Go to Brad De Long's Home Page
Consider the following from Allan Meltzer, a critique of a line of work on "Equipment Investment and Economic Growth" that I have been pursuing with Larry Summers:
...Professor Lawrence Summers and Bradford De Long claim to have evidence that investments in machinery produce returns to society far beyond the returns to investors.... The state, however, can supplement private investments, or subsidize them, and capture the returns for society....
Alas, it isn't so. Subsequent research showed that Summers and De Long were misled by the presence of Botswana in their data set. During the sample period they used, Botswana invested heavily in mining machinery to exploit its diamond mines.
Botswana had the highest growth rate and the largest share of spending on equipment investment, so machinery investment and growth appeared to be strongly related. Excluding Botswana, one of sixty countries, showed the results to be spurious.
Now take a look at a selection from the very first thing we wrote about equipment investment and economic growth, taken from the section, "Sample Selection Issues", where we discussed which observations should and should not be in our data set:
Results using the entire 61-nation sample are somewhat sensitive to outliers. The exclusion of Zambia, for example, raises the adjusted R<super>2</super> in the regression underlying Figure VI from 0.29 to 0.44; the exclusion of Botswana would reduce the adjusted R<super>2</super> from 0.29 to 0.21. Inclusion or exclusion of these two countries can move the equipment share coefficient between 0.21 and 0.31, although the coefficient remains significant at conventional levels.
....[I]t is worth pointing out that [the large 61-nation sample] omits two outlier nations with large identifying variances that would significantly strengthen our findings. Both Singapore and Taiwan have had high equpment quantities, low equipment prices, and rapid productivity growth in the post-World War II period. Neither Singapore nor Taiwan is in our sample.... The inclusion of these two observations would strengthen our conclusions.
Let's run, backwards, through what Meltzer said:
Claim: "Excluding Botswana... showed the results to be spurious"
Truth: Whether Botswana was left in or taken out of the sample, the results still held: "remain[ed] significant at conventional levels"
Claim: "Summers and De Long were misled by the presence of Botswana in their data set."
Truth: As we wrote in our very first paper on "Equipment Investment and Economic Growth", the surprisingly strong association between equipment investment and economic growth is there--whether Botswana and other low-income outliers are included in or excluded from the data set.
Claim: "Subsequent research showed..."
Truth: We were the ones who flagged the effect on our statistical study of the inclusion in our sample of Botswana (and of Zambia, and of Tanzania, and the omission of Singapore, and of Taiwan) as important issues. The implication that we did not understand what was going on in our dataset until "subsequent research" pointed it out to us is wholly false.
In the long run it will become clear whether countries with high rates
of machinery investment grow rapidly because the social returns to investment
in machinery and equipment are astronomically high, or because of some one
of the other factors Larry Summers and I discussed in our articles: perhaps
machinery investment is high in fast-growing countries because investors
forecast fast growth and high profits, and channel investment into such
countries; perhaps machinery investment is high in fast-growing countries
because savings are high whenever income is growing rapidly, and savings
have to be used for something; perhaps machinery investment is high in fast-growing
countries because governments that make investment profitable are taking
many other steps--educating the population, controlling corruption, reforming
the tax system--that promote growth.
I think that the preponderance of the evidence is that machinery investment does have very high social returns, but I agree that the case is not conclusive.
But I think much more strongly that comments like Allan Meltzer's--which could not have been written in good faith had he done his homework and actually read our article--poison the well of economic debate.
Go to Brad De Long's Home Page
Associate Professor of Economics Brad De
Long, 601 Evans
University of California at Berkeley; Berkeley, CA 94720-3880
(510) 643-4027 phone (510) 642-6615 fax