Subject: | |
From: | |
Reply To: | |
Date: | Thu, 18 Feb 2021 13:46:25 -0500 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
I’m detecting a surprising lack of understanding of probabilities when it comes to simulation models. All computer models, especially those aiming to map real events like whether of medicine, report their results with a probability. So when the weather model says it will be windy, it likely says that there’s an x% chance of this occurrence. What does it mean? It means that when 100 data points that match the current weather were fed into the model, x% of the time the model reported winds and it matched its prediction with the subsequent observations. But 100-x% of the time IT DID NOT. Does it make the model worthless? Of course not. Humans just have a hard time with imprecision - we want to know for sure whether it will rain or not, not an 80% assurance that it will.
Someone else made a point on this thread about lack of data about COVID virus and its treatments. I think that this is grossly uninformed. We don’t know everything about COVID, but we know a hell of lot of it, including its full gene structure, and that knowledge allowed us to develop & test vaccines in record time. The amount of computing power (aka models) that has been thrown at COVID is unprecedented and we’ve only begun to scratch the possibilities of the boundaries between the real world and computer simulations.
For modeling skeptics, I highly recommend “The Patient Equation” by Glenn De Vries, which basically postulates that we’re on the cusp of unprecedented insights into biology (and medicine) based on the data we’ve started to collect (biomarkers) and the computer analysis & simulations that we’re now able to apply to them. He provides a lot of examples of how this is already being done in practice. Full disclosure, Glenn De Vries is the co-CEO of Medidata purchased by the company I work for. 60% of the COVID vaccine trials are being run using Medidate software.
Przemek
> On Feb 18, 2021, at 12:39 PM, Allen Dick <[log in to unmask]> wrote:
>>
>
> "All models are wrong, but some are useful".
>
>
***********************************************
The BEE-L mailing list is powered by L-Soft's renowned
LISTSERV(R) list management software. For more information, go to:
http://www.lsoft.com/LISTSERV-powered.html
|
|
|