Data Analyst · Step 5 of 8 · 11 min
The statistics you actually need
You can pick the right average, describe change correctly, and avoid the most common false conclusions.
Mean, median and spread
The mean is the total divided by the count. The median is the middle value when sorted. When a few values are very large, like a handful of huge orders or high incomes, the mean gets pulled up and the median better describes a typical case.
Always report spread too. Two stores with the same average daily sales can be very different if one is steady and the other swings wildly. The range, percentiles (like the 25th and 75th) or the standard deviation all describe spread.
Percent change versus percentage points
If conversion goes from 2% to 3%, that is a rise of 1 percentage point, and a 50% increase. Both are true; they mean different things. Say which one you mean, and show the starting value.
Small bases exaggerate change. Going from 2 complaints to 4 is “up 100%”, but it is still two complaints.
Correlation is not causation
Ice cream sales and sunburn rise together because of hot weather, not because one causes the other. When two numbers move together, look for a third factor, for reverse cause, or for coincidence.
Be honest in wording: “customers who used the app bought more” is a finding; “the app made customers buy more” needs stronger evidence, usually an experiment.
Sample size and simple A/B tests
Results from 20 people bounce around far more than results from 20,000. Before celebrating a difference, ask how many observations sit behind each number.
An A/B test randomly splits people into two groups that see different versions, then compares an outcome. Randomness is what lets you talk about cause. Decide the metric and the sample size before starting, run it for the planned time, and do not stop early the moment it looks good. Free online calculators help estimate sample size and significance.
Try it
Catch the misleading claim
- 1.Use the prompt below to get five short business claims with hidden statistical problems.
- 2.For each, write what is wrong before reading the answers.
- 3.Check your reasoning against the explanations.
Act as a statistics tutor for a beginner data analyst. Give me five short, realistic business claims, each with one hidden problem: mean vs median, percent vs percentage points, correlation vs causation, tiny sample size, or stopping an A/B test early. Do not reveal the problems until I reply with my answers.