Covid and the lockdown effect: semi-parametric inference about incidence and R in England

The accepted narrative in the UK has been that only the full Covid lockdowns averted the catastrophe predicted by epidemic modelling. But a simple spline based deconvolution applied to daily Covid death data suggested in early May 2020 that infections were actually in decline days before the first lockdown. Statistical inference based on epidemic models, told a different story, of surging infections up until the eve of lockdown (e.g. Flaxman et al. Nature).

However that story collapses when the models' rigid parametric assumptions about how R changes are replaced  by something more flexibly data driven.  A later prominent epidemiological modelling  study used a detailed age and regionally structured epidemic and health service model to infer R and incidence from multiple data streams on deaths, hospitalization and testing in England over most of 2020. It also concluded that the first 2 English lockdowns  were essential to achieving control. Again this result does not survive replacement of strong parametric assumptions with data driven semi-parametric alternatives.

This talk will discuss these results and the methods permitting inference with the semi-parametric model versions. In 2021 more direct incidence reconstructions from two major UK statistical surveillance surveys confirmed the semi-parametric analysis: incidence appears to have been in substantial decline well before all three English lockdowns.