A comparative analysis of technical efficiency of smallholder tobacco and maize farmers in Tabora, Tanzania

Introduction

The agricultural sector in Tanzania plays an important role for the overall economy through its significant contributions to rural employment, food security, and provision of industrial raw materials for other sectors in the country. The performance of the overall Tanzanian economy has been driven by the performance of the agricultural sector, due to its large share in the economy. Agriculture in Tanzania employs the majority of the poor, and has strong consumption linkages with other sectors. In 2011, the sector contributed approximately 51 percent of foreign exchange, 75 percent of total employment and 27.1 percent of the Gross Domestic Product (GDP) dropping from 47 percent in 2004 (World Bank, 2013). Smallholder farming dominates agricultural production, and a large proportion is for subsistence. Since poverty is predominantly a rural phenomenon, and agriculture is a major economic activity for rural population, it follows that success in poverty reduction depends critically on performance of the agricultural sector. In terms of growth, the sector has achieved significant success in recent times as it has grown at an average of 4.1 percent from 1998 to 2007 (Kuboja, Kazyoba, Lwezaura and Namwata, 2012). One of the cash crops that help generate foreign exchange earning in tanzania is tobacco. Tanzania ranks as a third African country- after Malawi and Zimbabwe - As a major producer and exporter of Tobacco. Tobacco is also consumed by Tanzanians with a prevalence rate of 10.8 percent.

In the major tobacco producing region of Tabora and other places government, extension agents and, companies are encouraging farmers to produce more tobacco by availing credit to purchase fertilizer and seeds. However with regards to efficiency in the production of tobacco there a lot to be desired. Setting the negative health consequence aside, if one is to undertake benefit cost analysis tobacco farming may not be the better option for small scale farmers. Tobacco is alabor intensive where farmers are in the field for 10 hours per day and with a 10 month period from land ploughing to harvesting. On the other the gestation periods for annuals such as maize or ground nuts is less than four months with relatively less labour input. In other words it is possible to two or more harvest per year for maize and groundnuts.

We thus hypothesize that faremers are better being in crops other than tobacco and that tobacco production is less efficient. This study aims at comparing production efficiency between tobacco and maize in the Tabora region of Tanzania . The aim is to investigate whether tobacco farmers are better off growing tobacco as compared to maize - the main staple diet of Tanzanians

Objectives and Motivation

1) Objectives

The main objective of this study is to empirically determine and compare the efficiency of tobacco and maize farmers in Tanzania. Specifically, the study seeks to:

• ➢ estimate a frontier production function for maize and tobacco and identify which is more efficient

• ➢ analyze the determinants of efficiency for the two crops.

2) Motivation

The motivation of this study emanates from the fact that both tobacco and maize production is important in the economy of Tanzania in general and in Tabora region in particular. The market value for one kg of tobacco is three times that of maize. On the other hand, tobacco farming is more labor time intensive and hazardous. It may not be sufficient to compare the gross revenue from tobacco with that of maize and conclude that farmers are better off cultivating tobacco/ Setting aside the negative health consequence of tobacco production and consumption one should also take the cost of production and compare the net revenue. Alternately one may compare efficiency in the production of maize and tobacco. This study will attempt to estimate frontier production function for tobacco and maize and compare the results. The study is motivated by the fact that there have been a few studies on technical efficiency in the Tanzanian agricultural sector (Msuya et al., 2006; Msuya et al., 2008). None of these has been in the area of tobacco production Therefore, an empirical study to investigate technical efficiency in Tobacco and Maize cultivation in Tabora is a necessary first step in the national effort. Findings from such a study will help to improve resource use efficiency in specific production areas/zones, boost production, increase the contribution of agriculture to GDP and enhance the earnings of small scale farmers in the study area.

Analytical Framework

This study employs the stochastic frontier production function as proposed by Battese and Coelli (1995). The application of the function is in accordance with the early applications of Aigner, Lovell and Schmidt (1977); and Meeusen and van den Broeck (1977). Aigner et al, (op.cit) originally developed the model to handle cross-sectional data. The use of this tool of analysis has gained prominence in econometric and applied economic analysis in the last two decades (Amos, 2007). In Tanzania, few studies have applied the tool in the analysis of production functions especially in the agricultural sector. Mbelle and Sterner (1991) applied the model to analyze the importance of foreign exchange in industries; Msuya and Ashimogo (2006) applied the technique to estimate the technical efficiency in Tanzanian sugarcane production using the case study of Mtibwa Sugar Company; Msuya, Hisano and Nariu (2008) used the method to estimate the level of technical efficiency of 233 smallholder maize farmers. Other notable studies include those of Battese (1992), Battese and Coelli (1993; 1995), Ahmad et al (2002), Daramola (2003), Ojo (2003), Amos (2007). All these studies were applied to the agricultural sector¹.

This study applies the stochastic frontier approach mainly because of the following reasons: Firstly, the method is capable of capturing measurement errors and other statistical noises influencing the shape and position of the production frontier (Battese, 1992; Msuya et al, 2008). Battese (op.cit) extensively described techniques (deterministic versus stochastic, parametric versus non-parametric) that could be used to measured relative efficiency. Secondly, the technique better suits an agricultural production largely influenced by random exogenous shocks as the one found in Tanzania. This technique assumes that farmers may deviate from the frontier not only because of measurement errors, statistical noise or any non-systematic influence but also because of technical efficiency.

Model Specification

Following Battese and Coelli (1992), the production function can be specified as follows:

$$Y_t=f(X_i,\beta )+e_i;i=1,2,\dots ,N$$ (1)

Where Y_i epresents the previous potential output level (harvest) from the farms, X_i is a (1×k) vector of inputs and other explanatory variables associated with the i^th farm. β is a (k×1) vector of unknown parameters. The error term i e is composed of two independent elements, i.e., e_i = v_i = u_i, with the v_i term being a random (stochastic) error, which is associated with random factors not under the control of the farmers. It is assumed to be independently and identically distributed as $N(0,{\sigma }_v^2)$, where ${\sigma }_v^2$ stands for the variance of stochastic disturbance. v_i. u_i captures technical efficiency and is a non-negative one sided component associated with farm-specific factors. It is distributed independently from and identically to v_i. If farmers achieve their maximum output, then they would be technically efficient and this means that u_i = 0. u_i is associated with the technical inefficiency of the i^th farm and defined by the truncation (at zero) of the normal distribution $N(z_i\delta ,{\sigma }_u^2)$ where z_i is a (1×m) vector of explanatory variables associated with technical inefficiency of production of farmers; and δ is an (m×1) vector of unknown coefficients.

Following Battese and Coelli (1992) the stochastic frontier production function can be specified in terms of the original values as follows:

$$In{Y}_i=f(X_i,\beta )\exp (v_i-u_i)$$ (2)

The model is such that the possible production Y_i is bounded above by stochastic quantity, f (X_i,β)exp(v_i –u_i), hence the term stochastic frontier.

The technical efficiency of an individual farm from the above specification can be defined in terms of the observed output to the corresponding frontier output, given the available technology (Amos, 2007). The technical efficiency is thus empirically measured by decomposing the deviation into a random component (u) (Ojo, 2003, Amos, 2007).

$$TE=\frac{Y_i}{Y_i^{\ast }}=\frac{f(X_i,\beta )\exp (v_i-u_i)}{f(X,\beta )\exp \left(v_i\right)}=\exp (-u_i)$$ (3)

Where Where Y_i is the observed output and $Y_i^{\ast }$ is the frontier output and w_i being an error term that follows a truncated normal distribution. This is such that 0 ≤TE ≤1. If farmers achieve their maximum output, then they would be technically efficient and this means that u_i = 0

Study Area

The data for this study were collected in Tabora a major tobacco producing region in Tanzania. The unit of observation are small scale farmers. Even though tobacco is the major crop cultivated, farmers are also engaged in the production of other crops especially maize-a major staple diet of Tabora is one of the regions in Tanzania and it is located in the central-western part of the country. With a population of about 2.2 million (National Census, 2012), the region is the 24^th most densely populated with 30 people per square kilometer and a land area of 76,151 square kilometres representing 9% of the land area of Mainland Tanzania. The climate of the area is highly favorable for the agrarian activities of the population who grows crops such as Maize, groundnuts, beans, cassava, and Tobacco. The annual rainfall is between 700 mm and 1000 mm, with the daily mean temperature around 230C (The Planning Commision of Tanzania, 1998).

The data for this study were collected from a randomly selected small scale farmers in 2013. Data were collected with the use of a structural questionnaire designed for collecting information on output, inputs, prices of variables, and some important socio-economic variables about the farmers. The sample size is 306

Table 1 presents as summary statistics of selected varriables

Table 1.

Summary statistics of respondents' characteristics

Variable	Observations	Mean	Percent
Quantity of Harvest (Kg)
➭ Tobacco	259	1022.69
➭ Maize	252	1176.26
Age (years)	134	58
Household Size (Number)	289	6
Farm Size (Acres)	306	9.6
Education level	306
➭ No Education	45		14.71
➭ Primary Education	226		74.83
➭ Secondary and above	32		10.46
Gender	306
➭ Male	227		74.19
➭ Female	79		25.81

Source: Survey data, 2013

From Table 1, the average age of a farmer involved in tobacco and maize cultivation in Tabora region is about 58 years old. In other words, farmers are mature and should be able to make rational decisions about the daily operations in the farms. The mean household size appears to be relatively high : mean acreage planted is also xxxx while mean harvest per acre is xxxx. Only 10.46 of the population appear to have a high level of education while 25.81 percent are female headed household – this is larger than the national average.

Measurement of Variables

Quantity of Output

This is measured by the amount –in kg- of each crop (Tobacco and Maize)

Inputs

Inputs in the production function include, area planted in acres, manpower, fixed assets and expense on fertilizer.

Socio-economic Characteristics

These variables include Gender, Age (years), Level of Education, Household Size, Farm Size (acres),. These variables will act as explanatory variables while estimating the equation on the determinants of efficiency

Method of Analysis

A two stage frontier production function will be estimated. In other words,the following C ob-Douglas frontier production function is estimated

$$InY={\beta }_0+{\beta }_{1}In{X}_1+{\beta }_{2}In{X}_3+v_i-u_i$$ (4)

Where:

In	Denotes natural logarithms;
Y	Total amount of harvest of each crop expressed in kilograms;
X1	Labor input in mandays*
X2	Area of land cultivated in acres
X3	Proportion of fixed assets used**
X4	Cost of fertilizer, pesticides and fungicides
vi	Independent and identically distributed random errors $N(0,{\sigma }_v^2)$. These are factors outside the control of the smallholders.
ui	Non-negative random errors or technical efficiency effects

The second stage of the analysis investigates farm-and farmer-specific attributes that have impact on smallholders’ technical efficiency. The inefficiency function can be expressed as:

$$u_i={\alpha }_0+{\alpha }_1{z}_1+{\alpha }_2{z}_2+{\alpha }_3{z}_3+{\alpha }_4{z}_4+{\alpha }_5{z}_5+{\alpha }_6{z}_6+w_i$$ (5)

Where:

αi's	Inefficiency parameters to be estimated
z1	Gender of the farmer (1=male, 0 female)
z2	Age of the farmer
z3	Dummy variable for smallholder level of education (1= if the farmer has formal education and 0 if otherwise)
z4	Household size (number of people staying together)
z5	Farm size in acres
z6	Air breath (feeling sick) of the person while currying tobacco -variable used only in the tobacco equation (1=feeling sick, 0=otherwise)
z7	Dummy variable assuming a value 1 if land is owned by farmer and 0 otherwise (rented)
wi	An error term that follows a half-normal or a truncated distribution

The Cobb-Douglas production frontier defined in equation (4) and the inefficiency model defined by equation (5) are estimated using the Maximum Likelihood (ML) method.

Results and Discussion

The Maximum likelihood estimation shows the presence of technical inefficiency effects in both tobacco and maize of small holder farmers in Tabora region. This is confirmed by the statistical significance of the coefficients of $In{\sigma }_u^2$ as well as the log-likelihood ratio test of the overall maximum likelihood estimation. The highly significant value of $In{\sigma }_u^2$ suggests the domination of the inefficiency components of the error term. This is true for both tobacco and maize. With the exception of land area all the other significant elasticities suggest values that are too small confirming the inefficiency in the production process

In general the results in Table 2 show positive relationship and statistical significance between the levels of output (tobacco and maize) and labor input, area of land cultivated, proportion of fixed assets used and cost of fertilizer. This scenario is expected as the level of output depends to a certain extend on the quantities of these factors used. However, this can only be up to a level that is considered optimal after which farmers will be operating at a sub optimal level (Amos, 2007).

Table 2.

OLS and MLE of the Production function for Tobacco and Maize cultivation in Tabora Region

	Tobacco		Maize
Variable	OLS	MLS (Half-normal)	OLS	MLS (Half-normal)
loglabor	0.134* (0.0768)	0.0184* (0.0438)	0.0654 (0.0960)	0.0385 (0.0728)
logarea	0.678*** (0.1580)	0.932*** (0.126)	0.648*** (0.1630)	0.972*** (0.0870)
logasset	0.0542 (0.0345)	0.171*** (0.0021)	0.026 (0.0467)	0.0111 (0.0398)
logfertilizer	0.00478 (0.0123)	0.0280** (0.0122)	0.0673** (0.0331)	0.0350*** (0.0112)
Constant	5.046*** (0.6820)	4.894*** (0.0840)	5.088*** (0.5700)	6.364*** (0.3830)
R-sq	0.431		0.277
F(4, 164)	6.38***		9.01***
$In{\sigma }_v^2$		−3.525 (0.3014)		−5.112*** (1.0370)
$In{\sigma }_u^2$		0.412*** (0.1090)		0.224* (0.1230)
σ v		0.0022 (0.0030)		0.0776 (0.4020)
σ μ		0.8137 (0.0443)		1.1186 (0.0686)
${\sigma }_S^2={\sigma }_v^2+{\sigma }_{\mu }^2$		0.6622 (0.0720)		1.2574 (0.1508)
λ = σμ/σv		0.0036 (0.0443)		14.4144 (0.9355)
LR test of σμ = 0		87.82***		29.58***
Observations	169	169	190	190

Notes:

^***p<0.01

^**p<0.05

^*p<0.1

2) Values in parenthesis re standard errors for the ML estimation and robust standard errors for the OLS regression

Levels of Technical Efficiency

Once we estimate the frontier production function and establish the existence of technical inefficiency the next step will be estimate the frequency distribution of technical efficiency (one minus inefficiency) indices. Table 3 presents the results Table 3 shows that the predicted technical efficiencies range between 0.000 and 0.9999 for tobacco farmers and between 0.003 and 0.91 for maize farmers. The mean efficiency for tobacco farmer is 73.9 percent while that of maize farmer is 76.8 percent suggesting that tobacco farmers are less efficient than maize growers. The table also shows the t-test results for equal mean efficiencies, with the null hypothesis of no significant difference in the mean technical efficiencies between tobacco and maize cultivation. The null hypothesis is rejected at a 1 percent level of significance showing that the mean technical efficiencies of tobacco is significantly lower when compared with maize. In other words tobacco farmers can produce the same output with only 73.9 percent of current inputs. The corresponding value for maize is 76.8 percent.

Table 3.

Frequency distribution of technical inefficiency estimates and two sample t-test with equal mean efficiencies

	Tobacco		Maize
Efficiency Level	Frequency	Percentage	Frequency	Percentage
<0.1	1	0.59	1	0.53
0.11-0.20	0	0.00	0	0.00
0.21- 0.30	0	0.00	1	0.53
0.31-0.40	1	0.59	5	2.63
0.41-0.50	2	1.18	9	4.74
0.51-0.60	3	1.78	10	5.26
0.61-0.70	25	14.79	25	13.16
0.71-0.80	62	36.69	61	32.11
0.81-0.90	45	26.63	31	16.32
>0.91	29	17.16	47	24.74
Observ.	169	100.00	190	100.00
Mean	0.7389		0.7683
Min.	0.0000		0.0000
Max.	0.9999		0.9926
	Two sample t-test with equal mean efficiencies
Null Hypothesis		H0: Difference in mean = 0
	t-value		−2.94***

Note:

^***p<0.01

**p<0.05

*p<0.1

The Efficiency Effect Model

The Efficiency Effect Model (equation 5) tries to identify the socioeconomic determinants of efficiency among tobacco and maize farmers in the study area. The results are given in Table 4

Table 4.

Determinants of Technical efficiency

Variables	Tobacco	Maize
Gender	0.0152 (0.0239)	0.0146*** (0.0363)
Age	0.0009* (0.0009)	0.0011*** (0.0014)
Noneduc	−0.0008 (0.0659)	0.0149*** (0.1070)
Primeduc	−0.0309* (0.0649)	0.0045*** (0.1000)
Hhsize	0.0017** (0.0049)	−0.0026*** (0.0070)
Farmsize	0.0009* (0.0016)	−0.0006*** (0.0019)
Airbreath	−0.0249** (0.0105)
Constant	0.7495*** (0.0985)	0.515*** (0.1860)

Note:

^***p<0.01

^**p<0.05

^*p<0.1; Standard errors in parentheses

From Table 4, age, primary educational attainment, household size, farm size and air breath (sickness caused by the process of curing tobacco) are the major determinants of efficiency of tobacco farmers; whereas only age, household size and primary educational attainment of farmers significantly caused inefficiency in maize cultivation. While variables such as no educational attainment, air breath reduce the efficiency level of tobacco farmers, other variables such as, primary educational attainment, household size and farm sizes were observed to increase the efficiency level of tobacco farmers. On the other hand, farm sizes and no educational attainment reduce the efficiency of maize farmers in the model. Other variables are observed to increase the efficiency of maize farmers.

These results are plausible for the age of farmers given that the majority of farmers are old and may not be willing to try or adopt new innovations or some of the farmers are less efficient in the supervision role of their farms. As concerns household size, the major reason while farmers have many household members is for the provision of farm labor. Thus the bigger the household size, the more labor is available for farming operations and hence increasing the efficiency of farmers.

More so, technical efficiency should increase with the level of education of the farmers. This is so because being educated or being able to read or write increases the possibility of learning new farming techniques that will likely increase the efficiency of farmers. The negative coefficient of primary educational attainment indicates that farmer's education is an important variable in enhancing maize cultivation in Tabora. Previous studies obtained similar statistically significant results (Msuya and Ashimogo, 2006; Amos, 2007; Msuya et al, 2008).

The signs on gender coefficient show male farmers are efficient in Tobacco and maize cultivation even though not significant. Some studies have found similar results (Kibaara, 2005; Msuya et al, 2008). However, other studies have also reported no statistically significant results of the effect of gender on efficiency (Tchale and Sauer, 2007). Therefore, this study contributes to the ongoing debate on the role of gender in smallholder efficiency.

Conclusion

The issue on whether farmers are better off by engaging in the production of tobacco as compared to other annuals and perennials has been addressed in many instances. When the earnings from tobacco are compared to the other crops such as maize the former appears to generate more earning. This scenario appears to be reversed when the corresponding input costs are taken into consideration. In other words when net earning is estimated on per acre or per manpower it appears that farmers in the study region are better off being engaged in non tobacco annual or perennial crops. This finding does not take into consideration various health hazards associated with tobacco production.

In this study we tried to compare production efficiency between tobacco and maize and were able to establish that engaging in the production of tobacco as compared to maize is not a worthwhile undertaking. Farmers in the Tabora region appear to be relatively more efficient by being engaged in the production of maize as opposed to tobacco.

When the determinants of efficiency equation was estimated for tobacco growers the effect of tobacco curing appears to significantly reduce efficiency. The findings from this study may enable policy makers to reconsider the prevailing notion that farmers are better off being engaged in the production of tobacco and that the foreign exchange earning fhe ountry will be enhanced