Table 2.

Site-specific Tjur's Coefficients of Determination for model predictions when observations from each site were withheld and treated as out-of-sample data

	Absolute effect	Adjusted effect
Dhaka, Bangladesh	0·6%	3·2%
Fortaleza, Brazil	0·3%	0·1%
Vellore, India	0·2%	0·8%
Bhaktapur, Nepal	14·1%	1·1%
Loretu, Peru	0·1%	1·4%
Naushero Feroze, Pakistan	13·1%	1·6%
Venda, South Africa	0·0%	0·2%
Haydom, Tanzania	0·3%	0·9%

Data shown are the proportion of the variability in the rotavirus outcome at each site explained by the final models.