Significance
Every series is compared to one reference series, axis by axis. The question the card answers is "does this profile sit significantly above or below the reference on this criterion?".
The card runs up to three independent tests. A General group holds the settings they share and switches each test on; each test then has its own group.
General
| Property | Description | Default |
|---|---|---|
| Activate significance tests | Master toggle | Off |
| Significance Level | 90 %, 95 % or 99 % | 95 % |
| Test On | Percentages or Means | Percentages |
| Variance Method | Pooled or Separate (percentages only) | Pooled |
| Test 1 / Test 2 / Test 3 | None, vs. Reference series or External reference | Test 1: vs. Reference series — Tests 2 and 3: None |
| Icon Size (px) | Size of the icons and texts | 12 |
Test 1, Test 2, Test 3
| Property | Description | Default |
|---|---|---|
| Reference series (regex) | vs. Reference series — picks the reference by its legend name | (empty) |
| Reference measure (regex) / Apply to measure (regex) | External reference — see External reference | (empty) |
| Match status | Read-only: what was resolved, or why nothing was | — |
| Display mode | Icon, Marker, Label background or Text | Icon |
| Positive / Negative color | Colors of the two directions | Green / red |
| Custom icon positive / negative | Base64 images replacing the default arrows — Icon mode | (empty) |
| Positive / Negative text | Texts written for the two directions — Text mode | + / - |
Test 1 keeps the settings of the former one-test card, so a report built before the three tests keeps its test unchanged.
What is compared
Only series against series, on the same axis. Two axes of a radar carry their own unit and their own scale, so a comparison across spokes would be meaningless — the card does not offer one.
The Percentages test is a two-proportion test. Proportions must be expressed on 0..1,
while Power BI measures usually carry them on 0..100: when either of the two compared
values exceeds 1, the pair is divided by 100 before the test. No manual conversion is
needed, and a measure already on 0..1 is left alone.
The Means test keeps the raw values and uses the Standard Deviation field. It has no variance method to choose, so that dropdown disappears.
Requirements
The Base field must be filled with the sample size behind each value. Without it the
vs. Reference series tests are disabled and Match status says so (an external reference
carrying its own ttest or signif needs no base). A vertex whose base is missing or zero on
either side of the comparison is left untested, as is one where either series has no value
on that axis.
Reference series
The reference is chosen by a case-insensitive regular expression on the legend name — for
example Total or ^Market$. The first matching series wins, and it is itself excluded
from the decoration: the reference is never marked against itself.
Match status reports the resolved series, an empty match, or an invalid expression, so a typo in the pattern never fails silently.
External reference
A test set to External reference compares each vertex with a measure of the Significance
reference role, evaluated by Power BI for that same vertex — last year, a benchmark, another scope —
without that data having to be in the chart. The measure returns a small JSON that tells the visual
either what to test ({"val", "n"}, {"val", "n", "stddev"}) or the result to display
({"ttest"}, {"signif"}). See External reference.
Display modes
| Mode | Result |
|---|---|
| Icon | A ▲ or ▼ just beyond the vertex, clear of both the marker and the data label, replaceable by your own base64 images |
| Marker | A dot in the color of the direction is drawn over the vertex, covering the series marker. With Show Markers off it becomes a dot on the significant vertices only |
| Label background | The data label of the vertex gets a colored background |
| Text | A short text (+ / - by default, or any letter or emoji) in the color of the direction, where the icon would be |
When several tests decorate the same vertex with icons or texts, they queue up outwards, beyond the vertex.
Label background needs the Data Labels card to be on — it colors the label rather than drawing anything of its own, and it replaces the background configured there on the vertices that came out significant.