というNBER論文が上がっている。原題は「Estimating the COVID-19 Infection Rate: Anatomy of an Inference Problem」で、著者はCharles F. Manski(ノースウエスタン大)、Francesca Molinari(コーネル大)。ノースウエスタン大の論文pdfにリンクしたタイラー・コーエンは、2人を最高の計量経済学者(top people with econometrics)と紹介している。

We think it misguided to report point estimates obtained under assumptions that are not well justified. We think it more informative to determine the range of infection rates and rates of severe illness implied by a credible spectrum of assumptions. In some disciplines, research of this type is called sensitivity analysis. A common practice has been to obtain point estimates under alternative strong assumptions. A problem with sensitivity analysis as usually practiced is that, in many applications, none of the strong assumptions entertained has a good claim to realism.
Rather than perform traditional sensitivity analysis, this paper brings to bear econometric research on partial identification. Study of partial identification analysis removes the focus on point estimation obtained under strong assumptions. Instead it begins by posing relatively weak assumptions that should be highly credible in the applied context under consideration. Such weak assumptions generally imply set-valued estimates rather than point estimates. Strengthening the initial weak assumptions shrinks the size of the implied set estimate.


  • 感染率
    • イリノイ:[0.001, 0.517]
    • ニューヨーク:[0.008, 0.645]
    • イタリア:[0.003, 0.510]
  • イタリアでの感染者の重症化率
    • 入院率:[0.001, 0.172]
    • ICU:[0, 0.02]
    • 致死率:[0.001, 0.086]
      • 上限は4月6日時点の確認感染者数の死亡率の0.125よりかなり低い