The Skill Explained
What Is Statistics?
Statistics is the science of learning from data under uncertainty: describing data, estimating quantities and deciding whether a difference is real or just chance. It is the mathematical foundation under data science and machine learning.
Why AI Employers Ask for It
Every data-driven decision rests on statistics: is this model really better, did the new feature change behaviour, how confident can we be in this estimate. Data scientists and research roles are expected to reason with statistics so their conclusions survive scrutiny, and in regulated fields such as pharma they must.
How It Works
Descriptive statistics summarise data: averages, medians, spread (standard deviation), distributions and correlations. They show what the data looks like before any model.
Inferential statistics go from a sample to a conclusion about a larger population. Confidence intervals give a range for an estimate. Hypothesis tests ask how surprising the data would be if there were no real effect; the p-value measures that surprise. Statistical power is the chance of detecting an effect that exists, and it depends on sample size.
Regression models relate an outcome to several factors at once. Bayesian statistics updates beliefs with data and expresses results as probabilities. Across all of it, the key skill is judgment: knowing the assumptions behind a method, spotting bias in how data was collected and never confusing correlation with causation.
What You Do With It at Work
- Explore and summarise a dataset before modelling.
- Test whether a difference between groups is statistically significant.
- Put confidence intervals around model metrics.
- Design studies with enough data (power analysis).
- Explain uncertainty clearly to people who make decisions.
Tools and Terms You Will Meet
- p-value
- How likely data at least this extreme would be if there were no real effect.
- Confidence interval
- A range that, with a chosen confidence, contains the true value.
- Statistical power
- The probability of detecting a real effect of a given size.
- Regression
- A model of how an outcome changes with several factors.
- Bias
- A systematic error, often from how data was collected.
Industry Example: Pharma
In a clinical trial, statisticians decide in advance how many patients are needed and how the primary endpoint will be tested, so the trial can show a treatment effect convincingly to regulators.
A Common Mix-Up
A p-value is not the probability that a result is true, and "statistically significant" does not mean "important". A tiny effect can be significant with enough data.
How Deep Do You Need to Go?
- Know itDescribe data with averages, spread and distributions.
- Use itRun hypothesis tests, build confidence intervals and fit regressions correctly.
- Master itDesign studies, choose methods for complex data and explain uncertainty to decision makers.