Search

If we accept Savage's set of axioms, then all uncertainties must be treated like ordinary probability. Savage espoused subjective probability, allowing, for example, the probability of Donald Trump's re-election. But Savage's probability also covers the objective version, such as the probability of heads in a fair toss of a coin. In other words, there is no distinction between objective and subjective probability. Savage's system has great theoretical implications; for example, prior probabilities can be elicited from subjective preferences, and then get updated by objective evidence, a learning step that forms the basis of Bayesian computations. Non-Bayesians have generally refused to accept the subjective aspect of probability or to allow priors in formal statistical modelling. As demanded, for example, by the late Dennis Lindley, since Bayesian probability is axiomatic, it is the non-Bayesians' duty to point out which axioms are not acceptable to them. This is not a simple request, since the Bayesian axioms are not commonly covered in our professional training, even in the Bayesian statistics courses. So our aim is to provide a readable exposition the Bayesian axioms from a close rereading Savage's classic book., Comment: 23 pages

I think we can agree that dealing with uncertainty is not easy. Probability is the main tool for dealing with uncertainty, and we know there are many probability-related puzzles and paradoxes. Here I describe a rather idiosyncratic selection that highlights the problem of accepting uncertain statements. Without going into a formal decision theory, there are simple intuitive rational bases for doing that, for instance based on high probability alone. The lottery paradox shows the logical problem of accepting uncertain statements based on high probability. The DNA evidence story is an example of the use probabilistic reasoning in court, where philosophical differences between the schools of inference -- the frequentist, Bayesian and likelihood schools -- lead to substantial differences in the quantification of evidence., Comment: 11 pages, no figures

In Hempel's paradox of the ravens, seeing a red pencil is considered as supporting evidence that all ravens are black. Also known as the Paradox of Confirmation, the paradox and its many resolutions indicate that we cannot underestimate the logical and statistical elements needed in the assessment of evidence in support of a hypothesis. Most of the previous analyses of the paradox are within the Bayesian framework. These analyses and Hempel himself generally accept the paradoxical conclusion; it feels paradoxical supposedly because the amount of evidence is extremely small. Here I describe a nonBayesian analysis of various statistical models with an accompanying likelihood-based reasoning. The analysis shows that the paradox feels paradoxical because there are natural models where observing a red pencil has no relevance to the color of ravens. In general the value of the evidence depends crucially on the sampling scheme and on the assumption about the underlying parameters of the relevant model., Comment: 10 pages, 0 figures

We use a logical device called the Dutch Book to establish epistemic confidence, defined as the sense of confidence \emph{in an observed} confidence interval. This epistemic property is unavailable -- or even denied -- in orthodox frequentist inference. In financial markets, including the betting market, the Dutch Book is also known as arbitrage or risk-free profitable transaction. A numerical confidence is deemed epistemic if its use as a betting price is protected from the Dutch Book by an external agent. Theoretically, to construct the Dutch Book, the agent must exploit unused information available in any relevant subset. Pawitan and Lee (2021) showed that confidence is an extended likelihood, and the likelihood principle states that the likelihood contains all the information in the data, hence leaving no relevant subset. Intuitively, this implies that confidence associated with the full likelihood is protected from the Dutch Book, and hence is epistemic. Our aim is to provide the theoretical support for this intuitive notion.

The Monty Hall puzzle has been solved and dissected in many ways, but always using probabilistic arguments, so it is considered a probability puzzle. In this paper the puzzle is set up as an orthodox statistical problem involving an unknown parameter, a probability model and an observation. This means we can compute a likelihood function, and the decision to switch corresponds to choosing the maximum likelihood solution. One advantage of the likelihood-based solution is that the reasoning applies to a single game, unaffected by the future plan of the host. I also describe an earlier version of the puzzle in terms of three prisoners: two to be executed and one released. Unlike the goats and the car, these prisoners have consciousness, so they can think about exchanging punishments. When two of them do that, however, we have a paradox, where it is advantageous for both to exchange their punishment with each other. Overall, the puzzle and the paradox are useful examples of statistical thinking, so they are excellent teaching topics.

Multiple testing problems arise naturally in scientific studies because of the need to capture or convey more information with more variables. The literature is enormous, but the emphasis is primarily methodological, providing numerous methods with their mathematical justification and practical implementation. Our aim is to highlight the logical issues involved in the application of multiple testing adjustment.

Attacks on the P-value are nothing new, but the recent attacks are increasingly more serious. They come from more mainstream sources, with widening targets such as a call to retire the significance testing altogether. While well meaning, I believe these attacks are nevertheless misdirected: Blaming the P-value for the naturally tentative trial-and-error process of scientific discoveries, and presuming that banning the P-value would make the process cleaner and less error-prone. However tentative, the skeptical scientists still have to form unambiguous opinions, proximately to move forward in their investigations and ultimately to present results to the wider community. With obvious reasons, they constantly need to balance between the false-positive and false-negative errors. How would banning the P-value or significance tests help in this balancing act? It seems trite to say that this balance will always depend on the relative costs or the trade-off between the errors. These costs are highly context specific, varying by area of applications or by stage of investigation. A calibrated but tunable knob, such as that given by the P-value, is needed for controlling this balance. This paper presents detailed arguments in support of the P-value.

The role of probability appears unchallenged as the key measure of uncertainty, used among other things for practical induction in the empirical sciences. Yet, Popper was emphatic in his rejection of inductive probability and of the logical probability of hypotheses; furthermore, for him, the degree of corroboration cannot be a probability. Instead he proposed a deductive method of testing. In many ways this dialectic tension has many parallels in statistics, with the Bayesians on logico-inductive side vs the non-Bayesians or the frequentists on the other side. Simplistically Popper seems to be on the frequentist side, but recent synthesis on the non-Bayesian side might direct the Popperian views to a more nuanced destination. Logical probability seems perfectly suited to measure partial evidence or support, so what can we use if we are to reject it? For the past 100 years, statisticians have also developed a related concept called likelihood, which has played a central role in statistical modelling and inference. Remarkably, this Fisherian concept of uncertainty is largely unknown or at least severely under-appreciated in non-statistical literature. As a measure of corroboration, the likelihood satisfies the Popperian requirement that it is not a probability. Our aim is to introduce the likelihood and its recent extension via a discussion of two well-known logical fallacies in order to highlight that its lack of recognition may have led to unnecessary confusion in our discourse about falsification and corroboration of hypotheses. We highlight the 100 years of development of likelihood concepts. The year 2021 will mark the 100-year anniversary of the likelihood, so with this paper we wish it a long life and increased appreciation in non-statistical literature.

Derived here is a single regression coefficient equivalent to Pillai's trace statistic in multivariate analysis of variance., Comment: 3 pages

To improve confounder adjustments, observational studies are often matched on potential confounders. While matched case-control studies are common and well covered in the literature, our focus here is on matched cohort studies, which are less common and sparsely discussed in the literature. Matched data also arise naturally in twin studies, as a cohort of exposure-discordant twins can be viewed as being matched on a large number of potential confounders. The analysis of twin studies will be given special attention. We give an overview of various analysis methods for matched cohort studies with binary exposures and binary outcomes. In particular, our aim is to answer the following questions: (1) What are the target parameters in the common analysis methods? (2) What are the underlying assumptions in these methods? (3) How do the methods compare in terms of statistical power?, Comment: Published in at http://dx.doi.org/10.1214/12-STS390 the Statistical Science (http://www.imstat.org/sts/) by the Institute of Mathematical Statistics (http://www.imstat.org)