Exploratory data analysis

Exploratory data analysis (EDA) is the first analytical pass over a dataset: an open-ended investigation of what the data actually contain before any formal model is fit or hypothesis is tested. The approach was championed by statistician John Tukey, who argued that analysts should let the data suggest questions rather than only confirming ideas they brought with them.

EDA combines two toolkits. Summary statistics describe the data numerically — measures of center (mean, median), spread (standard deviation, range, quartiles), and relationships (correlation). Visualizations reveal structure at a glance: histograms and box plots show distributions and outliers, scatterplots expose relationships between variables, and time-series plots surface trends and seasonality. Along the way, analysts check data quality — missing values, duplicates, impossible entries — and note whether distributions are skewed or contain anomalies that would distort later analysis.

EDA occupies a specific place among the types of analytics. It is closely related to descriptive analytics (what happened), but its purpose is forward-looking: to generate hypotheses, guide the choice of models, and prevent errors that come from feeding unexamined data into predictive or prescriptive techniques. A single outlier or a badly skewed variable, caught during EDA, can change the entire direction of an analysis.

The CMA Part 1 exam covers exploratory data analysis in its technology and analytics section. Candidates should be able to define EDA, distinguish it from descriptive, predictive, and prescriptive analytics, and recognize the summary statistics and chart types used to explore a dataset.