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

Daniel E Platt; Laxmi E Parida; Pierre Zalloua

A response to an article in Nature 27 stated: "A well-known lawyer, now a judge, once grouped witnesses into three classes: simple liars, damned liars, and experts. He did not mean that the expert uttered things which he knew to be untrue, but that by the emphasis which he laid on certain statements, and by what has been defined as a highly cultivated faculty of evasion, the effect was actually worse than if he had." The statement was applied to the specific issue of expert forensic testimony. The statement has been restated as "lies, damn lies, and statistics." The message serves as a warning that statistics collected for certain purposes may not be suited to other purposes. That unsuitability does not reflect any attempt at obfuscation. Specifically, in this case, the use of testing, positive test counts, etc are tilted towards identifying patients who are likely to have specific treatment needs, and to try to identify contacts to stop epidemic spread. These uses render the reported statistics problematic for modeling. Physiological parameters based on identified patients may be biased in terms of the patients who were identified, and the methods by which they were identified. Further, protocols shifted as previously unrecognized community spread and asymptomatic individuals were recognized to be significant contributors to viral spread.