Top and Bottom Coding at LIS
Jörg Neugschwender (LIS)
In this note, Jörg Neugschwender compares the previous top and bottom coding procedures used at LIS with alternative measures. These sensitivity analyses for top and bottom coded incomes were motivated by the many additions of datasets from emerging economies with income distributions that are generally more unequal than those of mature economies. Using the Gini Index, Jörg compares the previous top coding procedure (10 times the median of household income) with alternative measures, such as 20 times the median and the detection of extreme values via the interquartile range. Likewise, using the Atkinson Index, he performs a sensitivity analysis for bottom coding techniques. The previous method (bottom coding at 1 % of the mean of equivalised household income) is contrasted against bottom coding at value 0 and the detection of extreme values via the interquartile range.
After carefully analysing the results, LIS is in the process of adopting a new practice of setting extreme income values for a bottom and top code for its Key Figures and Data Access Research Tool (DART). The new procedure first defines the interquartile range (IQR) of the logarithm of the income and detects values which could possibly be assessed as outliers (3 times below or above the IQR). The new measure keeps inequality measures much closer to the non-top and bottom coded measures as compared to the previous approach but serves to smoothen inequality trends to a reasonable amount by consistently reducing the influence of extreme values for within- and between-country comparisons.
Full article is available here.