Estimate of Covid-19 prevalence using imperfect data

Witold R Rudnicki; Radoslaw Piliszek

Both Model 2 and Model 3 are drastic simplifications of the model described by the Eq. 3. Nevertheless, in the exponential growth phase they have one significant advantage over more realistic distributions, namely they require fewer parameters, and therefore are less prone to overfitting. Effectively, they represent the effects of convolution of distribution of time to death and exponential growth. All parameters for the models, namely the time offset between test result and death Ï„ and time span for deaths [T min , T max were derived for the reference country by fitting models 2 and 3. The quality of the fit between the models and data is estimated using the square of the correlation coefficient between number of deaths estimated by the model and the true data. The final form of the models is then obtained as a weighted average with Model 1: