Ranking questions
How InviziPoll's ranking question works: the pick-in-order control respondents get, partial rankings, the mean-rank leaderboard, and why no single person's ordering is ever shown.
On this page
Results read as a leaderboard ordered by mean rank, not as a bar chart of first places. An option nobody puts first can still be the one everybody puts second, and a chart that counted only first places would show it as a zero.
A ranking question asks people to put options in order: "Put these in the order you would fix them." It answers a question the other types cannot — which of these matters most — without asking a respondent to rate each one in isolation and hope the ratings compare.
Setting one up
Add a question in the poll wizard and pick Ranking. You supply between 2 and 8 items, each up to 60 characters, and the prompt itself up to 200 characters. A new ranking question starts with three placeholder items for you to replace.
Eight is a real ceiling, not a soft one. Beyond about six items most people stop ranking carefully and start ranking to finish, and the upper bound is also what keeps the entropy of a full ordering finite — see below.
What a respondent sees
Two lists. Everything starts in a pool of items that have not been ranked yet, and the respondent moves the ones they care about across into Your order, then moves them up and down.
Every control is an ordinary button in the tab order, working on Enter or Space. There is no drag-and-drop: a drag-only reorder is unusable for keyboard and switch users, and a question every employee is asked to answer is the wrong place for that gap.
Nothing is pre-filled, on purpose. Showing the full list already in your declared order would mean a respondent who touched nothing submitted your order as their judgement, and one who moved a single item submitted a full ordering of items they never considered. Both invent an opinion.
Partial rankings are expected
Because items start unranked, ranking three of six is a normal answer rather than an error, and it is treated as one.
An option's mean is taken over the people who actually ranked it. Giving an unranked option a worst-place score would invent a judgement the respondent declined to make, and the size of that error would depend on how many items they skipped. So the results carry a count alongside every mean: a mean drawn from two people should not read like a mean drawn from two hundred.
How the results read
After the poll clears its anonymity threshold, the panel is a table of options in mean-rank order, best first:
| Column | Meaning |
|---|---|
| Mean rank | Average position, lower is better |
| Ranked by | How many people ranked it at all |
| First choice | How many put it first |
An option nobody ranked shows Not ranked instead of a mean and sorts to the bottom, rather than being given a number it did not earn.
The aggregate CSV carries the same five columns, one row per option. So does the trend snapshot for a recurring series, and so does the PDF export.
No single person's ordering is ever shown
A full ordering of eight items is one of 40,320. A raw ordering can therefore identify someone more precisely than a paragraph of free text, and a minimum-response threshold would be no defence against a single one of them.
What protects it is that no surface produces an ordering at all. Every number you can read is computed per option across every respondent, exactly the way a multiple-choice question's counts are. There is no verbatim panel for rankings, no per-response column in any CSV, and no per-response field in a snapshot or a shared slice. The two per-response exports that do exist — shuffled word cloud answers and open text verbatims — are specific to those types by design.
Branching does not fire on a ranking
A branching rule pointing at a ranking question never fires, and that is the intended behaviour rather than a bug to work around.
Branching matches a single exact answer value. For a ranking, every item the respondent moved across is part of the answer, in an order the rule cannot see, so "the answer includes this option" would be true of nearly everyone and would tell you nothing. A useful rule would need a different predicate — ranked X first — which does not exist. Branch on a single-choice question instead and keep the ranking for the detail.
Progressive disclosure does work normally: a ranking counts as answered once at least one item has been moved into the order, so the next question appears when you would expect it.