Problems in identifying the dynamics of biological systems from very short records

Surprisingly little attention has been paid to the fact that many identification experiments yield very short records. In biological modelling and some other areas, records may well be not merely too short for asymptotic analysis to be of much interest, but short enough to prevent statistical analysis based on testable noise statistics and short enough for the influence of individual observations to be crucial in experiment design. The paper discusses characterisation of uncertainty in short records by bounds rather than statistics, and its suitability for such purposes as assessing the adequacy of the output sampling schedule. Examples from biological growth-control modelling and medical drug-tolerance testing are given.