Creating a 2000 IES-LFS Database in Stata

PROVIDE Project Technical Paper 2005:1

The model explains about 70.5% of the variation of food expenditure. All coefficients are
significant, while the positive sign of all the coefficients makes economic sense. This model
was now used to estimate the expected values of the approximately 350 households that failed
to report any food expenditure (missing or zero values), but did report a value for total
expenditure. The model suggests that a two-person household earning R250,000 per annum
will spend about R1,909 per month on food, while a two-person household earning R60,000
per year will spend about R812 per month on food. These results appear to be realistic. The
average food expenditure rose by R7,096 per annum to R7,162 per annum as a result of the
imputation. The total expenditure levels were adjusted accordingly.

Section 2.4.3 looked in some detail at average tax rates per expenditure decile and found
that tax was grossly underreported in the IES 2000. The tax imputation was done in a similar
way. However, rather than fitting the regression model to the entire sample, only those
households whose total expenditure was higher than the median total expenditure were
included. The progressive tax system in South Africa is such that only about half of
households are eligible for tax. A regression model was run with the following double-log
function form:

ln(T_i) = a +b.ln(X_i)+c.ln(W_i)+ε_i .

Parameter a is a constant term, while b and c are coefficients. The independent variable, Ti
is the tax level (logtax), Xi the total expenditure (logtotexp) and Wi the number of working
adults per household (logW). This variable was found to be more significant than the
household size variable. Various other independent variables were also tested, but none of
these were significant. These included race dummy variables, location dummy variables,
sector of employment (formal/informal) dummy variables, occupation code dummy variables
and so forth. The regression model results appear below:

The regression results indicate that the model explains about 50.5% of the variation in
reported tax expenditure. The positive coefficient for variable logtotexp shows that as total
expenditure, which is a proxy for income, increases, more tax is paid. Furthermore, because
the coefficient is greater than one, a higher marginal tax rate is payable as households moves

49

© PROVIDE Project

<<   51   52   53   54   55   56   57   >>

More intriguing information