Descriptive Statistics: Summarizing and Interpreting Data

Descriptive statistics organize observed data so a reader can understand a distribution before making broader claims. Counts, percentages, averages, spread, and charts answer different questions. A number is useful only when the variable, unit, population, denominator, and period are clear. A careful analysis starts by checking what was measured and who is in the dataset, then selects summaries that fit the scale and shape of the data. Description does not, by itself, establish a cause or guarantee that a sample represents everyone of interest.

Inspect the data before summarizing it

List the variables and how each was recorded. A category such as department differs from a numerical measure such as wait time. Check missing entries, impossible values, duplicates, and inconsistent units. A wait recorded in minutes in one file and hours in another will produce a polished but wrong average if merged without correction. Keep a record of exclusions and corrections so the summary can be reproduced.

Define the denominator for every percentage. “Forty percent responded” means different things if it refers to all invited people or only those who opened a survey. Report counts alongside percentages, especially for small groups. An apparently large percentage change may represent only a few observations. When the dataset is a sample, explain the recruitment path and who may be missing before applying the description to a wider population.

Choose measures of center with the distribution in mind

The mean uses every numerical value and can be informative when the distribution is reasonably balanced. The median identifies the middle observation and is less affected by extreme values. The mode identifies a most common value or category. None is a universal “best average.” If a few very long service delays raise the mean, reporting only the median may hide the worst experiences; reporting only the mean may imply that a typical person waited longer than most did. Show both when they answer useful questions.

Consider the scale. A mean of codes assigned to categories has no substantive meaning. A mean of a rating scale may be useful with caveats about how respondents interpret its steps. Round only at the reporting stage so intermediate calculation does not distort results. State the units and time frame next to the number rather than leave a reader to infer them from a table title.

Describe variation and shape

Range gives the minimum and maximum but can be dominated by one outlier. Quartiles and the interquartile range describe the middle half of observations. Standard deviation summarizes spread around the mean under its own assumptions and is harder to interpret without the distribution. A histogram or box plot can reveal skew, clusters, and extreme values that one summary statistic conceals. Inspect whether an outlier is a genuine observation, a recording error, or a different process; do not delete it merely because it is inconvenient.

Compare groups using the same definitions and period. A department with more cases may have more adverse events in raw count while a lower event rate. Differences in case mix may also matter. Descriptive comparison can identify a question for investigation but should not be written as evidence that group membership caused the outcome. Note uncertainty from small numbers and missing data even when the task is “only descriptive.”

Work through a service-wait example

Suppose ten appointment waits in minutes are 5, 6, 7, 8, 9, 10, 11, 12, 13, and 39. Their sum is 120 minutes, so the mean is 12; the median is the average of the fifth and sixth values, 9.5. The 39-minute wait pulls the mean above what most people experienced. The range is 5 to 39. Both the typical wait and the long tail matter if the service promises timely access for everyone.

Before concluding that the service improved, compare the same type of appointment over comparable periods and check whether delayed patients abandoned the queue and disappeared from the records. If the next month’s median falls but the longest waits rise, the decision may depend on whether the goal is typical speed or protection against extreme delay. A chart of the full distribution and a count of waits above a defined threshold make the trade-off visible.

Use charts to answer a question

A bar chart can compare category counts; a histogram can show a numerical distribution; a line chart can show values over time. Choose a scale and baseline that do not exaggerate a small difference. Label axes, units, period, and denominator. Use color and text so the chart remains interpretable without relying on hue alone. A table may be better when readers need exact values or when only a few observations exist.

Avoid mixing counts and rates in one visual without clear labels. If groups differ in size, show a rate with the underlying count. For time series, annotate a change in definition or data collection; an apparent trend may be a measurement break. Every visual should support a sentence about what was observed and a separate sentence about its limit.

Report conclusions with a defensible scope

State the dataset, cleaning choices, summaries, and key pattern. Distinguish what the numbers describe from what remains unknown about explanation or generalization. If a sample is voluntary, the descriptive statistics accurately summarize respondents but may not characterize nonrespondents. If an outlier is consequential, report it and investigate rather than hide it behind an average. Descriptive statistics are strongest when they make both ordinary experience and important variation visible enough for a reader to ask the next question.

Keep calculations reproducible. Record the data version and any grouping or filtering decisions, and check a few results by hand or with an independent method. A chart copied into a report can become stale when the underlying data change. Verify the final numbers and labels against the dataset used for the conclusion.