Speakers
Description
The Athlete Biological Passport (ABP) uses repeated blood measurements to identify unusual changes in an athlete's biological profile. We study conditional false-alarm probabilities in Bayesian haemoglobin screening, using the normal--lognormal (NL) model of Sottas et al. as a reference. In this model, the posterior predictive distribution has no closed-form expression, and the predictive limits used to detect unusual haemoglobin measurements must be computed numerically.
We therefore consider two analytically tractable models. Our main alternative is the normal--inverse-gamma (NIG) model, which retains uncertainty in the athlete-specific mean and variance and provides conjugate posterior updates and a Student's $t$ predictive distribution with exact screening limits. We derive an integral representation of its conditional false-alarm probability and establish its convergence to the nominal level under stationary Gaussian observations. The fixed-variance (FV) model serves as a benchmark, providing an exact false-alarm expression and showing that incorrectly fixing within-athlete variability can lead to persistent miscalibration.
The analytical advantages of NIG require a different joint prior specification, including dependence between the athlete-specific mean and variance. Numerical comparisons with NL show broadly similar behaviour, although differences in finite-sample false-alarm probabilities arise from the different prior assumptions. NIG thus provides an analytically tractable alternative, although its empirical suitability relative to NL requires validation using athlete data.
| Pracovisko fakulty (katedra)/ Department of Faculty | Katedra aplikovanej matematiky a štatistiky |
|---|---|
| Tlač postru/ Print poster | Budem požadovať tlač /I hereby required to print the poster in faculty |