Counts of disasters and their impacts are a big deal these days. For instance, such data is central to evaluating progress with respect to the United Nations Sendai Framework on Disaster Risk Reduction. Disaster time series are also increasingly used as political footballs in climate advocacy and opposition. Separating out scientifically-accurate information from that which is not can be difficult.
This post provides a brief users’ guide to oft-cited disaster databases.
I’ve been studying, researching and publishing on these (and other) disaster datasets for almost 30 years. One thing I’ve learned is that we need better data on disasters — collected using consistent methods and with a global scope. As one paper concluded in 2022:
Reliable and complete data held in disaster databases are imperative to inform effective disaster preparedness and mitigation policies. Nonetheless, disaster databases are highly prone to missingness.
Disasters are one area where philanthropists have a huge opportunity to invest some modest funds to enable the collection and assessment of robust data with global significance.
Below, I use a stoplight to indicate of some oft-cited databases, data quality in the following categories:
research quality (green);
significant problems (yellow), and;
beware (red).
Munich Re and Swiss Re Insured and Total Losses
Munich Re and Swiss Re are the world’s leading reinsurance companies. They also administer global time series for insured and total disaster losses. Their tallies of insured losses are the most robust of any global disaster dataset, as there are legal and regulatory reporting requirements for these figures. Their reporting of total losses requires a greater degree of judgment and estimation — so I’d urge a bit more caution with these numbers. The methods and underlying data from each are not generally available, which can make interpretation sometimes difficult. That said, Munich Re and Swiss Re are at the head of the class. Green light.
Disclosure: I collaborated with Munich Re 15+ years ago (no funding to me) and back then they provided me with proprietary data for research.
EM-DAT
The Centre for Research for the Epidemiology of Disasters (CRED) in Belgium has been funded for decades by the U.S. Agency for International Development to maintain a database of global disasters, called EM-DAT. The dataset is widely used and much more often misused. For instance, their time series shows a huge increase in disaster counts to 2000, which CRED officials have long attributed to improved reporting and data collection. Since 2000 the data recently showed a decrease — but recently I noticed that 2022 disaster counts changed significantly without any announcement or notification. I emailed EM-DAT to ask what happened and they said that they had incorporated the results of a July 2023 study on excess heat mortality in Europe in 2022. This significant methodological change is problematic because such data did not exist before 2022 and in 2022 only exists for Europe. EM-DAT also explained to me that, “Occasionally, we make spontaneous event additions or modifications that are not part of our data collection/validation routine.” Such methodological discontinuities create real risks of introducing biases in the time series. I have encouraged EM-DAT to maintain methodological consistency and when they do introduce new methods, to account for and disclose them in parallel, so that consistent time series can be used to track actual trends.
I urge extreme caution in using EM-DAT — as important as it still is — and greater sophistication in its use. I plan to use it in research and my writing here with much greater precision (e.g., phenomenon by phenomenon). Be careful. Deep yellow.
NOAA Billion Dollar Disasters
Here is what I wrote last January:
Eleven years ago this week, I wrote my first critique of the so-called “billion-dollar disaster” count promoted by the National Oceanic and Atmospheric Administration (NOAA). The dataset and how it is used represents one of the most spectacular abuses of science you will ever see — and obviously so, it is not even close. Yet, the U.S. government and big media outlets continue to promote the dataset, while experts who know better stand by silently.
My views have not changed. The Billion Dollar dataset may be good for clicks and simplistic and misleading media reporting, but it can’t be used to say anything meaningful about disaster trends in the U.S., much less climate. I wish NOAA — a crucially important science agency — would just clean this up. Red light.
All the Rest
There are others who provide public estimates of disaster losses over time (I’ll ignore the proprietary ones here, as I have no basis to evaluate them). In 2020, back when I wrote for Forbes, I took a deep dive into one such dataset and found that it reported more than a trillion dollars more than Munich Re in total losses over 2010-2019. That’s — a lot. These datasets change retrospectively, methods are not published, press releases tout ever-bigger numbers as if looking for headlines and their company names in press. No thanks. Red light.
This is a nerdy, wonky post mostly for those who work in the weeds working with disaster data or interpreting it in their work. At the same time, it is important for everyone to see the challenges in grappling with such datasets and how a seemingly simple variable like “global disasters” can be so challenging to quantify. Thanks for reading and supporting THB!
If this has been covered elsewhere here, my apologies. Has Roger or anyone seen/analyzed this piece at the WaPo: https://www.washingtonpost.com/climate-environment/interactive/2023/pakistan-extreme-heat-health-impacts-death/ ? My guess is that there are places that will be more adversely affected than others by even modest changes in climate, while there are also places that may well benefit [a favorite topic of David Friedman]. The health risk, if real, strikes me as an important one, and also one that might be easier to disentangle from some of the partisan polarity. I'm not Team Fauci on this one; I am just curious.
Thanks for this, helpful.
As noted, statistics can be made to show anything you want.
And.
Learned a new word.
"missingness"
That was in an actual technical paper? How about "incomplete".