How many cases studies needed?

[seabbs]

Having a single case study (i.e Germany) greatly simplifies the work but may make the results less generalizable. Having 2 (i.e Germany and the UK) may improve matters. More may be better but obviously dramatically increases analysis degrees of freedom.

[jbracher]

In which sense does the addition of countries add analysis degrees of freedom? Because variant data for each country need to be pre-processed individually?

[seabbs]

In the sense, we have more to summarise and draw inferences from + compute burden of course. Currently thinking using your data from Germany and then scraping data for 3 other countries with relatively late introductions (as earlier data doesn't seem to be available with a permissive license)

[seabbs]

Currently, we have data covering quite a few countries included in the ECDC hub. See:

evaluate-delta-for-forecasting/data-raw/observations/covariants.org/process-obs.md

Line 65 in b4acd63

Could either try and use all of them (or at least those with semi-complete data) but this would potentially be quite a heavy computational burden or choose a subset.

If choosing a subset then which? Personally, interested in the UK but sadly historical data coverage is poor as covariants data only starts from the end of June (report date not sample taken date). Otherwise could try taking a representative sample or choose based on other criteria such as interest.

Whilst developing the analysis will likely pick a small subset just for speed of computation etc but can expand down the line as needed.

Other options:

• Use a subset that have good historical data for the main retrospective analysis and another set (i.e including the UK) for data availability sensitivity analysis.
• For countries with high Delta burdens early could use expert derived reporting schedules (i.e we guess) to approximate when we think data was available. Not that keen on this option.

[seabbs]

Added four countries for now with the plan to consider all available with a sufficient case burden to make forecasting sensible.