Using One’s Eyes

A reader writes, in reference to CDC Covid-19 data on fatalities and excess fatalities, and how the estimates change over time as new data come in:

I too have an inference that the daily data HAS NOT changed enough in the provisional period to change trend trajectory.

Provisional data extends way back (for the version released 9/23, footnotes indication the provisional nature of the data extends back to February 1st!), but usually one can compare vintages to see that big revisions go back for about a month and a half from date of release. Here’s a graph of the data referred to, the estimated excess fatalities according to various vintages of CDC data:

Figure 1: Excess fatalities, 9/23 vintage (blue), 9/16 vintage (tan), 9/9 vintage (green), 9/2 vintage (red), 8/25 vintage (brown). Note excess fatalities differ from CDC series which are bounded below at zero. Source: CDC , various vintages, and author’s calculations.

Using the 8/28 vintage of CDC data, excess fatalities are declining as of 8/8; using 9/23 vintage, excess fatalities are rising or flat as of 8/8…

What about CDC tabulation of reported Covid-19 fatalities?

Figure 2: Covid-19 fatalities, 9/23 vintage (blue), 9/16 vintage (tan), 9/9 vintage (green), 9/2 vintage (red), 8/25 vintage (brown). Source: CDC , various vintages, and author’s calculations.

Using the 8/28 vintage of CDC data, officially designated Covid-19 fatalities are declining as of 8/1; using 9/23 vintage, excess fatalities are rising or flat as of 8/1…

For me, this suggests that one should not be blinded by one’s priors. One should look at the data, consider the limitations of that data, and proceed accordingly, sometimes with caution. Doesn’t mean throw away the data. It means think about the data.