Understanding Bases
Overview
The base of a cell answers a simple question: how many respondents is this figure computed on? It is the denominator behind every percentage and the weight behind every mean. Because the visual re-aggregates data across a column hierarchy and across rows, getting the base right requires understanding a few subtleties. This page explains them and gives a decision guide for every layout.
Percentages and means are ratios — they stay correct even if the base is scaled. The base row, on the other hand, is an absolute count, so it is the value most sensitive to how rows and columns are aggregated. If your means look right but the base row looks wrong, this page is for you.
The two base rows
The visual can display two different base rows (see Totals & Subtotals to switch them on):
| Row | Reads the series | Meaning |
|---|---|---|
| Base | the base / count series (weighted count in a mean table, count in a % table) | The weighted base used as the denominator / weight. |
| Unweighted Base | the unweighted base series | The raw number of respondents (effectifs), unweighted. |
Both are configured in Percentage Series or Mean Series. They are aggregated the same way — only the source series differs. If the two disagree in a way that is not explained by weighting, an aggregation setting is wrong.
When significance testing is active, every cell is also tested against something — a complement, a total, a reference column. The Cell Tooltip can show that third figure, the compared base, without adding a row to the table.
How a base is aggregated
A base value shown on a total or subtotal is built by combining leaf cells along two independent axes.
1. Across columns (the hierarchy)
For a nested column hierarchy such as Direction › Sous-Direction, the base of a parent column (the subtotal) is the aggregate of its child columns. This is always correct: a respondent belongs to exactly one child, so the children partition the parent.
2. Across rows
This is the subtle axis. The base of a column is also combined across the table's rows, and here the correct operation depends on what the rows are:
- Rows are disjoint respondent groups (e.g. Age, Gender, Region). Each respondent appears in exactly one row → the bases add up. Summing across rows is correct.
- Rows are repeated items rated by the same respondents (e.g. a battery of 60 satisfaction items, each scored on a scale). Each respondent appears in every row → summing would count them once per item, inflating the base by the number of rows.
If a base (or subtotal) is exactly your correct value × the number of rows, the base is being summed across repeated items. Example: a Direction subtotal showing 5 760 when the detail sums to 120, with 48 item rows → 120 × 48 = 5 760. The means are still correct because the inflation cancels in the ratio.
The Rows are repeated items toggle
Because the two situations above are indistinguishable from the data model alone, a single switch controls the row axis:
Setting: Rows are repeated items (don't sum base)
Location: Table contents card
Type: Toggle
Default: Off
- On — rows are treated as repeated items answered by the same respondents. The base is not summed across rows; the per-item base is used for column totals and subtotals. Use this for rating batteries, note-by-note tables — any case where every respondent appears in every row.
- Off (default) — rows are treated as disjoint respondent groups. The base is summed across rows. This is the most frequent case (demographics, or any variable where each respondent belongs to a single row).
This toggle only affects the base figures (both the weighted Base and the Unweighted Base, at leaves, subtotals and the grand total). Percentages and means are unaffected.
Quick decision guide
| Your rows are… | Example | Toggle |
|---|---|---|
| Rated items / measures answered by the same people | 60 satisfaction items, each on a 0–10 scale | On |
| Disjoint respondent groups | Age brackets, Gender, Region, Seniority | Off |
| A single row (one measure) | one KPI over the columns | Either (no row aggregation happens) |
Layout matters: Hierarchical vs Side-by-Side columns
The column layout changes how the base is presented, but the row rule above is the same in both.
Hierarchical columns (nested)
Column variables are nested (e.g. Direction › Sous-Direction). You get:
- a base for each leaf (Sous-Direction),
- a base for each subtotal (Direction), aggregated from its leaves,
- a grand-total base.
With Rows are repeated items On, every level shows the respondent base; with it Off, every level is summed across rows.
Side-by-Side columns
When Side by Side Columns is enabled, each column variable is flattened into its own independent group placed next to the others. Each group shows the marginal base of that variable's modalities.
- For a single-response variable, the modality bases within a group add up to the same overall total across every group.
- Means are re-aggregated from the leaf values (weighted by the base) for each flattened modality — a transversal modality's mean is the base-weighted average of that modality across all parents.
The Rows are repeated items toggle applies identically here: it decides whether the base shown per side-by-side column is summed across rows or not.
Percentages vs Means
The base concept is the same for both table types; only which series feeds it differs.
| Percentage table | Mean table | |
|---|---|---|
| Displayed figure | a % (share of a base) | a weighted average |
| Weighted base comes from | the base series | the count series (also the mean's weight) |
| Unweighted base comes from | the unweighted base series | the unweighted base series |
| Subtotal value | re-summed / re-based from children | weighted mean re-aggregated from children (native Power BI subtotals used when available) |
In a mean table, the base doubles as the weight: a subtotal mean is Σ(meanᵢ × baseᵢ) / Σ(baseᵢ) over its children. This is why an inflated base never corrupts the mean — the same factor appears in the numerator and denominator.
Troubleshooting checklist
- A subtotal base is a large multiple of the detail → rows are repeated items; make sure
Rows are repeated itemsis On. - A demographic table shows a base that is too small (looks like one group) → rows are disjoint groups; turn
Rows are repeated itemsOff. - Base and Unweighted Base differ by more than weighting → check that both series are mapped correctly in Percentage Series / Mean Series, and that the unweighted base series is the raw respondent count.
- Means are correct but bases are not → this is expected when the row axis is misconfigured; fix it with the toggle — the means will not change.