4 81 TRANSFERRING TARGETED MAXIMUM LIKELIHOOD ESTIMATION INTO SPORT SCIENCE causal relationships in non-experimental data, the proper quantification of uncertainty, etcetera. The challenge is to prevent the specification of uninterpretable coefficients in miss-specified parametric models (e.g., GLMs) where different choices of such missspecified models yield different answers [2], [20]. In contrast, the targeted learning method (e.g., TMLE) aims to construct confidence intervals for user-specified target parameters by targeting the estimates retrieved from data adaptive estimators (e.g., machine learning) while relying solely on accurate statistical assumptions. This approach can reduce differences in statistical analysis results as model choices are automated, allowing for consistent estimates regardless of the researcher conducting the study [21]. The Targeted Learning methodology focuses on the art of causal modelling [2]. Causal modelling is a technique used to provide a formal model for and express assumptions about data-generating processes [22]–[24]. Currently, the four main approaches used for causal modelling are (i) Graphical models, (ii) potential-outcome models, (iii) sufficient-component cause models, and (iv) structural equations models [22]. These approaches offer complementary perspectives and can be used together to enhance causal interpretations [25]. With our paper we aim to introduce a roadmap to use the TMLE methodology in the field of sports science. As such, we introduce causal inference as a new tool in the sports scientists’ toolbox. 3. MATERIALS AND METHODS We adhere to the causal roadmap as a procedure to structure scientific research [22], [26]. This roadmap takes the form of seven steps: (i) specifying the knowledge of the system to be studied using a causal model, (ii) specifying the data and their link to the causal model, (iii) specifying the target causality, (iv) assessing identifiability, (v) stating the statistical estimation problem, (vi) estimation, and (vii) interpretation. By following this roadmap, we create a clear distinction between the knowledge about the system under study and about the assumptions that need to be made to answer the research questions. TMLE is part of this procedure and is applied in the estimation step. The present work adheres to this general structure and is what we see as the blueprint for performing TMLE in sports science. 3.1. Specifying the knowledge of the system to be studied using a causal model The first step in this roadmap is to define the knowledge about the system under study. Knowledge, in this case, is actual, fundamental knowledge about the system, and should not rely on assumptions on the underlying model. One way to define this system is by using a causal graph representation, which depicts the causal relations of the system
RkJQdWJsaXNoZXIy MjY0ODMw