People, a wise one of you once said, “If it ain’t broke, don’t fix it”.
Sadly it appears that this principle was not followed in the “upgrade” of the PWT to versions 8.0 and 8.1. At least according to this NBER working paper by Pinkovskiy and Sala-I-Martin.
They use satellite measures of nighttime lights as a measure of true income. They assume that these measures are correlated with true income and that the measurement error in the satellite method is uncorrelated with the measurement errors in the PWT methods. They then evaluate different versions of compiled datasets on how well they explain the satellite measure.
They report that, “in particular, we find that the PWT 7.1 chain-based GDP series substantially outperforms the constant-price series in both PWT 8.0 and PWT 8.1, the two most recent vintages of the PWT.”
Perhaps even more surprising, they find that, “the World Development Indicators are as good, and often better, measures of unobserved true income as are any recent vintages of the Penn World Tables.”
One reason for this may be that, “In particular, and again unlike the PWT, the WDI take national growth rates directly from the national accounts without any further editing, while the PWT performs some editing of the growth rates before using them.”
The paper also has a fascinating discussion of what approaches are better for measuring cross-country income differences versus producing comparable measures of country growth rates, along with a discussion on possibly average different measures in a way to produce an overall more accurate estimate.