Productivity and Efficiency of Agricultural Extension Package in Ethiopia

Gezahegn Ayele* Mekonnen Bekele**,Samia Zekeria***

Submitted to: The Ethiopian Economic Association

To be presented on

The Fourth Annual Conference of the Ethiopian Economy

June 2006

May 16, 2006

Addis Ababa, Ethiopia

Abstract

In an objective to see the impact of extension package program on productivity and efficiency of Ethiopian agriculture, the study used data collected in 2001 by Ethiopian Development Research Institute. The estimated total factor productivity using Tornqvist total factor productivity index shows that the total factor productivity of maize, teff and wheat extension farmers on average exceed that of non-extension while there are no clearly discernable differences in efficiency. Determinants of both total factor productivity and technical efficiency are identified. Most importantly, the role of agro-ecologies in influencing both total factor productivity and efficiency is significant, implying the need for considering agro-ecology differences in introducing extension package technologies.

1. Background

It is always claimed that Agriculture remained to be the mainstay of Ethiopian economy despite the dismal performance of the sector. Various factors were held responsible for poor performance, despite the attempts were made to modernize it. In an effort to change the living standard of the population and to transform agriculture, the government declared the Agricultural Development Led Industrialization (ADLI) in 1993, recognizing the country’s economic problem is deep rooted in agriculture. The measure is expected to improve the manufacturing sector simultaneously.

One of the major programs in Ethiopian agriculture is the extension package that provides with modern agricultural technologies and intensifies agriculture. The major outcome of the free market economy after 1991 is Ethiopian peasants can produce from their holdings and sell at the price they choose. Despite the introduction of new technologies in the 1970s & 1980s, the fact that farmers were discouraged to market their output freely constrained output & productivity i.e. the prevalence of technology in agriculture may not attain its target goal unless the land tenure, social, institutional, cultural and commercial, etc grounds are appropriate (Todaro, 2003). In this regard, the objective of the extension package is given the land tenure, institutional and commercial grounds, it is possible to provide and direct the farmers with the appropriate technology and skill so that the level of productivity will rise and bring in income. Otherwise food self-sufficiency is difficult to achieve both in the long-term and short. Given an irreversible trend of declining size of cultivated land, the only feasible way to raise production is to increase land productivity if China doesn’t want to rely on large-scale imports to feed her huge and still growing population (Yao and Liu, 1998). Similarly, due to land shortage, cropping systems in Africa is in transition from farm abundant to land constrained (Reardon et al 1996). Evidences therefore, suggest the need for rising productivity and the adoption of various alternative strategies.

Since 1995/96-cropping season when PADETS became operational, fertilizer and improved seeds have witnessed widespread & increasing rates of adoption, despite the removal of all input subsidy since 1997/98. Between 1995 and 1999, the consumption of fertilizer increased from 35,272 to 2,168,756 quintals. In the same period, improved seed application rose from 11,043 to 177,783 quintals. The number of participating farmers leaped from 31,256 to 3,731,217 covering nearly 40% of the farming population The value of credit, which began at 8.1 million, has reached 150.2 million. Demonstration plots in the fields of farmers covered by the package rose to at 3,807,658. In terms of its spread in hitherto unknown areas, adoption rates of new varieties & fertilizer, diffusion and increased yield rates resemble green revolution in cereals (Tenkir, et al., 2004).

The coming into the scene of the level of technology has to change the production frontier of farmers. While some indicators of adoption levels have been treated to some extent, however, its effect on the level of productivity has not been sufficiently treated by researchers. This study therefore fills the gap thereby looking into the impact of the technology packages on the productivity & technical efficiency of the farmers.

2.Objective of the Study

The study has the following general and specific objectives.

1) To assess the total factor productivity the performance of extension participating farmers in comparison to non-extension farmers.

2) To estimate the technical efficiency for both extension and non-extension farmers1 and identify determinants.

3. Conceptual Frameworks

Total Factor Productivity (TFP)

The economic theory of production has provided the analytical framework for most empirical research on productivity. The cornerstone of the theory is the production function, which postulates a well-defined relationship between output and factor inputs. Productivity can be achieved from two sources; first, through technological change of using improved practices of production such as ploughs, fertilizers, pesticides, improved seeds, etc which pushes the production frontier upward; and second, if the farmer has got further skills in using the existing techniques of production, productivity will increase.

Measuring productivity is conceptually better understood when total factor productivity (TFP) is measured empirically. Total factor productivity is the ratio of aggregate outputs to aggregate inputs. Some studies use interspatial measures of total factor productivity based on Divisia Index as defined by Denny and Fuss (1980), where efficiency is estimated for different kinds of land contracts. The TFP approach is found to be suitable for cases where the complexity and diversity of smallholder farming, like in Ethiopia, is large; it also makes comparison possible among different farming systems. The superiority of the method of TFP over the conventional method of measuring land and labour productivity emerges from the fact that the later is misleading if there is high substitutability between inputs (Gavian and Ehui, 1996). Within the TFP methods, there are different kinds of measurements that need to be seen from various methodological perspectives.

Most of the empirical literature focused on productivity of individual factor productivities in Africa such as labour and land productivities and some of those studies got strong evidence that fertilizer and improved seeds are associated with higher yields; and considerable yield variability across fields within a given technology type (Howard et, 1999). Reardon et al 1996, for instance, discussed returns per labour day and output per hectare of wheat maize and soybeans are generally low for some African countries and the yields differ by crop, zone, technology and farm size; determinants of productivity according to this evidence are many.2 Moreover, they indicated that policy reform (exchange rate, interest rate and market liberalization) is necessary but not a sufficient condition to spur productivity.

Another study in Ethiopia found that tenure difference in terms of “rented-in” and “owned” has significant effect on sorghum and wheat output while there is no significant impact on teff and maize output (Abebe & Negussie, 2005). The same study revealed that at regional level land fragmentation (number of parcels) and land conservation has positive relationship with the sample crops yields, remarking a possible difference at zonal level in the case of land fragmentation. An empirical study using discriminant analysis of participants and non-participants in extension package program in Oromia region indicates that the yields of maize and wheat from plots of National Extension Package participants as compared to non-participants in the study area is found to be as high as 50% for maize and 39% for wheat compared to yields of the same crops from the non- participant farmers, with insignificant difference for teff & sorghum (Samia & Habe, 2005). However, most of those studies conducted in Ethiopia, have focused on the technical efficiency, and not so much on factor productivity. When we come to the indices, one of the indexes of measurement of TFP is the Tornqvist quantity index, shown below.

Tornqvist quantity index:

Q_st=, is the Tornqvist quantity index where:

= arithmetic mean of output shares;

= arithmetic mean of input shares; P_it and Q_it are price and quantity of commodity i at time t respectively.

The estimated value of the index tells us the direction of change of TFP (Collie, 1998). In this study, after estimating the TFP, we will run linear regression model to identify factors influencing TFP; as in many studies in Ethiopian agriculture (Gezahegn, 2002).

Technical Efficiency

In estimating the frontier, we use the model derived by Battese & Coelli (1993,1995):

Y_i = F (X_i; b) + e_{i ;} e_i = V_i - U_i; where U_i ³0

Where, Y_i: output of the farm i=1,2,…N

F (…): is the production technology

X is vector of N inputs

b is vector of unknown parameter to be estimated

e_i is the error term with two components of:

V_i : is non-negative error term(due to the decision or action of the farmer);

U_i: the technical inefficiency component (factors out of control of the farmer / decision maker.

U_i = ådZ_i + w_i, U_i ³0; where Z_i factors affecting the technical efficiency of the farm and d is parameter.

The symmetric random error Vi accounts for random variations in output because of factors, such as, measurement error, exogenous shocks; etc, which is not under the control of the farmer and it is assumed to be independently and identically distributed as N (0,s²_vi). Moreover, the asymmetric non-negative random error, U_i measures technical inefficiency relative to the SF and is assumed to be to be independently and identically distributed non-negative truncations (at zero from below) of the N (m, s²_ui) distribution. The variance parameter of the model is parameterized as:

= + and g =; 0<g<1:

= y_i -u_i =f(x_i; b)-v_i , after finding the estimates of u_i and v_i;

Where,

: is the observed output of the i^th farm household adjusted for the stochastic random noise captured by u_i; this equation is used to derive the technically efficient input vector and to derive algebraically.

The model we use matters in measuring the efficiency of firms (Liu, 2005). There are two common functional specifications the Cobb-Douglas stochastic frontier & the translog. Cobb-Douglas is production function is criticized for its rigidity flexible despite the multi-colinearity problem. The functional form of the stochastic frontier is determined by testing. Thus, the frontier models estimated are defined as:

y_it = β_o + + V_it- U_it (Cobb-Douglas) ; and

y_it = β_o + + + V_it- U_it (Translog).

We select the appropriate model specification through tests. Wald tests are commonly used to test the null hypothesis of no inefficiency, i.e., that the variance of the one-sided process is zero. However, additional Monte Carlo experiments show that the size properties of this test are very weak (STATA, 2003). The estimation of truncated-normal distribution stochastic frontier model and the log likelihood test makes continuous iteration and attaches the maximized iteration, which is used to calculate the log likelihood statistics. The likelihood-ratio test statistic λ = -2{log [Likelihood (H_o)]– log [Likelihood (H₁)]} has approximately adistribution with q equal to the number of parameters assumed to be zero in the null hypothesis; it is compared with the critical values of the_distribution and decided between the two models. The power of the LR test is increased by testing jointly the null hypothesis that γ = δi =0, for all i, meaning that neither the constant term nor the inefficiency effects are present in the model; since g takes values between 0 and 1, any LR test involving a null hypothesis which includes the restriction that γ = 0 has been shown to have a mixed χ² distribution, with appropriate critical values (Kodde and Palm, 1986) in quoted in Piesse, et al (2002).

The technical inefficiency effect term is distributed N (m_i,s ²_v) where m_i can be specified and defined as:

m_i= βo + Z_ij ; where Z_j are socioeconomic & infrastructure variables which are identified in the literature or taken from the observation of the researcher.

The estimation of the inefficiency model has two approaches. The first is simultaneous equation modeling (Battese & Coelli, 1995) and the two-stage modeling (discussed above). The advantage of the simultaneous equation technique over the two stages is that it incorporates farm specific factors in the estimation of the production frontier because those factors may have a direct impact on efficiency (Wadud, 2002). The estimates for u_i & v_i are found from the SF model & the technical efficiency predictors by replacing parameter by their maximum likelihood estimates. We use the maximum likelihood estimation to identify the determinants, considering the choice of model is controversial (Batsse et, 1995).

Data

The May 2001 survey data generated by EDRI was combined with the survey conducted during the 2001/02 by the Central Statistical Authority’s (CSA) on Ethiopian Agricultural Sample Enumeration extension package data, mainly for the yield data.

The sampling method used is systematic sampling, which includes random selection of farm households in the order of zones from the four major regions of the country, which represents over 85% of the population. Two woredas were selected from each zones depending on the level of adoption of modern input technology. Within the woredas, PAs are randomly selected through stratified sampling, and then farm households are selected randomly through systematic sampling. As the objective of this study is to estimate Total Factor Productivity (TFP) and Technical Efficiency (TE), we took the sample for each crop out of the national sample of 1921 farm households (Agricultural Extension Survey of EDRI of the 2002); until we get complete data set on the variables required to estimate TFP differential between extension & non-extension farmers (Tornquvist index), we continuously reduced the sample size and finally we arrive at the matching sample size for each crop i.e. 115 for maize, 55 for wheat and 112 for teff. For technical efficiency estimation, however, the sample size is relatively larger i.e. 186 for maize & 244 for teff.

4. Analysis of Data

4.1. Descriptive

Table-1 depicts the socioeconomic characteristics of sampled households. Household size has no significant effect between extension and non-extension farmers, with higher mean for extension farmers in case of wheat and teff, which are relatively more labour and technology intensive. The number of livestock the households own has mixed features; first, for maize and wheat farmers, the average number of livestock is higher for adopters, which may show the need for higher wealth income to purchase the input packages. On the other hand, the number of livestock in case of teff is higher for non-extension farmers, but outliers affect it (standard deviation is 12.10 for Extension farmers compared to 35.44 to that of non-extension farmers). The higher mean livestock ownership in some cases implies there is disincentive to adopt modern technologies.

Table-1: Extension & Non-extension Farmers: Quantitative Variables

Corp/variable	Extension				Non-extension farmers
Maize (Maximum N=115)	Mean		SD	Median	Mean	SD	Median
Age of Head	41.94	14.33		40.00	44.00	14.00	40.00
Household size	6.28	2.17		6.00	6.0	2.00	6.00
No of Male	3.50	1.93		3.00	3.00	1.00	3.00
No of Female	2.97	1.66		3.00	3.00	1.00	3.00
No of livestock	5.44	4.68		4.00	4.48	3.48	3.00
Wheat (Maximum N= 55)
Age of Head	43.21	13.89		44.00	45.64	16.72	43.00
Household size	5.75	2.09		6.00	5.32	1.65	5.50
No of Male	3.14	1.53		3.00	3.00	1.29	3.00
No of Female	2.63	1.45		2.00	2.32	1.11	2.00
No of livestock	5.40	4.17		5.00	4.33	3.00	3.07
Teff (Maximum N=112)
Age of Head	47.0	13.00		45.00	45.00	14.00	46.00
Household size	6.00	2.00		6.00	5.00	2.00	5.00
No of Male