Table 2. The predictive power of the gamma model.

Removed	Predicted	Observed					# SD
AB		18.66	14	1.06	23	0.81	0.37
AD		18.82	13	1.10	20	0.78	0.14
AF		18.07	14	1.08	40	0.81	1.26
AG		22.50	14	1.11	48	0.80	1.46
AH		20.29	15	1.20	27	0.75	0.34
BD		17.77	15	1.17	49	0.80	1.01
BE		15.58	15	1.15	21	0.78	0.34
BG		17.31	15	1.21	42	0.78	1.58
BH		19.47	15	1.18	48	0.77	0.11
BI		16.52	15	1.06	30	0.85	2.05
CD		17.52	12	1.08	20	0.76	0.26
CE		11.56	13	1.05	59	0.85	5.30*
CH		19.49	15	1.16	29	0.78	0.08
DF		19.30	15	1.11	22	0.82	0.79
DH		18.48	15	1.08	38	0.81	0.41
DI		15.40	15	1.19	20	0.76	0.62
EH		16.54	15	1.20	27	0.76	0.06
EI		12.74	14	1.06	31	0.77	0.30

Each analysis consisted of removing one of the 18 double mutants, fitting the model to the remaining data, and predicting the fitness of the removed double mutant. For simplicity, we assumed the same shift () for each analysis. The last column gives the magnitude of the difference between the observed and predicted values as a number of standard deviations. A “*” indicates a significant difference at a 5% significance level from the model predictions with 15 degrees of freedom. A difference greater than 2.13 standard deviations is significant.