Attention checks are a blunt instrument
They catch the inattentive and the careless in the same net, and the people they remove most reliably are the ones you paid most to recruit.
Outline. Each heading below carries one line saying what the section has to establish. Expand each into two to four paragraphs, put at least one number or worked example in the piece that nobody else could publish, then delete this note.
A trap question measures whether somebody read one sentence. It gets used as though it measures whether somebody is honest, and those are different claims about a person.
What a trap question measures, precisely
One sentence, once. State the claim it can support and the claim it cannot.
Failure is not evenly distributed, and the skew is expensive
Mobile, senior, late-evening respondents fail more. Those are the recruits you paid for.
A fourth grid is an instrument problem, not a respondent problem
Straight-lining as a statement about questionnaire length.
Three checks that separate intent from tiredness
Internal consistency across time, response-pattern entropy, and the profile-history contradiction check.
Removing the response without removing the person
The two-class rule and what each class costs. Reference the fraud and inattention piece rather than repeating it.
Measuring your own false positives
Without an appeal route there is no reversal rate, and without a reversal rate there is no evidence the screening works.
-
Fraud and inattention are not the same problem
One is somebody trying to take money from you. The other is somebody answering your survey on a phone at a bus stop. Treating them alike costs you the second one.
5 min read -
Device signals: what they prove and what they only suggest
The same signal that catches an operator running forty accounts also flags a family sharing one laptop.
1 min read -
Consent that survives an audit
A consent record has to answer four questions years later. Most consent flows can answer two of them.
1 min read
