"Historical Trend" display in Indicator table is misleading

[RoniRos]

In the "historical trend" sparklines shown in the main covidcast dashboard, the Y axis is automatically scaled to use the entire available height to span the minimum and maximum values over the displayed period of about a month. As a result, even tiny changes (e.g. -0.21%) seem meaningful, even though they are not.

Solution: impose a minimum span of the Y axis, perhaps based heuristically on historical [min,max] values.

[krivard]

Could you link to the page where you see the problem and provide a screenshot please?

From your description though, it sounds like this won't be easy to resolve. An absolute minimum isn't feasible due to the widely varying scales of our different indicators (fractions to millions), and a minimum per-indicator would either slow page load times considerably or require a bunch of new infrastructure to compute, store, and serve regularly-updated and recency-weighted range criteria for a dozen (if just the main dashboard) or hundreds (if CTIS is included) of signals.

if you still think it's worth it, we'll need to find someone to build the backend part of that system. sam can probably handle the changes to the frontend pretty easily though.

[RoniRos]

https://delphi.cmu.edu/covidcast/ (yes, the entry page). See below:

I realize that calculating appropriate [Min,Max] may be complicaated/costly.

How about manually setting a fixed minimum Y-axis span per signal? Coarse, but will alleviate the most extreme nonsense, like the -0.21% example above. I volunteer to produce these constants for the ~12 signals on the main dashboard.