Box Plot

Most graphs are misleading because they display too much. The box plot graph misleads by displaying just enough; that’s precisely why the box plot is worth learning.

The box plot graph reduces your entire dataset to the five lines and one rectangle needed to communicate everything important: Where’s the center? How spread out is this data? Which data points are behaving abnormally? You can’t compare groups as quickly on any other graph, and you can’t find outliers as clearly either.

By the time you’re done reading this tutorial, you will have learned how to interpret box plots, create them in Python, and even spot misleading graphs.

What Is a Box Plot?

It was John Tukey who created it. Created in the late 1960s and early 1970s, it was officially published in his book “Exploratory Data Analysis” in 1977 — although a preliminary version existed since 1970. What he had done was solving a particular problem – how to understand a disorganized set of information fast without graphing it entirely. His solution was the so-called five-number summary, i.e., minimum, Q1, median, Q3, maximum, represented in a form of a rectangle with lines extending out of its edges.

This is still the definition of a box plot.

A graphical representation of a dataset using only five values.

Within Exploratory Data Analysis (EDA), box plots are used everywhere since they are designed to be compared. One can put ten box plots in a row and instantly spot differences in behaviors among their groups. Of course, other graphs can do it too – but not that fast and not with outliers identified.

One should mention one aspect of a box plot right away – it does not include all data. It only provides a summary. This feature makes a difference that is more significant than most explanations suggest; it will become relevant when discussing the limitations of box plots.

The 5 Components of a Box Plot

The box sitting in the middle is where most of the action lives. Each part encodes something specific about your data.

Component	What It Shows
Minimum	Lowest non-outlier value — the bottom whisker endpoint
Q1 (25th percentile)	Bottom edge of the box; 25% of data falls below this line
Median (Q2)	Center line inside the box; half the data above, half below
Q3 (75th percentile)	Top edge of the box; 75% of data falls below this line
Maximum	Highest non-outlier value — the top whisker endpoint

The difference between Q1 and Q3 is the interquartile range (IQR):

IQR = Q3 – Q1

IQR spans the middle 50% of your data. Seriously, it is the IQR that determines nearly everything else about this plot: whisker length, outliers, and most of the conclusions you can draw when looking at it.

The whiskers. The whisker on each side extends to the furthest data point which is less than 1.5 times the IQR away from the edge of the box. Anything further out is marked as a separate point – your outlier.

What’s with the 1.5? Well, this is what Tukey found to be useful: the factor of 1.5 includes approximately 99.3% of data values in a perfect normal distribution – an idealistic notion since real-life data usually deviates from perfection. Rare enough to raise flags, yet common enough to be more or less meaningful – that was Tukey’s decision, made based on experience, not some divine rule of statistics.

How to Read a Box Plot

Reading a box plot is a four-part skill. Each part answers a different question about your data.

Reading the Center (Median)

Your reference point is the median line. This will tell you where “average” lies within this data set.

Here is the key – it is not only the position of the median that counts, but its location relative to the box. A central placement of the median means that the distribution is symmetrical. However, if the line is pulled towards Q1, there is a rightward skew because of the presence of large values.

Salary distribution is perhaps the most obvious example here. The majority of employees earn medium salaries, while a few executives skew the average considerably to the right. This becomes instantly apparent when plotted on a box plot but invisible on a bar chart.

Reading the Spread (IQR)

High Variability in the Middle 50% = Tall Box, Short Box = data is tightly clustered.

This is all you need to know. If you have a tall box in exam score data, then you can be certain that students’ performance was highly inconsistent; some students performed well while others didn’t. A small box indicates consistent clustering of scores.

There is no right or wrong. It all boils down to the nature of the measure.

Reading the Whiskers

Whiskers display the span of normal values after identifying the outliers. If the whiskers are unequal, then there is an indication of asymmetry. Long upper whiskers with short lower whiskers indicate that the data tapers off to the higher side; the data is more dispersed on the upper end than on the lower end.

Spotting Outliers

Dots outside of the whiskers are usually misunderstood in any box plot. They look wrong to our eyes and we think there must be an error.

An outlier, in simple terms, is just a number that is farther than 1.5 × IQR from the edge of any box. Period. Before removing a data point, you should consider whether it makes sense in the context in which it was gathered. A one-off spike in app usage cannot be a recording error – perhaps it was due to a link on Techcrunch.

How to Create a Box Plot (Step-by-Step with Python)

Manual Calculation Method

Consider the following data set – marks obtained by 10 students in an examination:

45, 52, 58, 61, 67, 70, 73, 81, 88, 94

Step 1: The data is sorted in ascending order here. It should always be sorted first.

Step 2: Calculate Q2 (the median) – average of the 5th and 6th terms: (67+70)/2=68.5

Step 3: Find Q1 – the median of the first five values: 58

Step 4: Find Q3 – the median of the second five values: 81

Step 5: Find the IQR. Q3−Q1=81−58=23

Step 6: Set whisker boundaries:

• Lower: 58−(1.5×23)=23.5
• Upper: 81+(1.5×23)=115.5

Step 7: Check if there are any outliers – minimum is 45 (greater than 23.5✅), maximum is 94 (less than 115.5✅). There are no outliers in this data set.

Five-number summary: 45 | 58 | 68.5 | 81 | 94

Python Code — matplotlib

In practice, though, nobody runs these calculations by hand once they’ve moved past learning the concept. Here’s the same dataset in Python using matplotlib:

python
import matplotlib.pyplot as plt

# Exam scores dataset
scores = [45, 52, 58, 61, 67, 70, 73, 81, 88, 94]

fig, ax = plt.subplots(figsize=(6, 5))
ax.boxplot(
    scores,
    patch_artist=True,
    boxprops=dict(facecolor='steelblue', color='navy'),
    medianprops=dict(color='red', linewidth=2),
    whiskerprops=dict(color='navy'),
    capprops=dict(color='navy'),
    flierprops=dict(marker='o', color='red', markersize=6)
)

ax.set_title('Exam Score Distribution', fontsize=13)
ax.set_ylabel('Score')
ax.set_xticks([1])
ax.set_xticklabels(['Class A'])
plt.tight_layout()
plt.show()

Here’s what you should be careful about: patch_artist=True fills the box (without it, the box will be transparent), medianprops makes the median red, so that it pops out right away when used in presentations, and flierprops specifies how outliers look. Default settings are good for exploratory purposes, but these settings are required for a first-time viewer of box plots.

Python Code — seaborn

Grouped comparison is done much more effectively in seaborn than in matplotlib – less coding, direct dataframe interaction from pandas, and good default settings for multi-group visualization. I have seen that data professionals in 2026 typically use seaborn when making group-based distribution comparisons.

python
import seaborn as sns
import pandas as pd
import matplotlib.pyplot as plt

# Scores across three classes
data = pd.DataFrame({
    'score': [45,52,58,61,67,70,73,81,88,94,
              55,60,63,68,72,75,79,83,87,91,
              40,48,55,62,70,74,78,85,90,96],
    'class': ['A']*10 + ['B']*10 + ['C']*10
})

sns.boxplot(
    x='class',
    y='score',
    data=data,
    palette='Set2',
    width=0.5,
    linewidth=1.5
)

plt.title('Exam Scores by Class', fontsize=13)
plt.xlabel('Class')
plt.ylabel('Score')
plt.show()

Four lines of logic. Three groups. Instantly comparable.

How to Compare Multiple Box Plots

This is where box plots earn their place. One box plot is a summary. Several side by side are a conversation between datasets.

Compare the Medians

Begin here. If the median of one set lies beyond the box of the other set, then there is likely a significant difference between the two sets worth exploring, even before conducting any statistical test.

When the medians lie within each other’s boxes, there is significant overlap between the two sets. Do not assume that they are different right away; look at their variability.

Compare the Spread

If two groups have a box which is tall and short respectively, the group with the tall box will have a higher variability compared to the one with a short box. However, this depends on the situation.

Data on the manufacturing process: Shorter box is good since the output is relatively constant. Data on sales performance: A shorter box may be an indication that your staff has reached their limits. The graph remains the same, but the measure varies. Two sets of data may be very similar in terms of medians but differ considerably in terms of IQRs.

Compare Skewness

Right skew: Median is nearer to Q1, upper whisker is longer. Left skew: Median is nearer to Q3, lower whisker is longer.

If comparing groups, it would be wise to see if they both have similar skews. When they do not have similar skews, this implies that they have distinct underlying distributions, which would matter when choosing the appropriate test to perform.

Compare Outliers

Additional outliers do not indicate bad data. Oddly enough, in the case of detecting fraud and monitoring anomalies, the outliers are actually the main reason to make the graph. The box is simply a supporting element to the dots.

Correct.

This reversal, where the “summary” is the backdrop and the “exception” is the message, is never explained to you at the beginning when you learn box plots.

When to Use a Box Plot (And When Not To)

Most articles on this topic get this exactly wrong. They explain all the things box plots do well and quietly skip past the situations where a different chart would have served you better. That omission causes real problems in practice.

Here’s the decision table no one else gives you:

Situation	Use Box Plot?	Use Instead
Comparing distributions across 3+ groups	✅ Yes	—
Single group, want to understand shape	❌ No	Histogram
Fewer than 20–30 data points	❌ No	Dot plot or strip chart
Suspect bimodal or multimodal data	❌ No	Violin plot or histogram
Need to show absolute frequencies, not distribution	❌ No	Bar chart
Comparing centrality quickly across many categories	✅ Yes	—
Presenting to a non-technical audience for the first time	⚠️ Caution	Bar chart with error bars

Regarding sample sizes: there is no absolute lower limit suggested by the statistical literature, but sample sizes under 20-30 start exhibiting instability in quartile estimates, leading to the potential for deceptive whiskers.

An eight-point box plot is not an abstraction. It is a conjecture disguised as a graph.

Box Plot Variations Worth Knowing

Notched Box Plot

The notch is a confidence interval drawn on the sides of the box, calculated at about 1.58 × IQR / √n.

The bottom line is straightforward: non-overlapping notches on box plots suggest different medians, but this is merely an informal observation, not a replacement for a rigorous statistical test. And overlapping notches mean you have not shown statistical significance, which is one of the frequent mistakes people make when presenting results using this kind of graphic. That might seem obvious except when it isn’t, and you have to explain it to a client.

Violin Plot — When It Beats a Box Plot

The violin plot simply plots the density function on both sides of a typical box plot. The violin plot reveals the shape of the distribution, which the box plot itself cannot achieve.

The violin plot is ideal to use if you think that your data may be multimodal. If, for example, you have a multimodal data set like two distinct groups of users who behave quite differently, it would appear normally within the box plot since the median and IQR cannot distinguish between the two different distributions within the data.

Vertical vs. Horizontal Orientation

Vertical boxplots are appropriate where the categorical variable is temporal in nature – months, quarters, years – with top-down reading. Horizontal boxplots make more sense when there are many categories to display, with category names that may be very long. There’s no other rationale here.

Common Mistakes When Reading a Box Plot

Mistake 1: Interpreting the box to indicate all the information. It contains 50% of the information. The other 50% lies outside the box. In the form of whiskers and outliers. You can have a box that appears very narrow yet contains data whose entire range extends well beyond the box’s borders, possibly even to three times its size.

Mistake 2: Automatically treating an outlier as an error. Outliers should always be investigated first before being removed. In health data, the outlier can be the most significant subject in the population sample. In e-commerce, the outlier can be your best customer.

Mistake 3: Reading a short IQR as “clean data”. A small IQR does not mean the dataset is clean. The box represents only the middle 50% of the data. The rest may have long whiskers and extreme outliers. Short box ≠ , neat dataset.

Mistake 4: Comparing medians without checking the spread. This occurs when two box plots have the same median lines but different IQRs. Equating them would be an analytical mistake that leads to poor decisions down the road.

Mistake 5 –  Skipping whisker asymmetry. When one whisker is three times as long as another, then your data is skewed. The degree of skewness will determine which tests you can do on it. This is no trivial matter – it’s a shape indicator contained within a line.

Frequently Asked Questions

Conclusion

Box plots don’t answer your questions, but they ask them. If the median is towards the outer edge of the box, that’s a question. If the whisker length of one group is double the length of another’s, that’s a question. If the outliers are clustered together on just one side that’s definitely a question.

The analysts who know how to leverage box plots are not those who can read them quickly. They are those who recognize what the graph cannot tell them from an analytical perspective, and seek it with another method.