Constant Sum Analysis.

A constant sum question asks respondents to spread numbers across a set of items — points, dollars, visits, percentages. It comes in two flavours: Constant Sum: Choice (allocate across one list of choices) and Constant Sum: Matrix (allocate across choices within each statement). The example throughout is a small restaurants study — small on purpose (n=10), so you can trace every calculation:

• "Q2. In the last 6 months, roughly how many times did you dine in or order from each of the following restaurants?" — a number per restaurant (Choice).
• "Q3. In the last 6 months, when have you eaten from each of the following restaurants?" — those visits split across meal occasions, per restaurant (Matrix).

(New to Analyse? Start with the Analysis Overview.)

How a constant sum answer works

Every cell of a respondent’s answer is one of three things, and the distinction drives every number below:

• A number — they allocated something to the item.
• A zero — they saw the item and gave it nothing.
• Blank — they never saw or never answered the item. Blanks are never counted, anywhere: blank ≠ zero.

Sum % — the default: share of the total

New constant sum tables open on Sum %: each item’s slice of all the points allocated. Our 10 respondents reported 129 restaurant visits in total, and Restaurant ABC took 37 of them — a 28.7% share:

The column adds to 100% — it’s a share-of-wallet read. And like Cell %, Sum % is orientation-proof: transpose the table (set the Choices slot to the other axis — see Table Slots) and every number stays put.

The Values menu

Constant sum gets the fullest Values menu in Analyse:

The first five work on the numbers respondents allocated; the familiar percentage family at the bottom (Cell %, Row %, Column %, Total %) works on people — e.g. Cell % shows 80% for Restaurant ABC because 8 of the 10 respondents allocated it anything. For those, see Values: Count, %, Cell %, Sum & Average. The number-based five:

• Sum — the raw total: 37 visits to ABC.
• Sum % — that total as a share of all points (above).
• Count — how many people allocated the item anything: 8 for ABC (this is also each row’s Total (n)).
• Average — Sum ÷ the people who gave it a non-zero answer.
• Average (incl. 0) — Sum ÷ everyone who was asked, zeros included.

The two Averages — and when each one is right

Average answers "how much, among the people who use it at all?" ABC’s 37 visits came from 8 diners — 4.6 visits each:

Average (incl. 0) answers "how much, across everyone?" The same 37 visits spread over all 10 people asked is 3.7:

Both are true; they answer different questions — and they can rank your items differently. Restaurant G has the most devoted customers (its 6 diners average 5.0 visits, the highest loyalty in the set), but ABC leads the market overall (3.7 visits per person vs G’s 3.0). Use Average for intensity among users; Average (incl. 0) for market-level per-head comparisons.

The Matrix flavour

Constant Sum: Matrix opens with Statements in the rows and Choices in the columns, and Sum % normalises within each statement — each restaurant’s row splits its own visits to 100% across the occasions:

Read across a row: ABC’s visits are 32% lunch, 27% breakfast, 24% dinner. Every Value works per cell the same way as above — ABC’s breakfast average is 2.0 among the 5 people who allocated breakfast visits, or 1.3 with zeros included. And the dimension slots work exactly as on any matrix — collapse the Statements slot to one restaurant (or the Choices slot to one occasion) and the freed axis goes to your filters; see Matrix Analysis.

Cut by audience

Cross filters and prefilters work as everywhere else: the + at the end of the table’s chips puts segments side by side (each showing its own sums, shares and averages), and + Prefilter above the chart narrows the whole view — see Prefilters & Cross Filters. With small constant sum bases, keep an eye on each row’s Total (n) as you cut.

FAQs

Why is Average higher than Average (incl. 0)?

Average divides by only the people who gave the item a non-zero answer; Average (incl. 0) divides by everyone who was asked. Unless nobody answered zero, the incl.-0 version always reads lower.

Which average should I report?

Ask what the number is for. Loyalty, intensity, "how much do users use it" → Average. Market sizing, per-capita comparisons, "how much does the average person account for" → Average (incl. 0).

How are people who never saw an item treated?

As blanks — they’re excluded from every calculation, including Average (incl. 0). Only respondents who were actually asked the item can count in its base.

Do Cell %, Row % and Column % measure points here?

No — those count people (anyone with a non-zero answer), just as on other question types. Sum % is the one that measures the points themselves.

Why do my tables show one decimal place?

That’s the dashboard’s Decimal Places setting (dashboard Settings ⚙, applies to every view) — worth raising above 0dp for constant sum, where averages like 4.6 vs 3.7 are the story.