This is the accepted version of a paper published in Behavior and Information Technology. This paper has been peer-reviewed but does not include the final publisher proof-corrections or journal pagination.

Corona virus disease 2019 (COVID-19) has proven to be the worst pandemic in modern times in terms of both mortality and infectiousness since the flu pandemic that took place in the early 20 th century, which is also known as the Spanish Flu. First being detected in China on December 8, 2019, the COVID-19 disease has spread swiftly into other countries and continents, which eventually led to its classification as "pandemic" by the World Health Organization (WHO) on March 12, 2020. 1,2 After the first confirmed case in Turkey was detected on March 11, 2019, the number of confirmed cases has increased rapidly and reached 95,591 as of April 21, 2020, according to the Ministry of Health-Turkey. 3

During this period, the total number of confirmed cases reached 148,067, according to figures reported by the Ministry of Health-Turkey. In our previous study, where we employed the SIR model to predict the progress of the COVID-19 pandemic, it was emphasized how imperative it is to forecast the pandemic's progression in the coming future to devise an appropriate policy response. 1 Besides predicting the future progress of the pandemic, an equally maybe more critical policy question concerns the timing for easing and eventually lifting limitations such as curfews and closure of schools and businesses. As of now, since there is no preventive vaccine or prophylactic drug for COVID-19, it is widely accepted that the transmission can only be reduced by isolation 127 127 127

The SIR model and its variants are widely used to predict the progress of COVID-19 worldwide, despite their rather simplistic nature. Nevertheless, robust estimation of the SIR model presents a significant challenge, particularly with limited and possibly noisy data in the initial phase of the pandemic. K-means algorithm is used to perform a cluster analysis of the top ten countries with the highest number of COVID-19 cases, to observe if there are any significant differences among countries in terms of robustness. As a result of model variation tests, the robustness of parameter estimates is found to be particularly problematic in developing countries. The incompatibility of parameter estimates with the observed characteristics of COVID-19 is another potential problem. Hence, a series of research questions are visited. We propose a SPE (“Single Parameter Estimation”) approach to circumvent these potential problems if the basic SIR is the model of choice, and we check the robustness of this new approach by model variation and structured permutation tests. Dissemination of quality predictions is critical for policy and decision-makers in shedding light on the next phases of the pandemic.