Beside the subject matter questions that people from social sciences may answer, from the statistical point of view you should take robust weighted estimates for the ranges. This means, weighted quantiles. Weighted because you deal with a survey conducted with a complex sampling design and you have to deal with sampling weights and robust (e.g. taking quantiles) since in such data sets you always have some outliers in the data (people with huge income compared to others). 

A starting point is to use something similar to the quintile share ratio, which is a very famous measure of inequality. So, something like (in R) laeken::weightedQuantile(x, w=w, c(0.2, 0.8)), to took the 0.2 and 0.8 quantile and consider sampling weights.

Another very famous measure of low income would be based on the at-risk-at-poverty rate. It is 0.6 * weightedMedian(x, w), so people with lower than 60% of the weighted median income are defined as to be of high risk of poverty.

Hi, when we have household data the standard procedure is to divide household income among household members, but not all members compute as one unit. OECD (and also EUROSTAT) add 1 for the first adult; 0'5 for the second adult; and 0'3 for each children. Hence, for a household composed of two adults and three children you will divide household income by 2'4 (1+0'5+0'3+0'3+0'3).

There is of couse some debate concerning this procedure, because by computing income in this way you may get a distorted picture of the real economic situation of households with children. To assume that a child consume only 1/3 than an adult may not be realistic. Having said so, this is a standard procedure.

As for how to determine what income is "low", there are some general rules, as the Dr. Templ indicates. OECD defines poverty as those with incomes below 50% of the mean national income. The EU defines poverty as those with incomes below 60% of the mean national income. Those with incomes around the median income would be defined as "middle" (the precise range is undefined), and much less agreement can be found to define "high" incomes.

In an analysis of political attitudes I aggregated into the category "middle income" all incomes from 60% of the national median to 40% above national median. And "upper-middle" those with income 40% above the median and more. I.e.: less than 60% of the median (low); from 60% of the median to 1.4 median (middle); higher than 1.4 the median (upper middle). It worked fine, however, I guess that the precise cut-points should depend on your topic.

All the best,

Inés

As Matthias Templ suggests, weighted quantiles is a very good way to make income ranges.

Thanks for your answers.

Dear Dr. Templ, why should i consider sampling weigths for income intervals? Why are these weights important for a microdata from the sample. Should I use these weights for the sample if I will not handle the population? or use these weights only for defining income ranges? Why do i need them-i will make predictions by using the SILC data-and this data is a sample data eventually. Would not it be more correct to make generalizations over the sample by setting income intervals for the sample? I got the results on weighted quantiles by laeken package with probs = c(0.2, 0.5, 0.8) as a,b,c for weighted quantiles. Should I use these ranges as income group ranges like this: 0-a, a-b, b-c, c+? Or would it be more correct to use c(0.2, 0.8)  to express income ranges as low-middle-high for

Hi Prof Calzada, can you give your project or article title that you have used political attitudes? When i take 60% of the weighted median income and 1.4 median for middle  income, i get very low cutpoints and i think these cutpoints are unrealistic for a personal income, it gives a very low value for lower threshold of the middle income. Instead of these methods, should i determine ranges for   income groups like i.e.: 0-1500 low, 1501-3000 middle and 3001+ high without using weights and only considering personal income variable?