A respondent who finishes too fast to have read the questions is a data-quality risk — but a single universal time cutoff is a blunt tool that removes genuine fast readers and misses paced speeders. This is a vendor-neutral guide to calibrated speeder detection: where speed is measured, what sets the threshold, and why timing is a flag to corroborate, not a verdict to act on alone.
CatalystMR Research Team. (2026). Speeder Detection: Catching Speeders Without Discarding Fast Readers. CatalystMR Methodology Papers. https://www.catalystmr.com/insights/methodology-papers/speeder-detection/
@techreport{catalystmr_speeder_detection,
author={{CatalystMR Research Team}},
title={Speeder Detection: Catching Speeders Without Discarding Fast Readers},
institution={CatalystMR}, year={2026}, type={Methodology Paper},
url={https://www.catalystmr.com/insights/methodology-papers/speeder-detection/}
}TY - RPRT AU - CatalystMR Research Team TI - Speeder Detection: Catching Speeders Without Discarding Fast Readers PB - CatalystMR PY - 2026 UR - https://www.catalystmr.com/insights/methodology-papers/speeder-detection/ ER -
Speeder detection identifies respondents who complete a survey too quickly to have read and answered with care. Very short response times are a well-documented indicator of low attention and reduced data quality — which is why timing is one of the most valuable quality checks in online and CATI-supported research. But it is also one of the easiest to get wrong: a blunt, universal time cutoff removes legitimate fast readers and lets calibrated, human-paced speeders through.
This paper sets out calibrated speeder detection. It explains why one threshold is not enough; maps the four timing layers where speed is measured; names the instrument factors that set a sensible threshold; argues that speed is a flag, not a verdict and must be corroborated before removal; and shows how calibration protects the genuine fast reader. It closes with a calibration maturity ladder. It is the timing deep-dive behind one of the five integrity questions of Paper No. 142, and the companion to the straight-lining study of No. 144.
It is tempting to detect speeders with a single rule — "remove anyone who finishes in under X minutes." It is also wrong. A five-minute tracker and a twenty-five-minute conjoint demand completely different standards, and a fixed cutoff applied across them either discards careful-but-quick respondents on the short study or waves through inattentive ones on the long study. The blunt threshold errs in both directions simultaneously.
A literate, motivated respondent on a short, familiar survey can answer accurately and quickly — and be wrongly removed, biasing the surviving sample toward the slow.
An inattentive respondent who keeps just above a fixed cutoff — or paces a long survey carelessly — passes the test while contributing low-quality data.
Response behaviour also varies within a single respondent across a survey — careful on some pages, hurried on others — so even one person rarely has a single "true" speed. Detecting too-fast responses page by page is more defensible than branding a whole respondent on one total-time number.2 The implication is the throughline of this paper: a threshold is something you calibrate, corroborate, and apply with judgement — not a universal number you set once.
"Too fast" is not one number; it is a pattern visible at several levels of a survey. A strong system reads timing at four layers, because a respondent who looks acceptable on total time can still race a critical grid, and a page-level stall can explain an otherwise-fast complete. Each layer catches something the others miss.
Overall length of interview against the modelled time the instrument should take.
The whole surveyTime spent on each page or screen — catching the page someone skipped through.
Per page / screenPace through matrix and grid questions, where careless speed most often hides.
Per gridTime taken on verbatims — a few seconds rarely yields a considered answer.
Per open-endBecause attention drifts within a survey, a page- and question-level view detects the specific too-fast responses a total-time average would hide — and protects a respondent who was simply quick overall but careful where it counted. The layers are read together, weighing where the speed occurred, not just how much there was.2
A defensible threshold is derived, not declared. It reflects how long this survey should genuinely take a careful respondent — which depends on the instrument, the audience, and the channel. The five factors below move the line; together they replace a universal cutoff with a threshold tuned to the study in front of you.
Rather than guess an absolute number, strong practice anchors the threshold to the distribution of the study's own timings — for example, flagging responses a set fraction below the median page or survey time — so the line is calibrated to how this instrument actually behaves, layer by layer, rather than to a number carried over from another study.2
A calibrated timing flag is strong evidence, but rarely conclusive by itself. Unless a completion time is flatly impossible — faster than the survey could be read at all — speed should be corroborated before a complete is removed. The strongest detection treats timing as one signal among several, and acts when they agree.
Speeding and straight-lining in particular tend to travel together — a respondent racing a grid is more likely to flat-line it — so a speed flag that coincides with straight-lining is far more convincing than either alone.1
There is one exception to corroboration: a time that is physically impossible — a multi-minute instrument "completed" in seconds — needs no second signal. Short of that, a lone speed flag should raise scrutiny, not trigger automatic deletion, precisely to avoid discarding the fast-but-careful respondent.
Ask: "Does a speed flag remove a complete on its own, or does something else have to agree?" "Speed alone, automatically" over-removes; "speed plus corroboration" is the defensible rule.
Speeder detection is usually framed as catching bad respondents. The other half — keeping good ones — matters just as much. Every legitimate fast reader wrongly removed is a real, qualified opinion deleted from the sample, and if the removed share leans a particular way, the deletion biases the result. Aggressive timing rules can quietly degrade data in the name of cleaning it.
The same disciplines that catch real speeders are what spare the fast reader: a calibrated threshold sets the floor where careful-but-quick is still plausible; a page-level view credits the respondent who was thorough where it mattered; and corroboration means a quick complete is removed only when other signals confirm carelessness. Detection that protects good data does both jobs at once.
Set thresholds conservatively enough to catch the impossible, but require corroboration before removing the merely fast — replacing borderline cases rather than deleting good ones outright.
Speeder detection matures in steps. The ladder below shows the progression from the weakest approach — a single number for every study — to the strongest, where calibrated, multi-layer timing is corroborated before anyone is removed. Each rung up reduces both errors: fewer real speeders missed, fewer fast readers lost.
Speeder detection is one of the most valuable data-quality checks in survey research and one of the easiest to do badly. A single universal cutoff fails in both directions — discarding the fast-but-careful reader while missing the paced speeder. The durable approach measures speed across the four timing layers, calibrates the threshold to the instrument, audience, and device, treats a timing flag as evidence to corroborate rather than a verdict to enforce, and removes the merely fast only when another signal agrees. Detection done this way protects the finding from careless completes and from over-zealous cleaning — catching speeders without discarding fast readers. Recognised conduct and service-quality frameworks let buyers ask for that rigour in consistent terms.3,4
CatalystMR is a global market-research panel and fieldwork partner specialising in hard-to-reach B2B, healthcare, and niche audiences. We review response timing at the survey and question level as part of a broader quality framework, calibrating thresholds to each instrument and corroborating speed flags before a complete is removed — protecting clients from fast, unreliable data without over-removing legitimate fast readers.
Borderline completes are replaced rather than silently deleted, so a study keeps both its quality and its planned base.
Compliance posture: our methodology is aligned to the ESOMAR Code and Guidelines and the ISO 20252 framework, and we are certified under the EU–U.S., UK, and Swiss Data Privacy Frameworks, with personal data siloed from response data.