Peak data calculation

[agnoski]

Checking data for Italy https://covid-gauss.site/country/italy it looks like the peak data are not well representing the gaussian estimation.

For istance, as for today, the peak is given at:

• date: 2020-04-06
• estimated new infections: 4162

When maybe:

• date: 2020-03-28
• estimated new infections: 5500

would be more meningful, as it is the peak of the gaussian curve too.

Not use that using the median on valuableData at this point is a good idea.

https://github.com/eatthatpie/covid-gauss/blob/6bb5cf53e4f723ac76ba06febc13be8e28ee645e/server/createEstimationReport.js#L109

What do you think?

[eatthatpie]

Hi, thanks for catching it.

Strange as it sounds but I didn't want for the results to rely only on abstract maths so I decided that the timespan, whose middle point will be treated as the peak, will not be the Gaussian curve itself but it will start with actual data and end with estimated values for the future. In that approach the peak date can surely differ from the peak of the curve.

However it seems that people are misled by the difference between the shape of the curve and the calculated peak. So maybe adding another label (indicator?) on the country page like "Gaussian curve peak and estimated infections for that day" (or sth like that, you get the idea) would solve the issue?

[agnoski]

If it's made on purpose, nevermind, so my was just a personal opinion. I think the median is weighing too much the tail of the gaussian, pushing the peaks on the right side, resulting in a "bad" estimation.

South Korea is another example where this approach tends to fail.

Not sure if adding another indicator would help or clarify, because I believe in that case, the "gaussian curve peak" would be more precise than the actual one, so..why having two different numbers if they are both estiamtions? But again, just personal thoughts.