GLM based auto-regressive process to model Covid-19 pandemic in Turkey

Abstract Objectives: Our objective is to propose a robust approach to model daily new cases and daily new deaths due to covid-19 infection in Turkey. Methods: We consider the generalized linear model (GLM) approach for the autoregressive process (AR) with log link for modelling. We study the data between March 11, 2020 that is the date first confirmed case occurred and October 20, 2020. After a month of the first outbreak in Turkey, the first official curfew has been imposed during the weekend. Since then there have been curfews each weekend till June 1st. Hence, we include intervention effects as well as some outlying data points in the model where necessary. We use the data between March 11 and September 15 to build the models, and test the performance on the data from September 16 till October 20. We also study the consistency of the model statistics. Results: Estimated models fit data quite well. Results reveal that after the first curfew daily new Covid-19 cases decrease 18.5%. As expected, effect of the curfew gets more significant once a month is past, and daily new cases cut down 24.9%. Our approach also gives a robust estimate for the effective reproduction number that is approximately 2 meaning as of October 20, 2020 there is still a risk for an infected person to cause 2 secondary infections despite all the interventions, preventions, and rules. Conclusion: The GLM approach for AR process with log link produces consistent and robust estimates for the daily new cases and daily new deaths for the data covering almost the first year of the pandemic in Turkey. The proposed approach can also be used to model the cases in other countries.