Participatory research, multi-round

Read this before you plan the study

The anonymity model cannot produce per-participant data. There is no participant identifier in the record, so no contribution can be linked to any other contribution by the same person, in the same round or across rounds.

That rules out, permanently and by design:

• Longitudinal analysis at the individual level. You cannot follow how one participant's view changed between round one and round three
• Subgroup analysis. You cannot break findings out by age, role, tenure, or any other attribute, because no attribute is attached to a contribution
• Any statistic that needs a denominator of people rather than of contributions
• Matched pre and post comparison

What you get instead is a group-level record: what was said, how the group organized it, and what the group did with it, across rounds. For a lot of participatory work that is the right unit of analysis. For a study designed around individual trajectories it is the wrong tool, and the honest move is to say so in the first conversation rather than at analysis time.

When to use it

A research team, a research office, or a community-based participatory research partnership needs community input that is genuinely community-shaped rather than researcher-shaped. The specific strength here is that the participants name the categories. In a standard qualitative workflow the researcher codes the transcript, and the coding frame is the researcher's. Here the grouping is drafted by the software, corrected by the participants in the room, and the record keeps both versions, so the analysis can show where the group overruled the machine.

The multi-round shape matters more than the single session. Round one surfaces what is there. The group chooses the direction for round two. Session memory carries the earlier patterns forward, and the question record shows why each round asked what it asked. That chain is the methodological contribution, and it is the thing a reviewer will want to see.

Do not use it when the design needs individual-level data, when the population is small enough that a contribution could identify its author, or when the research question requires a probability sample and generalizable prevalence estimates. Also do not use it as a substitute for interviews when what you need is depth from a few people; this instrument is built for many voices at once and shallow-but-wide is what it produces.

The question

Each round gets its own question. The first one sets the frame, so it should be broad enough that the group can take it somewhere the research team did not anticipate.

Seeds for round one:

• What does [topic] look like from where you sit that it might not look like from outside?
• What do people get wrong about this, in your experience?
• What has been tried here before, and what happened?
• What would have to be true for this to work in your community?

The question that fails: any question written from the literature review. If the spark contains the framework the team already intends to test, the rounds will confirm it, and the participatory claim in the resulting paper will not be defensible.

Session settings

• Participants: cap at expected headcount plus ten, per round
• Anonymous participation: on
• Raffle: off. It defaults on. In a study context, collecting names and contact details alongside contributions undermines the anonymity claim you are about to make to an ethics board
• Blind gather: on
• Blind then reveal voting: on
• Shuffle order: on
• Access after finish: on, so participants can see what the group produced, which is a participatory obligation and not just a courtesy

Run of show

One round. Repeat with a new question per round.

Ends at: Evolve, every round, because the direction the group chooses is what the next round is built from. Resolve belongs to the study's final session, where the group endorses or reserves on what the rounds produced.

Rounds are advanced by the facilitator. There is no automated campaign mode that opens round two on a schedule, so build the between-rounds interval into the study timeline as facilitator work.

What the record actually contains

Export from the session overview, as CSV or JSON. Precisely this, and nothing more:

• Contributions. The text, the pattern it was assigned to, the question it answered, and a timestamp. Contributions that were assigned to no pattern are marked as such and are included
• Patterns. The name the group settled on, the AI's draft name beside it, a flag for whether the group changed it, the number of contributions in it, up, down and starred totals, comment count, key themes, and the narrative summary
• Comments. The text, the pattern it was on, and a like count
• Questions. The question text for each round

Rounds are told apart by their question, which appears on every contribution row. There is no round-number column.

Not in the export: any identity, any per-participant identifier, any link between two contributions by the same person, any demographic attribute, and any vote attributable to a voter. Votes and comments are per-pattern aggregates.

The round-to-round genealogy, meaning which direction the group chose and the facilitator's stated rationale for each round's question, lives in the synthesis report and its PDF rather than in the CSV or JSON. If that chain is part of your methods section, plan to take it from the report.

Moderated content is excluded from the export. If a contribution was hidden during the session, it is absent from the file and it is not counted in the pattern totals. That is deliberate, and it is worth knowing when a count in the export differs from a count you wrote down during the session.

Ethics and consent

Written for the application, not for marketing. Confirm each point against the current product before you submit, and do not copy claims forward from an old application.

• Processing location. Data is stored and processed on infrastructure in the United States. There is no Canadian or European data residency option today. If your institution requires domestic processing, this instrument does not meet that requirement and no configuration changes it
• AI processing. Contributions are sent to a commercial AI provider in the United States for the grouping step. Any text a participant writes goes to that provider
• Anonymity. Contributions are not linkable to an identity from what is stored. Participants are not asked for a name or an email to take part
• Withdrawal. A participant can delete their own contributions from the device they used to take part. If they clear that browser's storage or switch devices, that route is gone, because the credential that proves the contributions were theirs is held on the device and nowhere else. Say this in the consent script; a withdrawal right you cannot honour is worse than one you scoped honestly
• What you cannot claim. This platform has not been through an ethics review. Do not present it as approved, certified, or validated by any body. It is a tool; the approval is yours to obtain

What the group leaves with

Per round: the patterns they named, what held up under scrutiny, what they said was missing, and the direction they chose next. Across the study: a record of how the community's own framing developed, which is usually more interesting to participants than the findings are.

Give the findings back between rounds. A participatory study that only reports at the end has borrowed the community's time without returning anything, and round three attendance will show it.

The failure mode

The researcher expected data the anonymity model cannot produce. It surfaces at analysis, when someone asks to break the findings out by subgroup or to track individual change across rounds, and the answer is that the data does not exist and cannot be reconstructed. At that point the study design is already built on it, the participants' time is spent, and the relationship with the partner organization is the thing that breaks.

The prevention is entirely in the first conversation, before the design is fixed. Say what the record contains and what it does not, in the words above, and get agreement that group-level analysis answers the research question. That is why the section at the top of this guide comes before everything else, and why the honest recommendation is sometimes a different instrument.