Baking Better With Cochrane-Orcutt (part 2)

Tackling serial correlation in generalised linear modelling of rolling 7-day new cases detected (rev 1.0)

Yesterday I talked about sausage rolls, tarts, tests, people tested and COVID case counts being a function of testing people. The spectre of serial correlation that had haunted my Pandemic En Croûte refused to go away and base model adequacy was called into question as we watched ρ-hat do a crazy hat dance.

To expect a pandemic to remain the same pandemic over a span of 2 years 4 months is asking a lot from a pandemic. Viruses are going to get jittery, people are going to get jittery and the government is going to get jittery. Definitions will change, data collection will change, methodologies will change and labs will get up to different things. Just because we can plot a daily time series of rolling 7-day new cases detected from Mar 2020 through to Jun 2022 doesn’t mean this series is homogeneous or makes any clinical sense.

Parallel Modelling (Bakewell tart method)

It is at this point that I need to bring in a sharp knife and slice the data into edible chunks, hopefully of uniform size, shape and colour. Simpson knew all about this back in 1951 but he was beaten to the idea by the legends that were Karl Pearson and crew in 1899.

Here is my slicing guide…