Challenges in Incorporating Exploratory Data Analysis into Statistical Workflow
Jessica R. Hullman, Andrew Gelman · Harvard Data Science Review · 2021
It is a pleasure to take part in such a lively discussion about the relationship between exploratory and confirmatory data analysis (EDA and CDA), and what is and is not feasible and likely to be valuable in supporting visual analysis.Graphics and data exploration have long been the ugly duckling of statistics even while they have become important aspects of data science, so we are thrilled to see the Harvard Data Science Review give this topic a prominent place of discussion.Several discussants point out that we offer a framework, more of a placeholder for a theory than a theory itself, and no novel graphical or analytical methods.And, indeed, our immediate goal in writing this paper is to not to develop or present new methods so much as to point to the potential value for integration of existing practical and theoretical ideas.To put it another way, when we write, "Designing for interactive exploratory data analysis requires theories of graphical inference," we do not claim to offer any comprehensive theories ourselves, beyond the meta-theory that such a theory would be useful, and a discussion of alternatives and their