Fig. 2: Effects of climatic variability and switches in dominant serotype on dengue incidence in Singapore.

A–D show posterior marginal effects and density plots for covariates in the final selected model. These include maximum temperature in °C (12 week running average), days without rain (12 week total), Niño 3.4 SSTA (12 week average with a 4 week lag) and weeks since switch in dominant serotype. These are shown on the relative risk scale displaying the median value and associated 95% credible interval and can be interpreted as the effect of the covariate on dengue incidence rate with all other parameters held constant. E compares the estimated mean yearly random effect, \({\gamma }_{a[t]}\), and associated 95% credible interval for a random effects only model including only weekly and yearly random effects \({\gamma }_{a[t]}+{\delta }_{w[t]}\) (in yellow), the final selected climate and serotype model including all climate and serotype covariates and random effects (in purple), a climate only model with random effects (in pink), and a serotype only model with random effects (in green). The estimated yearly random effect from the random effects only model indicates whether dengue incidence was higher or lower for that year than the overall mean incidence. We would expect the estimated yearly random effects for covariate models to be closer to 0 (indicated with a dashed line) when covariates are able to account for interannual variability in dengue incidence. These estimates are based on 1201 observations of reported dengue cases, as shown in Fig. 1.