Written and maintained by CASRAI Editorial Board
Last updated
Most SPSS tutorials teach one descriptive-statistics procedure — usually Descriptives — and leave people running it on every variable regardless of type. SPSS actually ships three separate procedures under Analyze > Descriptive Statistics for exactly this job: Frequencies, Descriptives, and Explore. Each computes a different set of statistics, expects a different kind of variable, and leaves out things the other two provide. Picking the wrong one either produces meaningless output (a mean of a nominal code) or misses the diagnostic you actually needed (a normality test, an outlier flag).
Frequencies: for categorical variables, and for counting distinct values
Analyze > Descriptive Statistics > Frequencies is the only one of the three that reports Valid Percent (the percentage after missing cases are excluded) alongside raw Percent and Cumulative Percent — the standard output shape for a categorical variable such as department, condition group, or a Likert response code. It is built for nominal and ordinal data: every distinct value gets its own row, with a count and percentage.
The Statistics button also exposes central tendency (mean, median, mode, sum), dispersion, and percentile options — which makes Frequencies double as a percentile engine for a scale variable when you uncheck Display frequency tables in the main dialog (otherwise a continuous variable with hundreds of distinct values produces a useless hundreds-of-rows table). Running those same statistics checkboxes on a genuinely categorical variable, however, produces numbers with no real meaning — a “mean department code” is not interpretable. Charts (bar, pie, or histogram with an optional normal curve) are available directly from the same dialog.
Descriptives: a fast summary table for continuous variables
Analyze > Descriptive Statistics > Descriptives is the quickest way to get a one-line summary — N, minimum, maximum, mean, and standard deviation — for several scale (interval/ratio) variables at once, all in a single compact table. The Options dialog adds S.E. mean, variance, range, and (unlike the default Frequencies statistics list) skewness and kurtosis with their standard errors.
What Descriptives deliberately does not give you: no median, no mode, and no percentiles at all — those checkboxes simply aren’t in its Options dialog. It also has no split-by-group option and runs no normality or outlier diagnostics. Its one extra feature the other two lack is the “Save standardized values as variables” checkbox, which writes a new z-score variable (prefixed Z) for each selected variable directly into the dataset. Use Descriptives when the goal is a fast, publication-style summary row for several scale variables, not a diagnostic.
Explore: normality, outliers, and diagnostics before you run a test
Analyze > Descriptive Statistics > Explore is the deepest of the three, and the one built for a specific job: deciding whether a scale variable is fit for a parametric test, or hunting down the case(s) skewing a distribution. Its Statistics output includes the mean with a 95% confidence interval, the 5% trimmed mean, median, variance, standard deviation, min/max, interquartile range, and skewness/kurtosis with standard errors — then, under Outliers, the five highest and five lowest case numbers by value, so you can go straight to the offending rows in Data View instead of guessing from a plot.
Checking Plots > Normality plots with tests adds a Tests of Normality table with both Shapiro-Wilk and Kolmogorov-Smirnov. SPSS applies a Lilliefors Significance Correction to the Kolmogorov-Smirnov row automatically — meaning it estimates the mean and SD from your own data and looks the D statistic up against the Lilliefors reference distribution rather than the standard K-S tables, which is the statistically appropriate correction for testing normality of unknown-parameter data (as opposed to a K-S test against a fully specified distribution). The same dialog also produces a boxplot, stem-and-leaf plot, and Q-Q plots. Put a categorical variable in the Factor List and Explore reruns the entire diagnostic block once per group — genuinely useful before a between-groups test, since it lets you check normality and spot outliers separately for each group rather than pooled.
Choosing the right procedure by variable type and goal
| Procedure | Best for | Gives you | Doesn’t give you |
|---|---|---|---|
| Frequencies | Categorical variables; counting every distinct value | Count, Percent, Valid Percent, Cumulative Percent; optional percentiles/charts | Confidence intervals, normality test, outlier flags |
| Descriptives | A fast summary row for several scale variables at once | N, min, max, mean, SD; optional variance, range, skewness, kurtosis; z-score save | Median, mode, percentiles, split-by-group, normality/outlier diagnostics |
| Explore | Checking assumptions or diagnosing one scale variable before/instead of a parametric test | Mean with CI, trimmed mean, median, IQR, skewness/kurtosis with SE, Shapiro-Wilk/K-S normality, 5 highest/lowest outlier cases, boxplot, split by factor | Compact multi-variable summary — one variable (or one per factor level) at a time is the point |
As a rule of thumb: reach for Frequencies the moment a variable is categorical, or when a scale variable’s percentiles are what you actually need. Reach for Descriptives when the deliverable is a clean summary table of several continuous measures and you already trust the data is well-behaved. Reach for Explore whenever you’re about to run a t-test, ANOVA, or regression and need to check normality and screen for outliers first — it is the only one of the three built to answer “is this variable safe to analyze this way,” not just “what does this variable look like.”
Building a publication-ready table from the output
A typical results section needs both kinds of output side by side: categorical demographic or grouping variables reported as n (%) from Frequencies, and continuous measures reported as M (SD) from Descriptives, each rounded to the convention your target journal or thesis style guide specifies (commonly two decimal places). If a continuous variable’s Explore diagnostics flagged non-normality or outliers, note that alongside the descriptive row rather than silently reporting the mean as if the distribution were unremarkable — the reporting conventions for descriptive statistics generally, including how which statistic fits which measurement scale, apply the same way regardless of which SPSS procedure generated the numbers. Copy each output table into your word processor via Edit > Copy on the pivot table (or export via File > Export for a Word/Excel-ready version) rather than retyping values by hand, which is a common source of transcription errors in write-ups.
Frequently asked questions
Can Frequencies produce a mean and standard deviation?
Yes — its Statistics button has the same central-tendency and dispersion checkboxes as Descriptives, plus percentiles. Use them on Frequencies only when running it on a scale variable specifically for the percentile output; running them on a genuinely categorical variable produces a number with no real interpretation.
Why doesn’t Descriptives show a median?
It’s not in the Options dialog at all — Descriptives is deliberately limited to mean-and-SD-family statistics. Get a median from Frequencies (with “Display frequency tables” unchecked so it doesn’t print every distinct value) or from Explore.
Does Explore replace Descriptives for a routine summary table?
It can, but it’s more than a routine table needs. Explore always produces a normality test and outlier listing, plus a boxplot and stem-and-leaf plot by default — the right amount of output when you’re deciding whether a variable is analysis-ready, more than you want for a simple M/SD row in a methods table.
Which procedure reports Valid Percent, excluding missing data?
Frequencies only. Its output table shows raw Percent (of all cases, including missing), Valid Percent (of non-missing cases), and Cumulative Percent side by side — a distinction neither Descriptives nor Explore’s summary statistics make.
Can any of the three split output by group automatically?
Only Explore, via its Factor List, which reruns the full diagnostic block once per group. Descriptives has no split option. For Frequencies, use Data > Split File first, since the Frequencies dialog itself has no factor field.
