The <scp>MEDDLE</scp> Data Analysis Guides as a Living Resource for Multiple‐Driver Marine Research

In trying to understand the consequences of a changing ocean, researchers working on marine multiple drivers have become adept at manipulating complex combinations of environmental drivers. Consider a study of future ocean conditions: the research question may require simultaneously controlling pH, nutrients, light, and temperature. Scientists quickly learn that relying solely on full factorial designs leads to an unmanageable number of experimental units and, even worse, too little statistical precision to yield meaningful results. However, simplified experimental designs, when done without an understanding of how these simplifications affect the questions that can be addressed, can slow or even impede building an understanding of complex organismal responses. To help the community tackle the challenge of designing multiple-driver experiments, the Scientific Committee on Oceanic Research (SCOR) established the Changing Ocean Biological Systems (COBS) project: Changing Ocean Biological Systems: how will biota respond to a changing ocean. Through COBS and its online resource MEDDLE (an acronym for Multiple Environmental Driver Design Lab for Experiments), we have developed web-based resources (https://meddle-scor149.org/), a Best Practice Guide (Boyd et al. 2019), led over twenty in-person and virtual international workshops (McGraw et al. 2023), and provided a review of experimental strategies (Boyd et al. 2018). These resources have helped researchers move towards tractable designs that answer the questions of interest. However, one of the key roadblocks to implementing these experimental designs remains statistical understanding. Many emerging researchers struggle to identify or apply the “right” statistical approach, or to interpret and present their results. Although most researchers have some knowledge of relevant statistical techniques (e.g., linear mixed models or generalized additive models), translating this into a clear understanding of when and how to apply them remains a persistent challenge. As a result, researchers often default to experimental designs they are comfortable with (e.g., a standard 2 × 2 ANOVA design), even if the design cannot answer their question. To help emerging researchers link experimental designs with appropriate statistical analyses, we developed a series of data analysis guides, available through the MEDDLE website (https://meddle-scor149.org/vignettes/). The MEDDLE guides serve as both broad tutorials to develop general statistical skills and exemplars for specific designs. Each data analysis guide was developed using concepts from the Best Practice Guide (Boyd et al. 2019), as well as our experiences delivering design workshops and interacting with scientists. Each guide includes embedded and commented R code, covering both data analysis and figure creation (Fig. 2). The guides are based on a standard R Markdown template that we developed and customized for each topic. Researchers who wish to contribute a new guide can request the template through the MEDDLE website or by contacting the corresponding author. While presented in a conversational manner, the structure is guided by best practice principles. The purpose of Steps 1–3 is to ensure that research is driven by a clearly articulated scientific question; a need for measurements, predictions, or hypotheses grounded in prior knowledge; and an experimental design that makes assumptions explicit and is appropriate for the question and scale of interest. While this provides a strong foundation, many emerging researchers move from design to data collection without a data analysis plan, often resulting in low-powered or superficial analyses that do not fully address the biological question(s). Steps 4–6 illustrate the benefits of integrating design with analysis planning, ensuring that data collected are sufficient, methods are appropriate and tractable, and results are biologically meaningful, transparent, and reproducible. The level of statistical literacy amongst marine researchers varies widely, from those who have only taken one formal statistics course to those with substantial training in statistical or mathematical modeling. A primary goal of our resources is to meet researchers wherever they are, improving core skills for those with less training while introducing advanced techniques for those with more training. Even experienced researchers may not use the most powerful techniques, or misapply statistical methods as experimental complexity increases. Our view is that statistical thinking starts with formal training but should be an ongoing process. That is, not only should a researcher's ecological knowledge advance over time, so should their statistical knowledge. Living resources such as the MEDDLE guides that provide community-led, context-specific training are an important part of the process. We undertook the development of these guides because of the common request from emerging researchers for more training in data analysis. A large challenge was deciding on scope. It was tempting to expand to the point where we were replicating a statistics textbook, or to focus on common statistical errors. In the end, we set the following boundaries for our resources: (1) keep the focus on multi-driver studies, (2) recognize that our examples would never be exhaustive or be able to replace formal training, and (3) include examples appropriate for different levels of statistical training. The importance of an appropriate statistical analysis cannot be overstated: the correct data analysis approach is essential for interpreting and communicating our findings to our scientific peers, communities we engage with, and policymakers. The MEDDLE data analysis guides provide a living resource that embeds statistical thinking as an integral part of the design process. The guides are by no means exhaustive, but provide practical examples that can help emerging and established researchers improve both design and analysis. There is room for growth; for example, the current examples focus on experiments that measure organismal responses at the level of whole-organism traits. Future guides could consider molecular or physiological responses, or analyze data from mesocosm or field studies. MEDDLE has been an important part of the COBS project since 2019; along with the workshops we offer, we expect the data analysis guides to be an integral part of what we offer in the future. Just as we hope these resources will benefit the community, we hope that the community will use them, adapt them, and ultimately contribute new examples so that our statistical tools can grow along with our knowledge of the complex challenges our future oceans face. This project was supported by national committees of the Scientific Committee on Oceanic Research (SCOR) and from a grant to SCOR from the US National Science Foundation (OCE-1840868) for the Changing Oceans Biological Systems project. Further support was provided by the Natural Environment Research Council (UK) Grant NE/X001237/1.