You've seen them in test score reports, salary surveys, and medical charts. They show up everywhere data gets summarized — but most people treat them like interchangeable labels. Percentiles. In practice, quartiles. They're not.
The short version: quartiles are just specific percentiles wearing a different name tag. Day to day, that's it. But the relationship runs deeper than a simple definition, and understanding it changes how you read any distribution.
What Is a Percentile
A percentile tells you what percentage of values fall below a certain point. That said, score in the 80th percentile on a standardized test? In real terms, you did better than 80% of test-takers. Simple Simple, but easy to overlook. Turns out it matters..
But here's where it gets slippery. Linear interpolation. The method Excel uses versus the one R uses versus the one your statistics textbook teaches. Because of that, the nearest-rank method. Even so, there are actually multiple ways to calculate percentiles, and they don't always agree. Same percentile label, slightly different numbers And that's really what it comes down to..
Percentiles slice data into 100 equal-ish chunks. The 99th percentile marks the value below which 99% fall. The 1st percentile marks the value below which 1% of observations fall. Because of that, the median? That's the 50th percentile — right in the middle.
What Is a Quartile
Quartiles do the same job with broader strokes. Second quartile (Q2) is the median — the 50th percentile. First quartile (Q1) marks the 25th percentile. They split data into four roughly equal parts. Third quartile (Q3) sits at the 75th percentile.
That's the whole relationship. So q1 = P25. Q2 = P50. Q3 = P75.
But wait — "roughly equal" matters. Now, rarely happens in real data. With small datasets or tied values, the splits aren't perfectly clean. In real terms, four groups of exactly 25% each? The quartile boundaries are estimates, just like percentile boundaries.
Why It Matters
You've seen a box plot. That box spans from Q1 to Q3 — the interquartile range (IQR). Because of that, 5 × IQR beyond the box. That said, the whiskers stretch to 1. Everything outside gets flagged as a potential outlier Worth knowing..
This visualization only works because quartiles anchor it. Practically speaking, your outlier detection would be wrong. If you misunderstood Q1 as the 20th percentile instead of the 25th, your box would be wrong. Your whole read of the distribution shifts.
Salary surveys use percentiles heavily. "The 90th percentile software engineer earns $180k.Which means " That's a percentile statement. But compensation bands? Worth adding: those often use quartiles. Because of that, "Our pay band spans Q1 to Q3 of market data. " Same data, different framing Turns out it matters..
Medical growth charts track kids by percentile. "Your daughter is in the 60th percentile for height."Children in the lowest quartile of weight-for-age had 3.Worth adding: they'll report quartile ranges. " Pediatricians think in percentiles. But research papers on growth patterns? 2x higher hospitalization risk That's the part that actually makes a difference..
The framing changes the conversation. Percentiles feel precise — "60th" sounds specific. Because of that, both are useful. Still, quartiles feel structural — "bottom quartile" sounds like a group. Neither is more "true.
How They Actually Work Together
Let's walk through a concrete example. Say you have 20 test scores, sorted:
42, 45, 48, 51, 53, 55, 57, 59, 60, 62, 64, 66, 68, 70, 72, 75, 78, 81, 85, 90
Where's the 25th percentile? The 75th? The quartiles?
With 20 values, the 25th percentile sits at position 0.In real terms, that's between the 5th and 6th values — 53 and 55. Interpolating: 53 + 0.Think about it: 25. 25(55 - 53) = 53.On the flip side, 25 × (20 + 1) = 5. 5 Simple, but easy to overlook. No workaround needed..
The 75th percentile: position 0.Plus, 75. Result: 72 + 0.Between the 15th and 16th values — 72 and 75. 75(75 - 72) = 74.75 × 21 = 15.25.
So Q1 ≈ 53.5, Q3 ≈ 74.Think about it: 25. Also, the IQR ≈ 20. 75 Small thing, real impact..
But different software might give you slightly different answers. INC uses n-1. Also, eXC uses that (n+1) method. PERCENTILE.Python's numpy.R's default (type 7) uses yet another formula. In real terms, excel's PERCENTILE. percentile has its own default.
The quartile-percentile relationship holds conceptually — Q1 is conceptually P25 — but the exact number depends on your calculation method. This isn't a flaw. It's just how estimation works with finite data.
The Five-Number Summary
This is where quartiles earn their keep. Plus, minimum, Q1, median (Q2), Q3, maximum. Five numbers that sketch an entire distribution.
From those five, you get:
- Center (median)
- Spread (IQR = Q3 - Q1)
- Skewness (compare median to midhinge = (Q1+Q3)/2)
- Tail behavior (compare min/max to 1.5×IQR fences)
Percentiles can do this too — P25, P50, P75 give you the same three quartiles. But the five-number summary is a quartile-native concept. It's taught that way for a reason: it's memorable, visual, and sufficient for most exploratory analysis It's one of those things that adds up..
Percentile Granularity When You Need It
Quartiles are coarse. Sometimes you need finer slices That's the part that actually makes a difference..
Income inequality metrics use the 90/10 ratio (P90/P10) or the 99/1 ratio. On the flip side, quartiles can't capture that. The Gini coefficient integrates the entire percentile curve — the Lorenz curve plots cumulative share against population percentiles.
Standardized testing reports percentiles because "84th percentile" communicates more than "between Q2 and Q3." College admissions care about the 95th percentile cutoff, not the top quartile That alone is useful..
Clinical trials track adverse events by percentile thresholds. Even so, "Grade 3 toxicity: lab value > P95 of reference range. " That specificity matters for safety monitoring The details matter here..
The rule of thumb: use quartiles for structure and overview. Use percentiles for precision and thresholds. They're the same measurement system — just different zoom levels.
Common Mistakes
Treating quartiles as exact 25% splits. They're not. With 10 data points, you can't split into four groups of 2.5 each. The boundaries fall between values. The groups are approximate.
Confusing percentile rank with percentile value. "She's at the 80th percentile" means her value corresponds to P80. Her rank is 80th percentile. But "she scored 80th percentile" is ambiguous — 80th percentile of what? The score distribution? The norm group? Always specify the reference population And that's really what it comes down to. Took long enough..
Assuming percentiles are equally spaced. The gap between P10 and P20 isn't necessarily the same as
the gap between P80 and P90. In real terms, in a normal distribution, the middle percentiles are densely packed, while the extreme percentiles are stretched thin. In a highly skewed distribution, such as wealth or city populations, the distance between P98 and P99 might represent a massive numerical jump, while the distance between P49 and P50 is negligible Nothing fancy..
Ignoring Outliers. Because quartiles and percentiles are based on rank rather than magnitude, they are "reliable" statistics. This is usually a strength, but it can be a trap. A dataset could have an extreme outlier—a value ten times larger than the rest—and the median and quartiles will barely budge. If you rely solely on these measures, you might miss a catastrophic error or a significant anomaly that a mean and standard deviation would have flagged immediately That alone is useful..
Summary: Choosing Your Lens
The choice between quartiles and percentiles is rarely about which is "right" and more about what story you are trying to tell.
If you are performing Exploratory Data Analysis (EDA), start with quartiles. Plus, they provide a high-level blueprint of your data's shape, allowing you to quickly identify skewness and the spread of the middle 50%. They are the building blocks of the box plot, the most efficient tool for visual comparison.
If you are performing Comparative Analysis or Threshold Setting, pivot to percentiles. When you need to define "normal" vs. "abnormal," or when you need to rank individuals within a massive population, the granularity of percentiles provides the necessary resolution.
When all is said and done, both tools serve the same master: the need to summarize complexity. By understanding the mechanics of how they are calculated and the specific contexts in which they excel, you can move beyond simply reporting numbers to providing meaningful, accurate insights into the distributions that govern our world.
