Real people, verified

A model is only as good as the people whose answers go into it, and the quality of online samples has become the central problem of survey research. StakeholderWatch invests a large part of its fieldwork budget in keeping respondents real.

How respondents are recruited

Respondents are recruited through established online panel bureaus with which StakeholderWatch has standing agreements. We work with several bureaus rather than one, select them against a verification standard, and review each partner's quality continuously. Where a client's stakeholders cannot be reached through general panels (employees, members, patients), the sample is drawn from the client's own population under an agreement that protects anonymity.

How every complete is checked

Every complete passes through StakeholderWatch's own checks before it enters the data: completion time against the questionnaire's expected duration, straight-lining and other low-effort patterns, internal consistency between related items, duplicate devices and sessions, and the response signatures associated with automated and farmed accounts. A complete that fails is removed and the removal is logged. Familiarity with the organisation is screened before any evaluation is asked.

No synthetic data

StakeholderWatch does not use large language models or any other generative method to create, impute or extend respondent answers, and does not use synthetic sampling to fill quotas or segments. Every record in a client's data is a complete from a person who answered the questionnaire. When a panel partner is found to have supplied synthetic or automated completes, the partner is dropped and the affected completes are removed from the data and from every estimate built on them. What goes into a model is what comes out of it, and we would rather report a smaller sample of real people than a larger one we cannot vouch for.

More on recruitment, screening and data protection on the Panels page.