh Bureau of Statistics, Dhaka,

Bangladesh.

BBS, 2000. Statistical Yearbook of Bangladesh, 1999. Bangladesh Bureau of Statistics, Dhaka,

Bangladesh.

BBS, 1992. Statistical Yearbook of Bangladesh, 1991. Bangladesh Bureau of Statistics, Dhaka,

Bangladesh.

Begum, S. 1985. ‘Women and technology: rice processing in Bangladesh’. In Women in Rice

Farming. Proceedings of a Conference on Women in Rice Farming Systems. 26-30

September, 1983, International Rice Research Institute (IRRI), Philippines. Aldershot:

Gower Publishing Co. Ltd.

Blackorby, C., Russell, R. R. 1989. Will the real elasticity please stand up? American Economic

Review, 79, 882–888.

34
Bozoglu, M., Ceyhan, B. 2007. Measuring the technical efficiency and exploring the inefficiency

determinants of vegetable farms in Samsun province, Turkey. Agricultural Systems, 94,

649-656.

Bravo-Ureta, B.E., Solis, D., Lopez, V.H.M., Maripani, J.F., Thiam, A., Rivas, T. 2007.

Technical efficiency in farming: a meta regression analysis. Journal of Productivity

Analysis, 27, 57-72.

Chiang, A.C. 1984. Fundamental Methods of Mathematical Economics. 3rd Edition. McGraw

Hill, Singapore.

Coelli, T.J. 2000. On the econometric estimation of the distance function representation of the

production technology. CORE Discussion Paper 2000/42. Universite Catholiqué de

Louvain

Coelli, T.J., Fleming, E., 2004. Diversification economies and specialization efficiencies in a

mixed food and coffee smallholder farming system in Papua New Guinea. Agricultural

Economics, 31: 229-239.

Coelli, T.J., Perelman, S., 1999. A comparison of parametric and non-parametric distance

functions: with application to European railways. European Journal of Operational

Research, 117: 326-339.

Coelli, T.J., Prasada Rao, D.S., Battese, G.E. 1998. An Introduction to Efficiency and

Productivity Analysis. Boston: Kluwer Academic Publishers.

Coelli, T.J., Rahman, S., Thirtle, C. 2002. Technical, allocative, cost and scale efficiencies in

Bangladesh rice cultivation: a non-parametric approach. Journal of Agricultural

Economics, 53, 607-626.

35
Coelli, T.J. 1995. Estimators and hypothesis tests for a stochastic frontier function: a Monte-

Carlo analysis. Journal of Productivity Analysis, 6, 247-268.

FAO. 1985. Women in developing country agriculture. Food and Agricultural Organization of the

United nations, Rome.

Frame, W.S., Coelli, T.J. 2001. U.S. Financial services consolidation: the case of corporate credit

unions. Review of Industrial Organisation, 18, 229-242.

Grosskopf, S., Hayes, K., Hirschberg, J. 1995. Fiscal stress and the production of public safety: a

distance function approach. Journal of Public Economics, 57, 277–296.

Hailu, A., Veeman, T.S. 2000. Environmentally sensitive productivity analysis of the Canadian

pulp and paper industry, 1959-1994: an input distance function approach. Journal of

Environmental Economics and Management, 40: 251-274.

Haji, J. 2007. Production efficiency of smallholders’ vegetable-dominated mixed farming system

in Eastern Ethiopia: a non-parametric approach. Journal of African Economies, 16:1-27.

Hasnah, Fleming, E., Coelli, T. 2004. Assessing the performance of a nucleus estate and

smallholder scheme for oil palm production in West Sumatra: a stochastic frontier analysis.

Agricultural Systems, 79, 17-30.

Irz, X., Thirtle, C. 2004. Dual Technological Development in Botswana Agriculture: A Stochastic

Input Distance Function Approach. Journal of Agricultural Economics, 55, 455-478.

Jacoby, H.G. 1992. Productivity of men and women and the sexual division of labour in peasant

agriculture of Peruvian Sierra. Journal of Development Economics, 37, 265-287.

Kabeer, N. 1994. Reversed Realities: Gender Hierarchies in Development Thought. London:

Verso.

36
Kaur, M., Sharma, M.L. 1991. Role of women in rural development”. Journal of Rural Studies, 7,

11–16.

Kumar, S. 2006. Analysing industrial water demand in India: An input distance function

approach. Water Policy, 8, 15-29.

Kumbhakar, S.S., Orea, L. Rodriguez-Alvarez, A. Tsonias, E. G. 2007. Do we estimate an input

or an output distance function? An application of the mixed approach to European

railways. Journal of Productivity Analysis, 27: 87–100.

Llewelyn, R.V., Williams, J.R. 1996. Nonparametric analysis of technical, pure technical, and

scale efficiencies for food crop production in East java, Indonesia. Agricultural

Economics, 15: 113-126.

Lovell, C. A. K., Richardson, S., Travers, P., Wood, L. L. 1994. Resources and Functionings: A

New View of Inequality in Australia. In W. Eichhorn (ed.) Models and Measurement of

Welfare and Inequality. Berlin: Springer-Verlag Press.

MoA, 2008. Agricultural Handbook of Bangladesh, 2007. Ministry of Agriculture, Government

of Bangladesh, Dhaka.

Morrison-Paul, C.J., Johnston, W.E., Frengley, G.A.G. 2000. Efficiency in New Zealand sheep

and beef farming: the impacts of regulatory reform. Review of Economics and Statistics,

82, 325-337.

Morrison-Paul, C.J.. Nehring, R. 2005. Product diversification, production systems, and

economic performance in U.S. agricultural production. Journal of Econometrics, 126,

525-548.

PC, 1998. The Fifth Five Year Plan (1997-2002). Ministry of Planning, Government of

Bangladesh, Dhaka.

37
Quisumbing, A.R. 1996. Male-female differences in agricultural productivity: methodological

issues and empirical evidence. World Development, 24, 1579-1595.

Rahman, 2009. Whether crop diversification is a desired strategy for agricultural growth in

Bangladesh? Food Policy, 34, 340-349.

Rahman, S. 2000. Women’s employment in Bangladesh agriculture: composition, determinants and

scope. Journal of Rural Studies, 16, 497-507.

Rahman, S. 2003. Profit efficiency among Bangladeshi rice farmers. Food Policy, 28: 487-503.

Rahman, S., Rahman, M.M. 2009. Impact of land fragmentation and resource ownership on

productivity and efficiency: the case of rice producers in Bangladesh. Land Use Policy,

26, 95-103.

Rahman, S., Routray, J.K. 1998. “Technological change and women’s participation in crop

production activities in Bangladesh”. Gender, Technology and Development, 2, 243–67.

Scott, G.L., Carr, M., 1985. The Impact of Technology Choice on Rural Women in Bangladesh:

Problems and Opportunities. World Bank Staff Working Papers No. 731. Washington,

D.C.: The World Bank.

Sharif, N.R., Dar, A. 1996. An empirical study of the patterns and sources of technical

inefficiency in traditional and HYV rice cultivation in Bangladesh. Journal of

Development Studies, 32, 612-629.

Sickles, R.C., Good, D.H., Getachew, L. 1996. Specification of distance functions using semi-

and on-parametric methods with an application to the dynamic performance of Eastern

and Western European air carriers. 2nd Biennial Georgia Productivity Workshop,

University of Georgia, October, 1996.

38
Stata Corp, 2003. Stata Statistical Software: Release 8.0. Stata Corporation. College Station,

Texas.

Stern, D. I. 2008. Derivation of the Hicks Elasticity of Substitution from the Input Distance

Function. Munich Personal RePEc Archive (MPRA) # 12414. @ http://mpra.ub.uni-

muenchen.de/12414/

Unnevehr, L.J., Stanford, M.L., 1985. “Technology and the demand for women’s labour in Asian

rice farming”. In Women in Rice Farming. Proceedings of a Conference on Women in Rice

Farming Systems. 26-30 September 1983, International Rice Research Institute (IRRI),

Philippines. Aldershot: Gower Publishing Co. Ltd.

Wang, J., Cramer, G.L., Wailes, E.J. 1996. Production efficiency of Chinese agriculture:

evidence from Rural Household Survey data. Agricultural Economics, 15, 17–28.

Weir, S. and Knight, J. 2004. Externality effects of education: dynamics of the adoption and

diffusion of an innovation in rural Ethiopia. Economic Development and Cultural

Change. 53, 93-113.

World Bank. 1991. Gender and Poverty in India. Washington, D.C.: The World Bank.

Zaman, H., 1995. ‘Patterns of activity and use of time in rural Bangladesh: class, gender, and
seasonal variations’. Journal of Developing Areas, 29, 371–88.

39
 Appendix

Appendix Table A1. Parameter estimates of the stochastic input distance function including
inefficiency effects.

Variables Parameters Coefficients t-ratio

Production Variables
Constant α0 -2.00*** -34.76
ln(Female labour/Land) α2 -0.19*** -10.85
ln(Male labour/Land) α3 -0.04 -1.38
ln(Fertilizer/Land) α4 -0.05*** -3.36
ln(Irrigation/Land) α5 -0.02** -2.49
ln(Pesticides/Land) α6 -0.02*** -3.16
ln(Animal power/Land) α7 -0.28*** -10.31
½ ln(Female labour/Land)2 α22 0.03** 0.15
½ ln(Male labour/Land)2 α33 -0.24*** -6.90
½ ln(Fertilizer/Land)2 α44 -0.01 -0.61
½ ln(Irrigation/Land)2 α55 -0.00 -1.03
½ ln(Pesticides/Land)2 α66 0.04*** 7.86
½ ln(Animal power/Land)2 α77 -0.12** -2.23
ln(Female labour/Land) x ln(Male labour/Land) α23 -0.03* -1.69
ln(Female labour/Land) x ln(Fertilizer/Land) α24 -0.01 -0.76
ln(Female labour/Land) x ln(Irrigation/Land) α25 0.01 1.05
ln(Female labour/Land) x ln(Pesticides/Land) α26 0.01** 1.97
ln(Female labour/Land) x ln(Animal power/Land) α27 -0.10*** -4.87
ln(Male labour/Land) x ln(Fertilizer/Land) α34 0.02 1.18
ln(Male labour/Land) x ln(Irrigation/Land) α35 0.00 0.75
ln(Male labour/Land) x ln(Pesticides/Land) α36 -0.01 -1.52
ln(Male labour/Land) x ln(Animal power/Land) α37 -0.02 -0.74
ln(Fertilizer/Land) x ln(Irrigation/Land) α45 0.01 1.15
ln(Fertilizer/Land) x ln(Pesticides/Land) α46 0.07*** 4.75
ln(Fertilizer/Land) x ln(Animal power/Land) α47 0.25*** 7.07
ln(Irrigation/Land) x ln(Pesticides/Land) α56 -0.00 -0.79
ln(Irrigation/Land) x ln(Animal power/Land) α57 -0.02** -2.31
ln(Pesticides/Land) x ln(Animal power/Land) α67 -0.05*** -3.82
ln(Traditional rice) β1 0.06*** 10.11
ln(Modern rice) β2 0.32*** 48.81
ln(Modern wheat) β3 0.02 0.97
ln(Cash crops) β4 0.05*** 15.26
½ ln(Traditional rice)2 β11 0.07*** 16.62
½ ln(Modern rice)2 β22 0.10*** 28.49
½ ln(Wheat)2 β33 0.03*** 3.15
½ ln(Cash crops)2 β44 0.02*** 10.02
ln(Traditional rice) x ln(Modern rice) β12 -0.02*** -14.35
ln(Traditional rice) x ln(Wheat) β13 -0.01** -1.96
ln(Traditional rice) x ln(Cash crops) β14 -0.00*** -2.74

40
Variables Parameters Coefficients t-ratio
ln(Modern rice) x ln(Wheat) β23 -0.01*** -5.91
ln(Modern rice) x ln(Cash crops) β24 -0.01*** -5.16
ln(Wheat) x ln(Cash crops) β34 0.00 0.89
ln(Female labour/Land) x ln(Traditional rice) τ21 0.01*** 3.91
ln(Female labour/Land) x ln(Modern rice) τ22 0.03*** 7.39
ln(Female labour/Land) x ln(Wheat) τ23 0.02*** 2.86
ln(Female labour/Land) x ln(Cash crops) τ24 -0.00 -1.20
ln(Male labour/Land) x ln(Traditional rice) τ31 0.03*** 3.28
ln(Male labour/Land) x ln(Modern rice) τ32 -0.06*** -5.80
ln(Male labour/Land) x ln(Wheat) τ33 -0.03* -1.93
ln(Male labour/Land) x ln(Cash crops) τ34 0.05*** 9.07
ln(Fertilizers/Land) x ln(Traditional rice) τ41 -0.01* -1.91
ln(Fertilizers/Land) x ln(Modern rice) τ42 -0.00 -1.14
ln(Fertilizers/Land) x ln(Wheat) τ43 0.00 0.19
ln(Fertilizers/Land) x ln(Cash crops) τ44 -0.01** -2.27
ln(Irrigation/Land) x ln(Traditional rice) τ51 -0.00*** -2.73
ln(Irrigation/Land) x ln(Modern rice) τ52 0.00** 1.98
ln(Irrigation/Land) x ln(Wheat) τ53 0.001** 2.22
ln(Irrigation/Land) x ln(Cash crops) τ54 -0.00*** -2.75
ln(Pesticides/Land) x ln(Traditional rice) τ61 0.00*** 2.64
ln(Pesticides/Land) x ln(Modern rice) τ62 0.01*** 2.97
ln(Pesticides/Land) x ln(Wheat) τ63 0.01** 2.46
ln(Pesticides/Land) x ln(Cash crops) τ64 0.01*** 4.30
ln(Animal power/Land) x ln(Traditional rice) τ71 0.01 1.06
ln(Animal power/Land) x ln(Modern rice) τ72 0.06*** 7.03
ln(Animal power/Land) x ln(Wheat) τ73 -0.02 -0.97
ln(Animal power/Land) x ln(Cash crops) τ74 -0.00 -0.57
Model diagnostics
Gamma γ 0.79*** 12.31
Sigma-squared σs2 0.09*** 3.10
Log likelihood -416.06
χ2(65,0.99) 1672.70***
Inefficiency effects function
Constant δ0 0.46*** 4.80
Female headed households δ1 0.00 0.01
Age of the farmer δ2 -0.00*** -3.40
Family size δ3 -0.03*** -4.03
Highest level of male education δ4 -0.02*** -3.75
Highest level of female education δ5 -0.01* -1.90
Share of women labour δ6 -0.50*** -3.37
Tenurial status δ7 0.21*** 5.87
Jessore region δ8 -0.03 -0.87
Herfindahl index of crop diversification δ9 0.13** 2.31
Note: *** significant at 1 percent level (p<0.01)
** significant at 5 percent level (p<0.05)
* significant at 10 percent level (p<0.10).

41