3 Descriptive Analytics and Exploratory Analysis
Learn how to prepare, summarize, visualize, and interpret business data while distinguishing observed patterns from causal explanations.
Purpose and scope
turns recorded business data into an account of what happened and how results vary. extends this work with summaries and visual displays that help reveal patterns, unusual observations, and questions to investigate. These approaches describe observed data; they do not by themselves establish causes or prove that a pattern will continue.
Before summarizing, state the business question—for example, “What was typical daily revenue by store last quarter?”—and check that the data and definitions fit it.
Prepare and check the data
A statistic is only as meaningful as the data and definitions behind it. Begin by identifying the : a row might represent one transaction, one customer, or one store-day. Mixing levels can produce invalid totals or comparisons; for example, summing a file that contains both transaction-level rows and daily totals may count some sales twice.
Check fields and definitions, including units, date ranges, currency, category labels, and how measures such as revenue, returns, and active customers are defined. Also look for missing, duplicated, or invalid values: check required fields, repeated identifiers, impossible dates, inconsistent category spelling, and values outside plausible ranges.
Do not automatically replace missing values with zero: “not recorded” and “none” different things. Investigate extreme values before correcting or excluding them. A genuine large order may be important, while a misplaced decimal may be an error.
Document filters, corrections, exclusions, and assumptions so that the results can be reviewed and reproduced.
Summarize distributions
Describe a quantitative variable through its center, spread, and shape. No single statistic captures all three.
Center: what is typical?
The is the sum of the values divided by the number of observations. Because it uses every value, unusually high or low values can pull it toward the tail. The is the middle value after sorting: half the observations lie at or below it and half at or above it, making it less affected by extreme values. The is the most frequent value or category and can help identify a common product, channel, or order size.
For daily revenue, in thousands of dollars, the values are . The is , while the is . The high value of raises the , so the better represents a typical day if that value is an exceptional but valid event. The remains useful when the total across observations is central to the business question.
Spread: how much do values vary?
The range is the maximum minus the minimum, so it is easy to calculate but depends on only two observations. Quartiles divide ordered data into four parts. The , calculated as , describes the spread of the middle half. Standard deviation summarizes how far values tend to lie from the and is expressed in the same units as the data. Because it is based on the and squared deviations, it can be sensitive to extreme values.
For the revenue example, using the -of-halves convention, and , so the IQR is thousand dollars. The IQR captures the compact middle of the data even though the overall range is thousand dollars. Quartile calculations can vary slightly across software conventions, so document the convention when precise reproducibility matters.
Shape: what does the distribution look like?
A distribution may be roughly symmetric, skewed to one side, have several peaks, or contain a long tail. If the and differ substantially, inspect the distribution and possible extreme values rather than reporting the alone. In right-skewed sales or order-value data, a small number of large observations can make the exceed the .
For categorical data, summarize counts and proportions. If of orders came through a mobile app, the count is and the share is . Include the denominator and time period: “ of orders in April” is more informative than “.”
Explore patterns with visuals
Visual displays can reveal features that a table of summary statistics hides. Match the chart to the question, label its units and time period, and keep comparisons on a consistent scale.
Histogram: Use it to examine a numeric distribution's center, spread, skew, multiple peaks, or possible outliers. A histogram groups observations into intervals, so its appearance depends partly on the selected bin widths.
Side-by-side box plots or dot plots: Use them to compare groups, including differences in medians, spread, overlap, and unusual values. Box plots compactly show the and middle of values.
Line chart: Use it to examine how a measure changes over time, looking for trends, recurring patterns, abrupt changes, or gaps.
Scatter plot: Use it to examine how two numeric measures relate, including the direction, form, and strength of an association, changing spread, and unusual points. Scatter plots can reveal linear or nonlinear associations, but association alone does not establish cause and effect.
Bar chart: Use it to compare category sizes, showing counts or percentages as appropriate.
For example, a store-level revenue comparison might show higher average sales in one region. A box plot can help reveal whether the difference reflects generally higher results across stores or only a few unusually large stores. A time plot may show that the difference occurred only during a short promotional period.
Interpret unusual values and group differences
An is an observation that stands apart from the rest of a dataset. A common box-plot convention flags values below or above . These are investigation flags, not automatic instructions to delete data. Check the original record and relevant business context: an unusual observation may be a recording error, a special event, or a real case the business needs to understand.
When comparing groups, examine more than their centers. Differences in sample size, spread, missingness, or mix of customers can change the interpretation. A higher average order value in one channel, for example, may reflect a different product mix rather than a channel effect. Descriptive comparisons identify what differs in the observed data; explaining why requires additional evidence and an appropriate study design.
Connect findings to decisions
A useful descriptive finding states what was measured, for whom, and over what period, then explains why the pattern matters and what further check or action is reasonable.
Across stores in May, daily revenue was ; stores had unusually high days. We will verify those transactions and compare results with the promotion calendar before using the overall for staffing plans.
This statement gives the scope, a robust summary, a limitation, and a next step. It does not treat a descriptive pattern as a proven cause or a forecast. When results will guide decisions, also consider data coverage: even a complete month may omit certain customer groups, channels, or locations.