What can we take away from COVID Models
Posted: Thu Apr 30, 2020 10:43 am
While it's fun to grab some popcorn and watch the culture wars between the doomsday preppers versus the cosmopolitan anti-vaxxers as a re-imagined WWF Saturday Night Main Event, it's been interesting to watch statistical models get thrown in the spotlight. I like to see how they're framed by officials and news outlets and received by the public as this is probably pretty similar to what we deal with in our industry.
Steve and Boyd, at last years IAI
Is it an imperfect overlay? Sure. Epidemiology gets the benefit of actual incidents being observed and having a contagion that actually spreads through a population in real-ish time. Fingerprints, in theory spread through a population, but at a much slower rate (birth rate to be exact) and it makes updating models subject to much different timelines. Additionally, there are more precise tests for identifying a virus, namely antibodies, so virology models one up us here too in terms of specificity whereas we get latent impressions instead of antibodies and our identification is constructed by Examiners.
All of that being said, there are some overlapping items. First, models are "..really just fancy estimates, and results vary by what factors the modelers put in. Some models updated on a daily basis may seem disconcerting to average folks searching for certainty." You are what you eat, or garbage in garbage out so to speak.
Secondly, statisticians are like photographers, they fall in love with their models. It's important to remember that statistical models can give a false sense of certainty. This goes back to the quote attributed to George Box that 'All models are wrong, but some are useful'. You could even apply the section of that link on parsimony to the OSAC articulation document if you really wanted to get fancy, but I digress.
Lastly, there's the politics of it all. I'm using politics as a shorthand for evangelism here. Especially as it relates to efficacy. What did it mean to flatten the curve and why would the semantics and goalpost now change?. Is there an equivalent to the statistical models in our own industry other than the fact that such complex models need evangelists to preach their advocacy or rebuke them?
What have you noticed?
Steve and Boyd, at last years IAI
Is it an imperfect overlay? Sure. Epidemiology gets the benefit of actual incidents being observed and having a contagion that actually spreads through a population in real-ish time. Fingerprints, in theory spread through a population, but at a much slower rate (birth rate to be exact) and it makes updating models subject to much different timelines. Additionally, there are more precise tests for identifying a virus, namely antibodies, so virology models one up us here too in terms of specificity whereas we get latent impressions instead of antibodies and our identification is constructed by Examiners.
All of that being said, there are some overlapping items. First, models are "..really just fancy estimates, and results vary by what factors the modelers put in. Some models updated on a daily basis may seem disconcerting to average folks searching for certainty." You are what you eat, or garbage in garbage out so to speak.
Secondly, statisticians are like photographers, they fall in love with their models. It's important to remember that statistical models can give a false sense of certainty. This goes back to the quote attributed to George Box that 'All models are wrong, but some are useful'. You could even apply the section of that link on parsimony to the OSAC articulation document if you really wanted to get fancy, but I digress.
Lastly, there's the politics of it all. I'm using politics as a shorthand for evangelism here. Especially as it relates to efficacy. What did it mean to flatten the curve and why would the semantics and goalpost now change?. Is there an equivalent to the statistical models in our own industry other than the fact that such complex models need evangelists to preach their advocacy or rebuke them?
What have you noticed?