You don't need a statistics degree to be a great analyst — but you do need the practical stats that stop you drawing wrong conclusions. Here's what actually gets used, explained plainly.
Averages — and which to use
"Average" hides three different measures:
- Mean — the sum divided by the count. Great for symmetric data; distorted by outliers.
- Median — the middle value. Robust to outliers — the right choice for skewed data like salaries or house prices.
- Mode — the most common value. Useful for categories. The classic trap: reporting the mean salary when a few huge earners pull it up — the median is more honest. Always ask which average fits the data.
Spread — how varied is the data?
An average alone can mislead; you need spread:
- Range — max minus min (simple, but outlier-sensitive).
- Standard deviation — the typical distance of values from the mean. Low = clustered; high = spread out. It's the key measure of variability.
- Variance — standard deviation squared (used in calculations). Two datasets can share a mean but behave completely differently — spread tells you how.
Distributions
The shape of your data matters:
- Normal distribution — the symmetric "bell curve"; mean = median, and known percentages fall within 1/2/3 standard deviations (the 68–95–99.7 rule).
- Skewed — a long tail one side (income is right-skewed); use the median. Plot a histogram early — it reveals shape, outliers and data problems instantly.
Percentiles and outliers
- Percentiles/quartiles describe position — the 90th percentile means 90% of values are below it. Great for SLAs ("95% of tickets resolved within…").
- Outliers — unusually extreme values. Decide whether they're errors (fix/remove) or real and important (investigate). Never delete data silently.
Put it to work
For any numeric column, calculate the mean and median, plot a histogram, and note the standard deviation and any outliers — then decide which average to report and why.