Population Growth and Technological Change: One Million B.C. to 1990

This 1993 Quarterly Journal of Economics paper by Michael Kremer builds and empirically tests a model of world population growth from one million B.C. to 1990 that combines the nonrivalry of technology – the endogenous-growth literature’s point that the cost of inventing a new technology does not rise with the number of people who will use it, so a larger population should spur more total invention – with the Malthusian assumption that the level of technology limits the population a given level of land and technology can support. In the simple benchmark version, in which each person’s chance of inventing something is independent of population, these two assumptions combine algebraically to predict that the growth rate of population is proportional to its level, which implies much faster than exponential population growth; Kremer then generalizes the research equation (following Jones 1992) to let research productivity depend on income and on the already-attained level of technology, and shows this generalized model remains consistent with the long historical record while also accommodating the roughly constant postwar growth rate of total factor productivity. Using population estimates assembled mainly from McEvedy and Jones (1978) and Deevey (1960) and spanning one million B.C. to 1990, a heteroskedasticity-weighted regression of population growth on the level of population finds a coefficient on population significantly different from zero and an intercept insignificantly different from zero (coefficient 0.524, standard error 0.0258, R^2 = 0.92, full sample), a finding that is robust to using three alternative historical population series (Deevey, Clark, Durand) and to restricting the sample to the period after 200 B.C. or to the regional series for Europe, China, and India. A second, cross-sectional test exploits the flooding of land bridges around 10,000 B.C., which cut off technological contact among the Old World, the Americas, mainland Australia, and Tasmania: by 1500, just before Columbus reestablished contact, technology and population density ranked exactly with land area (and hence initial population) across these regions – a ranking Kremer calculates would occur by chance only 1 time in 24 (1 in 120 if Flinders Island is included) under the null that technological change is independent of population. The model is explicitly a theory of the historical record rather than of the present: a fuller version in which population adjusts to income at finite speed, and in which population growth eventually declines as income rises, is needed to reconcile the model’s mechanism with the twentieth-century demographic transition.

Michael Kremer | Macro Paper Warehouse

Population Growth and Technological Change: One Million B.C. to 1990