Written and maintained by CASRAI Editorial Board
Last updated
A boxplot summarizes a variable’s distribution in five numbers — the median, the two quartiles that form the box, and the whiskers that reach out to the data — and it does one thing most other SPSS charts don’t: it flags individual cases as outliers or extreme values automatically, by case number, without you having to compute anything by hand. SPSS builds one from three different places (Chart Builder, Legacy Dialogs, and the Explore procedure), and the outlier-flagging convention it uses — a small circle for one tier of unusual points, an asterisk for another — is easy to misread if you don’t know what distinguishes the two.
What the box, whiskers, and flagged points actually represent
The box spans the interquartile range (IQR): its lower edge sits at the 25th percentile (Q1), its upper edge at the 75th percentile (Q3), and the line inside it marks the median, not the mean. The whiskers extend from each edge of the box out to the most extreme data point that still falls within 1.5 × IQR of that edge — not to a fixed multiple, but to whichever real case is farthest out while still inside that boundary, which is why whisker length varies case by case rather than always looking symmetric. Any point beyond the whisker is plotted individually and, in SPSS, sorted into one of two tiers rather than lumped together as a single undifferentiated “outlier” category.
SPSS’s two-tier convention: outliers vs extreme values
SPSS distinguishes points beyond the whisker by how far beyond they are, using the same 1.5 × IQR unit as the threshold, applied twice:
- Outliers — a point more than 1.5 IQR but no more than 3 IQR from the box edge. SPSS marks it with a small circle (○) and labels it with its case number.
- Extreme values — a point more than 3 IQR from the box edge. SPSS marks it with an asterisk (*) and, again, labels it with its case number.
The case number next to each flagged point is the row number from the Data View grid (or, if you’ve set a case label variable, that label instead) — not a rank or a value. To find out which respondent, sample, or record it actually is, go to Data > Go to Case and enter that number, or double-click the point in the Chart Editor, which selects the corresponding row directly. Because the case number is the whole point of flagging — it turns “there’s an outlier” into “case 47 is an outlier, go look at it” — always keep the label visible; don’t turn off case labeling to declutter a busy chart, or you lose the one thing a boxplot does that a simple summary statistic can’t.
This 1.5-IQR / 3-IQR split is a convention, not a statistical test: a flagged case isn’t automatically wrong, and an unflagged one isn’t automatically fine for every purpose. Treat both tiers as “go look at this case,” not “delete this case.”
Simple boxplot: one variable, or one variable split by groups
A simple boxplot draws one box per variable or per group and is the right default when you’re not yet crossing two categorical factors.
Chart Builder
Go to Graphs > Chart Builder. In the Gallery tab choose Boxplot, then drag the Simple Boxplot icon onto the canvas. Drop your scale (interval/ratio) variable onto the y-axis drop zone; if you want one box per group, drop a categorical variable onto the x-axis drop zone as well. Click OK.
Legacy Dialogs
Go to Graphs > Legacy Dialogs > Boxplot. In the dialog that opens, pick the Simple icon, then choose Summaries for groups of cases if you’re comparing one variable across levels of a categorical grouping variable, or Summaries of separate variables if you want one box per variable instead (e.g., three related scale items side by side, each as its own box). Click Define, assign the variable(s), and click OK.
Explore
Go to Analyze > Descriptive Statistics > Explore, move your scale variable into Dependent List and, optionally, a categorical variable into Factor List. Click Plots, and under Boxplots choose Factor levels together (one boxplot chart with a box per factor level, side by side) if a Factor is set. Explore is the most efficient route when you’re already running it for normality diagnostics — the same dialog also produces the Shapiro-Wilk test and Q-Q plots — and want the boxplot as one more piece of the same output rather than a separate chart run. It’s also one of the three routes covered in the guide to choosing between Frequencies, Descriptives, and Explore for a given variable type.
Clustered boxplot: two categorical factors at once
A clustered boxplot adds a second categorical variable so you can compare a distribution across two grouping dimensions simultaneously — for example, a score broken out by both treatment condition and sex, or by site and time point.
Chart Builder
In the Gallery’s Boxplot section, drag Clustered Boxplot onto the canvas instead of Simple. Drop the scale variable on the y-axis, the first categorical variable on the x-axis, and the second categorical variable onto the Cluster on X drop zone (it appears once you hover a variable near the top of the plot area, or you can drag it directly onto the “Set color” legend slot). SPSS then draws one cluster of boxes per level of the x-axis variable, with the boxes inside each cluster colored/grouped by the second variable.
Legacy Dialogs
Go to Graphs > Legacy Dialogs > Boxplot, pick the Clustered icon, and choose Summaries for groups of cases. Assign the scale variable to Variable, the primary grouping variable to Category Axis, and the second grouping variable to Define Clusters By. Click OK.
A clustered boxplot gets crowded fast once either factor has more than four or five levels, or the sample per cell drops low enough that a “box” is really just a handful of points — at that point a simple boxplot run separately per subgroup, or a different chart type entirely, usually reads more clearly than forcing everything into one cluttered cluster chart.
Reading a boxplot for what it’s actually for
A few things worth checking every time, not just when a chart looks obviously odd:
- Median position within the box. A median sitting off-center toward one edge, combined with unequal whisker lengths, is a visual flag for skew — useful as a first pass before running a formal normality check, and a cheaper first look than building a full histogram with a normal-curve overlay if all you need is a quick sanity check on shape.
- Box width itself carries no information in a standard SPSS boxplot (unlike some other software’s variable-width boxplots that scale width to sample size) — don’t read anything into how wide one box looks next to another.
- A short or invisible whisker on one side usually means the data are bounded there (a floor of zero on a count variable, a ceiling on a bounded scale) rather than a data-entry problem.
- Multiple flagged cases at the same case number across several boxplots (e.g., the same respondent flagged as an outlier on three different scale variables) is worth checking as a single problematic record rather than three unrelated coincidences.
A boxplot split by group is also a natural check to run before a group-comparison test, since it puts spread and skew on the screen where a p-value alone won’t: see the independent-samples t-test guide, or the Mann-Whitney U test guide when the boxplot itself is the reason you’re reaching for a non-parametric alternative.
Frequently asked questions
What’s the difference between an outlier and an extreme value in SPSS?
Both are points beyond the whisker, but outliers are between 1.5 and 3 times the IQR from the box edge and are marked with a circle, while extreme values are more than 3 times the IQR from the box edge and are marked with an asterisk. Both carry a case-number label.
Does SPSS ever remove flagged outliers automatically?
No. Flagging is purely visual and informational — SPSS never excludes or downweights a case in any subsequent analysis because a boxplot marked it. Any exclusion is a decision you make separately, and it belongs in your reported methodology if you make it.
Why does my boxplot show no whiskers or flagged points at all?
That happens with a small sample, a variable with very little spread, or a variable that’s actually discrete/categorical rather than continuous — check that the variable’s measurement level and actual values are what you expect before assuming the chart is broken.
Can I get a boxplot split by more than two grouping variables?
Not in one chart through the standard dialogs. A clustered boxplot handles two categorical factors; for a third, either use the Small Multiples / panel option in Chart Builder to facet by the third variable, or split the file (Data > Split File) and generate separate charts per subgroup.
Where do I find which record a flagged case number corresponds to?
Go to Data > Go to Case and enter the case number shown next to the point, or double-click the flagged point directly in the Chart Editor, which selects the matching row in the Data View grid.
