Covid-19

Last Updated: 19-Feb-2022

Overview

• Covid-19 was not just a bad flu season. Examples of bad flu seasons as reported at the time include winter 1999-2000 (as seen by the monthly spike in January, 20000) and winter 20003-4. The statistics can't provide a cause for the increase in deaths starting in April of 2020 (maybe it was Fentanyl overdoses ...), but the actual deaths compared to expected deaths was much worse than even bad flu seasons in the past 20 years.
  
  
• The total "excess" deaths roughly matches the reported Covid deaths. California reported around 50,000 Covid deaths in 2021 and saw a bit under total 55,000 deaths above those expected by the model.
  
  Again, the model can't reveal the cause of these excess deaths, but it is reasonable to assume that either Covid or the effects of the lockdown to fight Covid are responsible. Since the Covid death numbers are similar to the deaths over expected it is reasonable to conclude that the bulk of these deaths were basically caused by Covid. In short, the reported Covid deaths are likely fairly correct.

Details of Analysis

Raw Death Totals by Month

A few things to note about the death totals by month for years 2000 - 2019:

• Each month of the year has a different 'typical' death count. December and January tend to have more deaths than other months. June - September tend to have the lowest typical death counts. Broadly, more people die during the winter months than other months. Fewer people die in summer.
  
  
• Different months have different year-to-year variation:
  
  

  2000-2019 Monthly Min and Max Death Counts (ordered by Max)

  Month	Min	Max
  January	21231	28546
  December	20219	25530
  March	20022	24979
  February	18809	22919
  May	18557	22432
  April	18207	22348
  October	18643	22339
  November	18318	22241
  June	17771	21483
  July	18082	21358
  August	18242	21135
  September	17602	21064

  
  Note that the 2000-2019 highest death total for August and September are lower than the lowest death total for January!
  
  
• The typical death rates climb from 2000 to 2019. This is not surprising because California's population was rising at the same time.

Build a Model

Compare Model to Actual

This year's wave of influenza has become widespread across the nation, overwhelming emergency rooms in cities from Boston to Los Angeles, filling hospital beds and forcing postponements of operations as staff members turn to treating the rising number of flu patients.

— New York Times, Jan 8, 2000

Looking back on the 2003-04 influenza season, federal health officials say it was rougher than the previous three seasons but not unusual for years when the predominant flu virus is A(H3N2).

—CIDRAP News, Apr 08, 2004

Model vs Actual T-Scores

Because some months (e.g. December, January, February) have more year-to-year variation we want to look at how far from the model a given month is TAKING INTO ACCOUNT the expected variation. T-scores do this and t-scores for the difference between the real deaths and the model predicted deaths are:

T-scores with a magnitude of greater than 2 are noted with a gray backgroud. If monthly death are 'normally' distributed then we would expect about one month in 20 to have a t-score greater than 2 or less than -2. For 2000-2019 there were actually fewer than the expected excess values. This is likely because deaths are correlated — people who die in one month can't die again and people who don't die get older and are more likely to die in future months.

In any event, it is clear that the deaths per month changed significantly in April of 2020. This was the single worst month dating back to January of 2000. May and June were almost as bad and July was substantially worse.

Deaths per month remained substantially above 'expected' until December 2021, which was still higher than expected but back into the range of high that was within typical variation.

2020 saw almost 55,000 deaths more than expected. 2021 saw a bit over 60,000 deaths over expected. From 2000 - 2019 the year with the most deaths over expected was 2017 which saw about 9,000 more than expected.