Replications of our Normalized Hurricane Damage Work

July 14th, 2008

Posted by: Roger Pielke, Jr.

This post highlights two discussion papers that have successfully replicated our normalized hurricane damage analyses using different approaches and datasets. Interestingly, both papers claim a trend in losses after normalization, but do some only by using a subset of the data – starting in 1950 in the first case and 1971 and the second case. Our dataset shows the same trends when arbitrarily selecting these shorter time frames, however, as we reported, we found no trends in the entire dataset.

If you’d just like the bottom line, here it is:

I am happy to report that Nordhaus (2006) and Schmidt et al. (2008) offer strong confirmatory evidence in support of the analysis that we have presented on adjusting U.S. hurricane losses over time. What do these studies say about the debate over hurricanes and climate change? Well, almost nothing (despite the unsuccessful effort by Schmidt et al. to reach for such a connection). There is no long-term trend in the landfall behavior of U.S. hurricanes, so it is only logical that there would also be no long-term trends in normalized damage as a function of storm behavior. Those looking for insight on this debate will need to look elsewhere. If it is to be found, such a linkage will be found in the geophysical data long before it shows up in the damage data, as we asserted at our Hohenkammer workshop.

Please read on if you are interested in the details.


The first paper is by the leading economist William Nordhaus (2006, PDF). His analysis uses the same original loss data but adjusts it for GDP rather than population, wealth, and housing units. His analysis uses our 1998 study which we updated in 2008. Nordhaus claims to have “verified our analysis” but he does leave a loose end (emphasis added):

Our estimates indicate that the time trend in the damage function is positive. For example, the time trend in the OLS full-sample equation found that normalized damages have risen by 2.9 (+0.76) percent per year, indicating increased vulnerability to storms of a given size.

This is contrary to Roger A. Pielke, Jr., “Are There Trends in Hurricane Destruction?” Nature, Vol. 438, December 2005, E11, who reports no statistically significant trend. Similar negative results were found in Roger A. Pielke, Jr. and Christopher W. Landsea, “La Niña, El Niño, and Atlantic Hurricane Damages in the United States,” Bull. Amer. Meteor. Soc., vol. 80, 2027-2033. Additionally, issues of comparability over time are non-trivial, as is discussed in Christopher W. Landsea, “Hurricanes and Global Warming,” Nature, Vol. 438, December 2005, E11-E12. The reasons for the difference in findings have not been resolved.

The reason for the difference in findings should be completely obvious: Nordhaus looks at 1950-2005, and Pielke (2005) and Pielke and Landsea (1998) both begin their analysis in 1900. Pielke (2005, PDF) reports:

For example, take the 86 storms causing at least US$1 billion in normalized damages, which removes a bias caused by small storms resulting in no damage in the early twentieth century (that is, not subjected to normalization). There is an average per-storm loss in 1900–50 for 40 storms (0.78 events per year) of $9.3 billion, and an average per-storm loss in 1951–2004 for 46 storms
(0.85 events per year) of $7.0 billion; this difference is not statistically significant. Adding Hurricane Katrina to this data set, even at the largest loss figures currently suggested, would not change the interpretation of these results.

The following figures illustrate this point quite clearly. The first figure is from Nordhaus:

Nord1.jpg

The second figure is from the data of Pielke et al. 2008 (PDF):

pietal1.jpg

Clearly, the datasets show the same trends. However, our entire dataset shows no trend, as we reported in the paper:

The two normalized data sets reported here show no trends either in the absolute data or under a logarithmic transformation: the variance explained by a best-fit linear trend line=0.0004 and 0.0003, respectively, for PL05, and 0.0014 and 0.00006, respectively, for CL05. The lack of trend in twentieth century normalized hurricane losses is consistent with what one would expect to find given the lack of trends in hurricane frequency or intensity at landfall. This finding should add some confidence that, at least to a first degree, the normalization approach has successfully adjusted for changing societal conditions. Given the lack of trends in hurricanes themselves, any trend observed in the normalized losses would necessarily reflect some bias in the adjustment process, such as failing to recognize changes in adaptive capacity or misspecifying wealth. That we do not have a resulting bias suggests that any factors not included in the normalization methods do not have a resulting net large significance.

So to summarize, Nordhaus (2006) and Pielke et al. (2008) reconcile perfectly.

A second study is just out by Silvio Schmidt et al. (2008, PDF) which applies a modified version of our normalization methodology to hurricane losses found in the Munich Reinsurance global loss dataset. This study finds no significant trend from 1950, using a log transformation of the dataset:

The trend analysis for the period 1950–2005 yields no statistically significant trend in annual adjusted losses. Even if the two extreme years, 2004 and 2005, are omitted from the trend analysis, no trend can be identified in which the explanatory variable time is significant. Thus, no conclusion can be drawn regarding a possible trend in the periods 1950–2005 and 1950–2003.

The paper does identify a statistically significant trend starting in 1971, but the significance disappears when Katrina is removed from the dataset. Once again, the Schmidt et al. analysis is perfectly consistent with our analysis. The following figure shows their log-transformed data from 1950:

schetal1.jpg

And the following graph is the same transformation applied to the data of Pielke et al. (2008).

pietal2.jpg

One notable difference is that the Munich Re dataset apparently has some gaps, as it reports a number of years with zero damage that our dataset shows the presence of damaging storms. From the graph is should be clear that any claim of a trend over the dataset depends upon 2004 and 2005. And even in this case Schmidt et al. were only able to identify a statistically significant trend by starting with 1971 (which they claim as the start of a cold phase, contrary to most studies that use 1970). The following figure from Schmidt et al. shows how close the two analyses actually are (the red curve is Pielke et al. 2008):

schetal2.jpg

The differences between the two analyses are very small, and I would guess, of no particular statistical significance over the time series of the dataset.

So I am happy to report that Nordhaus (2006) and Schmidt et al. (2008) offer strong confirmatory evidence in support of the analysis that we have presented on adjusting U.S. hurricane losses over time.

What do these studies say about the debate over hurricanes and climate change? Well, almost nothing despite efforts by Schmidt et al. to reach for such a connection. There is no long-term trend in the landfall behavior of U.S. hurricanes, so it is only logical that there would also be no long-term trends in normalized damage as a function of storm behavior. Those looking for insight on this debate will need to look elsewhere. If it is to be found, such a linkage will be found in the geophysical data long before it shows up in the damage data, as we asserted at our Hohenkammer workshop.

2 Responses to “Replications of our Normalized Hurricane Damage Work”

    1
  1. Indur Goklany Says:

    FWIW, your findings were also essentially confirmed in an analysis I reported in “Potential Consequences of Increasing Atmospheric CO2 Concentration Compared to Other Environmental Problems.” Technology 7S: 189-213 (2000) [available at http://members.cox.net/imgrant/Goklany%202000%20Technology.pdf ]. That analysis, less sophisticated than yours, indicated upward trends when property losses for 1900-1997 were measured in real dollars, but there was no trend when they were measured in terms of nation’s cumulative fixed tangible reproducible assets (a surrogate for national wealth) (for 1926-1997). A similar analysis for 1929-2004 for the book, “The Improving State of the World: Why We’re Living Longer, Healthier, More Comfortable Lives on a Cleaner Planet” (Cato Institute, Washington, DC, 2007) once again showed no upward trend when losses were measured as a fraction of wealth. But in the latter case, I used state personal income as a surrogate for state wealth. Also I aggregated losses over non-overlapping periods for the analysis to calculate losses per year. I am not now convinced this was the best approach, however.

  2. 2
  3. Roger Pielke, Jr. Says:

    Thanks for the link Goks!