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

The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/2020.04.14.20064501 doi: medRxiv preprint and Transit stations. The subsequent GAM-analysis with these predictors reveals highly significant dependencies on infection density and first infection. In particular, the positive regression coefficient for first infection (0.02, 95 %-CI: [0.01,0.03]) indicates that infection rates decay the stronger the earlier the first infection in this region has been detected. At first glance, one might think that an earlier start leads to an earlier decay due to the natural progression of the infection curve. However, the 'natural progression' would be a further increase for a long time, which only saturates or decreases to changes in behaviour. The latter can be caused either by intrinsic (psychological) or extrinsic factors (sanctions). Since sanctions (such as curfews) have been realised approximately at the same time everywhere in Germany (with only differences of some days), the observed decay in connection with early detected infections is also most probably a result of the above mentioned psychological factors. A highly significant negative value for infection density (-97.2, 95 %-CI: [-135.6,-58.9]) suggest again that high local infection densities lead to stronger current decays of infection rates. Thus, both effects are most probably related to the above described awareness of the thread. Furthermore, the positive regression coefficient for Age (0.022, 95 %-CI: [-0.002,0.045]) indicates that in regions with more people older than 65, infection rates decay less. This might be explained by the fact that younger people may show a more mobile behaviour (e.g. due to their work) so that sanctions (such as curfew) might affect them in a stronger way. Other factors might be that older people are possibly less frighten (due to more life experience), and it also could be more difficult for them to substantially change their habits. The negative value for Transit stations (-0.012, 95 %-CI: [-0.027,0.002]) finally indicates that an increased use of public transport leads to a decrease of recent infection rates. This result is counter-intuitive, and we therefore think that here, possibly a random correlation was detected. Indeed, this result is statistically non-significant and movement data have been introduced only at the coarse federal statelevel. For future work, such data with a finer spatial resolution (such as currently available for the UK https://www.gstatic.com/covid19/mobility/2020-04-05_GB_Mobility_Report_en.pdf) would allow for more reliably estimates.