Mixed Effects Modelling of Excess Mortality and COVID-19 Lockdowns in Thailand

Accurate mortality data are critical for understanding the impact of COVID-19 and learning lessons from crisis responses. But published statistics risk misrepresenting deaths due to limited testing, underreporting, and lack of subnational data, especially in developing countries. Thailand experienced four COVID-19 waves between January 2020 and December 2021, and used a color-coded, province-level system for lockdowns. To account for deaths directly and indirectly caused by COVID-19, this paper uses mixed effects modelling to estimate counterfactual deaths for 2020-2021 and construct a monthly time series of provincial excess mortality. A fixed effects negative binomial and mixed effects Poisson model both substantiate other studies' estimates of excess deaths using subnational data for the first time. Then, panel regression methods are used to characterize the correlations among restrictions, mobility, and excess mortality. The regressions show that mobility reductions modestly curbed mortality immediately upon imposition, suggesting that aversion of non-COVID deaths was a major aspect of the lockdowns' effect in Thailand. However, the estimates are imprecise. An auto-regressive distributed lag model suggests that the effect of lockdowns was through reduced mobility, but the effectiveness appears to have varied over the course of the pandemic.