The state of Egypt’s economy is grim. Foreign investment is really low, and there is a lack of well-paying employment opportunities, especially for the young. Median annual income is mere 3900 Egyptian pounds per capita ($558), unemployment rate exceeds 13 percent, and an estimated 13.7 million people (or 16.5%) suffer from food insecurity. These problems in turn affect the country’s political stability. The revolution that overthrew Hosni Mubarak in spring 2011 was partially attributed to people’s sense of misery and growing inequality in the society. Surveys of sentiments confirm that the aversion to inequality has intensified in the past decade among all social groups in Egypt.
It is thus surprising that economic inequality in Egypt is reportedly quite low by international standards. In 2009, the Gini coefficients for inequality in incomes and expenditures were 32.9 and 30.5, respectively, lower than in neighboring countries, or even in southern Europe and North America. The Gini, however, is susceptible to various statistical problems, including households systematically failing to report their true incomes or respond to the survey altogether, or outlying income observations. It is worth investigating whether these issues may be responsible for the low value of the Gini in Egypt, and whether a more accurate statistic can be derived. In a new study (Top Incomes and the Measurement of Inequality in Egypt, World Bank Policy Research Working Paper No.6557), Paolo Verme and I examine the robustness of the Gini and attempt to correct it for several types of biases.
In the Egyptian Household Income, Expenditure and Consumption (HIEC) survey, all households reported their incomes and expenditures fully. Hence, systematic non-reporting of these items is not a problem there. This speaks of the high quality of survey practices and great effort at the country’s Central Agency for Public Mobilization and Statistics.
Still, 1,778 out of 48,635 contacted households (or 3.7%) did not respond to the survey at all. If rich households tended to respond to the survey less often than poor households, the estimated income distribution and Gini coefficient may be biased downward. On the other hand, the incomes of responding households contain several extreme values. If these high incomes are reported inaccurately or represent atypical people, the values may inflate the measures of wealth and inequality in the population. We investigated these two possibilities jointly, to evaluate how important both problems are and how they play out in conjunction with one another.
Indeed, survey nonresponse is related systematically to income in Egypt, since geographic areas with higher distributions of income tend to have higher nonresponse rates. To deal with the resulting bias, we used information on income distributions and nonresponse rates from across many geographic areas to estimate the relationship between households’ income and likelihood of survey participation. Then we assigned weights to each household in inverse proportion to their inferred response rate, to give more weight to households who are rarer. This analysis indicated that rich households are significantly less likely to participate, and are thus represented less adequately in the HIEC survey than poorer households. A small number of top-income households in particular have a much lower likelihood of participation than the rest of the population. Correcting for the underrepresentation of the richest households, we estimate a higher effective distribution of incomes and higher inequality. The Gini rises by approximately 1.3 percentage points, to 34.2 for income and 31.8 for expenditure per capita.
The above analysis suggests that high incomes are underrepresented in the Egyptian household survey, and should be assigned a greater weight to offset a downward bias to the Gini. On the other hand, economic literature suggests that high incomes may be reported inaccurately, or may represent people who are not typical of the population. This would suggest that top income observations should be removed or assigned a lesser weight. A visual inspection of the Egyptian data supports this conjecture to some degree. A handful of income observations are significantly higher than the following observations. Whether these differences are due to a natural diffusion in the population, or due to measurement and sampling issues is unclear but can be evaluated.
If we suspect the extreme income observations to be inaccurate or non-representative of typical high-income people, we can try replacing them with values more representative of the population, and estimating the Gini using this corrected income distribution. It is standard among economists to model high incomes as following a particular right-skewed distribution – Pareto distribution. By comparing the actual high incomes with those predicted under the Pareto distribution in Egypt, we can estimate the extent of the bias due to non-representative extreme values. By comparing the shape of the top-income distribution in Egypt with those in other countries, we can comment on how likely it is that the extreme observations should be excluded in our measurement of inequality.
In fact, replacing the actual top incomes with their Pareto estimates does not affect our inequality estimate. Moreover, the shape of the top-income distribution in Egypt is close to the average of those in surveys worldwide. Hence, extreme observations in the Egyptian survey are in line with those worldwide, and appear representative of the true population of rich households. This suggests that there is little or no bias to the measurement of inequality due to atypical top-incomes.
All in all, we have verified that inequality in Egypt is low even after accounting for survey non-participation by rich households, and for the presence of extreme-income observations. Correction for the non-response bias increases the Gini for income to 34.2, but this is still lower than the non-corrected values in the region, or even in many industrialized nations. Now, 1.3 percentage points is a fairly small bias correction. This is due to two properties of the HIEC survey and income distribution in Egypt. First, the number of non-responding households is low compared to other countries, implying that few high-income households are missed. Second, the original uncorrected measure of inequality is low, implying that reweighting of individual households does not affect the distribution of incomes much. For both reasons, there is relatively little bias in the Egyptian survey sample to correct. Of course this methodology is likely to yield greater bias corrections in other countries and samples.
In sum, economic inequality in Egypt is indeed low. The reported perception of injustice in the Egyptian society must thus be explained by other manifestations of social inequality such as inequality of assets, social immobility, or income distribution based on unfair principles. The goal of future research should be on measuring these alternative sources of social perceptions of inequality.
