In this paper I compare parametric and nonparametric regression models with the help of a simulated data set. Doing so, I have two main objectives. The first one is to differentiate five concepts of simplicity and assess their respective importance. The second one is to show that the scope of the existing philosophical literature on simplicity and model selection is too narrow because it does not take the nonparametric approach into account, S112–S123, 2002; Forster and Sober in The British Journal (...) for the Philosophy of Science 45, 1–35, 1994; Forster, 2001, in Philosophy of Science 74, 588–600, 2007; Hitchcock and Sober in The British Journal for the Philosophy of Science 55, 1–34, 2004; Mikkelson in Philosophy of Science 73, 440–447, 2006; Baker 2013). More precisely, I point out that a measure of simplicity in terms of the number of adjustable parameters is inadequate to characterise nonparametric models and to compare them with parametric models. This allows me to weed out false claims about what makes a model simpler than another. Furthermore, I show that the importance of simplicity in model selection cannot be captured by the notion of parametric simplicity. ‘Simplicity’ is an umbrella term. While parametric simplicity can be ignored, there are other notions of simplicity that need to be taken into consideration when we choose a model. Such notions are not discussed in the previously mentioned literature. The latter therefore portrays an incomplete picture of why simplicity matters when we choose a model. Overall I support a pluralist view according to which we cannot give a general and interesting justification for the importance of simplicity in science. (shrink)

My aim in this paper is to show how the problem of inflated effect sizes corrupts the severity measure of evidence. This has never been done. In fact, the Winner’s Curse is barely mentioned in the philosophical literature. Since the severity score is the predominant measure of evidence for frequentist tests in the philosophical literature, it is important to underscore its flaws. It is also crucial to bring the philosophical literature up to speed with the limits of classical testing. The (...) Winner’s Curse is one of them. The problem is that when a significant result is obtained by using an underpowered test, the severity score becomes particularly high for large discrepancies from the null-hypothesis. This means that such discrepancies are very well supported by the evidence according to that measure. However, it is now well documented that significant tests with low power display inflated effect sizes. They systematically show departures from the null hypothesis H0 that are much greater than they really are. From an epistemological point of view this means that a significant result produced by an underpowered test does not provide evidence for large discrepancies from H0. Therefore, the severity score is an inadequate measure of evidence. Given that we are now aware of the phenomenon of inflated effect sizes, it would be irresponsible to rely on the severity score to measure the strength of the evidence against the null. Instead, one must take appropriate measures to try and avoid using underpowered tests by setting a threshold for the sample size or by replicating the results of the experiment. (shrink)

It has been claimed that if statistical power and p-values are both used to measure the strength of our evidence for the null-hypothesis when the results of our tests are not significant, then they can also be used to derive inconsistent epistemic judgements as we compare two different experiments. Those problematic derivations are known as power approach paradoxes. The consensus is that we can avoid them if we abandon the idea that statistical power can measure the strength of our evidence. (...) In this paper however, I put forward a different solution. I argue that every power approach paradox rests on an equivocation on "strong evidence". The main idea is that we need to make a careful distinction between the evidence provided by the quality of the test and the evidence provided by the outcome of the test. Both provide different types of evidence and their respective strength are to be evaluated differently. (shrink)

The p value is the probability under the null hypothesis of obtaining an experimental result that is at least as extreme as the one that we have actually obtained. That probability plays a crucial role in frequentist statistical inferences. But if we take the word ‘extreme’ to mean ‘improbable’, then we can show that this type of inference can be very problematic. In this paper, I argue that it is a mistake to make such an interpretation. Under minimal assumptions about (...) the alternative hypothesis, I explain why ‘extreme’ means ‘outside the most precise predicted range of experimental outcomes for a given upper bound probability of error’. Doing so, I rebut recent formulations of recurrent criticisms against the frequentist approach in statistics and underscore the importance of random variables. (shrink)

In this paper I articulate a fictionalist solution to the closure problem that affects constructive empiricism. Relying on Stephen Yablo’s recent study of closure puzzles, I show how we can partition the content of a theory in terms of its truthmakers and claim that a constructive empiricist can believe that all the observable conditions that are necessary to make a part of her theory true obtain and remain agnostic about whether or not the other truthmakers for the other parts of (...) her theory obtain. This can be done even though she asserts her theory as if it was wholly true. (shrink)

In this paper, I compare theory-laden perceptions with imputed data sets. The similarities between the two allow me to show how the phenomenon of theory-ladenness can manifest itself in statistical analyses. More importantly, elucidating the differences between them will allow me to broaden the focus of the existing literature on theory-ladenness and to introduce some much-needed nuances.

The main point of this paper is to underscore that tests with very low power will be significant only if the observations are deviant under both H0 and H1. Therefore, the results of those significant tests will generate misleadingly high severity scores for differences between H0 and H1 that are excessively overestimated. In other words, that measure of evidence is bound to fail in those cases. It will inevitably fail to adequately measure the strength of the evidence provided by tests (...) with low power. (shrink)

The main point of this paper is to underscore the link between simplicity and truth in an unsupervised machine learning context. More precisely, we argue that parametric and dimensional simplicity are not indicators of truth but the methodological principle that urges us to pay attention to such notions of simplicity is truth conducive. The truth that we are looking for are specific geometrical shapes and we know which algorithm can find which shapes provided that we pay attention to parametric and (...) dimensional simplicity. (shrink)

In 1983, in an open letter to the journal Nature, Karl Popper and David Miller set forth a particularly strong critical argument which sought to demonstrate the impossibility of inductive probability. Since its publication the argument has faced many criticisms and we argue in this article that they do not reach their objectives. We will first reconstruct the demonstration made by Popper and Miller in their initial article and then try to evaluate the main arguments against it. Although it is (...) possible to conceptualize logically the idea of induction, it is shown that it is not possible on traditional Bayesian grounds. (shrink)

In 1983, in an open letter to the journal Nature, Karl Popper and David Miller set forth a particularly strong critical argument which sought to demonstrate the impossibility of inductive probability. Since its publication the argument has faced many criticisms and we argue in this article that they do not reach their objectives. We will first reconstruct the demonstration made by Popper and Miller in their initial article and then try to evaluate the main arguments against it. Although it is (...) possible to conceptualize logically the idea of induction, it is shown that it is not possible on traditional Bayesian grounds. (shrink)

The curse of dimensionality is one of the most prominent challenge that data scientists face when trying to make valuable inferences. It is an epistemic problem that hits particularly hard in "Big Data" research contexts, where the volume of the data set is particularly large. The way in which we tackle with this problem sheds light on the notion of scientific evidence. Yet, it is virtually absent from the current philosophical literature. In this paper, I aim to broaden the focus (...) of that literature by showing that the dimensions in which the data are embedded are an integral part of scientific evidence. This is an aspect of scientific evidence that is often hidden behind the traditional observation/theory dichotomy that we often encounter in philosophy of science. Dimensions are abstract objects. They are not observables like tables and chairs, nor are they entries in a data set. Ultimately, I aim to show that empirical adequacy is not merely a matter of finding the model that best fit the relevant observational data. It is also a matter of finding the model that best fit the relevant observational data inside the relevant dimensions. (shrink)

Machine learning is a scientific discipline that can be divided into two main branches: supervised machine learning and unsupervised machine learning. In this paper, we aim to show just how simplicity matters in unsupervised contexts. This is important because unsupervised machine learning algorithms have barely received any attention in philosophy. Yet, there is a direct link between simplicity and truth in unsupervised contexts that we do not find in their supervised counterparts. This has thus far evaded philosophical discussions on simplicity.

Classical statistical inferences have been criticised for various reasons. To assess the soundness of such criticisms is a very important task because they are widely used in everyday scientific research. This is one of the reasons why the philosophy of statistics is an exciting field of study. In this paper, I focus on two such criticisms. The first one claims that the use of the p-value violates the principle of total evidence. It is a thesis that has been defended by (...) Elliott Sober and Bengt Autzen. The second one says that the result of classical tests does not only depend on the data but on the sampling plan of the experimenter also. The underlying criticism of course is that the sampling plan is not part of the evidence and that classical tests therefore violate PTE. The intentions of the experimenter should not affect the result of an inference. My aim is to show that both criticisms are unsound. Doing so, I hope to clarify the concept of p-value and the nature of the evidence in classical statistical tests. The point of my paper is to show that the identification of the evidence on which those criticisms rest is inadequate. (shrink)

For model-based frequentist statistics, based on a parametric statistical model ${{\cal M}_\theta }$, the trustworthiness of the ensuing evidence depends crucially on the validity of the probabilistic assumptions comprising ${{\cal M}_\theta }$, the optimality of the inference procedures employed, and the adequateness of the sample size to learn from data by securing –. It is argued that the criticism of the postdata severity evaluation of testing results based on a small n by Rochefort-Maranda is meritless because it conflates [a] (...) misuses of testing with [b] genuine foundational problems. Interrogating this criticism reveals several misconceptions about trustworthy evidence and estimation-based effect sizes, which are uncritically embraced by the replication crisis literature. (shrink)