Open science is an academic movement to transform scientific research into a more open endeavor. For the purposes of this paper, we define open science as the infrastructure of a “statistical revolution,” which indicates the transformation of fundamental ideas in psychometrics. The four basic components of open science are: (1) open access, (2) open data, (3) data-centric science, and (4) civil science. With regard to open data and open access, we surveyed the present situation of psychological research and considered how it is associated with the statistical revolution internationally and domestically in Japan. To date, the introduction of open science to academic psychology circles has arrived relatively late in Japan. The future prospects of psychology in open science are discussed, especially with regard to what to do and how to do open science as a researcher.
In recent years, research data through advanced network environments and sensors has dramatically increased, and scientific research methods, called “fourth paradigm” and “data centric science,” have been attracting attention. From the viewpoint of research fraud prevention, the momentum of disclosing data on research results is increasing. These activities are widely called “open science.” To activate open science, it is necessary to promote daily data sharing inside and outside the research community. This paper introduces cases of open science on institutional and technical aspects, and discusses future prospects.
The use of Bayesian statistics in psychology has been of recent focus. This paper discusses the effectiveness and usefulness of “Bayesian statistical modeling” in psychological studies by exploring its differences with traditional methodologies used in psychological studies. First, we explain differences between two trends in Bayesian statistics: hypothesis testing using Bayesian statistics and Bayesian statistical modeling. Second, Bayesian estimation and probabilistic programming languages may make it easy for psychologists to use statistical modeling. Third, the three advantages of studies using Bayesian statistical modeling in psychology are demonstrated. These advantages include developing mathematical explanations of behavioral mechanisms, valid estimation of psychological characteristics, and improved transparency and replicability of the data analysis. Fourth, it is argued that Bayesian statistical modeling and traditional psychological methodology could coexist by influencing each other.
In the last decade, psychology faced a serious crisis, called the reproducibility problem. To solve the problem, several methodological and institutional changes have been proposed and implemented such as the promotion of replication studies and publication of negative results, the introduction of a preregistration system in academic journals, and the implementation of novel statistical methods for suppressing false-positive results. In this paper, based on a previously proposed formal model of population dynamics of scientific discovery, I first explain that these changes are insufficient for avoiding the reproducibility problem. Based on two examples in social psychology, I also discuss that what is necessary for fundamental reformulation of psychology is rigorous modeling, which has already been widely applied in many scientific fields outside psychology.
Studies in clinical psychology, especially abnormal psychology, typically use null hypothesis statistical testing to examine behavioral data that were gathered through cognitive tasks. This paper identifies the problems associated with using conventional research methods and discusses the advantages of using a computational approach to examine mental disorders. The computational approach includes cognitive modeling that estimates latent parameters that cannot be directly observed from behavioral data. Cognitive modeling makes it possible to explain and predict behavioral data in a logically valid way and contributes to the assessment of mental disorders. This paper introduces the best practices in cognitive modeling by using probabilistic reversal learning task as an example. In conclusion, this paper provides examples of research studies that use cognitive modeling in clinical psychology and discusses future directions, including parameter estimation by hierarchical Bayesian estimation and hierarchical Bayesian inference model as a cognitive model.
The purpose of this paper is to review the development and contribution of the Bayesian approach to higher-level human cognition. In doing so, this paper introduces several representative studies of the Bayesian approach, including causal reasoning or hypothesis testing, and argues the following: (1) the Bayesian approach is based on the theoretical assumptions of judgment and decision-making studies that are not themselves derived from Bayesian approaches, and (2) the Bayesian approach reflects a transition in the views on rationality in human thinking. Based on these arguments, this paper also proposes that, rather than emphasizing its explanatory power, the Bayesian approach can be further developed by considering fundamental problems in higher-level cognition.
Clinical trials are prospective studies that evaluate the effectiveness of interventions on humans under certain circumstances. The frequentist approach in which the sample size is strictly controlled before tests is common in clinical trials. However, clinical trials that use an adaptive Bayesian method in which the trial can be flexibly stopped, based on data accumulated during the course of the trial, has been recommended to reduce costs and to meet ethical requirements in the fields of medicine and medical device development. In the field of psychological interventions, the current situation is that application based on Bayesian framework is often extremely poor. Therefore, this paper outlines the design method of clinical trials by the adaptive Bayesian method, discusses the benefits of and problems in its application to psychological intervention research, and provides examples of its virtual application.
The Bayes factor has a basic and crucial role in Bayesian evaluation of psychological hypotheses and models. It forms a fundamental part of the advancement of psychological science. Its computation has been a major challenge, although recent advances in numerical estimation methods such as bridge sampling may allow the application of the Bayes factor to a wide range of practical research contexts. The objective of the current paper is to provide psychological scientists an introductory tutorial of the ideas and recent developments concerning the Bayes factor. Some running examples are presented and a few practical application methods are also discussed.
Many problems have been reported concerning null hypothesis significance testing (NHST), although it is used in most psychological research. The authors believe that this situation will not change, at least not in the near future. Before shifting to a Bayesian approach, there are several things that must be done in psychological research that uses NHST. Among them, sample size planning (SSP) is especially important. In practice, whether significant results exist primarily depends on sample size, and many studies have indicated that SSP is vital. However, the use of SSP is rare in psychological research. Most psychological research basically depends on the significance of NHST; therefore, psychologists should conduct SSP before collecting data. We describe the present situation of SSP in psychological research, discuss related topics, argue for the importance of SSP, and present future directions for SSP in psychological research using NHST.
In this paper, I comment on the current rise in the application of Bayesian statistics within the field of psychology, as the user views it. The merit of Bayesian statistics has recently been emphasized as a last resort to overcome the weakness of the null hypothesis significance test (NHST), which is a main issue in questionable research practices (QRPs). However, in the end, most psychologists are only users of statistics, and Bayesian statistics is only one of many tools available to them. If this is true, the issues caused by misunderstanding the p-value may also appear in Bayesian statistics as a form of posterior hacking. The “open” principles of psychology provide new hope. However, strange feelings arise in my insides when I consider the open-data policy. For example, unfairness or “free rider” problems caused by the cost asymmetry between data makers and users may soon erupt. However, it may be important for psychologists to dive into the Bayesian world rather than avoiding entering it.
In this article, I have selected and will comment on three topics from the papers in this special issue: power analysis, Bayesian hypothesis testing, and Bayesian statistical modeling. The major problem in power analysis is the quantitative evaluation of effect size, and this problem is shared in the estimation of effect size. Bayesian hypothesis testing is theoretically superior to classical null hypothesis testing. However, setting prior distributions properly continues to be difficult to overcome. Bayesian statistical modeling changes the role of statistics in psychological research and is promising in many aspects. However, the ease of computation can mask the necessity of careful thinking in modeling, which includes setting prior distributions. I also will discuss future statistics education in the era of rapid developments in methodology, as documented in this special issue.
In this paper, the features of Bayesian hypothesis evaluation and statistical modelling are compared with those of the frequentist paradigm. Bayesian statistics enjoys the recent development of computers and software for data analysis. However, such benefit may cause two main problems. One problem is the neglect of effort to specify the details of statistical models or prior information, each of which is carefully considered in traditional statistical data analysis. The other problem is the futile exploration of models owing to flexible manipulation of probabilistic programming languages. Furthermore, for a fair comparison, this paper provides some situations that are suitable for Bayesian and frequentist statistics, respectively. The message of this paper is the importance of quantifying hypotheses and constructing statistical models as clearly as possible in a subjectively interpretable manner.