How To Find Five Number Summary

You're staring at a dataset. You've got the standard deviation. Maybe it's test scores, maybe it's daily sales figures, maybe it's the number of steps your fitness tracker logged last month. You've got the mean. But something feels missing — like you're looking at a blurry photo and can't quite make out the shape.

That's where the five number summary comes in.

It doesn't replace the mean. Still, it doesn't replace the standard deviation. What it does is give you the skeleton of your data — the five landmarks that tell you where the bulk of your values sit, where the extremes live, and whether things are skewed in a way the average alone will never show you.

What Is the Five Number Summary

The five number summary is exactly what it sounds like: five specific numbers that describe a dataset's distribution. No more, no less. They are:

• Minimum — the smallest value
• First quartile (Q1) — the 25th percentile
• Median (Q2) — the 50th percentile, the middle value
• Third quartile (Q3) — the 75th percentile
• Maximum — the largest value

That's it. Which means five numbers. But those five numbers let you reconstruct the box plot, spot outliers, compare distributions side by side, and answer questions like "where does the middle 50% of my data actually fall?

The quartiles need a closer look

Quartiles split your ordered data into four roughly equal chunks. Consider this: q1 marks the boundary between the bottom 25% and the rest. Which means q3 marks the boundary between the top 25% and the rest. The median sits exactly in the middle.

But here's where people get tripped up: there's more than one way to calculate quartiles. They use different definitions. EXC and QUARTILE.Now, another one. percentile with default settings? Day to day, python's numpy. Excel's QUARTILE.R's default? The method you learned in intro stats (the "inclusive" or "exclusive" method, or the "Mendenhall and Sincich" method, or the "Tukey" method) might give slightly different answers than what your software spits out. In practice, iNC functions? Yet another.

The differences are usually small — often just a fraction of a unit — but they exist. If you're comparing results across tools, this matters.

Why It Matters / Why People Care

The mean tells you the center of mass. The five number summary tells you the shape.

Imagine two classes took the same exam. In practice, both have a mean score of 78. Class A has a five number summary of 52, 70, 78, 86, 98. Class B has 22, 60, 78, 90, 98. Same mean. Completely different stories.

Class A is clustered — most students scored between 70 and 86. The mean hides all of this. On the flip side, class B is stretched — a long left tail dragging the minimum down to 22, a wider spread in the middle 50%. The five number summary lays it bare.

This is why box plots exist. On top of that, a box plot is the five number summary drawn as a picture. The box runs from Q1 to Q3. The line inside is the median. In real terms, the whiskers stretch to the min and max (or to 1. 5×IQR, depending on convention). You can glance at two box plots and instantly see which distribution is tighter, which is skewed, which has outliers.

Real-world example: housing prices in a neighborhood. The median stays grounded. Q1 and Q3 tell you what a "typical" house actually costs — the range where half the homes sit. And the mean gets pulled up by the one mansion on the hill. If you're a buyer, that's the number you care about.

How to Find the Five Number Summary

Let's walk through it step by step. I'll use a small dataset so you can follow the logic, then talk about how it scales.

Step 1: Order your data

This is non-negotiable. The five number summary requires sorted values. Smallest to largest.

Say we have: 18, 3, 12, 7, 25, 14, 9, 21, 16

Sorted: 3, 7, 9, 12, 14, 16, 18, 21, 25

Nine values. Odd count. Good.

Step 2: Find the minimum and maximum

Easy. Min = 3. Max = 25.

Step 3: Find the median (Q2)

With nine values, the median is the 5th value — the one with four values on either side. That's 14.

If you had an even count — say ten values — the median would be the average of the 5th and 6th values. Always.

Step 4: Find Q1 and Q3

This is where the method choice appears.

Method A (Tukey / "inclusive" — common in box plots): Split the data at the median. Include the median in both halves if the count is odd. Then find the median of each half.

Lower half (including median): 3, 7, 9, 12, 14 → median = 9 → Q1 = 9 Upper half (including median): 14, 16, 18, 21, 25 → median = 18 → Q3 = 18

Method B (Mendenhall & Sincich / "exclusive" — common in textbooks): Split at the median. Exclude the median from both halves. Then find the median of each half.

Lower half (excluding median): 3, 7, 9, 12 → median = (7+9)/2 = 8 → Q1 = 8 Upper half (excluding median): 16, 18, 21, 25 → median = (18+21)/2 = 19.5 → Q3 = 19.5

Method C (linear interpolation / percentile method — what Excel's QUARTILE.INC and Python's default do): Treat position as (n+1)×p. For Q1, p=0.25. For Q3, p=0.75.

n=9. That said, q1 position = 10×0. 25 = 2.Worth adding: 5 → average of 2nd and 3rd values = (7+9)/2 = 8 Q3 position = 10×0. 75 = 7.5 → average of 7th and 8th values = (18+21)/2 = 19.

Notice: Method B and C gave the same result here. They won't always.

Step 5: Write it out

Using Method A (Tukey): 3, 9, 14, 18, 25 Using Method B/C: 3, 8, 14, 19.5, 25

Both are "

Both are perfectly valid, but the “right” choice hinges on what you’re trying to communicate and the tools you’ll use downstream.

When to lean toward Method A (Tukey’s inclusive split)

• Exploratory data analysis and visualization – Tukey designed this version specifically for box‑plot construction, so it gives you a box that aligns with the visual you’ll draw.
• Small sample sizes – When n is odd, including the median in both halves preserves the central tendency in the box, which often feels more intuitive.
• When you want a symmetric split – If the dataset truly has a natural middle value, Method A keeps that value represented in both Q1 and Q3 calculations.

When to prefer Method B or Method C (exclusive or percentile‑based splits)

• Statistical textbooks and formal reporting – Many introductory statistics courses teach the exclusive split because it mirrors the idea of “splitting the data into two independent halves.”
• Consistency with software defaults – Excel’s QUARTILE.INC (Method C) and Python’s numpy.percentile with interpolation='linear' (also Method C) are widely used in research and industry. Choosing Method B lets you replicate those results when you need a textbook‑style median of each half.
• Large datasets – With many observations, the differences between the methods become negligible, and the percentile approach (Method C) scales cleanly because it treats quartiles as empirical percentiles.

Computing the Interquartile Range (IQR)

Once you have Q1 and Q3, the interquartile range is simply:

[ \text{IQR} = Q3 - Q1 ]

Using the two summaries we just generated:

The IQR is the core of outlier detection. Plus, 5·IQRor **above**Q3 + 1. A common rule (also Tukey’s) flags any observation below Q1 − 1.5·IQR as a potential outlier.

• Method A:

  Continue exploring with our guides on what is 83 kilos in pounds and how to measure the diagonal of a rectangle.

  • Lower fence = 9 − 1.5·9 = −4.5 → no low outliers.
  • Upper fence = 18 + 1.5·9 = 31.5 → no high outliers.

• Method B/C:

  • Lower fence = 8 − 1.5·11.5 = −9.25 → none.
  • Upper fence = 19.5 + 1.5·11.5 = 36.75 → none.

If you later add a value like 40, Method A would label it an outlier (since 40 > 31.5) while Method B/C would also flag it (40 > 36.And 75). The exact thresholds differ, but the qualitative conclusion—there’s at least one extreme point*—remains the same.

Sketching the Box Plot

1. Draw a horizontal (or vertical) axis and mark the minimum and maximum values (or the fences if you’re showing outliers separately).
2. Place a box whose left (or bottom) edge is Q1 and right (or top) edge is Q3.3. Draw a line inside the box at the median (Q2).
3. Add “whiskers” extending from the box edges to the most extreme data points that are not flagged as outliers.
4. Plot outliers as individual symbols (e.g., circles or asterisks) beyond the whiskers.

The visual instantly conveys spread, central tendency, skewness, and the presence of extreme values—exactly why box plots are such a powerful exploratory tool.

Software Quick‑reference

Python (NumPy) – np.percentile and np.quantile

NumPy’s percentile (and the newer quantile) let you request any quantile with a choice of interpolation algorithms:

import numpy as np

# data
x = np.array([...])          # your vector of observations

# linear interpolation – the default and the one used by pandas
q1_lin = np.percentile(x, 25, interpolation='linear')
q3_lin = np.percentile(x, 75, interpolation='linear')

# other interpolations (all available in `np.quantile` as well)
q1_low  = np.percentile(x, 25, interpolation='lower')   # floor of the interval
q1_high = np.percentile(x, 25, interpolation='higher')  # ceil of the interval
q1_mid  = np.percentile(x, 25, interpolation='midpoint')# average of the two bounds
q1_near = np.percentile(x, 25, interpolation='nearest') # nearest data point

np.Consider this: quantile (introduced in NumPy 1. 22) is a thin wrapper that simply calls percentile with the same keyword arguments, so you can write np.25, 0.Also, quantile(x, [0. 75]) for a one‑liner.

Why the choice matters – When the sample size is small, the interpolation method can shift Q1 or Q3 by a full data‑point step. For large‑N data the differences shrink to less than a thousandth of a standard deviation, making the linear (default) approach the practical norm.

R – quantile() and the type Argument

R’s built‑in quantile() implements seven different algorithms (types 1‑7). The most common are:

x <- c(...)               # your vector
quantile(x, probs = c(0.25, 0.75), type = 7)   # default – matches numpy linear
quantile(x, probs = c(0.25, 0.75), type = 1)   # nearest‑rank, akin to Method B

If you need to reproduce a published analysis that cites a specific type, simply set the argument accordingly; otherwise, type = 7 (the R default) aligns well with the linear interpolation used by NumPy and pandas.

Other Major Platforms

Choosing a Method in Practice

1. Document the algorithm – Write the

2. Consider sample size and data distribution – For small datasets (e.g., <30 observations), the choice of interpolation can noticeably affect quartile estimates. In such cases, methods like type 1 (nearest-rank) or type 2 (averaging ranks) may be more appropriate to avoid over-smoothing. Conversely, large datasets (N > 1,000) benefit from linear interpolation, as the differences between methods become statistically negligible. Always assess the distribution shape (e.g., skewed, heavy-tailed) to determine whether extreme quantiles (e.g., 5th/95th percentiles) require special handling.

3. Prioritize reproducibility and collaboration – When sharing code or results with colleagues or reviewers, specify the quantile method explicitly. To give you an idea, in R, include type = 7 in your quantile() calls; in Python, use np.quantile(..., method='linear') to avoid ambiguity. This practice ensures that others can replicate your analysis precisely, regardless of default settings in their local environments.

4. Align with domain-specific conventions – Certain fields have established norms for quantile computation. Here's one way to look at it: financial risk models often use the "exclusive" method (similar to R’s type 1) to compute Value at Risk (VaR), while engineering standards might favor linear interpolation for tolerance intervals. Consult domain guidelines or prior literature to match the expected approach.

A Final Note on Practical Implementation

In most day-to-day scenarios, sticking with the default linear interpolation (NumPy’s method='linear', R’s type=7, or pandas’ quantile()) is sufficient. On the flip side, when precision matters—such as in academic research, regulatory reporting, or cross-platform comparisons—understanding the nuances of each method becomes critical. Consider this: tools like Python’s scipy. stats or R’s quantile() allow granular control, so take advantage of them to match your analytical goals.

At the end of the day, the key takeaway is this: quantiles are not just numbers on a report. Consider this: they are the building blocks of summary statistics, visualizations, and decision-making processes. By thoughtfully selecting and documenting your quantile method, you confirm that your analysis remains both accurate and transparent in a world where data speaks louder when it’s consistent.

Conclusion
The seemingly minor choice of interpolation method for quantiles can have tangible effects on statistical conclusions, especially in small or sensitive datasets. While defaults like linear interpolation offer a solid starting point, practitioners must remain cognizant of alternative approaches—whether in R’s suite of type arguments, Python’s method parameter, or platform-specific nuances in SAS, MATLAB, or Julia. By documenting choices, considering sample characteristics, and aligning with domain standards, analysts can work through these subtleties with confidence, ensuring their work stands up to scrutiny and delivers actionable insights. In the end, it is not just about calculating quartiles—it’s about building trust in the numbers that shape decisions.