Research Paper NBER Working Paper Series 14247
Estimating trends in US income inequality using the Current Population Survey: the importance of controlling for censoring
There is intense public interest in the spread of incomes in society between the rich and the poor and those in the middle, and whether income gaps have been growing larger over time. Although it is widely agreed that income inequality in the USA rose during the 1980s, there is less agreement about what happened during the 1990s and later. There are two main sources of data that researchers might use to investigate this topic. The first is the income information collected from respondents to large-scale surveys of households, and the second is administrative data on income that are kept by the US income tax authorities. Both sources have some advantages and some disadvantages, and so complement each other.
We analyze trends in US income inequality between 1975 and 2004, using data from the March Current Population Survey (CPS), the source most commonly used to address this topic. An important limitation of the CPS data that are made available to researchers (the ‘public use’ data) is that they are censored (‘topcoded’). In order to protect respondent confidentiality and for other reasons, the very richest incomes reported in interviews are not included in the data; instead they are replaced in the public use data with some lower value (the ‘topcode’). Not only does censoring reduce measured inequality, but changes over time in topcoding practices complicate assessments of trends in inequality. To investigate the effects of censoring problems in public-use CPS data, we secured unprecedented access to CPS internal data. Although there is some censoring in the internal data as well, its prevalence is relatively small and, moreover, we employ a statistical technique that enables us to impute plausible values to the small number of the very richest people for whom actual incomes are not included in the internal data. So, with access to the public use and internal CPS data, we can examine how trends in income equality changed over time, and check the robustness of our findings using a number of different data series. We also use several methods for summarizing inequality itself.
We find, as others have, that income inequality increased during the 1980s. However, we also show that the upward trend in income inequality slowed significantly after 1993.
To investigate the robustness of the slowdown finding, we compared our CPS-derived results with those derived from information about incomes derived from administrative record data from the US Internal Revenue Service and reported in research by Thomas Piketty and Emanuel Saez. The focus is on trends in top income shares - the share of total income that is held by the richest 1%, 2%, 5%, and so on. We show that our CPS estimates of trends in top income shares match the estimates of trends reported by Piketty and Saez, except for within the top 1% of the distribution. Thus, we argue that, if income inequality in the USA has increased substantially since 1993, such increases are confined to this very highest income group.