Skip to main content

Table 2 Bayesian Information Criterion (BIC) computed for Models (810)

From: Practical identifiability in the frame of nonlinear mixed effects models: the example of the in vitro erythropoiesis

Model

\(2ln(\widehat{{\mathcal {L}}})\)

\(\# \theta _F\)

\(\# \theta _R\)

BIC

(8)

1761

\(2^\text {a}/1^\text {b}\)

\(4^\text {a}/5^\text {b}\)

\(1778^\text {a}/1775^\text {b}\)

(9)

1801

3

4

1822

(10)

1759

\(2^\text {a}/1^\text {b}\)

\(8^\text {a}/9^\text {b}\)

\(1784^\text {a}/1782^\text {b}\)

  1. Since the definition of the BIC depends on the decomposition of individual parameters between fixed parameters and random parameters [37, 38], the computation of the BIC is ambiguous for Models (8) and (10). In these models, \(\rho _C^{pop}\) determines jointly the population values of \(\rho _C\) (which is fixed), \(\delta _{SC}\) and \(\delta _{CB}\) (which are random). Consequently, we can compute two different values of the BIC, depending on whether we consider \(\rho _C^{pop}\) as the fixed effect of a fixed parameter (a), or as the fixed effect of two random parameters (b). In practice, this consideration does not seem to affect the outcome of the selection