Altamens

Altamens Research · Evidence review · methods guide

Faster Is Not Always Better: Reading Speed and Accuracy Together

A practical guide to interpreting timed mental-task scores when quicker responses can come with more errors.

Altamens10 min read

Abstract

Timed mental tasks often report speed, accuracy or a score that combines them. Those summaries answer different questions. A faster session is not unambiguously better if it contains more errors, and a single combined score can hide which component changed.

This methods guide reviews primary methodological research on the speed–accuracy trade-off and works through five hypothetical sessions. In the example, inverse efficiency and rate-correct scoring rank a 60%-accurate fast-guessing session above a 95%-accurate baseline. That arithmetic is not evidence that either metric always fails; it shows why raw speed and accuracy should remain visible and why any combined score needs an explicit purpose, formula and validity check.

Key findings

  • Report accuracy and a robust summary of correct-response times separately before interpreting a combined score.
  • Integrated scores encode choices about how much speed and accuracy matter; no reviewed metric is unconditionally immune to speed–accuracy trade-offs.
  • Compare like with like: the same task version, instructions, timing rule, response device, exclusions and opportunity window.

1One task can produce two moving outcomes

In choice tasks, people can often respond sooner by accepting a higher risk of error, or wait for more information and answer more accurately. The speed–accuracy trade-off is therefore not a nuisance unique to games; it is a recurring feature of decision behaviour and a central issue in reaction-time methodology.1

A fastest-time label answers which valid response or session was quickest under the scoring rule. Accuracy answers how often responses were correct. Neither alone establishes that overall performance improved. If a session becomes 150 ms faster while accuracy falls from 95% to 75%, the data describe a trade-off—not a single, self-evident improvement.

2Keep the two components auditable

For each comparable session, retain the trial count, number correct, number incorrect, omissions or timeouts, and the response-time distribution. State whether response time is calculated for correct trials only and how anticipatory, interrupted or otherwise invalid trials are handled.

  • Accuracy: correct responses divided by eligible attempted responses, with omissions reported separately when they can occur.
  • Speed: a prespecified summary—often the median or a trimmed mean—of eligible correct-response times, plus the number of observations behind it.
  • Context: task/version, difficulty or item mix, instructions, time limit, device/input method and date or session order.
  • Comparability: the same definition and conditions across sessions; otherwise label the change rather than merging unlike results.

A mean can be strongly affected by a few very slow responses, while a median answers a different question about the typical response. Choose the summary before inspecting which option makes a result look best. For a fixed-duration task, also record the opportunity structure: a person who attempts more items can make more errors and still produce more correct answers.

3Worked example: the ranking depends on the rule

The downloadable example contains five invented session summaries. A careful baseline has 950 ms mean correct-response time and 95% accuracy. A fast-guessing-risk session has 580 ms and 60% accuracy. These are teaching values, not observations of people, Altamens accounts or game telemetry.

On speed alone, the fast-guessing-risk session looks best. On accuracy alone, it looks worst. Its inverse efficiency score is 967 ms per unit proportion correct, compared with 1,000 for the careful baseline; its rate-correct score is 1.034 versus 1.000 correct responses per second. Both simple integrated calculations therefore rank the 60%-accurate session slightly above the 95%-accurate baseline.

A third hypothetical session is faster while preserving similar accuracy: 790 ms and 94%. It improves both components relative to baseline, so its interpretation is much less dependent on a trade-off formula. This is the practical value of keeping the two outcomes visible: dominance on both is clearer than a composite win produced by compensation.

Download the worked examples CSV, standard-library calculator and methods and claim sheet. The files contain no participant or customer data.

4What common integrated scores actually do

The inverse efficiency score (IES) divides mean correct-response time by proportion correct; lower is better. The rate-correct score (RCS) divides proportion correct by mean correct-response time in seconds; higher is better. For the same two inputs they are reciprocal transformations apart from units, so they do not supply independent confirmation.

Simulation work by Liesefeld and Janczyk found that IES, RCS and the linear integrated speed–accuracy score can place unequal weight on speed and accuracy as accuracy changes. In those simulations, the balanced integration score (BIS)—accuracy z-score minus response-time z-score—was comparatively less sensitive to the imposed trade-offs.3

BIS introduces a different dependency: its value is relative to the observations used to calculate the means and standard deviations. Add sessions, change the reference group or mix task versions and earlier BIS values can change. A standardised score therefore needs a frozen reference definition and should not be presented as a self-contained property of one session.

5What the methodological evidence supports

A 2017 simulation comparison found that some combined measures can be useful for detecting effects, while also recommending inspection of the underlying speed and accuracy data rather than treating a composite as a replacement for them.2

The 2019 diffusion-model simulations favoured BIS over IES, RCS and LISAS for relative insensitivity to the simulated trade-offs, but that result is conditional on the studied models and scoring definitions.3

A broader 2021 set of four simulation studies reached a more cautious conclusion: none of five integrated measures was unconditionally insensitive to speed–accuracy trade-offs. The paper instead showed that design—specifically mixed-condition repeated-measures arrangements—could distribute trade-off effects so the effect of interest remained detectable.4

Later work also showed that two superficially similar linear measures, BIS and LISAS, can differ in consequential ways. The practical lesson is not to search for a universal winner, but to state the estimand, preserve the component outcomes and test the chosen score under conditions relevant to its use.5

6A six-step interpretation rule

  1. Check comparability: same task/version, instructions, time limit, difficulty, response device and eligibility rules.
  2. Report trial counts, accuracy and omissions before any speed summary.
  3. Report a prespecified correct-response-time summary and its valid observation count.
  4. Classify the direction: both better, both worse, or a speed–accuracy trade-off.
  5. If using a composite, publish its formula, units, reference set, direction and valid operating range beside the components.
  6. Describe the observed pattern without attributing learning, transfer, cognitive change or health significance unless an appropriate design supports that claim.

For personal progress, a simple display can show accuracy and median correct-response time as two aligned trends. Mark changes to task, device or rules. A combined score may be added for a defined use, but it should not erase the evidence needed to understand why it moved.

7Limitations, provenance and review status

This is a narrative methods review, not a systematic review or meta-analysis. The worked examples are deterministic arithmetic chosen to expose interpretation differences; they are not simulations of a population and do not estimate how often any pattern occurs. We did not test the measures against Altamens games or identify a universally valid score.

No Altamens user records, accounts, analytics or game telemetry were accessed. Literature search, drafting, calculations, source checks and publication preparation used an OpenAI automated assistant. The assets received automated consistency checks, but the paper has no independent human methods review and is not journal peer reviewed.

References

  1. 1.Heitz RP The speed-accuracy tradeoff: history, physiology, methodology, and behavior. Frontiers in Neuroscience 8:150, 2014. Peer-reviewed narrative review. DOI: 10.3389/fnins.2014.00150.. https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2014.00150/full
  2. 2.Vandierendonck A A comparison of methods to combine speed and accuracy measures of performance: A rejoinder on the binning procedure. Behavior Research Methods 49(2), 653–673, 2017. Simulation comparison. DOI: 10.3758/s13428-016-0721-5.. https://pubmed.ncbi.nlm.nih.gov/26944576/
  3. 3.Liesefeld HR, Janczyk M Combining speed and accuracy to control for speed-accuracy trade-offs(?). Behavior Research Methods 51(1), 40–60, 2019. Standard diffusion-model simulations. DOI: 10.3758/s13428-018-1076-x.. https://pubmed.ncbi.nlm.nih.gov/30022459/
  4. 4.Vandierendonck A On the Utility of Integrated Speed-Accuracy Measures when Speed-Accuracy Trade-off is Present. Journal of Cognition 4(1):22, 2021. Four simulation studies; publisher links a correction. DOI: 10.5334/joc.154.. https://journalofcognition.org/articles/10.5334/joc.154
  5. 5.Liesefeld HR, Janczyk M Same same but different: Subtle but consequential differences between two measures to linearly integrate speed and accuracy (LISAS vs. BIS). Behavior Research Methods 55(3), 1175–1192, 2023. Methodological comparison; corrected article. DOI: 10.3758/s13428-022-01843-2.. https://pubmed.ncbi.nlm.nih.gov/35595937/