Do these groups really differ?
Averages by category always differ a little. The useful question is whether the category explains a meaningful share of the variation, or whether the gaps are what chance would produce anyway.
The example compares unit price across product categories in illustrative order data bundled with the tool.
1. Start with the shape of the split.
In Breakdown, pick a category (Destination, Channel, Machine…) and a number. Bars show how the number divides across the category’s values. For totals the share of the whole is printed beside each bar; for averages a thin line marks the overall average so you can see who sits above or below it.
The sentence above the bars says who leads and what the top three bring, so a long tail does not hide the point.
2. A callout answers it for you.
Below the bars, KPI Master tests whether the groups differ more than chance would explain and tells you either “Category makes a real difference” with the share of the variation it accounts for, or “Little difference between groups”. That share is η² (eta squared): the fraction of the number’s ups and downs explained by which group a row belongs to. As a rough guide, 2% is the smallest the tool reports, about 6% is medium and 14% or more is large.
A difference is reported only when both the share is meaningful and the test is clear (p below 0.01). A large dataset can make a tiny gap statistically clear, which is why the share matters as much as the test.
3. Limits worth knowing.
- Rows should be independent. Repeated readings of the same customer, machine or day make a difference look more certain than it is.
- Groups with very few rows are unreliable. A single reading in a group tells you nothing about the group.
- The picture shows the nine largest groups and folds the rest into “Other”.
- A difference between groups does not say why. Check what else varies with the category (region and channel often overlap).
4. Two ways to ask.
Use Breakdown to cycle through categories and numbers with one tap, or Connections → Compare any two columns to pick a category and a number directly: you get the bars, the share of variation it explains and a plain “clear difference / no clear difference”. Findings such as “Channel matters most for revenue” on the Overview open into the same facts.