Baking Better With Cochrane-Orcutt (part 3)

Assessing the influence of mass viral testing programmes on case positivity: are we experiencing a testdemic? (rev 1.1)

In part 2 of this series I declared that “until I can get my Durbin-Watsons to rise I shall continue to improve the recipe”, so I went back to basics in Chicken Or Egg? only to discover some interesting structure within the rolling 7-day count of new cases detected with cases mysteriously rising/falling a few days before test activity rises/falls, and doing the same a fortnight after tests. With head-scratching also hitting a peak, Nadal retiring with a nasty abdominal injury and Cochran-Orcutt unable to solve the issue of serial correlation, I decided to start over again with plain old daily counts of new cases detected.

Whilst we’ve all seen how cases rise and fall over time (even though these ‘cases’ are merely positive test results and not medical cases proper) few of us have set our eyes on the autocorrelogram for this time series. Yes, there is such a thing as an autocorrelogram, and here is the Wiki entry proving its existence and mentioning its worth in revealing hidden patterns within data.

Below is autocorrelogram for the first order difference of newCasesBySpecimenDate for England for the period 30 Jan 2020 - 7 Jun 2022. I hope that subscribers recall that we take the first order difference in order to remove the autocorrelation (meaning self correlation) arising from a data series ascending or descending in value over time. Folk whose head is hurting at this point may want to watch this short YouTube clip.