To add to the above, the distribution of population-level COVID outcomes has fat tails (the system evolution is non-ergodic), so computing point estimates with

To add to the above, the distribution of population-level COVID outcomes has fat tails (the system evolution is non-ergodic), so computing point estimates with ensemble average “expected values” is junk modeling. And computing the (correct) time averages requires owning up to the uncertainty and fatness of tails (which is a very hard thing to model), so people have the lamppost bias of doing the easy (but useless) calculation. An alternative way to explain the meaning of “large uncertainties” is that you can fudge input numbers slightly to get whatever output decision you want, so whoever is making the policy decision is likely to keep fudging numbers till the outcome “feels right”. At that point, the modeling is just dangerously misleading pseudo-science. An intuitive approximation of the decision based on time averages is to “play it safe” in the presence of fat negative tails. “How safe tho?” is a very hard question to answer because of the uncertainties involved (as parent post explained).