The World's Greatest Instrument is still a Gibson Les Paul!

and not (as some would have it) settler mortality.

So says David Albouy in his new NBER working paper (which is on its second revision at the AER).

Here is the abstract:

In a seminal contribution, Acemoglu, Johnson, and Robinson (2001) argue property-rights institutions powerfully affect national income, using estimated mortality rates of early European settlers to instrument capital expropriation risk. However 36 of the 64 countries in their sample are assigned mortality rates from other countries, typically based on mistaken or conflicting evidence. Also, incomparable mortality rates from populations of laborers, bishops, and soldiers – often on campaign – are combined in a manner favoring their hypothesis. When these data issues are controlled for, the relationship between mortality and expropriation risk lacks robustness, and instrumental-variable estimates become unreliable, often with infinite confidence intervals.

The AJR paper has been very influential, so this paper is potentially very important and well worth reading (here is a link to an ungated version).

The indictment is strong. Here are some details:

The historical sources containing information on mortality rates during colonial times are
thin, which makes constructing a series of potential European settler mortality rates challenging. AJR construct their series by combining the mortality rates of soldiers (Curtin 1989, 1998), laborers (Curtin 1995), and bishops (Gutierrez 1986). Researchers have been eager to use this new series, particularly given its promise as an instrumental variable for institutions. Currently, over twenty published articles, and many more working papers, use AJR’s settler mortality data. This paper argues that despite AJR’s ingenuity and diligence, there are a number of reasons to doubt the reliability and comparability of their European settler mortality rates and the conclusions which depend on them. First, out of 64 countries in their sample, only 28 countries have mortality rates that originate from within their own borders. The other 36 countries in the sample are assigned rates based on AJR’s conjectures as to which countries have similar disease environments. These assignments are based on weak and sometimes inaccurate foundations. Six assignments are based upon AJR’s misunderstanding of former names of countries in Africa. Another sixteen assignments are based on a questionable use of bishop mortality data in Latin America from Gutierrez (1986), which are based on 19 deaths. Additionally, AJR use the bishop rates multiplied by a factor of 4.25, a procedure that appears to contradict evidence in their own sources. At a minimum, the sharing of mortality rates across countries requires that statistics be corrected for clustering (Moulton, 1990). This correction noticeably reduces the significance of AJR’s results. If, in the hope of reducing measurement error, AJR’s 36 conjectured mortality rates are dropped from the sample, the empirical relationship between expropriation risk and mortality rates weakens substantially, particularly in the presence of additional covariates. Second, AJR’s mortality rates never come from actual European settlers, although some settler rates are available in their sources. Instead, AJR’s rates come primarily from European and American soldiers in the nineteenth century. In some countries, AJR use rates from soldiers at peace in barracks, while in others, they use rates from soldiers on campaign. Soldiers on
campaign typically have higher mortality from disease, and AJR use campaign rates more often in countries with greater expropriation risk and lower GDP. Thus, AJR’s measures of mortality artificially favor their hypothesis. In a few countries, AJR use the maximum mortality rates of African laborers, although these do not appear comparable with average soldier mortality rates. Controlling for the source of the mortality rates weakens the empirical relationship between expropriation risk and mortality rates substantially. Furthermore, if these controls are added and the conjectured data are removed, the relationship virtually disappears. Additional data provided by AJR in their Response (2005) do not restore this relationship.