Ferguson Model is BS. Are All Models For COVID Like This?

Amazon
NIIB16

Go to company page Amazon

NIIB16
May 8, 2020 20 Comments

Neil Ferguson — the guy who advised the UK Government, and recently was exposed in a quartine-breaking mistress scandal — wrote a model over the last 15 years that uses over 450 magic number parameters, and no ground truth controls. Its original source code is still unpublished, but a cleaned-up version was done by Microsoft engineers and published a few days ago.

https://github.com/mrc-ide/covid-sim

It is full of bugs and cannot predict any where near the known results of any country, especially Sweden’s current numbers, without significantly overestimating. It is complete crap, and serves to show that Ferguson has been absolutely wrong on all past models of epidemics and pandemics.

I know scientists do not typically write production-ready software, but this one is way over the top! —especially given the real source code is unpublishable. It begs the question: Are all COVID models crap?

Here’s a Twitter thread that explains more: https://twitter.com/thatkatyagirl/status/1258431484018466817

#Covid19Models

comments

Want to comment? LOG IN or SIGN UP
TOP 20 Comments
  • WeWork
    RPvk36

    Go to company page WeWork

    RPvk36
    No, I heard climate change models are waaaay more accurate
    May 8, 2020 0
  • Most “science” is like this.

    Scientists talk about events that happened millions of years ago based on a few samples, with 1$ budget. They can run accurate simulations on the big bang, and write endless books about dark matter.

    Meanwhile they can’t find how a virus spreads in present time with millions of samples and billions in budget. They can create TED talks about microscopic beings that existed in the ice age but can’t find a lost 300 ton plane that was connected to the internet.

    Most “scientists” are full of sh*t.
    May 9, 2020 0
  • Intel
    Fiolpl

    Go to company page Intel

    Fiolpl
    The general public is too dumb and uneducated to understand statistical model outcomes and probability. The only thing they pay attention to is the worse case end of the range, particularly because that’s all the media publishes about. The primary lesson is to let individuals choose their own risk assessment based on available data instead of a tops-down one size fits all central planning. Just like about every other situation.
    May 10, 2020 10
    • Airbnb
      d🧜‍♀️

      Go to company page Airbnb

      d🧜‍♀️
      Not sick people. Healthy people that don’t want to have a chance of becoming sick. Perhaps they live with others that are high-risk, they are high risk themselves, or they just prefer to play it safe.

      Does an employer have the right to compel those people to put themselves at risk for additional revenue?
      May 12, 2020
    • Intel
      Fiolpl

      Go to company page Intel

      Fiolpl
      Healthy working-age people have a 1 in 10,000 chance of dying, so yes, you have to come to work in order to receive a pay check. CA allows people to take 12 week medical leave if you so desired.

      If you have an at risk person at home, you are going to have to find a new job. Luckily the government is fronting you the equivalent of $50K/year for a few months to do so.
      May 12, 2020
  • Airbnb
    vghT6x

    Go to company page Airbnb

    vghT6x
    No one has accurate models, but if your competing model is “the models are all exaggerating” that’s no better
    May 9, 2020 1
  • Google
    mRYT33

    Go to company page Google

    mRYT33
    I agree his code is poorly written. However, his forecast (accounting for lockdown measures) is still reasonably accurate. Check out Imperial College’s report #9.

    There’s some value in having the right answer, even if the way you got there is hacky.
    May 9, 2020 2
    • Amazon
      NIIB16

      Go to company page Amazon

      NIIB16
      OP
      From Katya, the Twitter feed I linked, who did a fair amount of analysis on the code:

      “All this wouldn't matter if it delivered reasonable results. However, no amount of fiddling with parameters deliver Swedish deaths of less than 90k. The model just consistently over counts infected and hence dead.”
      May 9, 2020
    • Google
      mRYT33

      Go to company page Google

      mRYT33
      That’s related to reproducibility. I’m referring to the actual forecast made months ago. Go read report #9. It’s pretty infamous now for having predicted 2M US deaths if no lockdown measures had been taken.
      May 9, 2020