Exploratory Data Analysis

Ethos

A core goal of EDA is to develop a detailed understanding of the data and to consider the processes that might produce such data. Tukey used the analogy of EDA as detective work because the process involves the examination of facts (data) for clues, the identification of patterns, the generation of hypotheses, and the assessment of how well tentative theories and hypotheses fit the data.

EDA is characterized by flexibility, skepticism, and openness. Flexibility is encouraged as it is seldom clear which methods will best achieve the goals of the analyst. EDA encourages the use of statistical and graphical techniques to understand data, and researchers should remain open to unanticipated patterns. However, as summary measures can conceal or misrepresent patterns in data, EDA is also characterized by skepticism. Analysts must be aware that different methods emphasize some aspects of the data at the expense of others; thus, the analyst must also remain open to alternative models of relationships.

If an unexpected data pattern is uncovered, the analyst can suggest plausible explanations that are further investigated using confirmatory techniques. EDA and CDA can supplement each other: Where the abductive approach of EDA is flexible and open, allowing the data to drive subsequent hypotheses, the more ambitious and focused approach of CDA is hypothesis-driven and facilitates probabilistic assessments of predicted patterns. Thus, a balance is required between an exploratory and confirmatory lens being applied to data; EDA comes first, and ideally, any given study should combine both.

Methods

EDA techniques are often classified in terms of the four Rs: revelation, residuals, reexpression, and resistance. However, it is not the use of a technique per se that determines whether it is EDA, but the purpose for which it is used—namely, to assist the development of rich mental models of the data.