arXiv subjects
Publications and source records attributed to G. D'Agostini.
In this note I go through the `proof' of frequentistic confidence intervals and show what it logically implies concerning the value of a physical quantity given an experimental observation (nothing).
This paper reviews the basic ideas behind a Bayesian unfolding published some years ago and improves their implementation. In particular, uncertainties are now treated at all levels by probability density functions and their propagation is performed by Monte Carlo integration. Thus, small numbers are better handled and the final uncertainty does not rely on the assumption of normality. Theoretical and practical issues concerning the iterative use of the algorithm are also discussed. The new program, implemented in the R language, is freely available, together with sample scripts to play with toy models.
Triggered by a recent interesting New Scientist article on the too frequent incorrect use of probabilistic evidence in courts, I introduce the basic concepts of probabilistic inference with a toy model, and discuss several important issues that need to be understood in order to extend the basic reasoning to real life cases. In particular, I emphasize the often neglected point that degrees of beliefs are updated not by `bare facts' alone, but by all available information pertaining to them, including how they have been acquired. In this light I show that, contrary to what claimed in that article, there was no "probabilistic pitfall" in the Columbo's episode pointed as example of "bad mathematics" yielding "rough justice". Instead, such a criticism could have a `negative reaction' to the article itself and to the use of Bayesian reasoning in courts, as well as in all other places in which probabilities need to be assessed and decisions need to be made. Anyway, besides introductory/recreational aspects, the paper touches important questions, like: role and evaluation of priors; subjective evaluation of Bayes factors; role and limits of intuition; `weights of evidence' and `intensities of beliefs' (following Peirce) and `judgments leaning' (here introduced), including their uncertainties and combinations; role of relative frequencies to assess and express beliefs; pitfalls due to `standard' statistical education; weight of evidences mediated by testimonies. A small introduction to Bayesian networks, based on the same toy model (complicated by the possibility of incorrect testimonies) and implemented using Hugin software, is also provided, to stress the importance of formal, computer aided probabilistic reasoning.
Beauty production in deep inelastic scattering with events in which a muon and a jet are observed in the final state has been measured with the ZEUS detector at HERA using an integrated luminosity of 114 pb^-1. The fraction of events with beauty quarks in the data was determined using the distribution of the transverse momentum of the muon relative to the jet. The cross section for beauty production was measured in the kinematic range of photon virtuality, Q^2 > 2 Gev^2, and inelasticity, 0.05 < y < 0.7, with the requirement of a muon and a jet. Total and differential cross sections are presented and compared to QCD predictions. The beauty contribution to the structure function F_2 was extracted and is compared to theoretical predictions.
Take a large bag of black and white beans, with all possible proportions considered initially equally likely, and imagine to make random extractions with reintroduction. Twenty consecutive observations of black make us highly confident that the next bean will be black too. On the contrary, the observation of 1010 black beans and 990 white ones leads us to judge the two possible outcomes about equally probable. According to C.S. Peirce this reasoning violates what he called "rule of balancing reasons", because the difference of "arguments" in favor and against the outcome of black is 20 in both cases. Why? (I.e. why does that rule not apply here?)
A quite old problem has been recently revitalized by Leonard Mlodinow's book The Drunkard's Walk, where it is presented in a way that has definitely confused several people, that wonder why the prevalence of the name of one daughter among the population should change the probability that the other child is a girl too. I try here to discuss the problem from scratch, showing that the rarity of the name plays no role, unless the strange assumption of two identical names in the same family is taken into account. But also the name itself does not matter. What is really important is `identification', meant in an acceptation broader than usual, in the sense that a child is characterized by a set of attributes that make him/her uniquely identifiable (`that one') inside a family. The important point of how the information is acquired is also commented, suggesting an explanation of why several people tend to consider the informations "at least one boy" and "a well defined boy" (elder/youngest or of a given name) equivalent.
The aim of this paper, triggered by some discussions in the astrophysics community raised by astro-ph/0508529, is to introduce the issue of `fits' from a probabilistic perspective (also known as Bayesian), with special attention to the construction of model that describes the `network of dependences' (a Bayesian network) that connects experimental observations to model parameters and upon which the probabilistic inference relies. The particular case of linear fit with errors on both axes and extra variance of the data points around the straight line (i.e. not accounted by the experimental errors) is shown in detail. Some questions related to the use of linear fit formulas to log-linearized exponential and power laws are also sketched, as well as the issue of systematic errors.
It is curious to learn that Enrico Fermi knew how to base probabilistic inference on Bayes theorem, and that some influential notes on statistics for physicists stem from what the author calls elsewhere, but never in these notes, {\it the Bayes Theorem of Fermi}. The fact is curious because the large majority of living physicists, educated in the second half of last century -- a kind of middle age in the statistical reasoning -- never heard of Bayes theorem during their studies, though they have been constantly using an intuitive reasoning quite Bayesian in spirit. This paper is based on recollections and notes by Jay Orear and on Gauss' ``Theoria motus corporum coelestium'', being the {\it Princeps mathematicorum} remembered by Orear as source of Fermi's Bayesian reasoning.
Testing hypotheses is an issue of primary importance in the scientific research, as well as in many other human activities. Much clarification about it can be achieved if the process of learning from data is framed in a stochastic model of causes and effects. Formulated with Poincare's words, the "essential problem of the experimental method" becomes then solving a "problem in the probability of causes", i.e. ranking the several hypotheses, that might be responsible for the observations, in credibility. This probabilistic approach to the problem (nowadays known as the Bayesian approach) differs from the standard (i.e. frequentistic) statistical methods of hypothesis tests. The latter methods might be seen as practical attempts of implementing the ideal of falsificationism, that can itself be viewed as an extension of the proof by contradiction of the classical logic to the experimental method. Some criticisms concerning conceptual as well as practical aspects of na\"ıve falsificationism and conventional, frequentistic hypothesis tests are presented, and the alternative, probabilistic approach is outlined.
The French Academy units of time and length - the second and the meter - are traditionally considered independent from each other. However, it is a matter of fact that a meter simple pendulum beats virtually the second in each oscillation, a `surprising' coincidence that has no physical reason. We shortly review the steps that led to the choice of the meter as the ten millionth part of the quadrant of the meridian, and rise the suspicion that, indeed, the length of the seconds pendulum was in fact the starting point in establishing the actual length of the meter.
The problem of inferring the binomial parameter p from x successes obtained in n trials is reviewed and extended to take into account the presence of background, that can affect the data in two ways: a) fake successes are due to a background modeled as a Poisson process of known intensity; b) fake trials are due to a background modeled as a Poisson process of known intensity, each trial being characterized by a known success probability p_b.
The issue of asymmetric uncertainties resulting from fits, nonlinear propagation and systematic effects is reviewed. It is shown that, in all cases, whenever a published result is given with asymmetric uncertainties, the value of the physical quantity of interest is biased with respect to what would be obtained using at best all experimental and theoretical information that contribute to evaluate the combined uncertainty. The probabilistic solution to the problem is provided both in exact and in approximated forms.
This report introduces general ideas and some basic methods of the Bayesian probability theory applied to physics measurements. Our aim is to make the reader familiar, through examples rather than rigorous formalism, with concepts such as: model comparison (including the automatic Ockham's Razor filter provided by the Bayesian approach); parametric inference; quantification of the uncertainty about the value of physical quantities, also taking into account systematic effects; role of marginalization; posterior characterization; predictive distributions; hierarchical modelling and hyperparameters; Gaussian approximation of the posterior and recovery of conventional methods, especially maximum likelihood and chi-square fits under well defined conditions; conjugate priors, transformation invariance and maximum entropy motivated priors; Monte Carlo estimates of expectation, including a short introduction to Markov Chain Monte Carlo methods.
Data obtained during five months of 2001 with the gravitational wave (GW) detectors EXPLORER and NAUTILUS were studied in correlation with the gamma ray burst data (GRB) obtained with the BeppoSAX satellite. During this period BeppoSAX was the only GRB satellite in operation, while EXPLORER and NAUTILUS were the only GW detectors in operation. No correlation between the GW data and the GRB bursts was found. The analysis, performed over 47 GRB's, excludes the presence of signals of amplitude h >=1.2 * 10^{-18}, with 95 % probability, if we allow a time delay between GW bursts and GRB within +-400 s, and h >= 6.5 * 10^{-19}, if the time delay is within +- 5 s. The result is also provided in form of scaled likelihood for unbiased interpretation and easier use for further analysis.
The end of LEP and SLC is a good moment to review the way to summarize search results in order to exploit at best, in future analyses and speculations, the pieces of information coming from all experiments. Some known problems with the usual way of reporting results in terms ``CL limits'' are shortly recalled, and a plea is formulated to publish just parametrized likelihoods, possibly rescaled to the asymptotic insensitivity limit level.
This analysis is based on the same ideas and numerical inputs of the recent paper by Ciuchini et al. on the subject. Some approximations are applied, which make analytical calculations applicable in most of the work, thus avoiding Monte Carlo integration. The final result is practically identical to the one obtained by the more detailed numerical analysis.
Within the Standard Model, a review of the current determination of the sides and angles of the CKM unitarity triangle is presented, using experimental constraints from the measurements of |ε_K|, |V_{ub}/V_{cb}|, Δm_d and from the limit on Δm_s, available in September 2000. Results from the experimental search for {B}^0_s-\bar{B}^0_s oscillations are introduced in the present analysis using the likelihood. Special attention is devoted to the determination of the theoretical uncertainties. The purpose of the analysis is to infer regions where the parameters of interest lie with given probabilities. The BaBar "95 %, C.L. scanning" method is also commented.
Criticisms of so called `subjective probability' come on the one hand from those who maintain that probability in physics has only a frequentistic interpretation, and, on the other, from those who tend to `objectivise' Bayesian theory, arguing, e.g., that subjective probabilities are indeed based `only on private introspection'. Some of the common misconceptions on subjective probability will be commented upon in support of the thesis that coherence is the most crucial, universal and `objective' way to assess our confidence on events of any kind.