What influences COVID-19 infection rates: A statistical approach to identify promising factors applied to infection data from Germany

Moritz Mercker; Uwe Betzin; Dennis Wilken

To calculate a third type of measure for recent changes in infection rates, a second GAMM was fitted to the infection data of each AD -similar to the above described GAMM approach, however with two differences: first, the time series has been restricted to the last 14 days, and second, the dependency on the date has been assumed to be (log-)linear instead of nonlinear. All other model specifications were as described above for the calculation of β 1 and β 2 . The advantage of the non-smooth (log-linear) incorporation of the date variable, here, β 3 (the slope of the average log-linear development) was represented by a corresponding date-dependent regression coefficient, and thus confidence intervals respectively standard errors could be computed, the latter representing the certainty of each estimate. Furthermore, after slight reformulations, β 3 can be directly interpreted, namely in terms of average percentage change in infection rates per day. A possible disadvantage of β 3 compared to β 1 and β 2 is that a shorter time series (and thus less data) have been used, which was necessary to select a time frame with nearly log-linear behaviour of infection rates corresponding to the model. All rights reserved. No reuse allowed without permission.