Don't Panic — Exploratory Data Analysis

Exploratory data analysis, or EDA, is the moment when you stop treating a dataset as a persuasive spreadsheet and start asking what is actually in it. Before models, tests, dashboards, and declarations of victory, there are rows, columns, types, and a fair chance that one of them is wearing a false moustache.

The central arrangement is pleasingly modest. Inspect the data. Look at one variable at a time. Compare pairs. Then look for interactions among several variables. Along the way, count missing values, duplicates, odd categories, and values that cannot possibly be true. The process is not a ceremony with a completion badge. It is a loop: notice a problem, understand it, make a cleaning decision, and look again.

Distribution is one of the first things to establish: where values gather, how far they spread, and whether a long tail, two peaks, or a boundary is trying to get your attention. A histogram gives the broad shape. A box plot brings the median, quartiles, and possible outliers into a smaller and more suspicious package. Neither chart knows whether an outlier is an error or an extraordinary real event. That inquiry remains inconveniently yours.

Next comes the relationship hunt. A scatter plot places two numeric variables together so curves, clusters, and stray points can stop hiding in separate columns. Grouped summaries and faceted plots ask whether the same pattern holds within a category. Correlation can show a linear association; it does not appoint itself head of causality, however much the number may enjoy the title.

The surprise is that cleaning and exploration are the same expedition. A null count, a date in the future, a duplicate row, or age = 999 changes what a chart means. Dropping every row with a null may remove a large part of the data and change the story. Record what you find and why you changed anything. Future-you is a perfectly reasonable colleague, but has an alarming memory.

Start with the intro for the map of distributions, relationships, patterns, and limits. Use the slides when you want the workflow and chart choices in one view. The cheatsheet keeps the inspection and plotting references close at hand. Then use the practice reference and exercise to turn observations into documented questions rather than a gallery of attractive accidents.