Psychology by BogdanovaThe Encyclopedia of Psychology for the Artificial Era

RESEARCH DOSSIER

Sensory Thresholds and Discrimination

Abstract. Sensory thresholds and discrimination describe the conditional relation between controlled physical stimulation and the probability of detecting or distinguishing sensory events. This fundamental critical narrative monograph examines absolute and difference thresholds, just-noticeable differences, Weber fractions, psychometric functions, response criteria, yes–no and forced-choice paradigms, classical and adaptive measurement procedures, Bayesian stimulus selection, internal and external noise, ideal-observer analysis, visual contrast masking, reference dependence, sensory modality differences, and the measurement effects of attention and adaptation. The evidence synthesis draws on foundational psychophysics, original visual, auditory, and tactile studies, historical primary sources, methodological simulations, and contemporary hierarchical psychometric research. It addresses parameter identifiability, small-sample error, calibration, reproducibility, ecological and clinical interpretation, and functional comparisons involving artificial sensory systems. The central conclusion is that a threshold has scientific meaning only together with the task, criterion, stimulus calibration, observer state, and uncertainty of its estimator. This is a critical narrative research monograph, not a new experiment, preregistered systematic review, or meta-analysis.

Keywords: sensory threshold; difference threshold; absolute threshold; just-noticeable difference; JND; Weber fraction; psychometric function; psychophysical discrimination; two-alternative forced choice; yes–no detection; adaptive staircase; QUEST; contrast discrimination; external noise; ideal observer; psychophysical measurement; perceptual sensitivity; Artificial Era; auditory intensity discrimination; frequency difference limen; tactile spatial acuity; vibrotactile masking

Sensory Thresholds as an Experimental Research Object

Sensory thresholds are operational locations on a relation between a controlled stimulus and a measurable response. The absolute threshold conventionally describes stimulus magnitude associated with a specified probability of detection; a difference threshold describes the separation between stimuli associated with a specified probability of correct discrimination. Neither number is meaningful without the response task, stimulus dimensions, reference level, viewing or listening conditions, and performance criterion. A threshold can shift when luminance, adaptation, temporal interval, prior probability, or instructions change. In this sense a threshold is a parameter of a person–stimulus–task system, estimated under controlled conditions rather than a free-standing constant of the senses. The methodological accounts of Klein (2001) and Kingdom & Prins (2016) make the operational relation between stimulus control and inference central to psychophysics.

A discrimination experiment asks whether a participant can distinguish two states of stimulation. In a visual task the states may differ in contrast, spatial frequency, direction, duration, or intensity; auditory experiments may vary frequency, level, temporal gap, or modulation; tactile experiments may manipulate pressure, orientation, or texture. It matters whether the participant identifies which interval contains a signal, which alternative has the larger intensity, or whether a difference was present at all. Each procedure imposes a different decision rule and different opportunities for bias. Green & Swets (1966) supplied the formal statistical language for understanding the role of uncertainty and decisions, while Wichmann & Hill (2001a) showed how binary performance must be fitted before the desired parameter is reported.

This monograph investigates the empirical meaning, historical development, measurement, statistical interpretation, and validity of sensory thresholds and discrimination. Its evidential core is controlled psychophysics, selected original studies, methodological comparisons, and modern quantitative inference. The methodological discussion is organized around the question of what exactly changes when an estimated threshold rises or falls. The scientific distinction between improved sensitivity, altered decisional criterion, attention, changed sensory normalization, and estimator artifact is preserved throughout. This is a critical narrative synthesis of published research, rather than a systematic review with exhaustive search protocol or a newly conducted experiment Geisler (2011) Prins & Kingdom (2018).

Absolute Detection and Difference Discrimination

Absolute detection is often described as finding the weakest stimulus that can be perceived. Experimental measurement refines that formulation: perception is inferred from a sequence of decisions whose reliability varies continuously with stimulus strength. When trials include a blank and a signal interval, the investigator may define threshold as the level giving a chosen proportion correct above chance. The level is different if threshold is defined by the frequency of saying 'yes' to a single interval, because false alarms and response willingness contribute. Green & Swets (1966) explains why performance and criterion require separate modeling. A nominal threshold measured with one decision arrangement should not be transported uncritically to another, even when stimulus equipment and physical units remain unchanged.

A difference threshold, frequently called a just-noticeable difference or JND, describes discriminability relative to a comparison or reference magnitude. In a two-interval discrimination task, the observer might report which tone was louder or which grating had higher contrast. The reference intensity, direction of increment or decrement, stimulus order, and uncertainty in memory for the standard all matter. A discrimination threshold around a nonzero pedestal may be smaller or larger than the absolute threshold for the same physical feature; consequently, detection and discrimination curves carry complementary information about coding. Nachmias & Sansbury (1974) provides a historically important visual example in which grating contrast discrimination was reported as potentially better than detection, inviting analysis beyond a simple monotonic account of signal strength.

Theoretical interpretation therefore requires describing the stimulus continuum precisely. Intensity in physical units, log intensity, decibels, Michelson contrast, Weber contrast, and relative increment are transformations with distinct permissible ranges. An absolute visual threshold at a given spatial frequency cannot be compared numerically with an auditory frequency JND without converting the measurement into a common inferential question. Lesmes et al. (2010) modeled contrast sensitivity across spatial frequency rather than treating a single threshold as a complete description of vision. This wider function illustrates why a sensory limit is best understood as conditional on stimulus dimensions and measurement context.

Weber Fractions and the History of Quantification

The investigation of differential sensitivity predates laboratory psychology as a distinct institution. Ernst Heinrich Weber's nineteenth-century experiments on tactile, weight, and related judgments established a research program connecting changes in physical magnitude with changes in discriminability. Gustav Theodor Fechner subsequently organized threshold and sensation concepts within mathematical psychophysics. The historical importance of this program lies in the transition from verbal descriptions of sensation to replicable stimulus manipulations and quantitatively expressible judgment relations. Modern accounts reconstruct those claims alongside the assumptions introduced when a JND is treated as a psychologically invariant step Kingdom & Prins (2016).

Weber's law is commonly expressed as ΔI/I = k, where I is a reference intensity, ΔI a discrimination increment at a specified response criterion, and k the Weber fraction. The relation approximates proportional discrimination over particular ranges and conditions. It is not a universal identity spanning zero intensity, all receptor systems, or every dimension of experience. When the baseline approaches the detection region, the fraction may depart from constancy; high levels may also introduce saturation or changed normalization. Because the empirical fraction is calculated using an estimated ΔI, its numerical value depends on how the JND was acquired and how observer uncertainty was modeled. Klein (2001) provides a broader account of psychometric estimation that helps distinguish law-like observations from artifacts of method.

Fechner's logarithmic proposal integrated small relative increments under assumptions about increments of sensation. Subsequent theories of sensory magnitude, probability judgment, and nonlinear transduction revisited the relationship between physical strength and experience. This monograph uses Weber and Fechner as historical foundations for experimentally measured discriminability, while leaving the larger intellectual history of quantification to its separately designated dossier. The continuing scientific issue is whether observed proportionality results from receptor coding, comparison noise, uncertainty, divisive gain control, task strategy, or several interacting mechanisms. For visual contrast, the quantitative models of Pelli (1985) and Foley (1994) show why psychophysical regularities do not uniquely identify an underlying physiological transformation.

The Psychometric Function and Threshold Convention

A psychometric function maps stimulus magnitude onto the probability of a designated response. For forced-choice detection it commonly rises from chance performance to near-perfect discrimination; for yes–no tasks the relevant lower asymptote and response bias require different interpretation. The threshold is the stimulus value at which the fitted function crosses a stated criterion, often a performance level midway between guessing and asymptotic performance. A 75-percent-correct threshold in two-alternative forced choice expresses a convention relative to chance at 50 percent; it must not be equated with a 75-percent detection-report rate in yes–no judgment. Wichmann & Hill (2001a) analyzed the fitting consequences of lapses and limited trial numbers, making explicit assumptions indispensable.

The shape of this function matters alongside its crossing point. Slope indicates how rapidly performance changes with stimulus strength under a selected stimulus transformation. Guess and lapse terms describe behavior near lower and upper performance limits, although they should not be mistaken for complete psychological mechanisms. A shallow estimated slope might reflect sensory variability, trial mixing, adaptation, attentional fluctuation, observer heterogeneity, or an unsuitable function family. Similarly, a displaced threshold can arise from a changed slope combined with an altered asymptote. Wichmann & Hill (2001b) addressed bootstrap confidence intervals and sampling, and Kontsevich & Tyler (1999) considered the simultaneous Bayesian estimation of slope and threshold.

Psychometric fitting transforms a series of responses into uncertain parameter estimates. Maximum likelihood, Bayesian posterior inference, bootstrap resampling, and simulation-based goodness-of-fit checks contribute different forms of evidence. Appropriate analysis requires recording the actual stimulus levels, trial outcomes, stopping rule, fitted link function, lapse policy, and threshold definition. A scalar JND without those ingredients cannot be audited or compared reliably. Prins & Kingdom (2018) advocates explicit model comparisons when research hypotheses concern changes in one or more parameters; Klein (2001) clarifies why apparently minor choices in function definition can produce substantive changes in interpretation.

Yes–No Reports, Forced Choice, and Response Criteria

In a yes–no detection trial, one observation leads to a categorical report. A participant must decide whether the observation is better explained by signal-plus-noise or noise alone. By varying instructions, stimulus prevalence, rewards, and feedback, experimenters can alter the willingness to say that a signal occurred. If they summarize only affirmative responses as detection, decisional changes masquerade as sensitivity changes. Signal-present and signal-absent conditions provide hit and false-alarm proportions that can be analyzed separately. Green & Swets (1966) formalized the consequences for distributions of internal evidence and decision boundaries. Threshold studies use this distinction to protect the interpretation of their fitted response probabilities.

In two-alternative forced choice, the participant compares responses to two defined alternatives or intervals. Under common symmetric assumptions, accuracy depends on relative evidence rather than one fixed yes–no boundary. Chance performance is 50 percent when there are two equally probable choices. Yet forced choice does not magically eliminate all response biases: interval preferences, unequal stimulus uncertainty, order effects, attention, and learning can influence choices. A four-choice task has a different chance level, and comparisons across task formats require a common model. Klein (2001) discusses distinctions between objective and subjective response procedures; Kingdom & Prins (2016) emphasizes that the experimental question determines the best procedure.

An informative protocol specifies whether participants receive feedback, whether standards are fixed or roving, whether signal and blank locations are balanced, and whether the same stimulation can be presented twice. These manipulations alter memory and decision processes independently of peripheral sensory transduction. Repeated measures of a detection threshold should therefore preserve response architecture as a controlled factor. Some modern proposals augment staircase tasks with blank trials to separate response tendency from sensitivity; their applicability remains dependent on how the extra conditions are modeled. The more general lesson from Green & Swets (1966) and Wichmann & Hill (2001a) is that thresholds inherit the assumptions of the behavioral observation model.

Classical Limits, Constant Stimuli, and Adjustment

The method of limits presents an ordered series of stimulus values and records where an observer begins or ceases reporting detection. It is efficient for bracketing a transition, but sequential expectancy, adaptation, and the dependence of a response on the preceding stimulus create potential bias. Ascending and descending series can reveal direction-specific shifts rather than cancel them automatically. The method of adjustment allows an observer to manipulate intensity until a subjective boundary is reached; it is convenient and intuitive, yet motor control, anchoring, stopping strategies, and knowledge of the adjustable range contribute to the observed setting. Kingdom & Prins (2016) treats these methods as options with distinct inferential costs.

The method of constant stimuli samples prespecified intensities across repeated, often randomized presentations. The resulting response proportions can support direct psychometric fitting and reveal unexpected departures from a presumed monotonic curve. Its cost is inefficient allocation when many trials fall far below or above an informative transition. The statistical comparison is conditional: if the threshold's likely location is well known, appropriately spaced fixed stimuli may perform effectively; if the range is broad, adaptive allocation may be advantageous. Simpson (1988) defended the efficiency of constant stimuli under particular simulations, while Watson & Fitzhugh (1990) challenged the representativeness of those simulation assumptions. Their exchange illustrates the dependence of methodological conclusions on the modeled observer.

An experimenter should predefine informative stimulus levels or an adjustment range, randomization, permitted retries, and criteria for invalid trials. Randomization helps reduce predictable sequence cues but does not remove adaptation or drift. Trial block design may deliberately interleave pedestal conditions to identify shared uncertainty, although this introduces further contextual effects. Classical procedures remain valuable as transparent benchmarks, especially when a psychometric function has multiple inflections or poorly understood asymptotes. The experimental problem is to align allocation with the inferential target, not to declare a universally superior procedure Klein (2001) Treutwein (1995).

Adaptive Staircases and Reversal-Based Estimation

Adaptive staircases change stimulus difficulty based on prior responses. A simple one-up/one-down sequence tends toward a response region near its rule's equilibrium; transformed up-down algorithms move after specified patterns of correct and incorrect answers and can target higher performance points. Levitt (1971) established widely used transformed methods in psychoacoustics. In practice, the investigator must report the rule, increasing and decreasing step sizes, number of reversals, initial level, stopping conditions, and the statistic extracted from the track. Otherwise an apparent threshold has no reproducible estimator definition.

A reversal is a change in the direction of stimulus progression. Averaging several reversal intensities is a convenient summary but need not yield an unbiased estimate of the chosen psychometric threshold, especially early in the track or with few reversals. García-Pérez (1998) analyzed both asymptotic and finite-sample properties of fixed-step forced-choice staircases, showing how nominal convergence targets may differ from actual behavior and how short tracks can be imprecise. A threshold estimate derived from a reversal average should therefore be distinguished from a parameter fitted to all trial-level outcomes. The latter may offer more information but introduces additional functional assumptions.

The adaptive rule also shapes stimulus history. Tracks that concentrate near threshold may amplify local adaptation, learning, and sequential dependence. Interleaved staircases obscure simple patterns but can introduce cross-condition carryover if references change frequently. Researchers need to simulate the chosen estimator under plausible lapse rates, response slopes, serial correlations, and nonstationary observers before relying on a nominal convergence probability. Treutwein (1995) reviewed the diversity and history of adaptive procedures; the methodological literature cautions that the performance of a staircase is a property of the entire design, analysis, and observer model.

Bayesian Adaptive Methods and Information Gain

Bayesian adaptive procedures treat threshold as an unknown parameter with a prior distribution, update its posterior after every response, and select new stimuli according to a rule. Original QUEST placed a new trial near the currently likely threshold under a chosen psychometric model Watson & Pelli (1983). This can reduce uninformative trials when model assumptions approximate the true observer. The prior, possible threshold range, fixed slope, lapse assumptions, and resulting posterior are substantive elements of the method. A narrow and incorrect prior may give an impression of precision while delaying discovery of the actual performance region.

Methods aimed at estimating multiple parameters allocate trials differently from those concerned only with threshold. Kontsevich & Tyler (1999) considered expected information gain for joint threshold and slope estimation; Watson (2017) generalized adaptive design through QUEST+, which can cover multiple stimulus dimensions, outcome types, and psychometric parameters. Their shared principle is experimental design by anticipated reduction of uncertainty. The optimal trial for learning a slope may lie away from the threshold-crossing region, and the optimal trial for a whole contrast-sensitivity curve may depend on uncertainty at several spatial frequencies. Procedure comparisons consequently require equivalent goals rather than a single count of trials.

Bayesian adaptive inference is only as reliable as its generative likelihood, sampling constraints, and treatment of unexpected behavior. Lapses, nonmonotonic response functions, a change of criterion across blocks, or a drifting threshold can invalidate the fixed-function assumptions that made the method efficient. An auditable protocol preserves the full posterior or sufficient data to reconstruct it, records each selected stimulus and observer response, and checks uncertainty calibration with simulations. Wichmann & Hill (2001b) motivates interval assessment, while Prins & Kingdom (2018) demonstrates the broader utility of contrasting hypotheses under explicit psychometric models.

Efficiency, Precision, and the Trial Budget

Method efficiency has several nonidentical meanings: mean squared error of the estimated threshold, systematic bias, posterior interval width, time per trial, resistance to lapses, and repeatability between sessions. Two procedures may rank differently on these criteria. Watson & Fitzhugh (1990) disputed Simpson (1988) on the relative efficiency of constant stimuli, partly because simulation assumptions determined the apparent winner. A scientifically useful comparison specifies the underlying psychometric family, allowed trial count, response noise, target performance, and risk of nonstationarity. Trial economy is valuable when subjects fatigue, but an extremely short experiment can return unstable individual estimates.

One primary visual study, Kontsevich & Tyler (1999), reported that under its typical task assumptions a threshold estimate could reach specified precision in substantially fewer trials than a slope estimate. This contrast matters because some manuscripts present a single well-estimated threshold as evidence that the entire psychometric function is known. Parameter covariance can make changes in slope or lapse rate difficult to distinguish from displacement at limited sample sizes. Wichmann & Hill (2001a) used simulation to show how ignoring stimulus-independent lapses biases fitting and how conventional asymptotic goodness-of-fit tests can misbehave with small psychophysical datasets.

An adaptive protocol should therefore be evaluated with simulated observers whose parameters span plausible experimental variation, and then checked against empirical test–retest data. Strong methods report errors and confidence or credible intervals for individual thresholds, not merely group means. Hierarchical inference may improve precision through pooling across repeated measurements, but pooled estimates embed assumptions about across-observer similarity. Zhao et al. (2021) and Zhao et al. (2024) illustrate modern hierarchical approaches to contrast sensitivity. Efficiency is ultimately a reproducibility and decision-quality problem: saving trials is valuable insofar as the scientific claim remains well identified.

Internal Noise, External Noise, and Ideal Observer Analysis

Discrimination depends on the relationship between signal variation and noise, including noise that originates in the environment, display, sensory apparatus, and observer's own response process. An external-noise manipulation adds controlled variability to the stimulus, permitting researchers to examine how performance changes when information becomes less reliable. Internal noise is inferred from behavior under a model; it cannot be read directly from a threshold alone. Ideal-observer analysis defines the best achievable performance for specified task information and decision assumptions Geisler (2011). Comparing an actual observer with that bound helps localize inefficiency while leaving open which biological processes contribute to the gap.

The relationship between external noise and a fitted threshold is diagnostic only when the external stimulus ensemble and expected optimal computation are specified. Adding spatial uncertainty may change the number of plausible signal locations, the strategy of allocating attention, or the task's computational burden. In a major visual treatment, Pelli (1985) developed an uncertainty model for contrast detection and discrimination that related a family of performance effects to the number of plausible signals considered by the observer. Its value lies in generating coordinated predictions rather than in establishing uncertainty as the sole explanation of contrast response nonlinearities.

An observer can be limited by additive variability, multiplicative variability, stimulus uncertainty, sensory saturation, or shifts in criterion; several different models may fit one threshold function. Factorial manipulation of mask intensity, signal location, task instruction, and baseline strength increases model discrimination. The empirical standard is to compare entire response functions and predicted interactions, rather than inferring a noise source from a single point on a curve. The formal caution in Prins & Kingdom (2018) and the ideal-observer approach of Geisler (2011) together encourage testing mechanistic alternatives with held-out or explicitly constrained predictions.

Contrast Detection, Pedestals, and Dip-Shape Functions

Visual contrast discrimination often measures the smallest increment in contrast distinguishable from a baseline or pedestal grating. Across pedestal levels, increment thresholds can produce a nonmonotonic profile: at some near-threshold pedestals the increment threshold falls below the absolute detection threshold, whereas at stronger pedestals it rises. The influential report by Nachmias & Sansbury (1974) made this relationship a major problem for simple stimulus-response pictures. A single monotonic threshold law cannot by itself account for the interaction of baseline contrast, increment, and task uncertainty. Interpretation requires the full threshold-versus-contrast function, plus carefully matched detection and discrimination procedures.

Several explanatory families are relevant. Nonlinear transducer accounts allow response gain to change with stimulus contrast; masking models incorporate excitatory and inhibitory interactions; uncertainty accounts examine candidate signals and decision distributions. Foley (1994) directly tested predictions of a widely used masking model and found empirical patterns that motivated a model including broadband divisive inhibition. Pelli (1985) provided a competing and complementary perspective grounded in signal uncertainty. These studies show that phenomenologically similar threshold curves can correspond to different hypothesized mechanisms, and that perturbing masker structure is more informative than fitting only an average increment threshold.

Contrast measurements require explicit geometry, mean luminance, temporal envelope, grating orientation, spatial frequency, retinal location, eye state, and display calibration. A pedestal effect may reflect local gain control, uncertainty reduction through a visible reference, temporal interactions, or an interaction between them. Models should generate predictions for several pedestal and mask combinations and be tested against uncertainty in each parameter. The psychometric slope and upper-asymptote terms are also relevant because a threshold shift can be partly statistical. The methods in Wichmann & Hill (2001a) support distinguishing meaningful shape differences from unstable threshold-crossing estimates.

Weber-Like Scaling, Gain Control, and Reference Dependence

Many sensory systems exhibit approximate proportional discrimination over a restricted intensity range, motivating Weber fractions as convenient descriptive summaries. A stable fraction can follow from a range of computational and neural architectures, including changing response gain, normalized coding, and multiplicative noise. In visual experiments, contrast discrimination may depart from simple proportionality near absolute detection and under masking Nachmias & Sansbury (1974) Foley (1994). Researchers should therefore treat a fitted Weber fraction as a conditional summary rather than as direct evidence for one receptor mechanism. The appropriate intensity transformation and local reference must be declared.

A roving-reference task changes the standard from trial to trial, which can reduce memorized category cues while adding working-memory and comparison demands. A fixed-reference design makes the comparison more stable but may allow learning of a sensory template or response criterion. These choices alter how the difference threshold relates to representations of standard and test values. For very small increments, assumptions about the equivalence of upward and downward discrimination should also be tested rather than presumed. Kingdom & Prins (2016) describes practical psychophysical distinctions, and Klein (2001) situates their statistical consequences in a broader analysis of perceptual performance.

Reference dependence connects psychophysical discrimination with adaptation and normalization. When recent stimuli shift sensitivity or decision weighting, the same physical test may produce different response probabilities. A strong design uses counterbalanced baselines, monitors stimulus history, and reports whether the theoretical object is an immediate local JND or a general performance curve across contexts. The resulting synthesis views Weber-like regularity as an empirical constraint on models. Mechanistic explanation requires interventions that separate gain change from altered noise and decisional evidence, consistent with the contrasts between Pelli (1985) and Foley (1994).

Modality-Specific Thresholds and Comparability

Thresholds in vision, hearing, touch, taste, and olfaction are all psychophysical estimates, yet their physical variables and task demands differ substantially. Visual contrast sensitivity varies with spatial frequency, luminance, eccentricity, adaptation, and temporal conditions; auditory level discrimination depends on spectral, temporal, and masking conditions. Tactile discrimination varies with receptive field geometry, contact force, skin site, and movement. Thresholds obtained with a convenient stimulus are therefore samples from a multidimensional performance space. The practical overview of Kingdom & Prins (2016) describes methods designed for distinct modalities without suggesting that the same numeric scale directly compares their sensory abilities.

The value of an integrated sensory methodology is to make task logic comparable while preserving modality-specific measurement. One can compare the reliability of a two-interval procedure, psychometric model, or adaptive rule across modalities, but interpretation of a physical threshold must be tied to the stimulus-generating apparatus. In visual applications, Lesmes et al. (2010) used a Bayesian method to estimate an entire contrast-sensitivity function more efficiently than measuring each frequency independently under the evaluated design. That experiment illustrates the benefits of treating stimulus frequency and contrast jointly instead of aggregating heterogeneous sensitivity values into one number.

Absolute limits are also conditioned by dynamic range and physical calibration. A laboratory monitor may display nominal contrast values that differ from emitted contrast; headphones may change their effective output with fit or transducer characteristics; tactile devices may have mechanical nonlinearity. Uncalibrated physical stimulation contaminates the psychological inference before any statistical fitting begins. Explicit device verification, environmental controls, and reference-condition documentation are integral components of construct validity. A multilevel comparison should state the physical scale, psychophysical criterion, expected measurement uncertainty, and practical functional meaning separately Klein (2001) Zhao et al. (2024).

Attention, Adaptation, Learning, and Threshold Instability

A threshold measured at the beginning of a session may differ from one measured after sustained exposure or repeated practice. Attention influences which evidence receives priority, while adaptation changes the relation between current and recent stimulation. Learning may improve a particular discrimination through changes in sensory weighting, decision strategy, or familiarity with task contingencies. A threshold trace observed across trials is therefore a time-dependent series potentially influenced by multiple processes. When the design assumes a stationary psychometric function, all such changes become potential model violations. Treutwein (1995) notes the historical variety of adaptive methods whose behavior must be interpreted under the actual response dynamics.

The distinction between learning and estimator convergence is particularly important. A staircase that descends toward weaker stimuli may appear to show improving performance even if the observer remains unchanged, because the algorithm is finding an unknown transition region. Conversely, a changing participant can make a stable adaptive posterior misleading. Researchers can separate these interpretations by monitoring performance at fixed probe stimuli, comparing independent tracks, using randomized anchors, and estimating time-varying parameters when warranted. Wichmann & Hill (2001a) demonstrates why lapse handling matters; Kontsevich & Tyler (1999) shows that parameter estimation and stimulus selection are intertwined.

Individual differences may involve peripheral sensitivity, attention control, health conditions, task comprehension, or response motivation. A group average can conceal multimodal or highly variable individual estimates. Test–retest designs and repeated conditions are indispensable before assigning stable sensory trait meaning to a threshold. Hierarchical models such as Zhao et al. (2021) can express variation across subjects and conditions, but their partial pooling introduces assumptions about common population structure. A comprehensive threshold interpretation accordingly reports within-person variability, possible practice trends, and measurement uncertainty instead of treating the earliest observed crossing as an immutable characteristic.

Auditory Intensity Difference Limens and Level Dependence

Auditory intensity discrimination asks how much the acoustic level of a comparison must change before listeners can reliably select the stronger of two stimuli. The physical reference can be specified in sound pressure or decibels, while the experimental comparison concerns the resulting probability of identifying a difference. These choices matter because intensity increments depend on reference level, spectral content, duration, background noise, and presentation order. The original experiments of Jesteadt et al. (1977) measured intensity discrimination for pulsed sinusoids across multiple frequencies and sensation levels; the article's sampling of these conditions is an important example of a threshold function rather than one general auditory JND. Its registered bibliographic record confirms the controlled dimensions, and modern interpretation should preserve the task context instead of converting every difference limen into a fixed auditory constant.

An intensity JND and subjective loudness growth address related but separable questions. Discriminability depends on the distribution and comparison of evidence whereas loudness functions summarize perceived magnitude under a scaling task. Hellman et al. (1987) tested the assumption that intensity discrimination could be derived from the slope of a loudness function by arranging pure tones in different noise backgrounds. Its abstract reports nearly similar intensity JNDs at a crossing of markedly different loudness functions. That result qualifies simple derivations of difference limens from magnitude scaling and calls for explicit modeling of noise, comparison, and task dependence.

In auditory work, the definition of sensation level also needs care: a pure-tone reference set some number of decibels above an individual's absolute threshold is not equivalent to a fixed sound pressure across listeners. Psychoacoustic laboratories must control earphone coupling, temporal gating, frequency response, listener training, and hearing status. Changes in reference level can affect the information available in a trial, but the observed direction and size of any improvement should be reported from data rather than assumed from Weber's approximation. Jesteadt et al. (1977) and Green & Swets (1966) support analyzing the entire threshold procedure and response contingencies before attributing an acoustic discrimination effect to peripheral transduction.

Auditory Frequency Discrimination and the Difference Between Dimensions

Frequency discrimination asks whether two sounds differ in pitch-related physical frequency under an experimentally defined comparison rule. It is a different construct from intensity discrimination, although both can be studied with two-interval judgments and adaptive stimulus selection. The original experiment by Wier et al. (1977) varied frequencies across the audible range and sensation levels in a forced-choice design, demonstrating the need to document both the reference tone and its audibility context. The bibliographic abstract identifies its controlled dimensions; the present synthesis does not infer unreported numerical thresholds or a universal monotonic relationship from the accessible summary.

Frequency JNDs depend on the characteristics of the acoustic signal, including tone duration, harmonics, temporal envelope, and background conditions. They may also change when observers employ learned cues or when physical level differences inadvertently correlate with frequency. To interpret a frequency result as frequency sensitivity, the protocol must randomize or control non-target dimensions and describe calibration. A threshold is thus attached to the specific discriminandum and available comparison information. The parallel studies Wier et al. (1977) and Jesteadt et al. (1977) help disentangle the shared method from the different sensory information that listeners must evaluate.

An ideal comparative program would estimate frequency and intensity discrimination in the same observers, across appropriately matched conditions, and test whether their within-person variation is explained by shared or dimension-specific factors. Test–retest data and trial-level response patterns would separate stable ability from nuisance variability, while mechanistic studies could manipulate temporal information or masking. Although a common psychometric function can summarize both auditory tasks, its slope and threshold should retain the proper units and experimental definition. Klein (2001) provides the general statistical rationale for this discipline; Wier et al. (1977) grounds it in a dedicated psychoacoustic research tradition.

Tactile Spatial Acuity and the Two-Point Threshold

The two-point threshold has historically been used to assess tactile spatial resolution by asking when two nearby skin contacts are reported as distinct. This task is sensitive to the criterion for responding 'two,' the details of stimulus placement, differences between one- and two-point contact patterns, skin site, applied force, and practice. Craig & Johnson (2000) reviewed these problems and documented the possibility of distinguishing two contacts from one at separations smaller than the traditional two-point threshold. Consequently the distance where a person begins producing a certain proportion of 'two' reports is an operational decision threshold, not automatically an invariant measure of spatial receptor resolution.

A stronger tactile experiment can compare alternative response tasks and exploit patterns that force judgments about spatial structure rather than merely the number of contacts. Randomized orientation discrimination, localization, and carefully controlled patterns can probe spatial acuity under more explicitly specified decision constraints. The review by Craig & Johnson (2000) identifies criteria and within- and between-person variability as serious interpretation problems and discusses the relation between psychophysics and neurophysiology. Different measures may rank skin locations or participants differently, which means a single generic tactile acuity score requires explicit construct validation before use.

The scientific lesson extends beyond touch. Any sensory threshold can conflate access to evidence with the categorical report used to reveal that evidence. Two-point tests make the confounding especially visible because the physical manipulation changes more than separation alone if force and number of contacts are not equated. Functional tactile abilities in object handling or accessibility interfaces will often require pattern recognition and active exploration in addition to passive spatial separation. This is consistent with the distinction between experimental threshold and practical behavior developed in Kingdom & Prins (2016), and with the methodological caution in Craig & Johnson (2000).

Vibrotactile Frequency Channels and Experimental Masking

Vibrotactile sensitivity can be measured by applying periodic mechanical stimulation to skin and estimating detection or discrimination as a function of frequency, location, contactor geometry, and adaptation. The physical stimulus must be described by displacement or force and the mechanical conditions at the contact surface. Gescheider et al. (2004) reviewed and updated a model of multiple information-processing channels of glabrous skin using forward masking and adaptation tuning curves. The accessible abstract describes distinct tuning characteristics associated with peripheral receptors and afferents, providing a mechanistic basis for examining differences across stimulus frequencies.

Masking creates an experimental probe of interactions among responses to test and preceding stimulation. If a masker selectively increases a threshold for some vibrotactile frequencies or conditions, the resulting curve constrains hypotheses about channel tuning and adaptation. Such a result is still measured behavior under a specified stimulus history; attributing it to an exclusive neural pathway requires physiological support. Gescheider et al. (2004) explicitly links the psychophysical model to peripheral receptor and nerve-fiber properties. Its value for threshold theory is the use of a multidimensional pattern of results rather than one sensitivity limit measured at an arbitrary frequency.

A related psychophysical study of hairy forearm skin Bolanowski et al. (1994) manipulated probe area and temperature and reported evidence consistent with multiple distinct information-processing channels. Because hairy and glabrous skin differ physiologically, numerical thresholds and channel assignments should not be generalized across skin types without comparable tests. This work also shows how stimulus temperature and contact geometry can affect measured sensitivity. Tactile psychophysics therefore requires mechanical and anatomical descriptions that play the role of spectral calibration in auditory experiments and luminance calibration in visual experiments. The physiological explanation must be evaluated against the full threshold-by-frequency pattern.

Fechner's Original Quantitative Program and Its Contemporary Limits

Fechner's two-volume Fechner (1860) established an early systematic relationship among physical stimulation, discrimination increments, and proposed measurements of sensation. The original 1860 edition is cataloged and digitally accessible at the ETH-Bibliothek and Wellcome Collection, which independently confirm its authorship, year, publisher, and two-volume publication. This primary historical record anchors the chronology of psychophysical research. The present monograph uses the bibliographically documented work as a foundational historical object while reserving detailed analysis of its original German chapters and textual variants for a dedicated historical dossier.

The conceptual move from a just-noticeable difference to a scale of experienced intensity requires assumptions about how comparison steps accumulate. Classical logarithmic interpretation depends on treating differential increments as meaningful units for an integrated quantity. Modern experiments challenge researchers to distinguish the empirical regularity of discrimination over a given range from the theory that produced its mathematical integration. A sensory code may show relative sensitivity without subjective magnitude following the same mathematical function in all contexts. The later quantitative tools of Green & Swets (1966), Klein (2001), and Pelli (1985) give independent methods for distinguishing response criteria, internal noise, and uncertainty from a putative scale of experience.

Historical continuity is best established through explicit conceptual and methodological lineages: nineteenth-century observers manipulated stimulus quantities and reported detection or difference; twentieth-century psychophysicists modeled variable trial responses and sensory uncertainty; contemporary researchers optimize stimulus placement and assess uncertainty of fitted response functions. The physical stimulus, the criterion, and the statistical estimator were progressively differentiated. Fechner's historical importance resides in making sensations a subject of quantitative experiment, while the limits of any one law are evaluated by the predictive behavior of modern evidence-based models. Fechner (1860) serves as a primary anchor to this trajectory, with Watson (2017) representing one later adaptive development.

Identification, Fitting, and Parameter Uncertainty

A threshold parameter is identified when different plausible parameter combinations yield distinguishable predictions for the actual trial design. Sparse sampling near one performance level may estimate a crossing point adequately but leave slope, lapse, and lower asymptote weakly identified. In an adaptive experiment, this problem can be hidden because stimuli are deliberately concentrated around a presumed transition. Wichmann & Hill (2001a) demonstrated through simulation how stimulus-independent errors affect maximum-likelihood fitting, and Wichmann & Hill (2001b) examined interval estimation. A complete report needs to show uncertainty about the threshold together with the fitted psychometric shape and observations that constrain it.

Likelihood methods require a stated response distribution and link function. For binary responses aggregated by stimulus intensity, a binomial likelihood may be appropriate if trials are conditionally independent; serial correlation, state changes, and overdispersion call for richer alternatives. The choice between logistic, Weibull, cumulative normal, or another family can influence inferred thresholds outside densely observed regions. Constraint bounds and treatment of chance performance likewise matter. Prins & Kingdom (2018) urges using formal model comparison for theoretically specific alternatives rather than interpreting a visually appealing fitted curve as mechanistic confirmation.

A model with many nuisance parameters can provide flexible fit but uncertain science if the experimental design does not constrain those parameters. Conversely, fixing slope or lapse at convenient values can create spuriously narrow threshold intervals when the assumptions are false. Posterior sensitivity analysis, parameter recovery, simulated observer tests, and residual diagnostics help identify these problems. Kontsevich & Tyler (1999) and Watson (2017) illustrate how design optimization depends on an assumed parameter space. Good inference reports which quantities were measured, which were assumed, and which remain weakly identified under the available trials.

Reproducibility, Calibration, and Psychophysical Quality Control

Reproducibility begins with the physical stimulus. For visual experiments, documentation should include display calibration, gamma correction, frame timing, luminance range, viewing distance, fixation and optical conditions. Auditory experiments require acoustic level, frequency response, headphone or speaker configuration, background noise, and temporal precision. For tactile and other modalities, force calibration, contact mechanics, preparation, and environmental variation may be decisive. The scientific question of a JND can only be answered after specifying the actual physical contrast experienced by the participant, not merely the software's requested parameter Kingdom & Prins (2016).

At the behavioral level, reproducibility requires the instructions, block structure, feedback, trial count, randomization policy, exclusion criteria, and exact stopping rule. Exploratory choices made after seeing an outcome can change the apparent effect of a manipulation. Replication should compare the same threshold criterion rather than only similarly named variables. Confidence intervals or posterior distributions should be archived with trial-level data whenever permitted. Wichmann & Hill (2001b) offers tools for resampling-based uncertainty, while García-Pérez (1998) shows that small-sample staircase behavior may diverge from its nominal asymptotic target.

Independent verification can include simulated response recovery and benchmark conditions with known or expected sensitivity ordering. Equipment drift and observer training should be checked on the timescale of data collection. In multicenter studies, site differences in hardware and instructions can dominate subtle population contrasts. Hierarchical modeling is a response to such structured variance, although it does not replace calibration. Zhao et al. (2021) and Zhao et al. (2024) represent contemporary work on analyzing contrast sensitivity across repeated or grouped measurements. Reliable psychophysics depends on transparent provenance from physical intensity through response records to final fitted estimates.

Study Comparisons and the Limits of Generalization

Apparent agreement between two studies may be superficial if one defines threshold by percent correct in two-alternative forced choice and another uses a yes–no report level. Even matched numerical criteria can imply different sensitivity when lapse rates, chance performance, and decision processes differ. Cross-study synthesis should map physical unit, stimulus reference, observer sample, task arrangement, and statistical estimand before comparing effects. Green & Swets (1966) and Klein (2001) give complementary conceptual and methodological bases for such harmonization. Without that mapping, a meta-analytic combination of nominal threshold numbers would risk mixing unlike constructs.

Sample composition matters because vision, hearing, and other modalities vary with age, clinical state, training, correction, and environment. Many foundational psychophysical experiments use modest, highly practiced samples whose precision within a specialized task may be excellent. Those results can be mechanistically revealing without establishing population norms or clinical screening performance. The clinical transfer of adaptive contrast sensitivity procedures therefore requires external validation and assessment of test–retest variability. Lesmes et al. (2010) explicitly discussed wider clinical validation as a future concern, and Zhao et al. (2024) addressed improved statistical inference for clinical-trial measurements.

Null results require interpretation as carefully as positive differences. A nonsignificant threshold shift may reflect a well-powered constraint on effect size, a noisy estimate, task insensitivity, or conflicting direction-specific effects. The scientific report should supply confidence intervals, plausible equivalence margins when used, and a reasoned account of generalizability. Studies of contrast masking Foley (1994) and uncertainty Pelli (1985) highlight why a strong explanatory test manipulates conditions expected to differentiate models. Cross-study synthesis becomes more informative when it compares these patterns and boundary conditions rather than counting the number of significant threshold comparisons.

The Role of Hierarchical Models and Contemporary Methods

Modern psychophysical analysis can estimate a curve across stimulus conditions while borrowing strength across repeated measurements. Hierarchical Bayesian models allow individual threshold functions to vary around a structured population distribution, potentially stabilizing estimates when single-session data are sparse. Zhao et al. (2021) applied such modeling to contrast sensitivity functions within a repeated-measures design. The gain in precision derives partly from shared structure, so the population assumptions and sensitivity to priors should be made visible. An individual result shrunk toward a group estimate is not identical to an independently measured subject-specific threshold.

A later application by Zhao et al. (2024) examined nonparametric hierarchical modeling of contrast sensitivity to improve statistical inference in settings including clinical trials. This expands the methodological vocabulary beyond reporting a threshold at a single spatial frequency. Yet the greater flexibility of an inferential system adds requirements for predictive diagnostics, identification checks, and reproducible code. Posterior uncertainty must be distinguished from uncertainty in stimulus generation and systematic bias. The most elaborate analysis cannot infer a reliable biological difference from inconsistently calibrated monitors or unbalanced experimental tasks.

Modern experimental design also unites adaptive placement with computational checks of parameter recovery. Watson (2017) showed how multidimensional Bayesian testing can select informative stimuli across several parameter types, while Prins & Kingdom (2018) explains how targeted hypotheses can be evaluated by comparing nested or competing response models. These methods enable a stronger program in which experimental variables are selected because their predicted effects differ among mechanisms. The distinctive contribution of contemporary practice is therefore increased explicitness about the data-generating model, decision uncertainty, and alternatives, with an accompanying obligation to audit assumptions.

Critical Evidence Appraisal and Theoretical Disagreement

The available literature supports several robust methodological propositions. Detection and discrimination are probabilistic in ordinary psychophysical measurement; thresholds depend on their declared criteria; adaptive methods allocate stimuli using observer responses; lapse and slope assumptions can bias fitted thresholds; and contrast increments reveal interactions that a scalar threshold cannot summarize Wichmann & Hill (2001a) Levitt (1971) Nachmias & Sansbury (1974). These findings are supported by original experiments, simulation studies, and formal analysis. Their precise magnitude and external validity differ across tasks, and they do not authorize treating a generic threshold as a direct readout of peripheral sensitivity.

Other claims are more model dependent. Divisive inhibition, nonlinear transduction, template uncertainty, sensory noise, and criterion shifts may produce related changes in psychometric curves. Pelli (1985) and Foley (1994) exemplify why alternative quantitative explanations remain relevant even when a model fits broad patterns. Direct causal localization requires interventions, time-resolved observations, neural measurements, or experimental manipulations that generate divergent predictions. Inferred noise parameters should not be reified as measured neural variance without additional evidence. The same caution applies to claims that an efficient algorithm automatically delivers a psychologically pure sensitivity estimate.

A critical synthesis should also recognize that classical methods have not been rendered obsolete by Bayesian allocation. Constant stimuli can expose irregular response functions, adaptive procedures can improve precision under good assumptions, and forced choice can reduce some criterion influences while introducing comparison demands. Simpson (1988) and Watson & Fitzhugh (1990) illustrate the consequences of differing simulation assumptions; García-Pérez (1998) demonstrates the finite-sample behavior that must be evaluated even for established staircases. The best design follows from the hypothesis, likely observer behavior, available trials, and intended generalization. This methodological pluralism is an experimentally testable stance grounded in explicit error criteria.

Accessibility, Screening, and Functional Interpretation

Sensory threshold measurement has practical significance for visual and auditory assessment, accessible interface design, and evaluation of environmental signals. A threshold obtained under ideal laboratory conditions describes the probability of an experimental decision at a specified stimulus level. Functional access to a sign, warning, subtitle, or acoustic alert also depends on clutter, divided attention, motor response, context, and the consequences of missing an event. The empirical objective should match the application: design verification may require a range of stimulus conditions and a reliable detection rate rather than a single averaged JND. The multidimensional contrast approach of Lesmes et al. (2010) helps illustrate this principle.

Clinical measurement requires additional controls, including validated norms, test–retest repeatability, age-appropriate tasks, sensory correction, and appropriate referral or interpretation procedures. A threshold inferred from an experimental monitor or short staircase is insufficient by itself for diagnosing disease. Research on hierarchical contrast sensitivity analysis Zhao et al. (2024) addresses one component of the problem—estimating uncertainty within an inferential model—but device reliability, norms, and clinical utility remain separate validation stages. Statements about patient care should rely on current professional evidence rather than direct conversion of psychophysical scores into treatment claims.

Applied psychophysics benefits when an experiment reports threshold alongside the slope of performance improvement, lapse rate, and a function over relevant operating conditions. An accessibility system may be usable even when a reference task returns a high nominal threshold if redundancy, multimodal cues, and task learning compensate. The opposite can also occur: a low laboratory threshold does not guarantee that a cue is noticed in distraction. A research program assessing actual function therefore combines psychophysical tests with ecologically valid tasks and records how users of different ages and abilities encounter stimuli Kingdom & Prins (2016) Geisler (2011).

Human–Artificial Measurement and Instrumental Comparisons

An artificial sensory system can be evaluated with detection and discrimination tasks analogous to those used in psychophysics, provided the target variable, sensor apparatus, noise ensemble, and decision outputs are operationally defined. A machine-vision classifier may have a contrast threshold at a chosen accuracy criterion, and a robotic auditory system may have a frequency-discrimination limit under controlled noise. These values characterize the tested model–hardware–task configuration. They do not imply that an artificial system experiences sensation in the human sense. The relevant scientific connection is comparability of input-output evidence and decisions, with ideal-observer benchmarks offering a common analytic vocabulary Geisler (2011).

Computational psychophysics can test how performance changes as a stimulus parameter varies, how stable an error function is under distribution shift, and how a system allocates uncertain evidence to alternatives. Threshold estimates are sensitive to preprocessing, training data, inference randomness, input discretization, and decision policy. A comparison with human observers must match available information and task constraints rather than merely compare headline percentage-correct values. Watson (2017) provides a general adaptive-testing framework that can inspire experimental sampling designs, although its psychometric assumptions need explicit testing for systems with discontinuous or nonmonotonic response profiles.

Within the Artificial Era research horizon, sensory discrimination is one possible dimension of comparative functional organization. Empirical claims concern measurable system behavior; broader philosophical interpretation belongs to the associated conceptual literature. New research can examine whether artificial response surfaces exhibit Weber-like regions, how output variance changes with external noise, and whether adaptive thresholds remain stable after model revisions or changes in sensor calibration. Such hypotheses are testable with controlled datasets, replicated inference runs, and fully disclosed criteria. A measured machine threshold is an instrumental property of the evaluated system, and its significance follows from the experiment rather than a presumed equivalence of biological and artificial mechanisms.

Research Design Recommendations and Open Problems

Future studies should estimate threshold uncertainty jointly with the behavior of the observer. Researchers can preregister the intended criterion, record trial-level physical stimulus values, include conditions that estimate lapse behavior, and simulate the proposed adaptive design under plausible nonstationarity. If a claim concerns the transducer mechanism, its design should manipulate pedestal levels, external noise, or mask structure to discriminate quantitative alternatives. If a claim concerns human function, test conditions should approximate the demands of real-world detection. These principles follow from methodological tensions identified by Wichmann & Hill (2001a), García-Pérez (1998), Pelli (1985), and Foley (1994).

Several scientific problems remain open across tasks. How should an experiment distinguish a genuinely changing sensory representation from strategic adaptation of decision rules? How can multiple stimulus dimensions be sampled efficiently when task responses depend on attention and fatigue? How robust are hierarchical threshold estimates when a study includes rare sensory deficits or multimodal participant distributions? And how should local JNDs be connected with continuous behavior in complex environments? The adaptive frameworks of Watson (2017) and the hierarchical studies of Zhao et al. (2021) and Zhao et al. (2024) provide workable tools, but each question also requires targeted intervention and independent validation.

Progress depends on separating identifiable scientific claims. Prediction of a psychometric curve can succeed while the proposed mechanism remains underdetermined; accurate threshold estimation can coexist with poor out-of-sample functional prediction; statistically detectable group shifts may have little practical magnitude. A strong experiment reports uncertainty in every relevant unit and identifies the inference level at which each result is supported. Collaborative open measurement protocols, shared stimulus calibration, and the preservation of raw response sequences can improve reproducibility. The critical standard is explicit correspondence among construct, experiment, fitted parameter, and claimed generalization Prins & Kingdom (2018) Klein (2001).

Integrative Scientific Synthesis

The central empirical insight of sensory psychophysics is that discriminability is measured through behavior under controlled physical stimulation, and the resulting threshold belongs to a defined measurement procedure. Absolute threshold, difference threshold, Weber fraction, psychometric slope, and response criterion answer different scientific questions. Their relations can illuminate sensory coding only when experimenters document the conversion from physical input to response probability. Classic psychophysical laws supplied the motivation for quantitative research, while signal detection and modern response modeling exposed the uncertainty and decision variables hidden behind apparent boundaries Green & Swets (1966) Kingdom & Prins (2016).

The combination of original contrast experiments and modern methodology establishes a distinctive explanatory structure. Near-threshold visual discrimination is influenced by baseline stimulation, masker conditions, and candidate-signal uncertainty; it can require models of nonlinear gain or noise rather than one fixed sensory limit Nachmias & Sansbury (1974) Pelli (1985) Foley (1994). Adaptive algorithms can reduce the cost of data acquisition, but their results depend on convergence rules, priors, lapses, finite samples, and observer dynamics Levitt (1971) García-Pérez (1998) Watson (2017). Consequently, the measured function and the inferential assumptions should be reported as inseparable components of the result.

A coherent research synthesis joins four levels: physical stimulus calibration, behavioral task architecture, statistical inference, and psychological or physiological explanation. Evidence grows stronger when several paradigms converge under common predictions, when independent replications preserve conditions, and when models survive tests designed to separate their mechanisms. No single threshold estimate can substitute for that evidential chain. The resulting framework makes sensory discrimination research useful across basic perception, methodological science, accessibility design, and comparative assessment of artificial sensory systems, while preserving the distinction between documented performance and philosophical interpretation Geisler (2011) Prins & Kingdom (2018).

Conclusion

Sensory thresholds and discrimination form an experimentally mature field whose core object is the relation between graded stimulation, uncertain sensory evidence, and task-specific decisions. Its historical contribution is the quantification of perception; its modern contribution is a family of reproducible methods for estimating response functions and testing models of sensitivity. Absolute detection and difference discrimination require explicit criteria and reference conditions. Their numerical estimates derive meaning from the experimental arrangement, not merely from the physical units in which they are expressed Green & Swets (1966) Klein (2001).

The evidence shows that threshold measurement is sensitive to task design, response criterion, internal and external variability, reference contrast, uncertainty, lapses, and the assumptions of adaptive estimation. Empirical debates over visual contrast and the efficiency of measurement procedures show why quantitative predictions across conditions offer stronger scientific tests than isolated threshold values. Contemporary Bayesian and hierarchical methods increase opportunities for informative stimulus allocation and uncertainty estimation while raising corresponding obligations to validate their assumptions Pelli (1985) Foley (1994) Watson (2017) Zhao et al. (2024).

The fundamental scientific standard emerging from this research is construct-to-measure correspondence. An investigator states precisely which sensory distinction is being studied, builds a calibrated task that observes it, estimates the corresponding psychometric quantities with defensible uncertainty, and tests competing explanations through independently informative manipulations. This standard connects nineteenth-century quantification with current computational measurement and creates a durable basis for research across biological and artificial sensing. The next stage of progress lies in stronger cross-paradigm identification, test–retest reliability, ecological transport, and transparent open evidence Wichmann & Hill (2001b) Geisler (2011) Prins & Kingdom (2018).

References

  • Bolanowski, S. J., Gescheider, G. A., & Verrillo, R. T. (1994). Hairy skin: Psychophysical channels and their physiological substrates. Somatosensory & Motor Research, 11(3), 279–290. https://doi.org/10.3109/08990229409051395
  • Craig, J. C., & Johnson, K. O. (2000). The two-point threshold: Not a measure of tactile spatial resolution. Current Directions in Psychological Science, 9(1), 29–32. https://doi.org/10.1111/1467-8721.00054
  • Fechner, G. T. (1860). Elemente der Psychophysik (2 vols.). Breitkopf und Härtel. https://doi.org/10.3931/e-rara-10879
  • Foley, J. M. (1994). Human luminance pattern-vision mechanisms: Masking experiments require a new model. Journal of the Optical Society of America A, 11(6), 1710–1719. https://doi.org/10.1364/JOSAA.11.001710
  • García-Pérez, M. A. (1998). Forced-choice staircases with fixed step sizes: Asymptotic and small-sample properties. Vision Research, 38(12), 1861–1881. https://doi.org/10.1016/S0042-6989(97)00340-4
  • Geisler, W. S. (2011). Contributions of ideal observer theory to vision research. Vision Research, 51(7), 771–781. https://doi.org/10.1016/j.visres.2010.09.027
  • Gescheider, G. A., Bolanowski, S. J., & Verrillo, R. T. (2004). Some characteristics of tactile channels. Behavioural Brain Research, 148(1–2), 35–40. https://doi.org/10.1016/S0166-4328(03)00177-3
  • Green, D. M., & Swets, J. A. (1966). Signal detection theory and psychophysics. Wiley. https://search.worldcat.org/title/890266
  • Hellman, R., Scharf, B., Teghtsoonian, M., & Teghtsoonian, R. (1987). On the relation between the growth of loudness and the discrimination of intensity for pure tones. Journal of the Acoustical Society of America, 82(2), 448–453. https://doi.org/10.1121/1.395445
  • Jesteadt, W., Wier, C. C., & Green, D. M. (1977). Intensity discrimination as a function of frequency and sensation level. Journal of the Acoustical Society of America, 61(1), 169–177. https://doi.org/10.1121/1.381278
  • Kingdom, F. A. A., & Prins, N. (2016). Psychophysics: A practical introduction (2nd ed.). Academic Press. https://doi.org/10.1016/C2012-0-01278-1
  • Klein, S. A. (2001). Measuring, estimating, and understanding the psychometric function: A commentary. Perception & Psychophysics, 63(8), 1421–1455. https://doi.org/10.3758/BF03194552
  • Kontsevich, L. L., & Tyler, C. W. (1999). Bayesian adaptive estimation of psychometric slope and threshold. Vision Research, 39(16), 2729–2737. https://doi.org/10.1016/S0042-6989(98)00285-5
  • Lesmes, L. A., Lu, Z.-L., Baek, J., & Albright, T. D. (2010). Bayesian adaptive estimation of the contrast sensitivity function: The quick CSF method. Journal of Vision, 10(3), Article 17. https://doi.org/10.1167/10.3.17
  • Levitt, H. (1971). Transformed up-down methods in psychoacoustics. Journal of the Acoustical Society of America, 49(2B), 467–477. https://doi.org/10.1121/1.1912375
  • Nachmias, J., & Sansbury, R. V. (1974). Grating contrast: Discrimination may be better than detection. Vision Research, 14(10), 1039–1042. https://doi.org/10.1016/0042-6989(74)90175-8
  • Pelli, D. G. (1985). Uncertainty explains many aspects of visual contrast detection and discrimination. Journal of the Optical Society of America A, 2(9), 1508–1532. https://doi.org/10.1364/JOSAA.2.001508
  • Prins, N., & Kingdom, F. A. A. (2018). Applying the model-comparison approach to test specific research hypotheses in psychophysical research using the Palamedes Toolbox. Frontiers in Psychology, 9, Article 1250. https://doi.org/10.3389/fpsyg.2018.01250
  • Simpson, W. A. (1988). The method of constant stimuli is efficient. Perception & Psychophysics, 44(5), 433–436. https://doi.org/10.3758/BF03210427
  • Treutwein, B. (1995). Adaptive psychophysical procedures. Vision Research, 35(17), 2503–2522. https://doi.org/10.1016/0042-6989(95)00016-X
  • Watson, A. B. (2017). QUEST+: A general multidimensional Bayesian adaptive psychometric method. Journal of Vision, 17(3), Article 10. https://doi.org/10.1167/17.3.10
  • Watson, A. B., & Fitzhugh, A. (1990). The method of constant stimuli is inefficient. Perception & Psychophysics, 47(1), 87–91. https://doi.org/10.3758/BF03208169
  • Watson, A. B., & Pelli, D. G. (1983). QUEST: A Bayesian adaptive psychometric method. Perception & Psychophysics, 33(2), 113–120. https://doi.org/10.3758/BF03202828
  • Wichmann, F. A., & Hill, N. J. (2001a). The psychometric function: I. Fitting, sampling, and goodness of fit. Perception & Psychophysics, 63(8), 1293–1313. https://doi.org/10.3758/BF03194544
  • Wichmann, F. A., & Hill, N. J. (2001b). The psychometric function: II. Bootstrap-based confidence intervals and sampling. Perception & Psychophysics, 63(8), 1314–1329. https://doi.org/10.3758/BF03194545
  • Wier, C. C., Jesteadt, W., & Green, D. M. (1977). Frequency discrimination as a function of frequency and sensation level. Journal of the Acoustical Society of America, 61(1), 178–184. https://doi.org/10.1121/1.381251
  • Zhao, Y., Lesmes, L. A., Dorr, M., & Lu, Z.-L. (2024). Using hierarchical Bayesian modeling to enhance statistical inference on contrast sensitivity. Translational Vision Science & Technology, 13(12), Article 17. https://doi.org/10.1167/tvst.13.12.17
  • Zhao, Y., Lesmes, L. A., Hou, F., & Lu, Z.-L. (2021). Hierarchical Bayesian modeling of contrast sensitivity functions in a within-subject design. Journal of Vision, 21(12), Article 9. https://doi.org/10.1167/jov.21.12.9