Last month I reported on a study published in the journal Energy Policy by Atif Ansar, Bent Flyvbjerg, Alexander Budzier, and Daniel Lunn on the economic costs of dams. Ansar et al. concluded that dams are subject to consistent cost and schedule overruns.
Since then I received an e-mail from Brian Sadden, who I think is an engineer who has worked in the hydropower industry. Sadden writes:
The paper [by Ansar et al.] is flawed in a number of ways. Most particularly the quoted figure of “90% cost overrun” is misleading and represents (if you read the study) the mean of a highly skewed distribution curve (see how significantly skewed the curve is on page 5 which I have highlighted – there appears to be one point somewhere near 5,000 percent overrun).
It is disingenuous to quote the mean from such a highly skewed, long tailed, curve. A careful researcher would be at pains to examine and document the 5,000 percent overrun which in itself must be very unusual in some way (maybe it’s a simple typographical error from some old documents, or the first estimate was made before some extreme inflationary period such as occurred in Argentina or Zimbabwe?). Even if the 5000 percent overrun is valid, it may be an overrun on the smallest and earliest project sampled- which would have little bearing on modern projects, but because of the parameter (percentage overrun) it assumes greater weight.
This outlier has a great influence on the curve so it is important to find out how the overrun became so large.
In any case, as statisticians understand, the mean of a highly skewed curve is important mainly in defining the extent of “skewness”. If one is trying to draw some realistic conclusions from the data contributing to such a skewed curve it is more proper to use the mode or the 50th percentile.
I suspect that the authors do understand this as they have quoted the 50th percentile figures in Table 8 on page 12, but then could not resist using the mean for their conclusion. Comparing the 50th percentiles in Table 8 indicates that historical cost overruns for dams and hydropower are very much “in line” with other infrastructure.
A second major problem with the paper is that the sample chosen (from 1934 to 2007) is a small sample and includes projects conceived and constructed in and through many worldwide and regional economic and financial crises such as: the Great Depression; World War II; the Korean War; the Vietnam War; Oil shocks in 1970s and 90s; the march of colonial independence; post-war construction and reconstruction; the Colombo plan; the Marshall plan; the creation of the World Bank, ADB etc.; the Cold War; miscellaneous political shocks; speculative bubbles; the privatization trend; the collapse of the Soviet Union; 1973-75 stagflation; 1980-82 stagflation; globalization, the 1998 financial crisis etc. How can one possibly neglect all these global currents in an analysis of major infrastructure projects, many of which are dependent on cross border lending, and thus global economic and financial conditions?
It is obvious that infrastructure planners need to be as prudent as possible when developing a country, and in a general sense the paper reinforces this – but is so selective in its conclusions as to be naive or scurrilous (I am not sure which). One comment is that “policy makers should prefer energy alternatives that require less upfront outlays”. A direct result of that preference would be a continued reliance on fossil fuels which in the long run do not have the same fuel price certainty that hydro does – and arguably would contribute substantially to climate change. Until there is a carbon tax, the full societal costs of fossil fuel generated power have not been taken into account.
I found it difficult to judge this criticism in all its details. I sent it over to Ben Flyvbjerg (the researcher who pointed me to the original study that I posted) to ask if he had any response to Sadden’s comments, and Flyvbjerg replied:
I suggest any complaint is sent to the editor of Energy Policy, so if we have a debate it will in the pages of the journal.
So maybe this will happen. In the meantime, since I posted the original paper in this space, I’m posting the criticism too, and you can make of it what you will.