Statistics for Data Science: What Beginners Need to Know
Understand variation, sampling and uncertainty through small examples.
Part of the Data Science track: modules, projects and first roles →
By the GullyMentor team · 5 min read · Published 2026-10-18 · Updated 2026-11-06 · Sources listed at the end. Not a promise about mentorship outcomes.
Statistics for data science means learning to tell a pattern in your data from a conclusion about the world. Start with distributions, sampling and uncertainty. Learn named tests only after you can explain what a mean, a median and a spread are hiding or showing.
What should you learn first?
Learn in this order: describe the data, question the sample, separate association from causation, then reason about uncertainty. Each step has a small exercise below. The full learning path sits in the data science track, and the data science roadmap shows where statistics fits among Python, SQL and modelling.
How do you describe a distribution?
Consider fictional delivery times of 20, 22, 23, 25 and 90 minutes. The mean is 36 minutes. The median is 23. Both are correct, but they tell different stories.
The mean is pulled up by one long delay. The median describes the middle observation. Neither number shows that one delivery took 90 minutes. Look at the full distribution and the tail rather than choosing whichever number makes performance look better.
A simple routine for any numeric column: - Count the rows and the missing values. - Report the minimum, median, mean and maximum. - Plot a histogram or a sorted list of values. - Ask whether the mean and median are far apart, and why.
How do you question a sample?
Ask how each observation entered the dataset. A survey answered only by enthusiastic participants may not represent everyone. Adding more respondents of the same kind does not remove selection bias.
Write down the population, the sampling method and the groups that are missing. If you cannot establish representativeness, narrow your claim. "Students who replied to this form" is an honest population. "All students" usually is not.
How do you separate association from causation?
Suppose students attending optional practice sessions score higher. They may have been more motivated or better prepared before attending. The association alone does not identify the effect of the sessions.
A suitable experiment or causal design needs extra assumptions and evidence. For beginner work, state the observed relationship and list plausible alternative explanations. Then name the data that would help tell them apart, such as earlier scores for the same students.
How do you reason about uncertainty?
An estimate from a small sample can change a lot from one sample to the next. Confidence intervals and other uncertainty summaries depend on the method and its assumptions. A narrow interval does not repair biased data.
A p-value is not the probability that a hypothesis is true. Do not treat a threshold as automatic proof of practical importance. The size of an effect and the cost of a wrong decision matter too.
What are the common mistakes?
- Reporting only the mean for skewed data.
- Comparing groups of very different sizes without saying so.
- Treating a missing answer as a neutral answer.
- Reading "significant" as "important".
- Claiming a cause from an observational comparison.
How do you check you got it?
Try this practice task with synthetic data. Generate two small groups with overlapping values, for example 30 values each from slightly different ranges. Compare mean, median and spread. Add one extreme observation to one group and repeat. Write down what changed and why.
You have understood the idea if you can predict, before running the code, which summaries will move most when you add the extreme value. The mean and the spread move a lot. The median barely changes.
Then write a paragraph that states your finding without overclaiming. Include the sample size, how the data was collected and what extra data would help. Ask a friend to find one sentence that claims too much, and fix it.
Worked example: is the delivery service getting slower?
Last week's times were 20, 22, 23, 25 and 90 minutes. This week's are 24, 26, 27, 28 and 30. The mean falls from 36 to about 27, so it looks like a big improvement. The median moves from 23 to 27, so the typical delivery got slightly slower. The honest summary is that the long delay disappeared, but the typical wait did not improve. With only five orders each, say that the sample is too small to be sure.
What should you build next?
Apply these habits to a real investigation. The exploratory data analysis workflow walks through a full notebook. The data science project ideas guide gives three projects built around sampling and uncertainty. This foundation also helps when you later evaluate machine learning models, because a model score is an estimate from a sample too.
Sources & further reading
Examples and practice recommendations are educational illustrations. Tool documentation and programme inclusions should be checked for your intended use.