Table 2. Effect of early growth (t^lag, µ and σ_b) and control (t̄^c, V^c and n^tip) parameters on the capacity of the model to predict survival.

Log-likelihood is computed either on bacterial load data (i.e. on the data set we used to fit the model) or on survival data. In this latter case, we used the probability of control predicted by the model as an offset in a binomial general linear model (glm) fitted to survival data. This glm includes an intercept which quantifies the difference between predicted control probability and observed survival. p-values indicate whether keeping a parameter constant for all bacteria significantly alters survival prediction.

		Complete data set			Intermediate survival
		Bacterial load	Survival		Bacterial load	Survival
	df	logLik	logLik	p-value	logLik	logLik	p-value
Full		−1243.119	−17.85860		−800.0603	−11.48770
Control	24	−1629.965	−21.68408	0.999	−839.5523	−17.18319	0.496
t̄c	8	−1660.396	−19.18941	0.954	−839.5791	−16.82534	0.030
Vc	8	−1643.985	−17.37663	1	−826.4296	−14.67760	0.173
ntip	8	−1634.382	−25.23397	0.064	−832.8896	−12.85596	0.603
Growth	24	−1798.630	−81.13355	6.512e-16	−854.1100	−11.37739	1
tlag	8	−1692.743	−16.49076	1	−835.3833	−11.30397	1
µ	8	−1705.175	−61.56015	1.564e-15	−836.3794	−11.75406	0.970
σb	8	−1704.156	−21.95270	0.415	−848.2682	−12.28487	0.810