Quantitative Editor, @FiveThirtyEight | PhD, @LSEGovernment | 
in 
, via 
| she/her
You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
Given the uncertainty around every component of #COVID19 forecasts, how can *anyone* do any useful modeling at all? The more research we did, the more confused I became. But our conversation with @drake_lab helped.https://fivethirtyeight.com/features/why-its-so-freaking-hard-to-make-a-good-covid-19-model/ …
Because there are huge ranges around most, if not all, of the parameters in the models, propagating the uncertainty through all of them (i.e. multiplying them all out) would result in a uselessly huge final range of outcomes. So how can we say anything meaningful?
Basically, many of the uncertainties are correlated; not all combinations of parameters are equally plausible or even possible. The thing is, it's hard to estimate those correlations from the data. But it's not just unhelpful to assume they're uncorrelated -- it's also wrong.
So if you think, like I feared, that the uncertainty means we can't say anything useful at all, that's not true. But this is where domain expertise and experience come in -- that's what helps modelers make informed decisions about how to narrow the range of outcomes.
Of course, even experts can disagree! Which explains why different research groups have come up with different forecasts. But that's what makes this topic a particularly iffy one for non-experts to weigh in on, as this excellent @W_R_Chase piece outlines:https://www.williamrchase.com/post/why-i-m-not-making-covid19-visualizations-and-why-you-probably-shouldn-t-either/ …
I mean, welcome to academic modelling, in many cases. The one thing going for you is that in many instances, errors aren't correlated and are normally distributed, so the chances of several data uncertainties all coming with extreme worst case true parameters is low.
IE the chances are good that your errors will cancel out to some degree, and the reasonable worst case scenario is somewhat better than, e.g., the 95th quantile on all of your error distributions.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
