Lies, Gosh Darn Lies, and Not Enough Good Statistics: Why Epidemic Model Parameter Estimation Fails

Daniel E Platt; Laxmi E Parida; Pierre Zalloua

First, consider the idea that tests may be broad enough to sample spread. When test numbers were low, the likelihood that targeted testing would reflect the general population was also low and sampling uncertainties large. Therefore, a lower bound on testing levels was applied. This cut samples prior to 3/20. Later, test ratios started to demonstrate a downwards bend. This shoulder was cut for samples beyond 3/30. New York doubling time was estimated from a regression between the log of positive test ratios versus time, yielding = 0.0471 ± 0.0095 with a doubling time of 14.7 ± 3.0 adjusting for testing counts. In the alternative scenario, positive samples reflect the proportion of symptomatic patients seeking medical aid, a possibility since the testing was so closely tied to diagnosed patients plus contact surveillance. A regression was performed on the cumulative positive counts shown in Figure 2c ) yielding = 0.1170 ± 0.0021 per day, with a doubling time of 5.9 ± 0.1 days.