Two coffee shops both average a 4.0-star rating. One earns straight 4s from everyone — reliable, unremarkable. The other is a war zone of 1s and 5s: people either adore it or storm out. Same average, opposite realities. The number that told you they were 'the same' threw away the most important thing about each shop: the shape of its reviews. Learn to see that shape — the distribution — and a column stops being a wall of numbers and starts telling you a story.
What you'll learn
What
A distribution describes how the values of a variable are spread out — where they cluster, how far they range, and whether they lean to one side. A histogram is the picture of a distribution: bars showing how many values fall into each range.
Why
A single average hides everything about spread and shape. Two datasets with identical means can behave completely differently. Reading a distribution is the first real move of exploratory data analysis (EDA) — it tells you where the data lives, whether outliers lurk, and which summary statistic to even trust.
Where it's used
Every EDA notebook opens with histograms. Response-time monitoring, income studies, A/B test results, quality control — all start by looking at the shape of the data, not just its centre.
Where this runs in production
Engineers never judge a service by its average latency alone — they look at the whole distribution, because a long tail of slow requests hides behind a healthy-looking mean.
Most learners do short sessions, a few do marathon ones. The distribution of session length — not the average — is what tells the product team how people really use the app.
Plot every runner's marathon time and you get a distinctive right-skewed shape with clusters near round-number goals. The distribution reveals behaviour an average would erase.
Centre — roughly where the values pile up (mean or median from the last lesson). Spread — how far the values range, from tightly clustered to widely scattered. Shape — the silhouette: symmetric, skewed to one side, one peak or several. Every distribution description is these three things.
Imagine asking a stadium crowd to stand in lanes marked by height — under 150cm, 150–160, 160–170, and so on — then looking down from above. The lanes in the middle are packed; the very short and very tall lanes have only a few people. That top-down view of how many people stand in each lane is exactly a histogram. The bumps and gaps you'd see are the distribution's shape.
reviews_A = [4, 4, 4, 4, 4] # everyone gives 4 stars reviews_B = [1, 1, 5, 5, 4] # love-it-or-hate-it # mean(A) = 4.0, mean(B) = 3.2 ... close-ish average, # but the SHAPES are opposite: A is one tight spike, # B is split into two clumps at the extremes.
The mean barely separates these, but their distributions are worlds apart. A is a single narrow spike (tiny spread); B is bimodal — two clusters at opposite ends. Only by looking at the shape do you see that B is polarising.
A skewed distribution is like a comet: a bright dense head where most values pile up, and a faint tail streaming off to one side. The direction the tail points is the direction of the skew. Right-skew has the tail pointing to the right (high values) — think incomes, where most people cluster low and a few billionaires stretch the tail far right. Remember: the skew is named for the tail, not the hump.
This connects straight back to the last lesson. In a symmetric distribution, the mean and median sit together in the middle. In a right-skewed distribution, the long tail drags the mean toward the high values while the median stays near the hump — so the mean overstates the typical value. Reading the shape first is how you know which average to trust.
Click each shape to learn how to recognise it and what it means for your analysis. Naming the shape is the first move of exploring any variable.
Select a type to see its full definition, operations, and data science usage.
Ten app reviews came in: 5, 4, 5, 3, 4, 5, 5, 2, 4, 5. How are the ratings distributed?
A histogram is just counts per bucket. Tally how many times each rating appears — that count is the bar height for that value.
Your task
Turn a list of numbers into a text histogram by counting how many fall into each bucket, then printing a bar of '#' for each count. This is exactly what a plotting library does under the hood. Fill the blanks and run it to see the shape.
2 # 3 ## 4 #### 5 ## 9 #
Write your solution in the editor on the right, then hit Run.
What does the height of a bar in a histogram represent?
A distribution has most values bunched at the low end and a long tail stretching toward high values. What is it called?
In a right-skewed distribution, how do the mean and median compare?
What is the most likely explanation?
Measure spread with the range: given a list of values, print the difference between the largest and smallest (max - min). A bigger range means a wider distribution.
Which factory is better, and why can't the mean tell you?
What is a distribution, and why do you look at one instead of just the mean?
A distribution describes how the values of a variable are spread out — where they cluster, how far they range, and whether they lean to one side. I look at it instead of just the mean because a single number throws away almost everything about the data. Two variables can share an identical mean while behaving completely differently: one tightly clustered and predictable, the other wildly variable or split into two groups. Looking at the distribution — usually via a histogram — tells me the centre, the spread, and the shape all at once. That's what reveals outliers, skew, and hidden subgroups, and it's why plotting histograms is the first thing I do in exploratory analysis before trusting any summary statistic.
How do you describe the shape of a distribution, and what does skew tell you?
I describe a distribution with three things: its centre (roughly where values pile up), its spread (how far they range), and its shape (the silhouette). For shape, the key categories are symmetric — roughly mirror-image around the centre, like heights — and skewed, where one tail is longer than the other. Skew is named for the direction of the tail: right-skew has a long tail toward high values, which is typical of income, prices, and wait times; left-skew has a tail toward low values, like scores on an easy test. Skew matters because it tells me which summary to trust. In a right-skewed distribution the long tail drags the mean above the median, so the median is the honest 'typical' value. I also watch for multiple peaks — a bimodal shape usually signals two different groups mixed together, which is a cue to split them rather than summarise them as one.
A histogram can look different depending on how you build it. What choices affect the picture, and how do you avoid fooling yourself?
The biggest lever is bin width, or equivalently the number of bins. Too few wide bins over-smooth the data and can erase real structure like a second peak; too many narrow bins turn random noise into fake features and make the shape look jagged and meaningless. So a histogram isn't an objective truth — it's a summary I designed, and I can accidentally design it to tell the wrong story. To avoid fooling myself I try a few bin widths and check that the shape is stable — genuine features like skew or bimodality persist across reasonable bin choices, while artifacts appear and vanish. I also mind the axis range and whether outliers are squashing everything into one bar. And I don't rely on the histogram alone; I cross-check with summary numbers like the median, percentiles, and the spread, and sometimes a boxplot, so the visual and the statistics agree before I draw a conclusion. The habit is to treat the first histogram as a question, not an answer.
Common Mistakes to Avoid
1) Judging a variable by its mean alone and ignoring spread and shape. 2) Getting skew backwards — the skew is named for the tail's direction, not the hump's. 3) Reporting a single average for bimodal data, landing in the empty valley that describes nobody. 4) Trusting one histogram without trying another bin width, so noise looks like structure (or real structure gets smoothed away). 5) Forgetting that a long tail can hide serious outliers behind a healthy-looking average.
Ask the AI Tutor
Try these prompts in the AI Tutor panel: • 'Describe a distribution in words and quiz me on its shape.' • 'ELI5: the difference between spread and centre.' • 'Give me two datasets with the same mean but different shapes and ask what differs.' • 'Show me a right-skewed vs. left-skewed example and let me name each.' • 'Interview mode: ask me why I plot a histogram before trusting the mean.'
Glossary
Distribution — how a variable's values are spread across their range. Histogram — a bar chart of counts per value-range, showing a distribution. Bin — one bucket (value range) in a histogram. Spread — how widely values range around the centre. Skew — asymmetry; a longer tail on one side (right/positive or left/negative). Symmetric — roughly mirror-image around the centre. Bimodal — two distinct peaks, often two mixed groups. Tail — the thin, stretched end of a distribution. Percentile — the value below which a given percent of the data falls (e.g. p95).
Recommended Resources
• Interactive: Seeing Theory (seeing-theory.brown.edu) — its animated chapter on distributions builds intuition beautifully. • Practice: sketch the histogram shape you'd expect for house prices, adult heights, and dice rolls, then check your reasoning. • Reference: preview pandas' .hist() and .describe() to see how you'll generate these shapes and read spread in code. • Next in DSM: you can read the shape of one variable — next you'll reason about how two variables relate, and the trap of confusing correlation with causation.
Recap
✓ A distribution shows how a variable's values spread out — described by centre, spread, and shape. ✓ A histogram pictures a distribution: bar height is the count of values in each bin. ✓ Skew is named for the tail's direction — right-skew (income, prices) drags the mean above the median. ✓ Two datasets with the same mean can have completely different spreads and shapes. ✓ A bimodal shape (two peaks) usually means two groups are mixed and shouldn't share one average. Next up: Correlation Is Not Causation. You can now read one variable's shape — next you'll reason about two variables together, and learn to resist the field's most common error: assuming that because two things move together, one must cause the other.
Run your code to see the output here.