Bibliography

(eng): Metrology as the science about measurement is highly intertwined with statistical point estimation. Evaluating and controling uncertainty of measurements and analyzing them by means of exploratory data analysis (EDA) or predictive data mining requires to exploit advanced tools of statistical estimation. The main focus of the chapter is devoted to nonstandard approaches to the analysis of measurements in two fundamental models, namely, the location model and linear regression. Robust regression methods, which are resistant to the presence of outlying (anomalous) measurements, are discussed here. An illustration of their performance over a real dataset related to thyroid disease and a Monte Carlo simulation reveal here the least weighted squares estimator, which has remained quite neglected so far, outperforms much more renowned robust regression estimators in terms of the variability. Further, Bayesian estimation in the location model is revealed here to have the ability to incorporate previous measurements in a very intuitive way. Finally, the chapter gives a warning that linear regression performed on data contaminated by measurement errors yields biased estimates and requires specific estimation tools for the so-called measurement error model.