1. Dietrich Vollrath writes,
the weighted variance of log capital and log coal per worker is either 0.0188 (if you use Clark’s index of capital) or 0.0381 (if you use Clark’s data on looms equivalents). Either way, this is only 2.92% or 5.90%, respecitively, of the total variance in real wages. A tiny fraction of variation in real wages is driven by differences in capital per worker, and the rest must be explained by technology, human capital, or something else. Clark has disposed of technology as an explanation, so it could be human capital. Clark eliminates big human capital differnces (at least in terms of age structure or experience), so it has to be “something else”. That something else is either local effects or culture, depending on your choice of terms.
This refers to international comparisons of productivity in the cotton industry. Clark is Gregory.
2. Gerben Bakker, Nicholas Crafts, and Pieter Woltjer write,
Compared with Kendrick, we find that labour quality contributes more and TFP growth less. For this period as a whole, TFP growth accounted for about 60% of labour productivity growth rather than the 7/8th famously attributed to the residual by Solow (1957).1 Contrary to secular stagnation pessimism, TFP growth was very strong both in the 1920s and the 1930s, at 1.7% and 1.9% per year, respectively – well ahead of anything seen in the last 40 years. Regardless, even though the 1930s saw the fastest TFP growth in the private domestic economy before WWII, it was not the most progressive decade of the whole 20th century in terms of TFP growth. Both 1948-60 and 1960-73 were superior at 2.0% and 2.2% per year, respectively
Pointers from Mark Thoma for both.
Keep in mind that in (2), they are starting with output per worker in the aggregate economy, and certainly there are problems measuring the numerator. Then you adjust for capital per worker, and that raises another measurement challenge. Then, in order to calculate you take a percentage change, which amplifies measurement error. Then, to compare growth rates across time periods, you take the difference in percentage changes, which amplifies measurement error yet further. I’m not criticizing these specific results, but just raising a general caution.