Practical difficulties in testing identifiability of linear structural models

Serious practical difficulties in obtaining meaningful parameter estimates have been encountered in the compartmental modeling of the human metabolic system using methionine test data. This is notwithstanding that the system was structurally identifiable, that the input was "sufficiently exciting," that the initial state of the system was known, and that the parameter values could be determined uniquely. The problems are stated in terms of an analysis of experimental data. A case is made for "regional" identifiability, namely the ability to locate the parameters to within a region of parameter space specified by the measurement accuracy.