Agriculture is considered as a strong option and fundamental instrument for spurring growth and

sustainable development, poverty reduction, and enhancing food security in developing countries
like Ethiopia.

ABSTRACT
This study will be examines the contribution of household livelihood diversification on poverty
in Kamba woreda, Gamo Gofa zone, Southern Ethiopia. To accomplish this, the study will be
use cross-sectional data for 186 household heads during 2018/2019. The primary data will be
collected using mulit-state sampling technique at first stage purpose sampling technique will be
used to select representative kebeles and in the second, probability proportional to size will be
employed to select representative sample from those kebeles, ultimately simple random
sampling will be used to select household heads. Supplementary, secondary data will be
collected from various sources. Data will be analyzed using a modified multidimensional
poverty index which is the most resent approach to poverty assessment, descriptive statistics
and econometrics model with STATA version 11/13. For descriptive statistics measures such as
central tendency and measure dispersion will be used to summarize some important
characteristics of the sampled households. The contribution of livelihood diversification on
poverty will be determined by using econometric model (Tobit regression model). The study
will be tried to determine the poverty status of the households’ using weighted deprivation
score (which is the proxy of multidimensional poverty index). Furthermore, the study will
determine the effects of socio-economic variables (age of the household head, educational
status of household head, sex of household head, household size, number of dependent and
livelihood diversification) and resource endowment (asset ownership and farm size) and
institutional factors (access to electricity) on poverty.

Key terms: Multidimensional poverty index, livelihood diversification, weighted deprivation

score, Tobit regression model

Tobit Regression Model

Measuring the effects of livelihood diversification on poverty is the main objective of this
study. To do this tobit regression model will be employed. The tobit regression model (limited
dependent variable model) is a statistical model proposed by (Tobin, 1958) to describe the
¿
relationship between a non-negative dependent variable y i and an independent variable(or
 vector) x i. The term tobit was derived from Tobins name by truncating and adding it by
analogy with the probit model (IESC, 2008).This model is applied when the dependent variable
is continuous, but its range may be constrained. It is also known as a censored normal
¿ ¿
regression model because some observation on y i (those for which y i ≤ 0) are censored (we are
not allowed to see them), (Maddala, 1992:338).
¿
The models suppose that there is a latent (i.e. unobserved) variable y i . This variable linearly
depends on x i via a parameter (Vector) β which determines the relationship between the
¿
independent variable (or vector ) x i and the latent variable y i (just as in a linear model). In
addition, there is a normal distributed error term ui to capture random influences on this relationship.
The observable variable y iis equal to the latent variable whenever the latent variable is above zero, and
zero otherwise.

The tobit regression model formulated as follow:

{
¿ ¿
yi= y i if y¿i > 0
0if y i ≤0

{
¿ ¿
y i =β x + ui if yi >0
yi= i

0 if y ¿i ≤0

ui ∈(0 , δ 2)…………………source: Maddala (1992.pp. 339-341).

We cannot use OLS estimation using the positive observations yi , because when the error term
ui does not have zero mean. Since observations with y ¿i ≤ 0 are omitted, it implies that only

observations for which ui >−β x are included in the sample. Thus the distribution ofui , is a
i

truncated normal distribution and its mean is not zero.

Therefore, methods of estimation commonly suggested for the above model is the maximum
likelihood (ML) method which is as follows;

Note that we have two sets of observations;

1. The positive value of y, for which we can write down the normal density function as usual. We
noted that ( yi− β x )/δ has a standard normal distribution.
i
 ¿ ui
2. The zero observation of y for which all we know is that y i ≤ 0 or β x +ui ≤ 0 . Since has a
i
δ
ui
standard normal distribution, we will write this as ≤(β ¿ ¿ x i)/δ ¿ . The probability of this can
δ
be written as (−β x )/δ , as F (−β x )/¿is the cumulative distribution function of the standard
i i

normal.

The density function of the standard normal by f (.) and the cumulative distribution function of
F (.) .

2
1 −t
f ( t )= exp ⁡( )
√2 π 2

Figure.2.4. Truncated norimal distribution

z

and F ( z )= ∫ f (t)dt
−∞

Using this notation we can write the likelihood function for the tobit model as

Maximizing this likelihood function with respect to β∧α , we get the maximum likelihood
(ML) estimates of these parameters(Ibid).
3.7. Statistical and Specification Tests
Before executing the final model regressions, all the hypothesized explanatory variables will be
checked for the existence of statistical problems such as multicollinearity problems. Basically,
multicollinearity may arise due to a linear relationship among explanatory variables and the
problem is that, it might cause the estimated regression coefficients to have wrong signs,
smaller t-ratios for many of the variables in the regression and high R 2 value. Besides, it causes
large variance and standard error with a wide confidence interval. Hence, it is quite difficult to
estimate accurately the effect of each variable (Gujarati, 2004; Woodridge, 2002).

There are different methods suggested to detect the existence of multicollinearity problem
between the model explanatory variables. Among these methods, variance - inflating factor
(VIF) technique is commonly used and will also employed in the present study to detect
multicollinearity problem among continuous explanatory variables and it was defined that VIF
shows how the variance of an estimator is inflated by the presence of multicollinearity
(Gujarati, 2004).

Mathematically, VIF for individual explanatory variable (Xi) can be computed as (ibid):

1
VIF ( Xi)= 2
1−R

Where: R2 is the coefficient of correlation among explanatory variables.

According to Gujarati (2004), the larger the value of VIF indicates the more collinearity among
one or more model explanatory variables. As a rule of thumb, if the VIF of a variable exceeds
10, which will happen if a multiple R-square exceeds 0.90, that variable is said be highly
collinear.

Alternatively, we can use the inverse of VIF (1/VIF) called Tolerance (TOL) as a measure of
multicollinearity. The closer is TOL of one explanatory variable (Xi) to zero, the greater the
degree of collinearity of that variable with the other repressors’. On the other hand, the closer
TOL of Xi is to 1, the greater the evidence that Xi is not collinear with the other repressors
(ibid).
The multi-stage sampling design will be used to select the sample households. In the first stage,
three kebeles (namely Balta yele, Mero and Maze), will be purposively selected according to
agro-ecological zone of each kebeles. In the second stage, the respondents from each Kebeles
will be drawn by using combination of simple random sampling and population proportional to
size(PPS)technique ,accordingly proportionate to size technique will be applied to determine
sample households size from each kebeles. Ultimately, a total of 186 sample household heads
will be selected by using simple random sampling technique.

From the total of 2077 households in sampled three kebeles, according to their agro-
ecological zone, about 186 total sample households will be selected through simple random
sampling technique after obtaining number of total sample size through a basic formula
provided by (Yamane,1967), and will be distribute total sample size for each sample kebele
through probability proportional to size of each kebele.

Thus to determine the required sample size by using Yemane formula at 95% confidence level,
degree of variability = 0.5 and level of precision =7% are used to obtain the sample size which
inference the true population.

N 2077
n= 2 = 2 = 186
1+ N ( e ) 1+ 2077((0.07) )

Allocating the sample across the kebeles will be done through N- proportional to size allocation
technique. Thus, to select representative sample from each sampled kebeles will be taken as
follows:

Table.3.1 N-Proportional to Size Allocation of Sample across Representative Kebeles

Kebeles Total Number of Total Number of sample households

households in each
kebeles
1. Balta yele 487 43
2. Mero 800 72
3. Maze 790 71
Total 2077 186
Source: Own computing from (KWFEDODISD, 2017/2018).
WDS of 33.3 percent - 49.9 percent( in this study.

Tobit model specification:

Household Size (HHs):

non-poor (discrete) and WDS >33% and above as poor (continuous).

the contribution of household’s livelihood diversification on poverty