High-tech models developed by quants have, once again, greatly underestimated risk. What will be the consequences?
In the waning days of April, as federal officials were declaring a public health emergency and the world seemed gripped by swine flu panic, two rival supercomputer teams made projections about the epidemic that were surprisingly similar – and surprisingly reassuring. By the end of May, they said, there would be only 2,000 to 2,500 cases in the United States.
May’s over. They were a bit off.
On May 15, the Centers for Disease Control and Prevention estimated that there were “upwards of 100,000″ cases in the country,..
Intrade prices, while still more accurate, were clearly marred in the early days of the epidemic by these supercomputer projections.
“Developed by quants”? As opposed to the startling predictive power of high-tech models which are not developed by quants, presumably.
The problem with a lot of these things is that the model is being used as a crutch to compensate for a lack of observable data. The same lack of data makes it difficult or impossible to fit the model. So don’t blame the quants or the model. Some things are just difficult or even impossible to do.
Is there some reason why we don’t have huge supply of parameter-free empirical distributions for epidemic risk?
The Science in Society blog (http://ssmag.wordpress.com) has posted a response (http://ssmag.wordpress.com/2009/04/26/new-swine-flu-outbreak/) to recent swine flu news, putting it in perspective with previous flu pandemics. Here is a post comparing flu intervention in 1918 and today: http://ssmag.wordpress.com/2009/04/27/flu-intervention-then-and-now/
An excerpt:
Just how important is starting countermeasures early, and what kind of interventions work? The tragedy of the Spanish flu provides a natural laboratory for public health measures, as cities throughout the US differed both in scale and timing of their interventions.
Medical science in 1918 was still getting on its feet. The majority of older physicians of the time were not educated under the scientific regimen of the Flexnerian revolution. The leading bacteriologists of the day mistakenly believed that influenza was a bacterial disease, and it was not until 1943 when it was recognized that a virus was responsible. As a result, medical intervention in the pandemic was of questionable value, not least because most of the best doctors had been drafted to serve in the military for WWI.
However, nonmedical interventions were also employed. These included quarantines, isolation of the sick in makeshift wards, closure of public gathering places such as churches and schools. Quick action (as measured by when flu cases rose to double the baseline number of cases) had a strong correlation with reduced mortality, and that maintaining the measures was important to keep the disease from spreading.
St. Louis, for example, closed schools and canceled public gatherings early, and maintained quarantines for over ten weeks, leading to a significantly lower mortality rate. However, not all cities were as proactive; the median duration of these interventions was only four weeks, insufficient to protect the population. Some cities were even counterproductive: Philadelphia hosted a military parade to promote war bonds, over the objections of numerous doctors and public health officials. (Anything similar happen on campus recently?) Soon afterwards, it became one of the hardest-hit cities in the US.
Some cool H1N1-related Google widgets: http://ssmag.wordpress.com/2009/04/30/google-on-h1n1-swine-flu/
Google.org Flu Trends also didn’t work.
WHO says 25,000 cases worldwide with 13,000 in the US thus far.
http://www.voanews.com/english/2009-06-08-voa18.cfm
CDC says 13,217 in the US as of June 5.
http://www.cdc.gov/h1n1flu/update.htm
Even the linked article explains there were only 7,415 official CDC cases at the time. The 100,000 number is, strangely enough, the result of a high-falutin’ model.
I love this blog, but sometimes opinions lead fact a bit too much.
Get them to work on the global warming model, STAT!
Brian O,
Official tallies are known to underestimate the number of cases. Not everyone who gets infected gets sick. Not everyone who gets sick goes to see their doctor. Not everyone who goes to see their doctor gets tested. Initially states were mandating testing for all suspected cases, but as number of cases became less manageable only high priority cases (e.g. pregnant women, hospitalized, asthmatics, etc.) were tested in some states.
Think of the last model you made yourself, and think how you would feel if the results were condensed to one single number without any context, and then criticized for it.
damn, are the bots getting ‘smarter’?
Thomas:
How did the CDC arrive at its estimate? Probably via another mathematical model describing the ratio of undetected cases to detected cases. While that model may have been valid for seasonal flu, where many people who feel sick don’t go to the doctor, who’s to say that the model is valid for H1N1, which had much more publicity around it?
In other words, how do we know the CDC isn’t overestimating the n umber of cases of H1N1 by the same proportion that the Indiana and Northwestern models understimated?
催情性藥å“
ç¿»è¯
醫å¸ç¾Žå®¹
æ±½è»Šå ±å»¢
ç«¥è£æ‰¹ç™¼
Comments on this entry are closed.