Box Plots

Precise definition

A box plot displays a median, lower and upper quartiles $Q_1,Q_3$, and whiskers under a stated convention. The interquartile range is $IQR=Q_3-Q_1$. In a common modified box plot, potential outliers lie below $Q_1-1.5IQR$ or above $Q_3+1.5IQR$, and whiskers extend to the most extreme non-outlier observations.

Notation and mathematical language

The box spans the middle 50% of ordered data, and its internal line marks the median. Quartile algorithms vary for finite samples, so software or course convention should be stated. A box plot does not show every data value or generally identify the mean.

Conceptual picture

Box length compares central spread; median position and unequal whiskers can suggest skew. Side-by-side plots efficiently compare location, variability, overlap, and flagged observations across groups, but they hide sample size unless annotated.

Conditions and key results

The 1.5-IQR rule flags observations for investigation; it does not prove they are errors or members of another population. Small samples make quartile and shape interpretations unstable. Different group sizes and measurement contexts must accompany comparisons.

A reliable strategy

1. Order the data and compute the median and quartiles using one declared convention.
2. Calculate $IQR$ and both fences; classify potential outliers without deleting them automatically.
3. Set whiskers according to the chosen ordinary or modified convention and label group sample sizes.
4. Compare centre and spread descriptively; use inferential methods for population claims and avoid causal language.

Fully worked example

Interpretation and application

Box plots summarize waiting times, scores, incomes, and lab measurements. Apparent group differences are descriptive association. Random sampling or experimental assignment and uncertainty analysis determine whether broader or causal conclusions are warranted.

Common mistakes

Verification and reasonableness

• Recount the ordered positions used for quartiles.
• Recompute fences and verify each whisker is the farthest data point inside them.
• Compare the plot with a dot plot or raw list to detect hidden clusters or sample-size differences.

Practice

1. If $Q_1=10,Q_3=18$, find $IQR$.
2. Find the upper 1.5-IQR fence.
3. Does a flagged point have to be removed?

Further deduction

Notched box plots sometimes approximate a confidence interval around the median, but notch formulas and assumptions vary. Overlapping notches are not a universal hypothesis test, and non-overlap is only an approximate visual signal under the software's method. A standard box plot without notches contains no sampling-uncertainty interval.

Two groups can have identical box plots and very different distributions because the five-number summary omits clusters and internal gaps. For small or moderate samples, overlaying jittered points or a dot plot preserves raw structure. The box plot is a compact comparison device, not a complete distribution fingerprint.

For strongly skewed data, the usual $1.5\,\mathrm{IQR}$ fences may label many legitimate tail observations as outliers. That label is a screening convention, not proof of error or a reason for automatic deletion. Investigate data entry, measurement, and subject-matter plausibility, then report sensitivity with and without influential points when justified. Transformations or adjusted box plots may describe skew more fairly, but the chosen method must be stated.

Related topics

• Percentiles
• Histograms
• One-Variable Data