Fraud detection
Fraud detection in research is the set of checks that identify participants who should not be in the data: bots, people who lie to qualify, duplicate accounts and respondents who answer without reading, mostly to collect incentives.
How it works
Fraud concentrates in online opt-in samples, where anyone can sign up for many studies. Pew Research Center found that bogus respondents in such polls can shift results, and that two common checks, one for speeding and one for attention, failed to catch most of them. Detection therefore layers methods: screening questions with consistency checks, device and location checks, identity verification, duplicate detection, open-ended answers read for sense, and review of response patterns.
Fraud harms qualitative studies too, because one invented participant can supply a vivid quote that shapes a finding. Checks at recruitment, before the interview, prevent the most damage.
A worked example
Purchase intent for a new snack reads 64 percent in a 1,000-person survey from an opt-in panel. Layered checks then remove 140 respondents: 60 fail a device check for duplicate accounts, 50 give contradictory screener answers and 30 write nonsense in open-ended fields. Of those 140, 130 had said they would buy, so intent among the 860 who remain is 510 of 860, or 59 percent.
Related
Automate this in Pulse
Fraud detection in research screens out respondents who are not who they claim to be, and every participant Pulse recruits goes through fraud checks that protect the findings from bogus respondents.