3.

4 Methods of Data Analysis

To address the objectives of the research and to analyze the data, both descriptive and
Econometric methods were employed. Accordingly, in the descriptive part, simple measures of
central tendency, frequency and percentages were used; and in the Econometric analyses, a
stochastic frontier approach were used to estimate the level of technical efficiencies and the
relation between clients level socio-economic and institutional variables and inefficiencies.

3.4.1 Efficiency Estimation

In this study the stochastic frontier approach was used for its key features that the disturbance
term is composed of two parts, a symmetric and a one sided component. The symmetric
component captures the random effect outside of the control of the decision maker including the
statistical noise contained in every empirical relationship particularly those based on cross-
sectional household survey data. The one sided component captures deviations from the frontier
due to inefficiency.

The biggest advantage of the stochastic frontier approach models is the introduction of a
disturbance term representing noise, measurement error and exogenous shocks that are beyond
the control of the production unit in addition to the efficiency component. Hence, Technical
efficiency measures obtained from stochastic frontiers are expected to reflect the true ability of
the clients given the services.

The assumption that all deviation from the frontier are associated with inefficiency, as assumed
in DEA, is difficult to accept, given the inherent variability of agricultural production due to a lot
of factors like weather, pests, diseases, etc (Coelli and Battese, 1995). Furthermore, because of
the low level of education and high illiteracy among farmers in developing countries, keeping
accurate records is not a common practice. Thus, most available data on production are likely to
be subject to measurement errors.

Smallholder farmers in Ethiopia in general and in the study area in particular are characterized
by low level of education and keeping of records is thus non-existent. Moreover, there is high
variability of agricultural production due to weather fluctuations. Therefore, within the stochastic
frontier framework, the stochastic efficiency decomposition methodology is chosen as more
appropriate for this study.

The stochastic statistical frontier method requires a prior specification of the functional form,
among others, Cobb-Douglas, Translog, etc. However, recent advances in developing new
functional forms have been dominated by efforts to conceive “flexible” forms. As a result,
flexible functional forms such as the Translog form are usually recommended rather than the
restrictive Cobb-Douglas form (Greene, 1980). The Translog function is the only one of the
flexible functional forms that is readily used for direct estimation of the production function.

In fact, in this study the likelihood ratio test was conducted to select the appropriate functional
form that best fits the data. The value of the generalized likelihood ratio (LR) statistic to test the
hypothesis that all interaction terms including the square specification is equal to zero (H O:

ij=0) is calculated as:

LR= -2(LC-LT) ------------------------------------------------------------------------------- (1)

Where:
LR= Generalized log-likelihood ratio;
LC= Log-likelihood value of Cobb-Douglas frontier; and

LT= Log-likelihood  value of Translog frontier.

This value is then compared with the upper 5% point for the 2 distribution and the decision is
made based up on the model result. If the computed value of the tes is bigger than the critical
value, the null hypothesis will be rejected and the translog frontier production function better
represents the production technology of farmers.

Simple Correlation Analysis was used for checking the presence of a serious problem of
multicollinearity and the variables were selected accordingly.

The farm’s technology is represented by a stochastic production frontier as follows:

Yi= (i; ) +εi--------------------------------------------------------------------------------(2)

Where Yi denotes output of the farm; i is a vector of functions of actual input quantities used by

the ith household;  is a vector of parameters to be estimated, and εi is the composite error term

(Aigner et al., 1977; Meeusen and Van den Broeck, 1977) defined as;

εi= vi-ui-----------------------------------------------------------------------------------------(3)

2
Where, Vi’s are assumed to be independently and identically distributed N( 0, σ ν) random

errors, independent of the μi’s; and the distributional assumption of the technical inefficiency

term, ui, was estimated using the likelihood ratio test.

The subtraction of i from Vi implies that the logarithm of production function is smaller than it

would otherwise be if technical inefficiency did not exist.

The above parametric models are estimated in terms of the variance parameters,

σ2μ
γ=
σ 2=σ 2 ν + σ 2 μ and σ 2 ν +σ 2 μ . The parameter  measures the discrepancy between

frontier and observed levels of output and is interpreted as the total variation in output from the

frontier attributable to technical inefficiency. It has a value between zero and one. The value of

zero indicates that the non-negative random variable, ui , is absent from the model while the

value of one shows the absence of statistical “noise” or exogenous “shocks” from the model and

hence low level of farm’s production compared to the” best “ practice (the maximum output) of

the other farm that is totally a result of farm specific inefficiency.

More formally, to test whether technical inefficiency is absent and hence the conventional
(average) production function is appropriate or not, we can use the likelihood-ratio test using the
log- likelihood values of the OLS and the MLE. The test is presented in the result and discussion
part.

Given the specification of the stochastic frontier production function, as defined by equations (5)

and (6), the technical efficiency of the ith farmer is

TEi = EXP (-ui)---------------------------------------------------------------------------(4)

Where, uis are non-negative random variables, which are assumed to be independently

2μ .
distributed with mean ui and variance σ i

The Cobb-Douglas full frontier in its logarithmic form is specified as follows:

ln y = + ln X +ε ---------------------------------------------------------------------(5)
i 0 i i i

The translog full frontier in its logarithmic form is specified as follows:

ln y = + lnX +½ lnX lnX + ε --------------------------------------(6)

i 0 i i ij i j i

ln y = + lnL + lnP + lnA + lnS +(v -u )---------------------(7)

i 0 1 i 2 i 3 i 4 i i i

Where;
Yi= Quantity of output of local maize production.
ln= Natural logarithm
L=Amount of labor in man- days
P= Draught power used in oxen days
A= Area planted with maize crops in ha.
S= Amount of seed used in Kg.
i= Parameters to be estimated

Vi= Symmetric component of error term;

Ui= Inefficiency component of error term; and

i = number of farmers (1-120 sample respondents)

The technical efficiency effect model (Coelli and Battesse; 1995) in which both the stochastic
frontier and factors affecting inefficiency are estimated simultaneously is specified as follows.
In Cobb-Douglas functional form:

lnyi = 0+1lnLi+2lnAi+3lnSi+vi-(δ0+δ1Lvi+δ2Xti+ δ3Edi+ δ4Fmi+

δ5Agi+δ6AgSq+δ7Fri+δ8Cri+δ9Ni+δ10OFFi+δ11Sei+δ12Pri+δ13FS

+W------------------------------------------------------------------------(9)

In Translog functional form:

lnyi=b0+b1lnLi+b2lnPi+b3lnAi+b4lnSi+1/2[b5lnL2i+b6lnP2i+b7lnA2i+b8lnS
2i+b9lnLixlnpi+b10lnLixlnAi+ b11lnLixlnSi + b12lnPixlnAi +
b13lnPixlnSi+b14lnAixlnSi]+vi -
(δ0+δ1Lvi+δ2Xti+3Edi+δ4Fmi+δ5Agi+δ6AgSqi+δ7Fri+δ8Cri+δ9Ni+δ10OFFi
+δ11Sei+δ12Pri+δ13FS ) +W----------(10)

Where,
Yi , ln, i, Li, Ai, Si, and vi are defined as in equation (7)

Lv=Livestock holding; Cr=Access to credit; Pr= Proximity; AGSQ= Age square

Xt=Extension contact; N=Fragmentation; Fm=Family size; Ag=Age of farmers

Fs= Fertility status; OFF= Off/non-farm activity; SEX=Sex of the household head Fr=Farm

size; Ed=Level of education; δi= parameter vector to be estimated and W i=Error term; and

others are squares and interactions terms

The ML estimates of technical efficiency effects of the model given above will be estimated
using a software package FRONTIER VERSION 4.1( Coelli, 1996) which is specifically
designed for the estimation of efficiency.