Table 2.

Calibration performance metrics for a 17-day calibration period quantifying how well the fits of the SEIR model with a Poisson error structure (<dist1>=1) and a negative binomial error structure (<dist1>=3) captured the early phase of the fall wave of the 1918 influenza pandemic in San Francisco. The metrics indicate that the negative binomial error structure better fits the data, especially regarding the 95% prediction interval coverage.

Model	Calibration period	MAE	MSE	Coverage 95% PI	WIS
SEIR model with Poisson error structure ( <dist1>=1 )	17	5.73	58.91	58.82	3.89
SEIR model with negative binomial error structure (<dist1>=3)	17	5.74	61.43	94.12	3.67