China
Africa
Explore chapters and articles related to this topic
Penalized Regression Models
Published in Taylor Arnold, Michael Kane, Bryan W. Lewis, A Computational Approach to Statistical Learning, 2019
Taylor Arnold, Michael Kane, Bryan W. Lewis
Classical methods for computing data-driven variable selection typically fall under a class of algorithms known as stepwise regression [21]. In backward stepwise regression, we start with a standard regression model using all of the variables. Variables are then removed from the model due to some selection criterion, such as removing variables with small T-statistics. The new model is then refit with the smaller set of variables and the selection criterion is applied once again. Eventually this leads to a standard regression model over a reduced set of variables. The forward selection variation uses the same approach, but starts with an empty model and iteratively determines the next most significant term to add, stopping when some metric (i.e., AIC or Mallow’s Cp) is locally optimized. Even when not used formally, the ideas behind stepwise regression are frequently applied in the medical and social sciences. Econometrics texts, for example, often promote a process by which an initial model is built, the fit and residuals are evaluated, and variables and interaction terms are iteratively added and removed as needed [178]. Hand-constructed stepwise regression can lead to erroneous conclusions if hypothesis tests from the final model are presented without correcting for the post hoc variable selection step [120]. Done maliciously—pejoratively known as p-hacking—this can have serious effects for the advancement of research and is currently a hotly debated topic of interest within statistics [64, 75, 123]. When used for purely predictive purposes, such concerns are avoided and stepwise regression is a perfectly valid approach. However, empirical results suggest that it often leads to non-optimal results and its ad hoc description makes it difficult to provide a solid theoretical treatment [153].
Editorial: Sport and exercise psychology
Published in Journal of Sports Sciences, 2022
Robin C Jackson, Paul Appleton, David Fletcher, Jamie North
In closing, we would like to remind you about several recent developments at the Journal of Sports Sciences. First, the Advisory Board for each section has been refreshed. We have added strength in depth to the Sport and Exercise Psychology board and are delighted to welcome 11 new colleagues onto our Advisory Board. These individuals take the place of six colleagues who made a valuable contribution to the review process over many years, for which we are very grateful. Second, we have introduced the option of Registered Reports. We strongly encourage authors to consider this submission option, which is designed to promote research transparency and help combat issues of publication bias, along with questionable practices such as p-hacking. For a full consideration of the issues and how registered reports attempts to address them, we encourage you to read the recent Journal of Sports Sciences editorial (Abt et al., 2021). Third, the Journal of Sports Sciences has appointed a separate Advisory Board in the area of research design and statistics to help Executive Editors and Associate Editors address any questions or concerns in this area. This reflects an increase in the number of submissions with complex research designs and statistical procedures, such as those involved in the analysis of “big data”.
Registered Reports in the Journal of Sports Sciences
Published in Journal of Sports Sciences, 2021
Grant Abt, Colin Boreham, Gareth Davison, Robin Jackson, Eric Wallace, A Mark Williams
So, what are the solutions? A number of recommendations have been made over the last decade to improve the rigour of research and to minimise the introduction of these problems into the literature (Caldwell et al., 2020; Munafò et al., 2017). Two of these are the use of preregistration and the Registered Report submission format. Preregistration allows reviewers, editors, and readers to transparently evaluate the capacity of a test to falsify a prediction (Lakens, 2019). Preregistration (ideally a priori) involves publishing a timestamped protocol that outlines all data collection procedures together with a data analysis plan (Nosek et al., 2018). The data analysis plan usually involves identifying what are termed “confirmatory” and “exploratory” analyses (Caldwell et al., 2020). Publishing the data collection and analysis protocols prior to data collection allows authors to minimise the temptation to engage in questionable research practices (e.g., p-hacking, HARKing). Preregistration encourages greater transparency and rigorous reporting of study methods and analyses (Toth et al., 2020). Although we strongly support the use of preregistration, it is still open to abuse and manipulation (Ikeda et al., 2019; Yamada, 2018). The Registered Report format takes preregistration one step further by formalising the registration process and allowing reviewer feedback prior to data collection.
Sports medicine and biomechanics – synergies and nuances
Published in Journal of Sports Sciences, 2022
Eric Wallace, Massimiliano Ditroilo, Timothy A Exell, Mark Robinson, Natalie Vanicek
Authors wishing to submit their research manuscripts to the Sports Medicine and Biomechanics section should first consult the Instruction for Authors on the Journal of Sports Sciences’ website. We would strongly encourage authors to consider the submission option of Registered Reports, which is designed to promote research transparency and help combat issues of publication bias, along with questionable practices such as p-hacking. For a full consideration of the issues, and how Registered Reports attempt to address them, we encourage you to read the recent Journal of Sports Sciences ‘Registered Reports’ editorial by Abt et al. (2021b). The aims and focus of the manuscript’s research must be clearly associated with one, or both, of the following two research areas: sport and exercise medicine and sport and exercise biomechanics. We specifically welcome studies with perceived high impact related to injury and rehabilitation; clinical studies in sport and exercise medicine; sport biomechanics; musculoskeletal and neuromuscular biomechanics; biomechanics modelling; sports technology; biomechanics of strength and conditioning; and sport-relevant gait and posture. In all cases, the research must have a robust study design and be underpinned with substantive human data. Studies of sex comparison of performance and, even more importantly, injury risk conducted only on the basis of one factor – sex, provide limited information. These studies should include detailed information on training status and motor skills of participants to make the comparison more meaningful. For example, a well-designed study looking at sex differences in injury risk during a particular task should go beyond the usual “sex, age, height and body mass” of participants and ensure a similar level of normalised muscle strength, training background and specific skills between the categories being compared. This would help understand whether the differences are truly down to sex or are instead the result of a different level of physiological/ neuromuscular/ biomechanical characteristics (Nimphius, 2019).