Why Covid Models Misled Policy: Limits, Assumptions and Unintended Trade-offs

Early in the pandemic, influential models helped shape urgent policy choices. But those same models — built on limited data and strong assumptions — have also been used as definitive evidence of what would have happened under alternative policies. A recent claim, reiterated during the Covid inquiry, that a UK lockdown introduced one week earlier in March 2020 would have saved 23,000 lives illustrates why we need a more cautious interpretation of model outputs.

Models, assumptions and real-world behaviour

Models are tools for exploring scenarios, not crystal balls. The Imperial College projection that an earlier lockdown would have dramatically reduced deaths rests on assumptions about how people would have behaved and how the virus would have spread without government intervention. Those assumptions rarely capture voluntary behaviour change — such as people reducing mobility, wearing masks and practising social distancing — which data show began before formal lockdowns were imposed.

For instance, the same modelling framework once projected roughly 35,000 first-wave deaths in Sweden without lockdown. Sweden did not impose a formal lockdown and recorded about 6,000 first-wave deaths — a stark divergence from that projection. Such discrepancies underline the uncertainty inherent in early pandemic modelling when key epidemiological parameters were still being established.

What experts say

Professor Simon Wood (University of Edinburgh) has argued that infections were already declining before the UK lockdown, which challenges the model’s implicit baseline that transmission would have continued unchecked. A paper making similar points is scheduled for publication in the Journal of the Royal Statistical Society.

One academic described modelling an active pandemic as like “doing engineering on a collapsing building” — a vivid reminder of how unstable the inputs were in the first months of Covid-19.

“All models are wrong, but some models are useful.” — George E. P. Box

Delay versus prevention: trade-offs over time

Another important limitation is that many models show interventions may delay deaths rather than prevent them outright. The study behind the 23,000 figure itself notes that bringing lockdown forward could have shifted deaths into later waves, potentially producing a larger second wave that would require renewed suppression measures. Indeed, one modelled scenario suggested first-wave deaths could have risen to about 132,800 if lockdown had been delayed by a week — a result that highlights how timing and long-term strategy interact in complex ways.

Put simply, suppressing transmission quickly without a clear plan for long-term containment or mitigation can lead to resurgence later — a pattern visible in subsequent waves across many countries.

Costs beyond direct Covid mortality

Models rarely capture the full societal costs of stringent interventions. Lockdowns can cause widespread disruption to healthcare services, delay non-Covid treatments, and produce severe economic effects — all of which have implications for population health. Professor Wood has warned that life loss linked to economic consequences may be substantial, arguing for circumspection when adopting measures that could amplify those harms.

Conclusion

Models remain essential for policy planning, but they should be presented and used with careful caveats. Policymakers need to balance epidemiological projections with evidence on behaviour change, healthcare disruption and economic impact. Treating model outputs as established fact risks obscuring uncertainty and the trade-offs that any intervention entails.