The five numbers a box plot actually shows
A box plot compresses a distribution into five key values: the minimum, first quartile (Q1), median, third quartile (Q3), and maximum — plus any points flagged as outliers, shown separately. The "box" spans from Q1 to Q3 (the middle 50% of the data), with a line marking the median inside it; "whiskers" extend out to the most extreme non-outlier values.
Why the box specifically spans Q1 to Q3
Q1 and Q3 (the 25th and 75th percentiles) bound the middle half of the data — this range is deliberately chosen because it's resistant to extreme values in a way the full min/max range isn't. A single extreme outlier can drag the min or max far from the rest of the data, but it has zero effect on Q1 or Q3, which are determined entirely by the data's middle section. This is exactly why the box (not the whiskers) is considered the more robust summary of "where most of the data actually sits."
The 1.5×IQR rule, worked through
The interquartile range (IQR) is simply Q3 − Q1 — the width of the box. The standard outlier convention flags any point further than 1.5×IQR beyond either edge of the box:
IQR = Q3 − Q1
lower fence = Q1 − 1.5 × IQR
upper fence = Q3 + 1.5 × IQR
Any value below the lower fence or above the upper fence is flagged as an outlier.
For example, with Q1 = 22 and Q3 = 28 (IQR = 6): lower fence = 22 − 9 = 13, upper fence = 28 + 9 = 37. Any value below 13 or above 37 gets flagged.
Why 1.5 specifically, and why it's a convention, not a law
The 1.5 multiplier is a widely adopted convention (popularized by statistician John Tukey), chosen because it flags a reasonable, useful proportion of extreme values for many real-world roughly-normal distributions — not derived from any deeper mathematical necessity. Some analyses use a stricter multiplier (3×IQR, for "extreme outliers" specifically) when a more conservative flag is wanted. Like the 0.05 p-value threshold, 1.5×IQR is a practical, widely used default — reasonable to use by default, but not sacred.
Why whiskers stop at the fences, not the actual min/max
The whiskers extend only to the most extreme data point that's still within the fences — not necessarily all the way to the fence value itself, and never beyond it. Anything beyond the fences is plotted as an individual point (an outlier), separate from the whisker line. This is why a box plot's whisker length can vary noticeably even for datasets with a similar IQR — it depends on where the actual non-outlier data points happen to fall relative to the calculated fence.
Common mistakes
- Treating every flagged "outlier" as a data error. The 1.5×IQR rule flags statistically unusual values — some outliers are entirely legitimate, genuinely extreme but correct data points, not mistakes.
- Assuming 1.5 is a mathematically derived, universal constant. It's a widely used convention; some contexts reasonably use a different multiplier for a stricter or looser outlier definition.
- Comparing box plots across datasets without checking sample size. A larger dataset is statistically more likely to contain a few flagged outliers even from a genuinely well-behaved distribution, simply due to having more data points overall.
FAQ
What's the difference between Q1/Q3 and the min/max in a box plot?
Q1 and Q3 bound the middle 50% of the data and are resistant to extreme values; min and max are the absolute extremes, which a single outlier can drag far from the rest of the data.
Why does the outlier rule use 1.5×IQR specifically?
It's a widely adopted convention (from statistician John Tukey) that works reasonably well for many real-world distributions — not a strict mathematical law, which is why some analyses use a different multiplier when appropriate.
Does a flagged outlier always mean bad or wrong data?
No — it means the value is statistically unusual relative to the rest of the dataset under this convention; it could be a genuine, correct extreme value or an actual data error, and distinguishing between them requires looking at the specific case.
Generate box plots with automatic outlier detection for every numeric column with the Box Plot Generator — entirely client-side.