Active cases vs new cases

[nikohansen]

I wonder what lead the choice to show active cases rather than new cases. I understand that active cases reflect the current problem at this point in time. New cases however have, AFAICS, two possibly more important features:

• they give the first indication whether a measure taken is actually effective, assuming that there is currently no effective therapy available but (many) ways to modulate spreading / number of new infections.

• the trend of new cases directly predicts whether the problem will persist in near future or go away. It is as simple as checking whether the derivative of new cases is above or below zero. Of course, the trend in new cases also predicts to a great extend the trend of active cases in a-week-or-so from now.

Roughly speaking, active cases are the less noisy but lagging indicator, while new cases are the more noisy but a leading indicator. I see more advantages in showing new rather than active cases, but why not show at least both?

I saw that

Unfortunately, they no longer report recovered, hence we cannot plot active, just confirmed

which would make it even more relevant to show new cases. AFAICS, cumulated confirmed cases do not really reflect any relevant aspect of the problem if we do not know the ratio of recovered cases and if the number of cases is small compared to the overall population (that is, if there cannot be a relevant effect on population immunity). Am I missing something?

[GaelVaroquaux]

I wonder what lead the choice to show active cases rather than new cases.

The goal was to show the "active", because this are spreading the pathogen. Anyhow, we can no longer show them, and we can only show the total confirmed. New cases are very noisy. We would probably need to apply a temporal smoothing.

[nikohansen]

Maybe you overlooked it, but this is how new cases and smoothed new cases (with window size three) look:

http://www.cmap.polytechnique.fr/~nikolaus.hansen/covid19-figure2.html

The noise level seems certainly not to be detrimental when considering the graph as a whole (versus only single data points which should not be considered reliable on their own).

The main problem as I see it isn't really noise, but that total cases carry less and less meaningful information to begin with (like when comparing, say, China total cases vs US total cases at this point in time).