TABLE 1.
Fit and Performance of negative binomial models of seasonal variables including and excluding one‐week‐lagged county‐level all‐cause school absence rates to predict weekly confirmed influenza cases in Allegheny County, Pennsylvania during the 2010‐2015 seasons
| Model validation | Leave 20% randomly sampled out data (n = 124) | Leave 52 randomly sampled weeks out (n = 106) | Leave 20% randomly sampled schools out (n = 159) | |||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Model a | Variables | df | ∆AICc b | Relative MAE b | df | ∆AICc b | Relative MAE b | df | ∆AICc b | Relative MAE b |
| 1 (Ref.) | Week, temperature, RH | 8.5 | 0.0 | 1.0 | 8.8 | 0.0 | 1.0 | 10.0 | 0.0 | 1.0 |
| 2 | Week, temperature, all‐cause absence rates | 7.3 | −4.0 | 0.97 | 7.2 | −4.0 | 1.05 | 7.4 | −4.0 | 1.0 |
| 3 | Week, RH, all‐cause absence rates | 7.9 | −1.0 | 1.23 | 8.0 | −1.0 | 1.2 | 9.5 | −1.0 | 1.27 |
| 4 | Week, temperature, RH, all‐cause absence rates | 8.8 | 2.0 | 0.95 | 8.9 | 1.0 | 1.0 | 10.3 | 0.0 | 0.95 |
Abbreviations: ∆AICc, change in Akaike's Information Criterion corrected for small sample size; RH, relative humidity.
a
Each model used negative binomial regression and used generalized additive models to estimate degrees of freedom for nonlinear (ie, spline) variables.
b
Changes in AICc and relMAE compared all models to the reference (model 1), a seasonal variables‐only model that contains calendar week, average weekly temperature, and average weekly relative humidity.