Evaluation of the 2022 West Nile virus forecasting challenge, USA

Parasites & Vectors

Table 2 Impact of model characteristics across county subsets

County subset	Covariate	2.5th percentile	median	97.5th percentile
(a) All counties	Bayesian	−0.020	−0.015	−0.009
	Regression	−0.013	−0.010	−0.006
	Climate	0.015	0.022	0.029
	Mosquito surveillance	0.009	0.021	0.038
	Demographic	0.018	0.029	0.045
	Any avian	0.009	0.015	0.022
(b) High caseload counties	Climate	0.083	0.175	0.267
	Mosquito surveillance	−0.151	−0.102	-0.045
	Demographic	0.265	0.544	1.270
	Any avian	−0.199	−0.138	−0.085
(c) Counties with historical cases	Regression	−0.023	−0.018	−0.014
	Climate	0.034	0.043	0.053
	Demographic	0.015	0.025	0.037
	Any avian	0.001	0.008	0.018
(d) Counties without historical cases	Climate	0.001	0.002	0.003
	Mosquito surveillance	0.002	0.003	0.005
	Demographic	0.002	0.003	0.004
	Any avian	0.002	0.003	0.003

Regression coefficients of individual model characteristics on weighted interval scores for models incorporating that model characteristic compared with the models that did not, determined by a Bayesian generalized linear model. Negative values indicate higher skill when the characteristic is included and positive values indicate lower skill when the characteristic is included. The median value of impact of these characteristics is shown, along with 95% confidence interval bounds. Analysis performed using (a) all counties (n = 3108), (b) high-caseload counties (n = 49), (c) counties with historical cases (n = 2054), and (d) counties without historical cases (n = 1054). Only covariates with significant coefficients included.

ISSN: 1756-3305