HPC · Science Deep Dive 5 October 2026 · revised 7 October 2026

Signal Detection Theory: Hits, Misses, False Alarms and Correct Rejections.

When each case is truly signal or noise, every yes-or-no call lands in one of four cells. Whether holding back was right depends on how well you see the signal, where you set your bar and what each outcome pays. Four results that show what this framework can measure and where its reach ends.

Section: Decisions · Probabilistic Thinking
Reading time: 24 min read
Sources: 48 · reviewed

01The Opening

Saying no is not proof of good judgment: until you know what was true

A smoke alarm that stays silent through every fire has never once cried wolf. It has also never done its job. The same trap waits wherever you must say yes or no before you know the truth. A no feels like caution, but once the truth is known it is either a miss or a correct rejection, and on its own it says nothing about the quality of the call. Signal detection theory begins there.

Keeping all four outcomes in view at once is the job of signal detection theory (SDT). It covers a narrow class of problem: each trial either contains a signal or does not, and an observer who cannot be certain must answer "something is there" or "nothing is there".^[5] Say yes when the signal is present and you score a hit; say no and you score a miss. Say yes to empty noise and you raise a false alarm; say no and you make a correct rejection.^[5]^[3]

The scheme travels: one review found that fields from medical imaging to weather forecasting share one set of testing problems.^[6] Wherever each case is truly signal or noise, a yes-or-no answer lands in one of these four cells.

History

The tidy matrix has a stranger history. A historical review traces it from a nineteenth-century Gaussian model of sensation to an insight of the early 1950s.^[7] That insight held that the senses produce sensations even with no stimulus present, so noise sits above the threshold of awareness, not below it.^[7]

Once noise can feel like signal, every answer becomes a judgment about whether the evidence is strong enough to count, and that bar can move. The idea reached print in two papers of the mid-1950s.^[8]^[9]

Hold on to that movable bar; everything later depends on it. Trained listeners in one study did as well using a rating scale, which uses several criteria, as answering yes or no, which uses one.^[10] The result fits the view that a yes-or-no answer is one position among many possible criteria.

The field began when noise turned out to feel like signal, and the rest follows from that.

Holding back is neither safe nor foolish by nature. Three separate quantities let you judge whether your rule was good: how well you can tell signal from noise (sensitivity), where you set the bar for saying yes (criterion), and what each outcome is worth (the payoff).

Most arguments about patience squeeze these three into one dial from cautious to reckless. SDT keeps them apart and lets the payoffs say where the criterion should sit.

Patience is a dial setting, and the right setting depends on what the four cells are worth.

02The Mechanism

How Signal Detection Theory Sets a Movable Bar

Optimal decision criterion placement maximizes expected reward and requires sensitivity to the category base rates (prior probabilities) and payoffs

Review of decision-criterion learning [16]

Picture two bell curves on one axis. One shows the evidence when only noise is present; the other shows it when a real signal is there. They overlap, and that overlap is the whole problem. No trial arrives labelled: you receive one point on the axis and must answer yes or no.

Two numbers govern everything after that. The distance between the curves, measured in units of their spread, is sensitivity, written d'. The position of the cutoff is the response criterion, written c. Under the standard equal-variance Gaussian model, you recover both from two rates you can count. The hit rate H is hits divided by all signal trials, and the false-alarm rate F is false alarms divided by all noise trials. Then d' = z(H) - z(F) and c = -(z(H) + z(F)) / 2, where z converts each rate into a normal-distribution score. A positive c marks a conservative observer who says yes sparingly.^[5]^[3]

Take a worked example of our own. Across 50 signal trials and 50 noise trials, the tally reads 40 hits, 10 misses, 12 false alarms and 38 correct rejections. That gives H = 0.80 and F = 0.24, so d' = 1.55 and c = -0.07: good discrimination and an almost neutral bar. Every uncited figure in this article comes from a model of this kind, computed by script; none is data about people.

One number grades how clear the signal is. The other grades how much evidence you demand before saying yes.

Illustrative, model-generated figure: equal-variance Gaussian curves with d' = 2.0 and the criterion at x = 1.0. Not data about any person.

Diagram

Move the bar and the answers change while the evidence stays the same. Hold d' at 2.0 and slide the criterion from lenient to strict: the hit rate falls from 0.977 to 0.500, and the false-alarm rate from 0.500 to 0.023. Every point on that sweep belongs to the same observer. Together the points form the receiver operating characteristic (ROC) curve.^[7]^[11]

With sensitivity fixed, no bar raises hits and correct rejections together. A falling false-alarm rate may mean sharper judgment or only a stricter bar, and without the hit rate beside it you cannot tell which. The area under the ROC curve sums up discrimination across every bar in one number: the probability that a random positive case outranks a random negative one.^[12]^[13]

Two points of method matter later. In a comparison of four recognition models, discrimination and bias stayed independent when measured with d' and c, but not with the alternative bias measure beta.^[14] For confidence-rating tasks, area-based measures proved the most consistent.^[15]

Read the ROC curve as the price list for caution: every extra rejection is paid for in hits.

That leaves the question of where to put the bar, and here the theory starts prescribing. Two inputs set the reward-maximising position: the base rate, how often signals occur, and the payoffs, what each outcome is worth. As a likelihood-ratio threshold, the best bar is beta* = [P(noise) / P(signal)] x [(V_cr + C_fa) / (V_hit + C_miss)]. V is the value of a correct response and C the cost of an error, both entered as positive amounts.^[3]^[16]

Read the formula as two knobs. Rare signals call for a stricter bar. Costly false alarms or valuable correct rejections tighten it further; costly misses or valuable hits loosen it. Nowhere does the formula call waiting a virtue. Once you state the base rate, the payoffs and the evidence distributions, it simply calculates the reward-maximising criterion.

Editorial pause

The best bar is not a personality trait. It follows from how rare the signal is and what each cell costs.

Take a screening task in which two cases in a hundred carry the signal, give the observer strong discrimination (d' = 2.0), and set the bar optimally for equal payoffs. That observer is correct on 0.9819 of cases but catches only 0.1721 of the real signals.

A rule that never acts is correct on 0.98 of cases and catches nothing. On accuracy alone, the careful observer and the one who stopped looking are nearly twins.

A high correct-rejection rate therefore proves nothing by itself. In this model the best bar grows stricter as signals grow rarer, so correct rejections and accuracy rise while the hit rate falls. Costly misses reverse the picture. At a base rate of 0.10, pricing a miss at four times a false alarm pulls the best beta from 9.0 down to 3.6 and lifts the hit rate from 0.461 to 0.640. Price false alarms four times higher instead, and beta climbs to 22.5 while the hit rate sinks to 0.289.

So neither popular slogan survives. Costly misses make the best policy act more often; costly false alarms make it reject more often. The theory supplies a benchmark, not a verdict on people, and whether people reach it is a question for experiment.^[16]

Editorial pause

Holding back is not automatically good. Base rates, payoffs and the hits it gives up decide what it is worth.

03Evidence

Where Signal Detection Theory Holds Up and Where It Does Not

01The claim

How confident to be

Support for SDT comes in three tiers.

Support for SDT comes in three tiers. The arithmetic, d' and c from a tally and the best bar from base rates and payoffs, is deduction under stated assumptions. Next comes observation: where people set their bars, and whether sensitivity and criterion come apart. Last is unsettled ground: the Gaussian assumptions and the origin of bias.

Criterion shifts replicate and the arithmetic is sound; the Gaussian assumptions and the origin of bias are still disputed. Use this framework as disciplined bookkeeping, never as proof that anyone decides well or badly.

Pooled estimate

62 effects

02How we measured

The six-criterion rubric

Studies scored on design, sample, rigour, causality, replication, citations.

We graded the five studies below on our own 100-point rubric from their abstracts, with citation counts from Crossref on 4 October 2026; the grades are not published metrics.

Every ranking is an editorial bet on what matters, and our weights are shown so you can challenge them.

Rubric weights

Design/30

Sample/20

Rigour/15

Causality/15

Replication/10

Citations/10

03Meta-analysis

Key finding

Moving Bar

Across 62 effects from 29 publications, a meta-analysis found a large conservative criterion shift on average when visual-search targets were rare, with no consistent change in sensitivity.^[1]

effects

04Direct test

Key finding

Unequal Spread

In two recognition-memory experiments, direct strength ratings spread less for new items than for studied ones, at an average ratio near 0.80, so equal variance is an approximation.^[2]

SD ratio

~0.80

The studies

The 5 Studies That Carry the Weight

Each study is scored out of 100 across six dimensions, from design and sample to replication and citations. Within their tasks, these papers show a criterion that moves with prevalence, base rate and payoff, and they show where the equal-variance assumption starts to give. In perceptual-categorization studies, base rates and payoffs moved the criterion, and people came close to the best bar under some conditions but not others. In visual search, target prevalence moved the criterion on average; in fingerprint comparison, it moved the criterion only when feedback followed every trial. In recognition memory, the Gaussian model is an approximation.

01Anchor

Pooled analysis of 62 visual-search effects from studies that varied how often the target appeared

Study 01 Evidence study 62 effects · drawn from 29 publications using robust variance estimation

Across 62 pooled visual-search effects, rare targets produced a large, reliable conservative criterion shift on average, while sensitivity showed no consistent change.^[1] The shift grew with the prevalence gap, held across the tested stimulus types, expertise levels and time constraints, and strengthened when feedback followed each trial.^[1]

Shows: in pooled visual-search studies, rare targets were associated on average with a conservative criterion shift, with no consistent pooled change in sensitivity.^[1]

It alone tests the theory's central split, criterion apart from sensitivity, across dozens of experiments. The result is an average within visual search; it does not bind other tasks, and no single mechanism behind it is settled.

Rubric breakdown

Design28/30

Sample17/20

Rigour13/15

Causality10/15

Replication8/10

Citations0/10

Total 76/100

The strongest studies, ranked by methodological weight.

Each study is scored out of 100 across six dimensions, from design and sample to replication and citations.

Anchor (Rank 1) Supporting

Rank Study Type Finding Score

Study 02

Pooled analysis of 232 experiments comparing recognition memory in younger and older adults^[4]

Meta-Analysis · The largest experimental base

Read through the theory's lens, older adults discriminated less sharply and used a different criterion, and both effects varied with the material.^[4]

Tracks d' and c as separate outcomes across 232 recognition-memory experiments.^[4]

No study here is larger, but comparing age groups cannot show what moves a criterion.

75/100

Study 03

Pooled re-analysis of 33 experiments that trained people to spot unreliable news^[17]

Meta-Analysis · The split tested in an applied setting

A hierarchical Bayesian model of 33 inoculation experiments found sharper discrimination between reliable and unreliable news, with no rise in response bias.^[17]

Across the pooled news-judgment experiments, discrimination improved with no increase in response bias.^[17]

The largest participant pool here, but a re-analysis of existing experiments on news judgment, not personal criterion setting.

68/100

Study 04

Two laboratory experiments that varied base rates and payoffs in a perceptual categorization task^[18]

Controlled Human Data · Base rates and payoffs under tight control

In Experiment 1, unequal payoffs produced conservative cutoffs and unequal base rates produced extreme ones; Experiment 2 found only small departures from optimal and no qualitative difference between the conditions.^[18]

In the two perceptual-categorization experiments, base rates and payoffs affected cutoff placement.^[18] The first showed conservative cutoffs under unequal payoffs and extreme ones under unequal base rates; the second found small departures from optimal with no qualitative difference between conditions.^[18]

Both factors were under direct control, but the evidence rests on one paradigm and a narrow task.

61/100

Study 05

Two recognition-memory experiments in which people rated memory strength directly^[2]

Controlled Human Data · An assumption tested head-on

On 20-point and 99-point scales, ratings spread more widely for studied items than for new ones; the ratio from the ratings tracked the ratio from ROC curves, and both averaged about 0.80.^[2]

In these recognition-memory tasks, direct ratings support unequal variance.^[2]

It tests a key assumption directly, but in only one task domain.

60/100

Reading the studies together

Together, these five studies support the split between sensitivity and criterion, and they show the criterion responding to circumstances. They do not show that people set it optimally in every condition, or that the equal-variance Gaussian model is exact.
The arithmetic cannot tell you whether people put the bar where the theory says. A review of criterion learning sums up the record: with skewed base rates people set nearly optimal criteria, and with skewed payoffs they fall short, even at the same optimal position.^[16] A follow-up found performance closer to optimal under base-rate than payoff changes, and only partial support for the two inputs combining independently.^[19]

In categorization experiments with biased payoffs, accuracy pulled the bar off the reward-maximising spot: observers told to maximise earnings still tilted their criteria toward accuracy.^[20] Biased payoffs split the reward-maximising and accuracy-maximising positions; biased base rates did not.^[21] Experience moves the bar too: recognition criteria tracked the probability that an item had been studied.^[22]

People clearly move the bar with conditions, but whether the new position is the best one depends on which input changed.

low-prevalence misses represent failures of perception when early search termination or motor errors are controlled
The low-prevalence effect shows the theory's strength and its seams together. Rare targets in visual search are missed far more often: miss rates were much higher at 1% to 2% prevalence than at 50%, and the authors read this as a criterion shift, not a loss of sensitivity.^[23]^[24] The flagship meta-analysis averages this literature into a large conservative shift, while noting that prevalence-induced concept change in other tasks can push the bar the other way.^[1]

Beneath that average, several mechanisms compete. At 98% prevalence the criterion moved the opposite way, and a separate cutoff governed when searchers quit and answered "absent".^[25] When people could correct themselves, many low-prevalence misses proved to be slips in executing the response.^[26] Yet with overall prevalence held at 50% and early quitting ruled out, rare targets still went undetected on 12% to 34% of trials while the searcher looked straight at them.^[27]

Feedback looks like the fix, and the data do not cooperate. Brief retraining at high prevalence with full feedback helped observers hold a good criterion in one series, yet across the meta-analysis immediate feedback enlarged the shift.^[24]^[1] In fingerprint comparison, prevalence moved errors only when feedback followed every trial, mainly through bias rather than sensitivity.^[28] One mammogram-search condition with computer-aided detection showed no low-prevalence effect at all.^[29]

The conservative shift is an average, not a law, and no single mechanism explains every rare-target miss.
A second weak point sits inside the Gaussian model: equal variance. Recognition-memory ROCs fit normal curves once new items are allowed to spread about 0.8 as wide as studied ones, and a major review catalogues the resulting regularities.^[30]^[31] Both unequal-variance SDT and a dual-process threshold account can reproduce those curves, though by one review's account, head-to-head studies have mostly favoured SDT.^[32]

Unequal variance undermines the quick arithmetic. A single-point d', computed from one hit rate and one false-alarm rate, carries a bias that larger samples cannot remove.^[33] In our model, with discrimination fixed and a signal spread of 1.25, the single-point d' drifts from 1.30 to 1.60 only because the bar moved. In simulations, d', A', percent correct and gamma could all carry unacceptably high Type I error rates.^[34] The supposedly nonparametric indexes carry assumptions too: area-ratio measures imply logistic distributions.^[35]^[36]

Rates at the edges need repair. A rate of 0 or 1 leaves the z-score undefined, and adding 0.5 to every cell gave less biased d' estimates in simulations than the 1/(2N) correction.^[37] Nor is A' a safe fallback: across 23 recognition experiments, this supposedly criterion-free index varied with the response criterion.^[38]

One pair of rates and one formula yield d' only by inheriting the model's assumptions, and the shortcut can mislead.
Where response bias comes from is still disputed. One family of models blames trial-to-trial jitter in the criterion: the decision noise model lets such jitter hide genuine criterion shifts, and one recognition experiment measured non-trivial criterion noise.^[39]^[40] A rival account, from three experiments with distribution-free tests, holds that bias moved the evidence distributions and left the decision rule unbiased.^[41] Neither account is confirmed across tasks.

Other tests favour the theory. A meta-analysis plus two experiments found that binary ROCs in recognition memory are curved, backing the classic case for SDT over threshold models.^[42] Another team tested five properties of SDT and found a strong empirical foundation for SDT models of recognition memory.^[43]

SDT comes through its challengers as a model with known approximations, not as a literal picture of the mind at work.
Diagnostics. One review scored six diagnostic fields with ROC analysis and noted that their test data are often of uncertain quality.^[6] 6 fields · diagnostic fields scored in a single review

Memory and witnesses. For police lineups, one proposal makes ROC analysis the yardstick, since procedures that cut false identifications often cut correct ones too.^[44] 86 conditions · pooled across 23 recognition experiments^[38]

Clinical research. Hallucinatory experiences tracked response bias across populations, while lower sensitivity appeared only in non-clinical samples.^[45] A psychosis meta-analysis pooled discrimination and response bias separately.^[46] 1,391 participants · across 15 auditory-verbal studies

04Stakes

A log containing only the moves you made is a matrix with half its cells blanked out

A decision to wait can be a miss or a correct rejection, depending on what was true

Signal detection theory does not tell you to act more or to wait more. It gives you a map. Wherever each case is truly signal or noise, every yes-or-no answer lands in one of four cells. In the standard model, sensitivity and criterion are separate quantities estimated from the same hit and false-alarm rates, and the right height for the bar comes from the base rate and what each outcome is worth. A correct rejection counts as a win only inside that frame: when signals are scarce and misses are costly, a long run of correct rejections can hide a collapsing hit rate.^[3]^[16]

A sharper question than "should I act?" asks four things: what the signal is, how good your information is, how often the signal appears and what each outcome costs. Temperament answers none of them.

If you make the same kind of call repeatedly, judge yourself against a standard, not a result. A rejection under a criterion set in advance follows your rule even if the opportunity later paid off; whether the rule was well set is a separate question. This series treats a profitable breach as a breach all the same, an editorial position rather than a finding of the theory.

None of this is grand. The theory measures; it does not claim that people are optimal or biased, though it gives you the words to say which. The parent guide, Strategic Inaction: Why Doing Nothing Is Sometimes the Highest-Performance Decision, takes up the practical side.

Score every cell, write down the payoffs, and review the bar when the odds, the stakes or the quality of your evidence changes.

The rare-signal trap

That observer is correct on 0.9819 of cases but catches only 0.1721 of the real signals.

01Claim

The matrix

Wherever each case is truly signal or noise, every yes-or-no answer falls into one of four cells, and under the equal-variance Gaussian model sensitivity and criterion can be estimated as separate quantities.^[5]

02Consequence

The rare-signal trap

Scarcity flatters caution. In the worked rare-signal model the best bar is strict, so near-perfect accuracy can sit beside a hit rate that misses most signals, and refusing every call scores almost as well.

03Lever

State the payoffs

In the model, the best bar follows base rates and what each outcome is worth,^[16] so write down the payoffs before you set the threshold.

07Bibliography

The bibliography.

48 sources · ~6h est. corpus read · 48 visible

Meta · 6 Review · 1 Journal · 41

Type

Sort

01 Meta

Caruso, V. D., Fox, G. Q., Carlisle, N. B., & Cox, P. H.2026

The role of feedback in the low-prevalence effect: a meta-analysis of criterion shifts in visual search

Cognitive Research: Principles and Implications

doi: 10.1186/s41235-026-00747-5
02 Journal

Mickes, L., Wixted, J. T., & Wais, P. E.2007

A direct test of the unequal-variance signal detection model of recognition memory

Psychonomic Bulletin & Review14(5) · 858-865

doi: 10.3758/bf03194112
03 Journal

Lynn, S. K., & Barrett, L. F.2014

"Utilizing" Signal Detection Theory

Psychological Science25(9) · 1663-1673

doi: 10.1177/0956797614541991
04 Meta

Fraundorf, S. H., Hourihan, K. L., Peters, R. A., & Benjamin, A. S.2019

Aging and recognition memory: A meta-analysis

Psychological Bulletin145(4) · 339-371

doi: 10.1037/bul0000185
05 Journal

Huang, Y., & Ferreira, F.2020

The Application of Signal Detection Theory to Acceptability Judgments

Frontiers in Psychology11 · 73

doi: 10.3389/fpsyg.2020.00073
06 Journal

Swets, J. A.1988

Measuring the Accuracy of Diagnostic Systems

Science240(4857) · 1285-1293

doi: 10.1126/science.3287615
07 Journal

Wixted, J. T.2020

The forgotten history of signal detection theory

Journal of Experimental Psychology: Learning, Memory, and Cognition46(2) · 201-233

doi: 10.1037/xlm0000732
08 Journal

Tanner, W. P., & Swets, J. A.1954

A decision-making theory of visual detection

Psychological Review61(6) · 401-409

doi: 10.1037/h0058700
09 Journal

Peterson, W., Birdsall, T., & Fox, W.1954

The theory of signal detectability

Transactions of the IRE Professional Group on Information Theory4(4) · 171-212

doi: 10.1109/TIT.1954.1057460
10 Journal

Egan, J., Schulman, A. I., & Greenberg, G. Z.1959

Operating Characteristics Determined by Binary Decisions and by Ratings

The Journal of the Acoustical Society of America31(6) · 768-773

doi: 10.1121/1.1907783
11 Journal

Green, D. M.1960

Psychoacoustics and Detection Theory

The Journal of the Acoustical Society of America32(10) · 1189-1203

doi: 10.1121/1.1907882
12 Journal

Hanley, J. A., & McNeil, B. J.1982

The meaning and use of the area under a receiver operating characteristic (ROC) curve

Radiology143(1) · 29-36

doi: 10.1148/radiology.143.1.7063747
13 Journal

Zweig, M. H., & Campbell, G.1993

Receiver-operating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine

Clinical Chemistry39(4) · 561-577

doi: 10.1093/clinchem/39.4.561
14 Journal

Snodgrass, J. G., & Corwin, J.1988

Pragmatics of measuring recognition memory: Applications to dementia and amnesia

Journal of Experimental Psychology: General117(1) · 34-50

doi: 10.1037/0096-3445.117.1.34
15 Journal

Kornbrot, D. E.2006

Signal detection theory, the approach of choice: Model-based and distribution-free measures and evaluation

Perception & Psychophysics68(3) · 393-414

doi: 10.3758/bf03193685
16 Journal

Maddox, W. T.2002

Toward a unified theory of decision criterion learning in perceptual categorization

Journal of the Experimental Analysis of Behavior78(3) · 567-595

doi: 10.1901/jeab.2002.78-567
17 Meta

Simchon, A., Zipori, T., Teitelbaum, L., Lewandowsky, S., & van der Linden, S.2026

A signal detection theory meta-analysis of psychological inoculation against misinformation

Current Opinion in Psychology67 · 102194

doi: 10.1016/j.copsyc.2025.102194
18 Journal

Maddox, W. T., & Bohil, C. J.1998

Base-rate and payoff effects in multidimensional perceptual categorization

Journal of Experimental Psychology: Learning, Memory, and Cognition24(6) · 1459-1482

doi: 10.1037/0278-7393.24.6.1459
19 Journal

Bohil, C. J., & Maddox, W. T.2001

Category discriminability, base-rate, and payoff effects in perceptual categorization

Perception & Psychophysics63(2) · 361-376

doi: 10.3758/bf03194476
20 Journal

Bohil, C. J., & Maddox, W. T.2003

On the generality of optimal versus objective classifier feedback effects on decision criterion learning in perceptual categorization

Memory & Cognition31(2) · 181-198

doi: 10.3758/bf03194378
21 Journal

Maddox, W. T., & Bohil, C. J.2005

Optimal classifier feedback improves cost-benefit but not base-rate decision criterion learning in perceptual categorization

Memory & Cognition33(2) · 303-319

doi: 10.3758/bf03195319
22 Journal

Rhodes, M. G., & Jacoby, L. L.2007

On the dynamic nature of response criterion in recognition memory: Effects of base rate, awareness, and feedback

Journal of Experimental Psychology: Learning, Memory, and Cognition33(2) · 305-320

doi: 10.1037/0278-7393.33.2.305
23 Journal

Wolfe, J. M., Horowitz, T. S., & Kenner, N. M.2005

Rare items often missed in visual searches

Nature435(7041) · 439-440

doi: 10.1038/435439a
24 Journal

Wolfe, J. M., Horowitz, T. S., Van Wert, M. J., Kenner, N. M., Place, S. S., & Kibbi, N.2007

Low target prevalence is a stubborn source of errors in visual search tasks

Journal of Experimental Psychology: General136(4) · 623-638

doi: 10.1037/0096-3445.136.4.623
25 Journal

Wolfe, J. M., & Van Wert, M. J.2010

Varying Target Prevalence Reveals Two Dissociable Decision Criteria in Visual Search

Current Biology20(2) · 121-124

doi: 10.1016/j.cub.2009.11.066
26 Journal

Fleck, M. S., & Mitroff, S. R.2007

Rare Targets Are Rarely Missed in Correctable Search

Psychological Science18(11) · 943-947

doi: 10.1111/j.1467-9280.2007.02006.x
27 Journal

Hout, M. C., Walenchok, S. C., Goldinger, S. D., & Wolfe, J. M.2015

Failures of perception in the low-prevalence effect: Evidence from active and passive visual search

Journal of Experimental Psychology: Human Perception and Performance41(4) · 977-994

doi: 10.1037/xhp0000053
28 Journal

Growns, B., & Kukucka, J.2021

The prevalence effect in fingerprint identification: Match and non-match base-rates impact misses and false alarms

Applied Cognitive Psychology35(3) · 751-760

doi: 10.1002/acp.3800
29 Journal

Kunar, M. A., Watson, D. G., Taylor-Phillips, S., & Wolska, J.2017

Low prevalence search for cancers in mammograms: Evidence using laboratory experiments and computer aided detection

Journal of Experimental Psychology: Applied23(4) · 369-385

doi: 10.1037/xap0000132
30 Journal

Ratcliff, R., Sheu, C. F., & Gronlund, S. D.1992

Testing global memory models using ROC curves

Psychological Review99(3) · 518-535

doi: 10.1037/0033-295X.99.3.518
31 Review

Yonelinas, A. P., & Parks, C. M.2007

Receiver operating characteristics (ROCs) in recognition memory: A review

Psychological Bulletin133(5) · 800-832

doi: 10.1037/0033-2909.133.5.800
32 Journal

Wixted, J. T.2007

Dual-process theory and signal-detection theory of recognition memory

Psychological Review114(1) · 152-176

doi: 10.1037/0033-295X.114.1.152
33 Journal

Verde, M. F., Macmillan, N. A., & Rotello, C. M.2006

Measures of sensitivity based on a single hit rate and false alarm rate: The accuracy, precision, and robustness of d', Az, and A'

Perception & Psychophysics68(4) · 643-654

doi: 10.3758/BF03208765
34 Journal

Rotello, C. M., Masson, M. E. J., & Verde, M. F.2008

Type I error rates and power analyses for single-point sensitivity measures

Perception & Psychophysics70(2) · 389-401

doi: 10.3758/PP.70.2.389
35 Journal

Macmillan, N. A., & Creelman, C. D.1996

Triangles in ROC space: History and theory of "nonparametric" measures of sensitivity and response bias

Psychonomic Bulletin & Review3(2) · 164-170

doi: 10.3758/bf03212415
36 Journal

Pastore, R. E., Crawley, E. J., Berens, M. S., & Skelly, M. A.2003

"Nonparametric" A' and other modern misconceptions about signal detection theory

Psychonomic Bulletin & Review10(3) · 556-569

doi: 10.3758/bf03196517
37 Journal

Hautus, M. J.1995

Corrections for extreme proportions and their biasing effects on estimated values of d'

Behavior Research Methods, Instruments, & Computers27(1) · 46-51

doi: 10.3758/BF03203619
38 Meta

Gardiner, J. M., Ramponi, C., & Richardson-Klavehn, A.2002

Recognition memory and decision processes: A meta-analysis of remember, know, and guess responses

Memory10(2) · 83-98

doi: 10.1080/09658210143000281
39 Journal

Mueller, S. T., & Weidemann, C. T.2008

Decision noise: An explanation for observed violations of signal detection theory

Psychonomic Bulletin & Review15(3) · 465-494

doi: 10.3758/PBR.15.3.465
40 Journal

Benjamin, A. S., Diaz, M., & Wee, S.2009

Signal detection with criterion noise: Applications to recognition memory

Psychological Review116(1) · 84-115

doi: 10.1037/a0014351
41 Journal

Balakrishnan, J. D.1999

Decision processes in discrimination: Fundamental misrepresentations of signal detection theory

Journal of Experimental Psychology: Human Perception and Performance25(5) · 1189-1206

doi: 10.1037/0096-1523.25.5.1189
42 Journal

Dube, C., & Rotello, C. M.2012

Binary ROCs in perception and recognition memory are curved

Journal of Experimental Psychology: Learning, Memory, and Cognition38(1) · 130-151

doi: 10.1037/a0024957
43 Journal

Kellen, D., Winiger, S., Dunn, J. C., & Singmann, H.2021

Testing the foundations of signal detection theory in recognition memory

Psychological Review128(6) · 1022-1050

doi: 10.1037/rev0000288
44 Journal

Wixted, J. T., & Mickes, L.2012

The Field of Eyewitness Memory Should Abandon Probative Value and Embrace Receiver Operating Characteristic Analysis

Perspectives on Psychological Science7(3) · 275-278

doi: 10.1177/1745691612442906
45 Meta

Medcalf-Bell, R., Moseley, P., & Barkus, E.2026

Hallucinations and auditory-verbal signal detection studies: A meta-analytic review

Psychological Medicine56 · e254

doi: 10.1017/s0033291726105339
46 Meta

Damiani, S., Cavicchioli, M., Guiot, C., Donadeo, A., Scalabrini, A., Grecuzzo, V., Bergamaschini, I., Provenzani, U., Politi, P., & Fusar-Poli, P.2024

The noise in our brain: A systematic review and meta-analysis of neuroimaging and signal-detection studies on source monitoring in psychosis

Journal of Psychiatric Research169 · 142-151

doi: 10.1016/j.jpsychires.2023.11.036
47 Journal

Evans, K. K., Birdwell, R. L., & Wolfe, J. M.2013

If You Don't Find It Often, You Often Don't Find It: Why Some Cancers Are Missed in Breast Cancer Screening

PLoS ONE8(5) · e64366

doi: 10.1371/journal.pone.0064366
48 Journal

Kantner, J., & Lindsay, D. S.2012

Response bias in recognition memory as a cognitive trait

Memory & Cognition40(8) · 1163-1177

doi: 10.3758/s13421-012-0226-0

THE HAND-OFF — Performance Scan: find your limiting factor. Three minutes, 24 questions, six systems — and your answers never leave your device.

Support for SDT comes in three tiers.

Studies scored on design, sample, rigour, causality, replication, citations.

Moving Bar

Unequal Spread

The 5 Studies That Carry the Weight

Pooled analysis of 62 visual-search effects from studies that varied how often the target appeared

The strongest studies, ranked by methodological weight.

Hidden Cells

The Accuracy Trap

Drifting Bars

Confused Dials

The protocol, as a sequence.

Define the Signal

Log Every Opportunity

Read the Rates Together

Move the Bar for a Reason

That observer is correct on 0.9819 of cases but catches only 0.1721 of the real signals.

The matrix

The rare-signal trap

State the payoffs

More from the Science Deep Dives