David Stasavage (NYU) was here on Monday, giving a talk based on work with Ken Scheve. It was a compelling talk, with some interesting issues; statistically, politically, normatively…
David and Ken have cross-national data on wealth shares (the proportion of a country’s GDP going to the top 1% or top 10% of the income distribution; the bigger this number, the more inequality), spanning most of the 20th century and about 14 advanced, industrial democracies. David had a very interesting graph, simply displaying the country-specific trends in the variable, all overlaid on each other; graphs like this are among the most simple yet most informative things you can do with these kinds of panel/repeated measures data, and I wish people would display them more often when working with that kind of data. The lack of color and the small size made it hard to see the specific country patterns that well, but in no small measure, that is precisely the point of the graph (i.e., the common time trend soaks up a lot of the variation in the data): I cut-and-paste this graph from the PDF of the paper as a JPG, below. We kept coming back to the graph in the discussion and for good reason: the inequality data start off high and relatively dispersed, but trend down from about the late 1930s onwards, hitting a more or less steady state through a lot of the 1970s and 1980s, with remarkably little dispersion; the recent, large, increases in inequality in the United States are clearly apparent, as well as a little more dispersion than elsewhere in the series, and a slight trend upwards from what looks like a global minimum around about 1980.
The JPG above is a thumbnail: you can click on it to get it up to about 780px wide in a separate window.
As the graph makes clear, the most compelling feature of the data is the strength of the common time trend: in response to a question I asked, David told us that about 70% of the variation in the data is attributable to a common time trend (i.e., the [tex]r^2[/tex] you get from simply regressing the dependent variable on a set of year-specific dummy variables is about .7). That is clearly apparent from the graph David showed us, but was really quite large relative to what I expected a priori. The cross-national variation is relatively small (and was really quite small in the 1960s-1980s). The covariates David had available for analysis largely pick up cross-national variation and display very little longitudinal variation, and so it is not surpising that they do very little in the analysis, with the possible exception of union density, which does trend around within countries (but the other controls are quite serially persistent, like dummies for presence/absence of universal suffrage, PR electoral system, centralized wage bargaining). It was interesting that tax rates and transfers weren’t in the analysis, or at least not that I remember.
Jim Fearon and I were quite struck by the conclusion that David gave us, and was vividly demonstrated by the graph I referred to above: that the cross-national variation in inequality observed in recent decades is small relative to the longitudinal variation one sees when taking a longer, historical view at the phenomenon. Political scientists’ interest in institutions (constitutional arrangements, electoral systems, etc) doesn’t seem to have much to do with the phenomenon at all, at least not relative to the strong, common time trend in the data.
Josh Cohen pointed out that even when the cross-national variation is at its minimum, we’re talking about a doubling of the level of the dependent variable across the countries in the data set: i.e., those kinds of differences across countries mean a lot, both normatively and politically. Fair enough, but if one was interested in trying to do something about levels of inequality, then the long-run historical analysis presented by David shows that it is more likely to be massive, supra-national processes that will generate shifts in inequality, rather than institutional or political innovations within any single country. An important exception is the recent, large increase in inequality in the United States over the last 15-25 years: the U.S. has clearly broken away from the pack, on a trajectory somewhat at odds with the rest of the data.
The data also had some other interesting features: missingness that they said they handled via multiple imputations. For all the reasons listed above it sounds like a great teaching data set and I hope to get my hands on it eventually.