Model-based estimation of annual per capita income based on the results of the labor force survey for the estimation of annual poverty statistics

This study focused mainly on the construction of predicting model for annual per capita income using the ordinary least squares method in estimation and with variables from the Labor Force Survey (LFS) as possible predictors. Since the National Statistics Office (NSO) conducts the LFS quarterly, the survey would be a very good source of information needed in constructing a model for per capita income which would make possible the estimation of annual per capita income every year, and consequently, the possibility of estimating poverty statistics annually. The potential of being able to estimate statistics on poverty, which are essential in poverty monitoring and in targeting beneficiaries for government and non-government programs to help the poor, resulted in the conduct of this study. The constructed model for the annual per capita income had the limitation of being appropriately interpreted only for the employed members of the population, since only a fraction of the total number of employed sample individuals was found to be valid for the analysis. Given that the government's approach in estimating poverty statistics in the country is by the income approach, it is valid that the employed persons are considered in modeling annual per capita income for use in estimating statistics on poverty. Employed members of the population are generally those that would have information on income, be it from salary, business or other sources. The selected model of per capita income yielded an adjusted R square of 51.80 percent. Significant predictors of annual per capita income includes the age of the households (HH) members as of last birthday, total number of hours worked during the past week, basic pay per day in the primary occupation, marital status of the HH member (i.e., whether married or not), highest grade completed of the HH member (i.e., whether at least college level or not), nature of employment in the primary occupation (i.e. whether permanent or not), underemployment status (i.e. whether unemployed or not), usual occupation (i.e. whether officials of government, supervisors, manager, professionals or not), kind of business in primary occupation (i.e. whether in the AFF sector or not), and kind of business in primary occupation (i.e. whether in industry sector or not). Age, total number of hours worked during the past week and basic pay per day in the primary occupation as well as the per capita income are in the logarithmic form when used in the model-building process. Using the predicting model of per capita income to estimate the poverty incidence among employed persons, it was found that, overall, 73.0 percent of the total employed individuals were correctly identified as either poor or non-poor using the predicted values of the annual per capita income. More than 50 percent (52.8%) of the observed poor employed individuals were correctly identified using the predicted per capita income while 85.6 percent of the total non-poor employed persons were correctly identified. The national level model-based estimate of poverty incidence among the employed was considered reliable with a coefficient of variation of 4.7 percent. However, computed estimates of poverty incidence in the regions resulted in coefficients of variations having values larger than 10 percent for almost all of the regional estimates. This result indicates that the obtained regional estimates of poverty incidence among the employed population, based on the fitted per capita income model have limitation in terms of reliability.