People complain, but a Gaussian is good enough for a rough approximation. The real curve is something like the derivative of a logistic, convoluted with something like a 15 day interval[1], and some noise, a lot of noise. Take a look at the SIR models and similar <a href="https://en.wikipedia.org/wiki/Compartmental_models_in_epidemiology" rel="nofollow">https://en.wikipedia.org/wiki/Compartmental_models_in_epidem...</a> , that are betters models. They are close enough to a Gaussian, but the Gaussian has a faster decay, so a Gaussian approximation underestimate the length of the tail.<p>I don't want to be mean, but the curve for Argentina is fucking inaccurate. [Hi from Argentina!] <a href="https://covid-gauss.site/country/argentina" rel="nofollow">https://covid-gauss.site/country/argentina</a> The expected peak is in the middle of May, the epidemy is not reducing here :(. Moreover, we made a huge mistake that effectively broke the quarantine last week, so I expect the peak to be earlier and higher.<p>Perhaps your fit is good enough for post-peak countries, but not for pre-peak countries.<p>[1] The important part is not how many people is diagnosed, but how many people is in the hospital. So it is necessary to model that the bad cases will need something like 2 weeks / 1 month of hospitalization.