Survey completion time has been used as a quality indicator for a long time. The standard implementation is a single threshold: responses completed faster than some fraction of median completion time get flagged or excluded. That approach captures the most obvious speeder cases, but it misses most of the interesting quality variation that timing data actually carries.
The more informative signal is not total completion time but the distribution of time between individual items: inter-item response times. The pattern of those intervals reveals different quality failure modes than a single total-time check does, and it does so in ways that are more interpretable and more defensible.
Why Total Completion Time Is a Blunt Instrument
A single completion-time threshold has one useful property: it is easy to implement and easy to explain. The problems emerge when you examine what it misses and what it incorrectly flags.
It misses the respondent who reads carefully, straightlines all items in under two minutes, but takes enough total time to fall above the speeder threshold. It misses the respondent who pauses the survey, leaves the browser tab open for 45 minutes, then returns and clicks through the remaining items in 30 seconds. Total time is high; actual engagement on the return visit is zero.
It also incorrectly flags fast legitimate respondents. Survey panel professionals, people who take dozens of surveys per month, and respondents answering in their primary language about a familiar topic can legitimately complete instruments much faster than naive respondents completing the same instrument. Excluding them based on total time removes valid data while keeping inattentive respondents who happened to take longer.
Inter-item timing disaggregates the completion into its components. It asks not how long the whole thing took but how long each individual response decision took, and what the pattern of those decisions looks like across the instrument.
What Normal Inter-Item Timing Looks Like
Attentive respondents show inter-item timing distributions that follow a few recognizable patterns. Items that are cognitively simple, familiar, or follow naturally from the previous item get faster responses. Items that are complex, involve genuine deliberation, or introduce a new construct get slower responses. This creates systematic within-survey variation that is instrumentally predictable.
A 20-item Likert battery on brand perceptions, administered in Arabic to urban Egyptian respondents, should show faster item-level times for recognition items ("Have you heard of this brand?") and somewhat slower times for evaluative items ("How well does this brand understand your needs?") where the respondent is doing actual cognitive work. The within-wave distribution of item times is not flat.
Inattentive respondents collapse this variation. Inter-item times become uniform, often in the 0.5-to-1.5-second range, because the respondent is clicking through without engaging item content. The regularity itself is diagnostic: human deliberation has variance; mechanical clicking does not.
The specific floor varies by language and item type. Arabic reading speed norms run slower than English norms on average, and individual item complexity compounds this. A threshold calibrated on English-language panel norms will over-flag Arabic-language completions if applied directly. The calculation needs to be grounded in the language and item characteristics of the specific instrument.
The Pause-and-Return Artifact
One of the harder timing artifacts to interpret is the long pause within a survey. A respondent who takes 8 minutes on one item before continuing may have been thinking carefully. Or they may have stopped to do something else and returned without repositioning their attention. Inter-item timing alone cannot distinguish these cases.
What distinguishes them is what comes after the pause. A respondent who pauses, returns with genuine attention, and then completes subsequent items at a thoughtful pace produces a different pattern than one who pauses and then clicks through everything in the next 90 seconds. The post-pause timing tail is informative in a way that the pause duration alone is not.
This is where treating timing as a single number breaks down. The quality-relevant information is in the shape of the inter-item time series, not in any single duration. A quality system that only captures total time or minimum inter-item time is discarding the most useful part of the signal.
Timing Across Language Contexts: Why Normalization Matters
A practical challenge in MENA and Sub-Saharan Africa fieldwork is that inter-item timing norms vary substantially across languages and scripts. RTL languages like Arabic and Farsi have different average reading-rate distributions than LTR languages. Scripts with greater visual complexity at typical screen rendering sizes may take slightly longer for comfortable reading. Translated instruments, where the translation is a close rendering of an English original rather than a natural expression, sometimes produce slower reading times because the phrasing is less idiomatic.
Applying timing thresholds calibrated on one language to surveys deployed in another is a methodological error that agencies commonly make because most QC tooling was built against English-language panel populations. The operational fix is per-instrument, per-language threshold calibration rather than universal defaults.
Minimum plausible item completion time can be estimated from word count, item complexity, and language-appropriate reading speed distributions. For a 15-word Arabic item on a five-point scale, an attentive minimum might be around 4 seconds. For a 6-word English item on a binary scale, it might be under 2 seconds. These are not the same threshold, and treating them as interchangeable produces different error rates in different deployments.
Combining Timing with Pattern Signals
Timing signals are most useful when combined with response pattern signals rather than applied in isolation. The combination reduces both false positives and false negatives.
A fast completion with high variance across items is not the same quality profile as a fast completion with zero variance. The first might be a legitimate fast responder with genuine opinions. The second is almost certainly not attending to item content. Treating both as equivalent based on timing alone is the error that produces wrongful exclusions.
Conversely, a slow completion with a completely flat response pattern is also a quality problem, just a different one. This profile often represents a respondent who is completing the survey on a slow device, scrolling carefully, and selecting the same option throughout without reading. Total time looks normal, timing threshold checks pass, but the pattern check catches it.
The most defensible quality framework uses timing as one weighted signal among several rather than as a stand-alone exclusion criterion. Reason codes from multi-signal scoring are also more useful to field managers than single-metric flags: "fast completion with flat pattern" is an actionable description; "below minimum time threshold" is not.
What Timing Cannot Tell You
It is worth being direct about the limits of timing-based quality signals. They are behavioral, not intentional. A very fast flat completion might be a bot, a professional speeder, or someone completing the survey on behalf of a family member who told them the answers in advance. The timing signal does not distinguish these cases. It tells you that something is anomalous about the response behavior; it does not tell you why.
Similarly, timing signals are specific to survey instruments with item-level timestamps. Many survey platforms still only record total completion time or capture timestamps with low resolution. Where item-level timestamps are not available, the full inter-item analysis is not possible, and quality checks must rely on response pattern signals alone. This is a data availability constraint, not a methodological preference, and it is worth being honest about which instruments in a given field setup actually support the analysis.
The goal at Besample is not to use timing as a magic number that settles the question. It is to include it as one component in a multi-signal quality score, calibrated to the specific instrument and population, that surfaces the flaggable cases while keeping false positives low enough that field managers trust the flags and act on them.