Cross-Sections Are History
Richard Easterlin (University of Southern California)
Although cross section relationships are often taken to indicate causation, and especially the important impact of economic growth on many social phenomena, they may, in fact, merely reflect historical experience, that is, similar leader-follower country patterns for variables that are causally unrelated. Consider a number of major advances (“revolutions”) in the human condition over the past four centuries – material living levels, life expectancy, universal schooling, political democracy, empowerment of women, and the like. Suppose that each has its own unique set of causes, and, as a result, a unique starting date and a unique rate of diffusion throughout the world. Suppose too that initially all countries are fairly closely bunched together on each variable in fairly similar circumstances. Suppose, finally, that the geographic pattern of diffusion is the same for each aspect of improvement in the human condition, that is, the same group of countries have a head start, and the follower countries in the various parts of the world fall in line in a similar geographic order. The result will be statistically significant international cross section relationships among the various phenomena, despite their being causally independent. The oft-reported significant cross-country relationships of many variables to economic growth may merely demonstrate that one set of countries got an early start in virtually every “revolution”, and another set, a late start.
Review by Chris Colvin
The “correlation equals causation” fallacy says that one thing preceding another does not imply causation. Be that as it may, inferring causality from time series data is significantly more plausible, under certain conditions, than from cross sections. A cross-sectional regression only shows the co-occurrence of different factors; to prove causality we also need to know about history. This is the argument made in an IZA Discussion Paper by USC’s Richard Easterlin distributed as part of NEP-HIS 2013-04-27 (since published in the journal Population and Development Review).
Easterlin’s particular beef is with purveyors of cross-country growth regressions. He notes that studies of actual historical experience of individual countries frequently disprove expectations about causation based on cross-sectional relationships. The fact that a certain group of countries enjoys high levels of per-capita GDP and high life expectancies does not mean the former causes the latter. Indeed, the fact even that these countries were the first to enjoy both high GDP and high life expectancies still does not prove causality.
Easterlin, famous for his Easterlin Hypothesis, instead argues that there could be unrelated factors causing GDP and life expectancy that cannot be picked up in a cross-sectional regression. The reason: cross sections register the results of history, not insights into likely experience. Co-occurrence at any one point in time does not imply causation. Per-capita GDP and life expectancy may be independent of one another and governed by advances in separate underlying technologies. The Industrial Revolution and the Mortality Revolution may be totally unrelated; each phenomenon must be analysed in its own right.
This is a short paper which I think offers an important contribution. It is especially useful as a teaching aid. Easterlin presents his argument in a clear and concise fashion that undergraduate students should easily grasp. His paper reiterates the importance of economic history in the teaching of economics, something which is noted to be lacking in many university syllabi by many of the authors of a great volume on the future of economics teaching edited by Diane Coyle. And when read in conjunction with e.g. Morten Jerven’s recent book on the unreliable nature of the statistics pertaining to growth and income in the Global South, the lessons of this paper can be used by students to themselves explore the problems of much of the empirical development literature of recent history.