Determinants of household activity choice, rural income strategies and diversification - Institutional Constraints to Small Farmer Development in Southern Africa - page 58

3.3 Model structure and data

This study employs a model that enables us to calculate the determinants of income
from different income generating activities practiced by households in the study area.
Classification of sources of income in this study follows almost similar classification as that
adopted by Fraser et al. (2003) in Guquka and Koloni villages. These villages are in the
same district with the present study area. The following classification of activities was used:
1. Own agriculture income (income earned from agriculture-crop, poultry and livestock).
2. Own business income (income from activities such as industry, transportation,
construction, and services).
3. Remittances (money or goods received from migrant household member by the
household).
4. Wage income (income earned from formal or informal wage employment, including
salaries, allowances, bonuses, and other kinds of remuneration).
5. Inheritance and grants - incomes without quid pro quo , such as pensions, transfers,
grants/subsidies, rents, and financial income.

With the view of tracking the determinants of income received by each household in
2008, separate simultaneous equations were modeled for each activity. The equations
were developed to estimate the effects of household members' characteristics, household
variables and community level variables on household income. The estimates of income by
activity are derived using the Probit model, and represented as:

Y j =β 0 + β i X i + μ i (1)

where:
Y j = the dependent variable representing income from each income category;
β 0 = the constant term;
β i = the vector of coefficients;
X i = the vector of explanatory variables;
μ i = the error term.

Increases in households income can only reduce poverty if it benefits the poor sect of
the society. An attempt was made to model poverty directly as function of household
characteristics, household assets and community characteristics (Fan et al. , 2002). In this
case household poverty is a binary variable representing either household whose annual
per capita income falls below the poverty line or above. Using ordinary least squares
(OLS) will result in biased estimates. Therefore, a Probit model is used to estimate the
poverty determination equation. The Probit model was necessary to avoid selection bias in
estimating income distribution within the communities (Yúnez-Naude and Taylor, 2001).