Probabilistic interpretation of data: a physicist's approach
Devinderjit Singh Sivia · Radiation Protection Dosimetry · 2014
Modern science, unlike its ancient philosophical ancestor, is based on empirical observation and measurement. The analysis of the data, to draw appropriate inferences, is then central to the whole endeavour of learning. Indeed, it forms the basis of sound decision-making. Despite its importance, the training that most of us receive in data analysis is shockingly poor. If not close to non-existent, it is either presented as a collection of disconnected cookbook recipes or taught as exercises in pure mathematics. Miller's book is the latest in a small but growing line of texts published in response to the above shortcomings. Their common features are that the authors are not statisticians and they adopt the Bayesian approach to probability. This entails the view that probabilities represent a state of knowledge, or a degree of belief, based on the information at hand, and predates the conventional or orthodox insistence that they must relate to the limiting frequencies of events in repeated trials. It is surprising, therefore, that Miller always tries to make a connection with the frequency interpretation even though it is not necessary.