When it comes to understanding inequality, the debate is frequently encumbered with a multitude of misunderstandings about data. When talking about wealth inequality, we see statements like “85 people own more wealth than the bottom half of humanity.” Wealth is routinely misunderstood to mean the money and things someone owns. It isn’t. Wealth is someone’s total assets minus their liabilities. It is common to have negative wealth. The peasant farmer in rural China who has managed to save $200 and is debt free, is “wealthier” than the high-income young M. D. who has a negative net worth due to substantial student loans (i.e., she owes more than she presently owns.) I recently wrote about this in The World’s Bottom 10%: 7.5% Live in North America and None Live in China … And Other True But Worthless Facts.
Then there is the constant citation of growing inequality in pre-tax and pre-transfer income in the U. S. (usually just stated as “income”), and the need to rectify it through redistribution. But if you only look at pre-tax and pre-transfer income, no amount of redistribution will have one penny of impact. We could transfer $100,000 to every household in the bottom half the income distribution and it wouldn’t matter because it would be income after taxes and after transfers. When we look at after-tax and after-transfer income, we see that there has been little change in inequality between those at the 95th percentile and those at the 20th percentile for the last twenty years. See my post, Is Income Inequality Really the Problem? It Depends on What You Call Income.
Today, Arnold Kling reviews Chasing the American Dream by sociologists Mark Robert Rank, Thomas A. Hirschl, and Kirk A. Foster. (See Kling's post: The Longitude of Well-Being) He cites a stat that shows that homeownership rates have remained fairly constant at about 67%. Kling then asks you what percentage of Americans aged 55 have owned a home? A) 50%, B) 70%, and C) 90%. Kling says he would have guessed 70% when in fact it is 90%. The 67% number is a cross-sectional piece of data, taking a “snapshot” of homeownership at a given point in time. The 90% number is a longitudinal piece of data, taking a “video” of homeownership for over a period of time.
… the question that I asked concerns what demographers refer to as longitudinal information. If you follow given individuals over a long period, what sort of cumulative outcomes will you observe? In particular, over a lifetime, how many people will at some point own a home? To answer a longitudinal question, you need to use longitudinal data. To instead use time-series cross-section data risks making serious errors.
Most of the conventional wisdom about relative economic well-being, including the famous studies by Thomas Piketty and Emmanuel Saez, commits the time-series cross-section fallacy. Rank, Hirschl, and Foster did not set out to debunk this fallacy or to attack the many economists guilty of it. Instead, they took what seemed to them a natural approach for studying the evolution of wealth and poverty: longitudinal data. The result, in my reading, is that, like the boy in the fable, they have in an innocent, unintended fashion exposed statistical nakedness among many economists who are regarded as experts on the topic of inequality.
Once you think about it, the truth about homeownership rates makes sense. At some point in our lives, nearly all of us have been renters. In addition, most of us are likely to "downsize" as we grow older, and in the process many of us may choose to rent.
Kling moves on to the authors’ discussion of how many years a household spends in poverty or in affluence between ages 25 and 60. Kling offers an interesting alternative.
I would be interested in what the data show if, rather than looking at the extremes, one does the opposite. That is, throw out each household's lowest and highest three years of income. For the remaining years of income, take the average relative to the poverty line. If this average is below 150 percent of the poverty line, call it low. If it is above 500 percent of the poverty line (which works out to about 200 percent of the median), call it high. Then calculate the proportion of households that have high, medium, and low incomes by this longitudinal measure.
This would produce a very different breakdown. For instance, suppose that, rather than quitting my job to start an Internet business, I had kept working and that my salary had continued to increase gradually until I reached age 50. In that case, under the authors' measure, our household would be in the bottom of the income distribution, because of the "poverty" of my graduate school years and my failure to achieve the income level that they require for "affluence." However, using my approach, my household would have been somewhere in the vicinity of the boundary between high-income and middle-income. That seems much more reasonable to me.
Overall, as with homeownership data, the longitudinal view of income paints a picture in which life-cycle variation and idiosyncratic factors play a role. This role is overlooked in discussions of inequality that commit the time-series cross-section fallacy.
As I read Kling’s piece, I began to wonder how many people have had pimples. My guess is that the answer approaches 100%. Yet we don’t see headlines about acne being experienced by more than 90% of people at least one year in their lives. We understand that for most people this is a temporary life-stage issue. The universe of people for whom acne is an ongoing problem is much smaller. The same is true with poverty. I’m intrigued by Kling’s idea of discarding outliers and looking at 90% of the data between the outliers.
The reality is that no one set of data, or particular lens, can tell us the whole picture about issues like poverty and inequality. We must look at the issues from multiple angles to get to the truth. But it is incumbent on us to be cognizant of what lens we are using at any given time and what that lens is actually showing; in this case, knowing the difference between a snapshot and a video.