TABLE 2.

All‐cause and cause‐specific absences model performance using three school‐based cohorts’ data to predict confirmed influenza cases in Allegheny County, Pennsylvania, USA during the 2007‐2008, 2012‐2013, and 2015‐2016 influenza seasons

Flu season	Cohort	Model a , b , c	In‐sample AICc (∆)	R 2	MAE	Relative MAE
2007‐2008	PIPP	Week‐only	183 (0)	0.97	8.9	1.0 (Ref.)
		All‐cause absence	215 (32)	0.44	158.6	17.8
		ILI‐specific absence	185 (2)	0.49	45.1	5.07
2012‐2013	SMART	Week‐only	161 (0)	0.93	14.4	1.0 (Ref.)
		All‐cause absence	187 (26)	0.98	11.2	0.78
		ILI‐specific absence	174 (13)	0.99	8.6	0.60
2015‐2016	SMART2	Week‐only	204 (0)	1.0	8.6	1.0 (Ref.)
		All‐cause absence	214 (10)	0.82	5.0	0.59
		ILI‐specific absence	206 (2)	0.84	18.6	2.17
Pooled analysis	PIPP, SMART, SMART2	Week‐only	664 (0)	0.35	51.6	1.0 (Ref.)
		All‐cause absence	665 (1)	0.30	52.4	1.02
		ILI‐specific absence	667 (3)	0.37	51.5	0.99

Abbreviations: ∆AICc: change in corrected Akaike's Information Criterion; ILI: influenza‐like‐illness; MAE: mean absolute error; PIPP: Pittsburgh Influenza Prevention Project; R ²: coefficient of determination; SMART: Social Mixing and Respiratory Transmission in Schools study; SMART2: Surveillance, Monitoring of Absences & Respiratory Transmission Study.

^a The week‐only model included only week of the year and absence models included weekly lagged absence rates from the previous week, week of the year, and average temperature.

^b SMART models included weekly lagged absence rates and week of the year.

^c Cross‐validation used leave 20% of schools out.