The linked will take you to a chart that shows a computation for the rate of Covid transmission by state. In theory, if the Rt is below 1.0, the virus will stop spreading. The creators of the model are Kevin Systrom and Mike Krieger, the co-founders of Instagram. Tom Vladeck is a data scientist and owner of Gradient Metrics.
I find it to be "interesting/useful" ... lots of caveats / discussion below.
The bars for each state are the 80% confidence interval for the model ... meaning there is an 80% chance that the actual number is in that range. So why can't they be more correct? Basically, the map/model is NEVER the "territory" that is what "model" means!
If you choose to dig into more detail, the Faq gives that ... the basic fact is that like all models, reality may be different from reality. This is true of ALL models! The "proof" of model correctness is in the rear view mirror, but must people reporting on models often don't tell you that. The reaction to Covid was based on models ... "masks are not effective at all", "we are going to be shut down for 2 weeks to flatten the curve", "masks are effective to protect others, not you", "it doesn't spread outside", "masks provide significant protection for you", "it does spread outside", etc Draw your own conclusions on how good the models have been.
Why look at the model? Basically for the same reason you look at the weather forecast -- to give you a general idea of how future weather looks TODAY ... the prediction may (and often is) be different tomorrow. Good models allow you to look back at how accurate they were in the past, and discuss a bit about updates to the model they have made, and some they would like to make. (this one does)
Although many statisticians don't like to admit this, this is the reason there are "lies, damned lies, and statistics". Statements like "women are better drivers!" based on them having less accidents are ridiculous, but often made. A "better" statistic would be per mile driven ... just using accident statistics would call a woman who never drives a good driver.
Getting a good statistic is HARD (election predictions should be a clue here). It gets REALLY complicated, and moves into the area of art rather than science. "Correlation is not causality" is also key. Drowning is strongly correlated (meaning the graph curves look alike) with ice cream sales and high temps. Of course neither of them CAUSE drowning.
The important factor here is that today statistics are often presented as "science" when they are in fact just mathematical TOOLS. Climate change is often portrayed as "settled science" of which there is no such thing. Science is by definition NEVER settled! If something is "Science", then it MUST be falsifiable -- the next experiment may show the "settled" to be wrong. This is doubly true for statistical models ... especially as they predict the future rather than look at the past.
So why is Rt a useful statistic? It is just "better" than number of cases which is essentially meaningless ... more cases resulting from more testing doesn't mean more people are getting it, it just means we are finding more. cases (this model attempts to account for that). We don't have good data on number of asymptomatic cases because we are not doing randomized testing. Assuming we are tracking asymptomatic cases (which I doubt), we MAY be able to get a better idea of how many people have had it but did not know it.
Naturally, even THAT doesn't really help us that much, because the "experts" claim that you don't get "much" immunity if you have had it ... therefore you still have to take protective measures -- for a disease that we believe you will be asymptomatic of in at least 50% of cases ... although we of course don't KNOW that either!
Socrates was the wisest man because he knew that he knew nothing.
No comments:
Post a Comment