Lesson 3 of 5 · 8 min · Shape and relationships

Distributions: normal, skewed and everything else

Covers: Python 3.14 statistics module and standard definitions. Updated 13 Nov 2026. Part of the Data Analyst roadmap.

What you will learn

  • Read a histogram and describe the shape of data
  • Understand the normal distribution and the 68-95-99.7 rule
  • Calculate and interpret z-scores
  • Recognise skewed, bimodal and long-tailed data
  • Choose statistics and charts that match the shape

In simple words

If you line up everyone in a class by height, most people stand near the middle, and a few are very tall or very short. Draw it as bars and you get a shape. The distribution is that shape: it shows which values are common, which are rare, and whether the data leans to one side. Once you know the shape, you know which average to trust and what counts as unusual.

Words to know

  • Distribution: how often each value (or range of values) occurs in the data.
  • Histogram: a bar chart of counts for ranges of values (bins).
  • Normal distribution: the symmetrical bell-shaped curve.
  • Z-score: how many standard deviations a value is from the mean.

Histograms

A histogram groups values into bins and counts how many fall in each:

import pandas as pd
import matplotlib.pyplot as plt

df["amount"].plot(kind="hist", bins=20)
plt.xlabel("Order amount (INR)"); plt.ylabel("Number of orders")
plt.show()

In Excel use Insert > Chart > Histogram; in a pivot table, group the numeric field into bins. The choice of bins changes the picture: too few hides detail and too many looks noisy. Try several. Describe a histogram by its centre, spread and shape (symmetric or skewed, one peak or two, any gaps or extreme values).

The normal distribution

Many measurements (heights, test scores of a large exam, errors in a machine) form a symmetric bell curve, centred on the mean. It is described entirely by two numbers: the mean and the standard deviation. The 68-95-99.7 rule gives the share of values within:

Distance from mean Share of data
1 standard deviation about 68%
2 standard deviations about 95%
3 standard deviations about 99.7%

If exam marks have a mean of 60 and an SD of 10, then about 68 percent of students scored 50 to 70, 95 percent scored 40 to 80, and only about 0.3 percent scored below 30 or above 90. In a normal distribution the mean, median and mode are all the same.

Z-scores

A z-score converts a value to "how many SDs above or below the mean":

z = (value - mean) / standard deviation

A score of 80 with mean 60 and SD 10 has z = (80 - 60) / 10 = 2. About 97.7 percent of students scored below that (the 97.7th percentile). A z-score of 0 is exactly average; negative means below.

Uses: comparing different scales (is 80 in maths better than 70 in English? Compare the z-scores using each subject's own mean and SD) and flagging unusual values (an absolute z above 3 is unusual in roughly normal data). It works only if the data is roughly normal and has no huge outliers distorting the mean and SD.

Skewed data

Many business values are not symmetric. They bunch near a minimum and have a long tail of big values:

  • Right-skewed (positive skew): a long tail to the right. Salaries, order values, house prices, delivery times, website session lengths. The mean is greater than the median.
  • Left-skewed (negative skew): a long tail to the left. Age at retirement, marks in an easy exam. The mean is less than the median.

For skewed data, report the median and percentiles, not just the mean and SD. Sometimes a log transform makes a right-skewed variable look more symmetric: np.log1p(df["amount"]). Compare skew with df["amount"].skew() (near 0 is symmetric; above 1 is strongly right-skewed).

Other shapes to recognise

  • Bimodal: two peaks. Often two groups mixed together (laptop and mobile users, freshers and experienced staff). A single mean describes neither group, so split the data and analyse each.
  • Uniform: all values equally likely, such as a random draw of 1 to 6 on a fair die.
  • Long-tailed or power-law: a few items account for most of the total (a few products make most revenue, a few customers make most orders). The "80/20 rule" is this idea. Use cumulative percentage charts (Pareto) to show it.
  • Discrete counts: orders per day or defects per batch (often skewed, with many zeros).
  • Binary: yes or no; summarise with a proportion.

Why shape matters

If the data is... Trust Be careful with
Symmetric, normal mean and SD, z-scores little
Right-skewed median, percentiles, IQR mean and SD, z-scores
Bimodal each group separately any single average
Long-tailed shares of total, percentiles the mean

The same logic applies to dashboards: an "average order value" tile on skewed data tells a manager something that is often not true of a typical order.

Samples and the central limit idea

A useful fact: if you take many random samples and calculate each sample's mean, those sample means are approximately normally distributed, even when the data itself is skewed, as long as samples are reasonably large. This central limit theorem is why averages and confidence intervals work in practice, and why you can compare averages of groups with standard methods even when individual values are skewed. You will use it when you learn about A/B tests and confidence intervals; here it is enough to know the idea.

Common mistakes

  • Assuming data is normal and applying the 68-95-99.7 rule to skewed data.
  • Using z-scores to flag outliers in skewed or small datasets.
  • Reporting one average for a bimodal dataset that is really two groups.
  • Choosing histogram bins carelessly and drawing the wrong conclusion.
  • Ignoring shape and going straight to the mean.

Try it yourself

  1. Draw histograms of an order-value column with 5, 20 and 50 bins and describe how the picture changes.
  2. Compute the mean, median and skew of that column. Is it right-skewed?
  3. Calculate z-scores for exam marks 45, 60 and 85 with mean 60 and SD 10.
  4. Find a column in your own data that is bimodal and split it into two groups.

Check yourself

What does the 68-95-99.7 rule say?

For normal data, about 68 percent of values are within one standard deviation of the mean, 95 percent within two and 99.7 percent within three.

What is a z-score of -1.5?

The value is one and a half standard deviations below the mean.

If the mean is larger than the median, what does that suggest?

The data is right-skewed, with a long tail of large values pulling the mean up.

Why is a single average misleading for bimodal data?

There are two groups with different centres, and the overall average may lie between them where few values actually are.

Interview questions

How would you describe the distribution of order values?

Plot a histogram, then describe centre, spread and shape: typically right-skewed with a long tail of large orders, so I would report the median and percentiles and use the mean cautiously.

What is a z-score and how would you use it?

The number of standard deviations a value is from the mean. It compares values across scales and flags unusual values when data is roughly normal.

What can you do about skewed data before modelling or reporting?

Use medians and percentiles, transform with a log, cap extreme values after checking they are errors, or segment into groups.

Summary

  • A distribution shows how values are spread; a histogram shows its shape.
  • Normal data follows the 68-95-99.7 rule; z-scores measure distance from the mean in SDs.
  • Business data is often right-skewed; use the median and percentiles; split bimodal data.
  • Shape decides which statistics and charts you can trust.

Going deeper

See how much of this you can explain

A free mock interview asks you new questions on Statistics for analysts for Data Analyst and scores every answer, with what to fix.