RESEARCH DOSSIER
Psychophysical Scaling and Magnitude Estimation
Abstract. Psychophysical scaling investigates the quantitative relations among calibrated physical stimuli, sensory discriminability, experienced intensity, and numerical judgment. This fundamental research monograph reconstructs Weberian discrimination, Fechner's logarithmic proposal, Stevens's power law, direct magnitude estimation and production, and the axiomatic requirements for meaningful subjective ratios. It synthesizes experimental findings concerning stimulus-range effects, category judgment, sequence dependence, and numerical estimation; evaluates signal detection, difference scaling, memory-based comparison, Bayesian inference, and stochastic sensory-code frameworks; and examines measurement invariance, validity, uncertainty, and generalizability. The central result of the synthesis is a structured evidentiary distinction between stimulus-response regularity and an identified sensory mechanism: reliable fitted functions do not alone establish the operations by which observers perceive, compare, and report magnitude. The analysis identifies critical experiments for connecting local discriminability with global estimates, separating perceptual encoding from numerical response, and validating psychophysical laws across methods and populations.
Keywords: psychophysics; psychophysical scaling; magnitude estimation; Weber's law; Fechner's law; Stevens' power law; sensory magnitude; ratio scaling; magnitude production; signal detection theory; category judgment; range-frequency model; difference scaling; experimental calibration; perceptual discriminability; Bayesian estimation; artificial perception
Psychophysical Scaling as a Measurement Problem
Psychophysical scaling studies lawful relations between physically specified stimulus attributes and an observer's judgments of their experienced magnitude. A stimulus intensity is measurable in physical units, whereas a sensation magnitude is a theoretical quantity inferred through behavior. Psychophysical experiments connect these domains by manipulating calibrated stimulation and collecting repeatable discrimination, matching, estimation, or production responses. The inference is conditional on an explicit task: assigning a number to loudness, choosing the brighter patch, and adjusting one light until it appears twice as bright provide different observations. Stevens's account of measurement scales and Fechner's early experimental program established complementary requirements for specifying permissible numerical interpretations and observing systematic behavior Stevens (1946) Fechner (1860).
The core research problem concerns the transformation from stimulation to judgment. A complete account must distinguish external intensity, effective receptor input, sensory encoding, remembered reference, comparison operation, linguistic or numerical response, and measurement error. A power function relating an intensity to a reported number describes an empirical pattern; it does not uniquely locate that pattern in sensory transduction. If two models predict the same ratings but different discrimination thresholds or cross-modal matches, the additional observations help identify the operative mechanisms. Recent statistical models explicitly relate reported magnitudes and discriminability through features of an internal representation and its variability Zhou et al. (2024). The same distinction is necessary whenever psychophysical findings are interpreted as psychological quantities.
This monograph examines classical scaling laws, direct numerical estimation, numerical production, cross-modal matching, category judgment, discriminability, axiomatic ratio claims, and computational models of magnitude. Its evidence base combines original theoretical and experimental studies with methodological analyses and recent model-based investigations. Sensory physiology, consumer judgment, numerical cognition, and artificial sensing enter only where they illuminate the validity or application of scaling. The central criteria are that an investigator can reproduce the stimulus, state the question put to the observer, estimate systematic effects and uncertainty, and specify transformations that preserve the intended interpretation. Practical methodological treatments by Gescheider and by Kingdom and Prins explain why those steps belong together in experimental psychophysics Gescheider (1997) Kingdom & Prins (2016).
From Weberian Discrimination to Fechnerian Measurement
Classical psychophysics grew from experiments on discriminating changes in weight, brightness, and other continua. Weberian observation concerns the increment of stimulation needed to detect a difference from a baseline, usually summarized by a ratio between the increment and the baseline over an appropriate operating range. Fechner's Elemente der Psychophysik, published in 1860, organized psychophysical thresholds into a research program that sought to quantify sensory magnitude from differential sensitivity. His model connected the measurable accumulation of just-noticeable changes to the idea of equally sized psychological increments Fechner (1860). The historical achievement was a procedure for converting observations into mathematical hypotheses with conditions that could be tested.
That conversion depends on an additional measurement assumption. Equal discriminability does not logically guarantee equal differences in sensation: the probability of distinguishing two stimuli depends on noise, adaptation, attention, decision strategy, and trial configuration. When sensitivity varies across intensities, integrating the reciprocal of a local threshold may produce a convenient coordinate on stimulus space, yet the resulting function inherits the original task and its assumptions. Treisman's investigation of noise and Weber's law demonstrates how stochastic considerations enter the interpretation of sensitivity Treisman (1964). Recovering a putative sensation scale therefore requires an account of how sensory noise and comparisons change with intensity, rather than a curve fit alone.
The Fechnerian tradition established several enduring distinctions. Detection at very low intensities involves a threshold region that can behave differently from suprathreshold discrimination; performance often changes gradually as the difference grows; and the physical reference must be calibrated rather than presumed. Later reinterpretations asked whether logarithmic and power expressions might be alternative approximations in limited ranges or results of different judgment procedures. Krueger's reconsideration of the Fechner–Stevens divide demonstrates that reconciliation requires specifying what each law actually measures Krueger (1989). The enduring historical question is how to connect local psychophysical discrimination, global subjective judgment, and the transformation rules that make their numbers scientifically meaningful.
Stevens and the Direct Estimation Program
S. S. Stevens changed psychophysical practice by asking observers to assign numerical magnitudes directly to suprathreshold experiences. Instead of constructing sensation from a sequence of differential thresholds, magnitude estimation treats an observer's numerical production as the principal measurement. In experiments on loudness, brightness, and other dimensions, participants could be instructed to give a value proportional to the apparent strength of a presented stimulus; alternatively they could adjust a stimulus to match a requested number. Stevens's 1956 study of loudness and his 1957 formulation of the power law defined this experimental strategy Stevens (1956) Stevens (1957). The procedure broadened psychophysics beyond detection to perceptually meaningful variations in strong stimulation.
Direct estimation has distinctive strengths. It samples a broad range of intensities in a single experiment, allows comparisons between sensory continua, and can reveal regularities that are not naturally expressed by one discrimination threshold. At the same time, an assigned numeral is an overt action using learned symbols, implicit conventions, remembered precedents, and personal response ranges. Numbers can be reproducible while their status as true ratios of experience remains unverified. Early work comparing trained and unpracticed observers showed why instructions and familiarity deserve explicit attention Stevens & Poulton (1956). Stevens and Galanter compared numerical and category methods across multiple continua, clarifying how the elicitation method affects observed scaling Stevens & Galanter (1957).
Later accounts recognized the broad historical significance of direct measurement while treating its assumptions as empirical claims to investigate. A within-participant function may be stable over repetitions even when the same person's exponent changes after the intensity range or modulus changes. Apparent intermodal differences can emerge from multiple sources: transduction, coding, memory, and response format. Stevens's mature synthesis described the power approach across perceptual, neural, and social applications, but extension to any new continuum still requires calibrated trials and independent tests Stevens (1975). The program's continuing contribution is an explicit experimental question: under what operations do numerical reports behave as if they preserve perceived proportions?
What a Numerical Scale Can Represent
Numbers attached to psychological judgments gain meaning from the relations preserved under permissible transformations. An ordinal scale preserves ordering; an interval scale preserves differences up to a positive affine transformation; a ratio scale preserves meaningful ratios under multiplication by a positive constant. Stevens formulated this classification in his influential account of the theory of measurement Stevens (1946). The distinction is operationally important: a participant's assignment of 20 rather than 10 shows a numerical doubling, but a claim that the subjective experience itself doubled requires additional behavioral structure. A regression on reported values is descriptive even when its functional form has the algebraic properties of a ratio scale.
Ratio claims can be investigated through systematic operations, not only through responses collected once. If a requested multiplication of experienced magnitude is represented consistently, one can test whether sequential multiplications commute, whether product relations compose, and whether observers make comparable judgments across different starting stimuli. Narens presented axiomatic treatments that distinguish behavioral observations from theoretical claims about sensation and about how numerical names are internally understood Narens (1996). Luce likewise formalized relations among intensity proportions, matching, and joint presentations Luce (2002). These approaches reveal what a credible ratio interpretation requires and how hidden assumptions can be separated from observable predictions.
The choice of a scale also constrains admissible statistics. Correlations between ordinal judgments may survive monotonic transformations whereas arithmetic means of the same ordinal codes generally do not carry invariant interval meaning. A normalized magnitude score can support comparisons inside one specified task without licensing the inference that another response instrument measures the same numerical quantity. Classical methodological work on the measurement of sensation emphasizes that operational construction and representational justification are distinct stages Laming (1997). An adequate report therefore states the intended measurement level, explains any transformations, displays variability, and tests whether the interpretation holds under changes in task context and stimulus set.
Sensation, Response, and Identifiability
Every magnitude report includes at least two transformations: an observer extracts a usable internal representation from stimulation and then converts that representation into an external response. A model might write an internal quantity as a function of stimulus intensity and the reported numeral as a further transformation of that internal quantity. Without constraints on one of the two functions, many combinations generate the same observed scale. The result is an identifiability problem: a fitted power exponent can summarize data while remaining ambiguous about which process produced it. Poulton's examination of alternative models for magnitude estimation makes this issue concrete by treating the task itself as part of the measurement process Poulton (1968).
Response functions are influenced by experimental history. The same physical level may receive different numbers when surrounded by weak or intense comparison stimuli, when examples are supplied, or when observers become accustomed to a particular range. Some of these effects are adaptive comparisons; others reflect the selection and reuse of numerical categories. The range-frequency framework of category judgment shows that reported locations can depend on both endpoints and the distribution of tested stimuli Parducci (1965). A robust scaling claim therefore requires interventions that alter the physical signal independently of the response vocabulary and interventions that alter the response context while holding physical stimulation fixed.
Scientific identifiability improves through triangulation. Magnitude estimation can be paired with production, cross-modal matching, paired comparisons, categorical ratings, and discrimination functions; tasks that share sensory information but differ in response structure provide useful tests. Replication across observers, sessions, and stimulus sets distinguishes stable within-person relationships from effects tied to one apparatus or instruction. Evidence that a mathematical model fits several procedures is stronger when its parameters are constrained in advance and its predictions survive held-out tests. The contemporary challenge is to explain which measured invariants belong to sensory representation, which belong to judgment and learning, and which remain dependent on a chosen scale Zhou et al. (2024) Narens (1996).
The Weber Fraction and Local Differential Sensitivity
Discrimination experiments estimate how perceptual performance changes as two stimuli become more different. If an increment ΔI producing a chosen criterion is approximately proportional to reference intensity I, the Weber fraction ΔI/I remains roughly constant. This expression is a local empirical relation, requiring a specified baseline, contrast direction, duration, sensory modality, and response procedure. It has traditionally motivated the idea of scale invariance: a proportional change rather than a fixed physical increment can govern detectability. Fechner's program relied on such regularity, while subsequent noise models clarified its dependence on a particular region of stimulus space Fechner (1860) Treisman (1964).
A threshold is conventionally defined by performance or response probability, not by direct inspection of a private subjective event. In a two-alternative forced-choice design a criterion might be chosen on the fitted accuracy function; in yes–no detection the same performance can depend on decision bias. Gaussian approximations, unequal variability across stimulus conditions, and guessing or lapses alter threshold interpretations. A single estimated Weber fraction therefore compresses both a psychometric function and a criterion selection. Kingdom and Prins provide practical analysis of fitting and experimentation that treats threshold estimation as statistical inference with design-dependent uncertainty Kingdom & Prins (2016).
Departures from a constant Weber fraction matter scientifically. Close to detection limits, additive internal noise or physical equipment noise can dominate; at high intensities adaptation, saturation, or altered strategies can change the ratio. Some quantitative domains exhibit performance improvements or deteriorations that do not track a simple proportional function. Recent experimental investigations of perceived numerosity have shown how manipulating priors and incentives can produce variability patterns at odds with a naïve Weber prediction Prat-Carrabin & Gershman (2025). Researchers should report the tested range and variation around fitted thresholds rather than extrapolate local approximations into universal laws.
Logarithmic Functions and the Fechner Assumption
A logarithmic scaling relation is commonly written S = a log(I/I0), where I is a positive stimulus intensity, I0 denotes a reference intensity, and S is a constructed magnitude coordinate. One route to this relation is to suppose that a locally equal just-noticeable difference represents an equal sensory increment, while the physical increment grows proportionally with baseline intensity. Integrating dS proportional to dI/I yields a logarithm. This derivation explicitly separates a discrimination regularity from the additional assumption about subjective increments. Fechner's original formulation provided the historical basis, and modern critiques examine whether that supplementary assumption holds across actual psychophysical tasks Fechner (1860) Laming (1997).
The logarithmic form transforms multiplication of stimulus intensity into addition on the constructed coordinate. Consequently, the difference between the responses to two intensities depends on their ratio, in the portion of the domain where the model applies. Its reference and additive origin matter: at the lower end, a mathematical logarithm does not describe an actual perceptual system without specifying the region of effective stimulation and detection. A visually convincing linear trend against log intensity can be generated by several underlying mechanisms, so parameter stability and predicted behavior at the edges are stronger tests than a high in-sample fit.
Claims about Fechner's law should be stated with a clear dependent variable. If an experiment fits a logarithm to ordinal category ratings, its evidence concerns the particular ratings. If it estimates subjective distance through discrimination, the result depends on noise assumptions and decision rule. Krueger reviewed attempts to reconcile logarithmic and power relationships, identifying the possibility that different procedures and approximations produce apparently incompatible laws Krueger (1989). The proper scientific conclusion is conditional: a logarithmic coordinate can organize a set of judgments, and further tests are needed before identifying that coordinate uniquely with experienced intensity.
Power Functions and Stevens Exponents
Stevens's empirical formulation is commonly written R = kI^n, with reported response magnitude R, positive stimulus magnitude I, scaling constant k, and exponent n. On logarithmic coordinates the model becomes log R = log k + n log I, making exponent estimation a regression problem subject to measurement error in both axes and assumptions about residual variability. An exponent below unity implies compressive numerical growth; an exponent above unity implies expansive growth, relative to the chosen physical variable. Stevens reported differing exponents across perceptual continua and proposed a broad psychophysical law Stevens (1957). Such differences motivated systematic comparisons of modalities, tasks, and individual observers.
Neither the exponent nor the multiplicative constant is a context-independent measurement of sensory physiology. Changing the physical reference variable, stimulus calibration, instructions, estimation modulus, or numeric response conventions can change the fitted relationship. Restricted stimulus ranges can make logarithmic and power models nearly indistinguishable. A straight line on double-log coordinates is useful descriptive evidence, but the independent prediction of magnitude production, comparison judgments, or discriminability provides stronger evidence for underlying scale properties. Teghtsoonian analyzed regularities and limits in relations among psychophysical exponents, illustrating the continuing problem of interpretation Teghtsoonian (1971).
The power law remains valuable as a compact operational model. Within a declared range, fitted parameters facilitate comparisons across participants and experimental conditions, and departures from power scaling can reveal response saturation or nonstationary judgment. Model estimates should include uncertainty, individual variation, residual diagnostics, and sensitivity to data transformations. When comparing theories, a shared dependent variable and common stimulus domain are essential: exponents from loudness ratings cannot automatically be treated as commensurate with those from duration production. Empirical results should be reported as task-specific response functions before advancing broader explanatory claims Stevens & Galanter (1957) Kingdom & Prins (2016).
Magnitude Estimation and Choice of Modulus
In magnitude estimation, each stimulus is presented and the observer produces a number intended to express its perceived strength. Some procedures identify a standard stimulus and assign it a prescribed modulus, often a convenient integer. Others allow free numerical estimation and normalize the resulting judgments later. The assigned reference anchors the response mapping and introduces memory and compliance requirements; it does not turn a subjective impression into a physical unit. Early loudness experiments made the technique influential because observers could generate graded responses across a substantial intensity range Stevens (1956) Stevens & Poulton (1956).
Good magnitude-estimation protocols specify how many practice trials precede the measurement, whether references are repeatedly presented, whether zeros or fractions are permitted, the randomization of intensities, and the response time allowed. Numerical production can be positively skewed and heteroscedastic; geometric averaging and transformations are sometimes used, but their appropriateness depends on the model of error. Repeated trials reveal the stability of an observer's scale and the possible effects of order. When the modulus changes, one should test the prediction that only the multiplicative constant changes, rather than assume the exponent is invariant. Poulton cataloged several sources of systematic bias in magnitude estimation Poulton (1968).
A scientifically rigorous study treats the number sequence as behavior in its own right. Observers differ in arithmetic fluency, willingness to use extreme numerals, interpretation of ratio instructions, and adaptation to a limited response range. Cross-session comparison needs an explicit linking procedure. Results are more informative when the design contains independent tasks, such as pairwise discrimination and magnitude production, or when a held-out range is used for prediction. The goal is to determine whether an observer reliably represents a sensory relation and translates it into numbers in a stable way, while stating what remains unresolved about subjective ratios Narens (1996) Gescheider (1997).
Magnitude Production, Matching, and Cross-Modality Tests
Magnitude production reverses the estimation operation: an observer is given a target number and adjusts a physical stimulus until the experienced magnitude is judged to correspond. If numerical estimation and production are exact inverses of a stable mapping, their fitted functions should agree after accounting for response noise and direction-dependent biases. This is a strong prediction because the task demands a different response operation. In practice, estimation can involve selecting a numeral from memory, whereas production demands motor control, iterative adjustments, and perceived stopping criteria. Differences between them are theoretically informative rather than automatic signs of an invalid observer Stevens (1975) Luce (2002).
Cross-modal matching compares the perceived strengths of stimuli belonging to different sensory dimensions, for example adjusting brightness to match the loudness of a sound under a specified instruction. Cross-modality designs can reduce reliance on numerical labeling, but they introduce a new equivalence judgment between qualitatively different experiences. The resulting match function describes how an observer accomplishes a particular task; whether it measures a common dimension of subjective magnitude is a substantive theoretical issue. Luce's psychophysical treatment derives predictions involving proportions, joint presentations, and matches that can be tested across appropriately designed experiments Luce (2002).
Formal properties of production and matching offer discriminating experiments. If doubling an instruction and then tripling it produces the same final judgment as the reverse order, that behavioral commutativity provides evidence relevant to ratio operations. Ellermeier and Faulhammer's loudness production research evaluated axioms underlying Stevens's ratio-scaling interpretation and found support for some properties but important violations of others Ellermeier & Faulhammer (2000). A newer investigation of cross-modal commutativity likewise illustrates why group-level patterns and heterogeneous individuals require separate analysis Kohler & Heller (2026). Taken together, these approaches turn the intuitive language of proportional experience into hypotheses about reproducible operations.
Stimulus Range and the Calibration of Judgment
Psychophysical measurements depend on the stimulus ensemble in which judgments are obtained. When the range widens while the physical value of one target stays fixed, observers may change its numerical estimate or category placement because the reference distribution used for comparison has changed. A range effect can arise through rescaling of the response space, adaptation of perceptual reference, or attention to endpoints; experimental results alone do not identify which source dominates. Poulton's review classified several task-dependent processes capable of changing magnitude-estimation functions Poulton (1968). Design reports should specify not just the minimum and maximum tested intensities but also their presentation frequency, spacing, and distribution across trials.
A critical distinction is between physical calibration and judgment calibration. Laboratory instruments may verify that a light, tone, or vibration is delivered at the prescribed level, but a participant who has just encountered extreme values can alter the subjective comparison standard. The apparent response to a repeated test stimulus can therefore shift despite unchanged physical input. Manipulations of the surrounding stimulus range and presentation history help distinguish contextual mapping from changes in sensory transduction. The range-frequency framework developed for categorical ratings predicts how relative rank and stimulus frequency jointly organize responses Parducci (1965). This is a concrete example of a formal model tied to a manipulable experimental variable.
Good experimental design crosses the stimulus range with other conditions rather than allowing context to change silently with treatment. Blocked and intermixed presentations may give different estimates because they establish different local reference distributions. Counterbalancing, repeated anchor probes, and trial-level modeling can reveal whether drift is gradual or follows specific stimuli. It is useful to test absolute physical values that appear in several experimental contexts, since these shared probes allow direct comparison of response mappings. Claims that an intervention changes sensory magnitude should remain proportionate to evidence that contextual response selection and criterion changes have been evaluated Petrov & Anderson (2005) Parducci & Perrett (1971).
Category Ratings, Partition Scales, and Relative Frequency
Category rating asks an observer to select among ordered labels or bins rather than generate an unrestricted number. Such scales can describe comparative judgments efficiently, but their category boundaries are shaped by the available response options and the distribution of presented stimuli. Parducci's range-frequency model proposes that judgment combines the relative position of a stimulus within the encountered range with its position in the cumulative frequency distribution Parducci (1965). Under this account, adding many low-valued stimuli can change the category assigned to an otherwise unchanged intermediate stimulus. These effects make the sample of stimuli a substantive experimental factor.
Parducci and Perrett investigated the effects of relative spacing and presentation frequency on category ratings, providing an empirical basis for examining stimulus-ensemble organization rather than assuming a fixed transfer from physical intensity to category label Parducci & Perrett (1971). Category scales and direct magnitude estimates also differ in mathematical assumptions: selecting category 4 instead of category 2 does not justify the inference of twice the subjective magnitude. A scale can reliably order participants or conditions while leaving distances between categories unspecified. The distinction should shape statistical analysis, graph labeling, and comparisons between rating instruments Stevens (1946) Stevens & Galanter (1957).
An informative comparison uses matched physical stimuli and systematically varied response formats. If the same stimuli yield stable pairwise orderings but different numerical distances under category rating and unrestricted estimation, the shared ordinal structure may be more secure than the precise spacing. Repeated reference stimuli and within-session counterbalancing permit investigators to estimate boundary movement. Formal models can treat category selection as discretization of a variable internal estimate, a memory-dependent comparison, or a mixture of both. Contrasting these alternatives requires predictions about the effects of frequency manipulation, trial order, and learning, not an interpretation inferred from category means alone Petrov & Anderson (2005).
Sequential Dependencies, Adaptation, and Learned Anchors
Psychophysical responses unfold through trials rather than arriving as independent samples from an unchanging function. Recent stimuli can affect perceived contrast, remembered comparison standards, and the number chosen on the next trial. A participant may use an immediately preceding stimulus as an anchor even when the task specifies a more distant standard. Adaptation and learning may also alter receptor or central encoding over longer intervals. Petrov and Anderson developed a memory-based anchor model addressing dynamics of category rating and absolute identification, demonstrating that sequences of decisions can be studied as an explicit computational process Petrov & Anderson (2005).
Trial-level dependencies matter for estimation. Suppose a physical intensity is sampled repeatedly after different predecessor values. If reports vary systematically with the predecessor, an analysis that pools all repetitions into one mean hides a potentially causal contextual variable. Designs can randomize transitions, balance local stimulus histories, and examine whether response carryover is related to the previous physical intensity or the previous chosen numeral. These predictors need to be separated because perceptual carryover and numerical habit can covary. A model with participant-specific memory parameters can reveal structure that would otherwise appear as unexplained response noise, but its parameters need independent validation.
Instructions, practice, and prolonged exposure shape the available reference repertoire. Training may stabilize numerical mappings or induce expectations that change responses across blocks. An observer who receives accurate feedback on selected stimuli may subsequently anchor other estimates to those known values. Recent preregistered numerical-estimation experiments modeled calibration through comparative judgments, associative memory, and anchoring-and-adjustment mechanisms Kaspersen et al. (2026). Such work is especially informative because the model connects manipulations of numerical reference information to observable changes in estimates, making response strategy a testable alternative to a purely sensory scaling account.
Loudness, Brightness, and Sensory Continuum Differences
Loudness was a central test domain for the development of direct magnitude estimation. Acoustic stimulus level can be specified by physical pressure measures, but the resulting loudness judgment depends on frequency, duration, spectral composition, adaptation, and the listener's hearing. Stevens used controlled loudness tasks to develop and evaluate direct numerical scaling, while subsequent studies examined whether unpracticed participants followed similar response functions Stevens (1956) Stevens & Poulton (1956). The empirical object is a stated acoustic configuration and a reporting procedure. Transferring an exponent from one sound or listening condition to another demands separate experimental support.
Brightness judgments illustrate the same methodological problem through distinct stimulus physics. A measured luminance value does not uniquely determine perceived brightness because retinal adaptation, viewing context, spatial arrangement, and contrast affect the information entering the judgment. A comparison of brightness with loudness may yield different slopes on a log-log graph, yet slope differences alone cannot show that the modalities contain fundamentally different kinds of subjective number. Stevens and Galanter compared ratio and category scales across perceptual continua, illustrating how modality and response procedure interact Stevens & Galanter (1957). The relevant comparison therefore keeps task rules and scale interpretation explicit.
Cross-modality production studies offer a further test by replacing some symbolic reports with adjustments. They still require observers to understand a comparison instruction and to decide when two qualitatively different experiences satisfy a common criterion. The formal account of intensity matching developed by Luce shows the value of specifying that criterion mathematically Luce (2002). Empirical work on loudness production axioms found mixed evidence for behavior required by strict ratio measurement Ellermeier & Faulhammer (2000). Modern synthesis should retain the robust finding that reproducible scaling functions exist in defined tasks while asking how much of the fitted slope is attributable to sensory representation, relational comparison, and response conventions.
Numerosity and Symbolic Estimation
Numerosity estimation provides a valuable extension of psychophysical scaling because the physical target is the count of discrete objects rather than the strength of a continuous sensory energy. Observers may report approximate counts of briefly displayed dot arrays, and their precision can depend on the actual number, arrangement, exposure duration, and visual features correlated with count. Because responses are expressed as numerals, the task directly joins nonsymbolic representation with learned symbolic mapping. The research question is whether regularities such as roughly proportional error reflect a sensory number representation, a comparative decision process, arithmetic familiarity, or several jointly contributing mechanisms Kaspersen et al. (2026).
Recent computational work treats estimation as a sequence of comparative judgments with memory and anchoring. Kaspersen and colleagues tested a model in preregistered experiments using briefly presented dot arrays and reported that it reproduced key behavioral signatures across exposure durations and calibration effects associated with veridical reference quantities Kaspersen et al. (2026). These results support the explanatory relevance of comparison and reference learning. They do not establish that every numerosity estimate, or every sensory magnitude judgment, is generated by this particular mechanism. Cross-task predictions, independent replications, and developmental tests can distinguish a broad numerical estimation process from paradigm-specific adjustments.
Other recent experimental approaches place prior beliefs and task rewards at the center of quantitative judgment. Prat-Carrabin and Gershman examined how Bayesian estimation can yield patterns of variability that depart from a simple Weber-law expectation, using controlled tasks that manipulate relevant contextual information Prat-Carrabin & Gershman (2025). Numerosity is consequently a useful stress test for any theory that equates a fitted psychophysical function with a unique internal sensory law. A defensible account explains how physical number, visual sampling, prior distributions, memory, incentive structure, and symbolic reporting combine to predict both central tendency and variability. These components can be manipulated independently and compared with alternative computational models.
Axiomatic Measurement of Subjective Ratios
An axiomatic account specifies which observable relations must hold for numerical judgments to support a particular measurement interpretation. In ratio magnitude estimation, the fundamental question is whether an observer's use of expressions such as twice as intense corresponds to a coherent ratio operation over sensory events. One cannot answer this question by pointing to a scale's numerical labels, because the labels themselves are part of the task. Narens distinguished behavioral primitives accessible to an experimenter from cognitive and psychobehavioral assumptions concerning the interpretation of numerals and sensation Narens (1996). This framework enables explicit tests of what constitutes ratio consistency.
Consider an experiment in which a reference stimulus is successively adjusted according to multiplicative instructions. If two sequential operations are carried out in different orders, a lawful ratio representation imposes relationships among their outcomes. Transitivity, commutativity, and compositional properties can be evaluated empirically where the formal theory applies. Failure of one property may reveal memory dependence, nonlinearity of the response operation, or a wrong measurement interpretation; success supports only the tested property and conditions. Ellermeier and Faulhammer investigated axioms of loudness production and reported a pattern in which some anticipated regularities were supported while others were violated Ellermeier & Faulhammer (2000).
Axiomatization strengthens psychophysics by making explanatory assumptions falsifiable. It can also clarify why a judgment's numerical value is underdetermined: the investigator may know that an observer consistently behaves as if some positive multiplicative operation were applied without knowing that its internal interpretation agrees with the printed numeral. Luce derived a theory connecting intensity proportions, joint presentations, and matches that gives researchers additional cross-procedure tests Luce (2002). Generalization must consider whether sensory dimensions, individual differences, and task histories affect the operations. An experimentally justified ratio scale would be a powerful result precisely because it survives multiple constraints beyond ordinary regression fit.
Signal Detection and the Separation of Sensitivity
Signal detection theory distinguishes the statistical separability of sensory evidence from the criterion used to produce a yes or no response. A participant's willingness to say that a signal was present may change with expected prevalence, reward, or instruction even when the sensory distributions remain stable. Green and Swets developed the classical statistical framework for studying this separation across psychophysical tasks Green & Swets (1966). Its direct relevance to scaling is methodological: judgments of subjective intensity can be affected by response policy, and changes in reported magnitude need not imply changes in sensitivity. Where possible, parallel discrimination measures provide an independent constraint.
In a typical yes–no experiment, hits and false alarms together indicate response behavior under signal and noise conditions. In a forced-choice experiment, correct comparison choices vary with the difficulty of the stimulus distinction, but memory for intervals and position preferences introduce their own assumptions. Psychometric curves summarize response probability across stimulus differences; their slope and location relate to the experiment's stimulus distribution and decision process. Magnitude estimates describe a different mapping, usually from a presented intensity to a freely produced number. Reporting both functions can expose whether conditions that change numerical judgments also affect the ability to distinguish stimuli.
An observer can show stable discrimination thresholds and context-sensitive magnitude ratings, or changing discrimination with comparatively stable numerical calibration. These dissociations are valuable constraints on mechanism. They also illustrate why a psychophysical law is strongest when it predicts several outcomes that vary differently under a controlled manipulation. In quantitative terms, discrimination often depends on changes in the signal distribution relative to internal variability, while magnitude reports may depend on the distribution's mean and a separate response transformation. The unified modeling proposal of Zhou, Duong, and Simoncelli explicitly addresses the relation between perceptual appearance and discriminability through aspects of a common representation Zhou et al. (2024).
Difference Scaling and Psychophysical Geometry
Difference scaling studies comparisons between perceptual intervals rather than the numerical magnitude of one stimulus. Observers may indicate which of two stimulus pairs contains the larger apparent difference, producing relational data from which a latent scale can be estimated under explicit assumptions. Maximum-likelihood difference scaling, developed by Maloney and Yang, formalized this approach and connected response probabilities to an inferred psychophysical scale Maloney & Yang (2003). It offers an important complement to direct magnitude estimation because it uses a comparative response that does not require the participant to produce a ratio numeral.
To derive a difference scale, the investigator specifies how observer choices depend on internal interval differences and their noise. A common modeling strategy assigns values to the stimulus levels, forms predicted differences between paired values, and links the difference of those intervals to choice probability. The fitting procedure recovers scale coordinates subject to identifiability constraints, and their interpretation depends on the comparison model. Distinct monotonic transformations preserve order but can alter interval structure, so the mathematical invariances must be stated. Stimulus selection should include contrasts that distinguish a logarithmic law, a power function, and alternative nonlinear mappings.
Psychophysical geometry extends this reasoning to multiple dimensions, where perceived dissimilarities can provide information about relational organization among stimuli. Multidimensional scaling reconstructs a configuration from judged or observed dissimilarities, but its metric, dimensionality, and transformation assumptions determine which coordinates are interpretable Borg & Groenen (2005). The connection to magnitude estimation is instructive: both infer latent structure from behavior, yet the observation model and allowable transformations differ. Validation should ask whether the recovered structure predicts new comparisons or transfers to a second task. A visual map or a well-fitting curve is evidence for a model of responses, not an unrestricted picture of subjective experience.
Stochastic Coding, Fisher Information, and Modern Unification
Contemporary models seek to explain reported magnitude and discrimination accuracy using the same stochastic internal code. Suppose sensory intensity I produces an internal response distribution with mean μ(I) and variance σ²(I). A judgment of magnitude may depend on the central tendency of that response, while local discriminability depends on how rapidly the distribution changes relative to its variability. Under specific distributional assumptions, Fisher information quantifies the local sensitivity of encoded responses to changes in stimulus value. Zhou, Duong, and Simoncelli proposed a unifying framework linking measures of perceived magnitude and discriminability through distinct aspects of a shared representation Zhou et al. (2024).
This perspective can explain why sensation scaling and threshold scaling should not automatically be treated as identical measurements. Two observer systems might have similar mean internal responses but different trial-to-trial variability, yielding comparable average magnitude reports and different detection thresholds. Conversely, response mappings can differ while preserving some discriminability relations. The empirical challenge is to estimate enough independent features of both judgment and comparison behavior to constrain the mean and variability of the proposed code. The shape of a psychometric curve alone cannot identify all aspects of a distribution, and a rating function alone rarely fixes internal noise. Model-based inference must state the distributional and decision assumptions explicitly.
Comparative modeling benefits from contemporary open studies that combine formal derivation with behavioral tests. The 2024 unified-framework work gives a testable strategy for relating two historically separate measurement traditions, rather than replacing detailed sensory mechanisms with a single numerical slogan Zhou et al. (2024). New probabilistic studies of numerosity likewise demonstrate how task priors can reshape estimate dispersion Prat-Carrabin & Gershman (2025). These findings establish useful model families and empirical patterns, but their generality to audition, touch, chemical senses, and complex social judgments remains a research question. Independent data, preregistered discriminating predictions, and uncertainty-aware parameter estimation determine how strongly the unification claim can be supported.
Estimation, Uncertainty, and Reproducible Experimental Design
Reliable scaling research begins with traceable physical measurements. For visual stimuli, investigators should document luminance or contrast calibration, color and spectral specifications when pertinent, viewing distance, display timing, ambient light, and adaptation. Auditory experiments require appropriate calibration of sound levels and delivery systems. The design must record precise instructions, standard stimuli, order, repetition, feedback, response device, and inclusion criteria. Practical psychophysics guides treat stimulus calibration and trial structure as foundational parts of the inference rather than as incidental setup details Kingdom & Prins (2016) Gescheider (1997). A law fitted to inaccurately delivered stimuli cannot be rescued by statistical sophistication.
Statistical analysis must align with the measurement model. Repeated judgments from one participant are dependent observations; participant-to-participant variation requires an explicit aggregation strategy. Because numerical estimation errors can be multiplicative or skewed, analysts should compare plausible error models, inspect residuals, and report whether a logarithmic transformation changes the estimand. Confidence intervals or posterior uncertainty are needed for exponents, thresholds, context effects, and matching parameters. Model comparison should incorporate complexity and predictive performance rather than simply choose the curve with the smallest training residual. If a study estimates multiple individual slopes, a hierarchical account may help distinguish population regularity from heterogeneous response mappings.
Reproducibility also requires preservation of what was actually done. A research record should include randomization code, equipment calibration dates, the full stimulus ensemble, exclusion decisions, parameter constraints, and analysis scripts where sharing is appropriate. Strong claims involve predictions on new intensities or held-out participants and an explicit comparison against rival response-process models. Recent preregistered experiments in quantity estimation illustrate how prospective predictions and shared analysis materials can strengthen inference about comparative judgment and anchoring Kaspersen et al. (2026). A peer-reviewed publication and an accessible DOI establish provenance; whether its data warrant a specific conclusion depends on the original design, observed uncertainty, and replication scope.
Development, Individual Variation, and Generalizability
Psychophysical scaling functions are conditional on the observer's sensory capacities, experience, language, and familiarity with the response system. A numerical magnitude task requires participants to understand comparative instructions and to map perceptual differences onto learned symbols. Research involving children, older adults, or individuals with sensory impairment therefore needs methods appropriate to the population and separate tests of instruction comprehension. Even with identical physical stimulation, age-related sensory changes or different reference histories may alter response distributions. Such differences should be investigated as independent hypotheses rather than treated as deviations from an allegedly universal adult exponent Kingdom & Prins (2016) Gescheider (1997).
Individual differences affect both the location and the shape of fitted scaling functions. One observer may use a narrow numerical response range; another may vary more in repeated assessments; a third may show large contextual shifts between sessions. Aggregating all numerical responses can obscure these differences and may generate a population curve that characterizes no individual. Estimation studies should examine within-observer stability before combining results, and report the sensitivity of group conclusions to heterogeneous slopes or lapse behavior. Studies of axiomatic magnitude operations also indicate the importance of testing individuals because the consistency of group averages can mask violations of proposed measurement properties Ellermeier & Faulhammer (2000) Narens (1996).
Generalization across cultures and languages presents further questions about numerals, ratios, and communicative conventions. A direct scaling procedure that relies on a verbal instruction to double a sensation may recruit learned mathematical concepts to different degrees across participant groups. More elementary discrimination or nonverbal matching tasks can help investigate what remains invariant when symbolic demands are reduced. Recent modeling of symbolic numerosity estimation is relevant because it explicitly connects nonsymbolic comparisons to associative learning and numeric expression Kaspersen et al. (2026). Broader claims require direct cross-population and cross-language studies with equated stimulus conditions, not extrapolation from a single convenience sample.
Model Comparison and the Quality of Evidence
A psychophysical model earns explanatory standing by predicting how observations change under controlled intervention. Fechnerian logarithms, Stevens power laws, category-based range effects, memory-dependent anchors, and stochastic sensory-code models can each explain portions of a data set under selected conditions. Their evidentiary value differs when a model predicts only an in-sample curve or also predicts effects of reference shifts, response format, discrimination, and trial order. Historical synthesis therefore distinguishes the origin of a law from the present evidence for its proposed mechanism. Reviews by Krueger and formal analyses by Narens identify the conceptual assumptions that a simple model fit cannot settle Krueger (1989) Narens (1996).
Evidence should be assessed at several levels. An original experiment establishes the performance observed in its tested sample and apparatus; an independent replication addresses whether the pattern persists under new implementation; a mathematical proof establishes consequences of stipulated axioms; and a model simulation demonstrates what a specific implementation can produce. None of these evidentiary forms automatically replaces the others. For example, a derivation connecting Fisher information to discrimination rests on assumptions about an internal response distribution, while experimental magnitude ratings supply a separate test of that framework Zhou et al. (2024). A valid synthesis identifies those assumptions and looks for empirical alternatives rather than treating theoretical elegance as proof.
High-priority uncertainties include changes in exponents across stimulus ranges, failures of ratio axioms, response dependence on recent trials, and the representational meaning of symbolic numbers. A comparison should quantify these effects, show uncertainty, and include conditions that discriminate between sensory and response-based explanations. Contradictory findings are useful when differences in materials, recruitment, task timing, or analysis explain why results diverge. Recent preregistered numerical-estimation research contributes evidence about comparison and calibration within a specific paradigm, whereas axiomatic loudness studies constrain claims about subjective multiplication in another Kaspersen et al. (2026) Ellermeier & Faulhammer (2000). Scientific integration preserves these differences in scope.
Applications in Human and Artificial Systems
Psychophysical scaling supports instrument design, sensory accessibility, audio and visual engineering, ergonomic interfaces, and the selection of stimulus intensities that users can meaningfully distinguish. A display may be calibrated in physical units, yet its effective usability depends on contrast, viewing environment, adaptation, and observer variation. A sound notification can be measurable in decibels while its experienced prominence depends on spectral content and background sound. Perceptually informed design therefore combines stimulus calibration with threshold tests, comparative judgments, and magnitude reports tailored to the relevant user population Gescheider (1997) Kingdom & Prins (2016). Engineering decisions should cite task-specific evidence instead of assuming a single exponent applies universally.
Numerical scaling also intersects with visual analytics and behavioral research. Rating instruments and product-intensity judgments can be influenced by the distribution of options shown, so experiments that manipulate presentation context are especially important Parducci (1965). Difference-scaling procedures can be used to estimate how people compare pairs of images or graded changes when explicit numerical magnitude reports are unstable Maloney & Yang (2003). The practical question concerns what the chosen measurement predicts: performance, preference, perceived intensity, or a different construct. Those dependent variables should retain separate labels and validation procedures rather than being merged by convenience.
In the Artificial Era, engineered perception systems offer a productive comparison class for testing measurement theory. Artificial classifiers, robotic sensors, and multimodal agents can be evaluated by precisely specified physical inputs, error functions, calibration, and response variability. A machine's numerical confidence, activation strength, or category probability is an internal or reported quantity defined by architecture and training; it cannot be identified with human sensation simply because both display monotonic scaling. The meaningful interdisciplinary project is to compare discrimination accuracy, uncertainty calibration, robustness under changes of stimulus range, and information used for response. Formal models of shared coding and psychophysical measurement provide experimentally tractable comparison tools Zhou et al. (2024) Stevens (1946).
Open Problems and a Research Agenda
The first unresolved question is how to relate local discrimination to global reports of intensity under identified assumptions. A satisfactory program should measure both in the same observers, vary stimulus noise independently of signal strength, and compare predicted response functions across sensory domains. It should explicitly assess whether the mean and variability of an inferred internal code suffice to explain apparent magnitude and threshold sensitivity. The framework developed by Zhou, Duong, and Simoncelli is a useful starting point because it states common-representation hypotheses that can be subjected to independent perceptual testing Zhou et al. (2024). Cross-modal transfer remains a demanding verification target.
A second problem concerns the cognitive operations that turn an estimate into a number. Studies should isolate the influence of response vocabulary, reference numerals, training, sequence history, frequency distribution, and task goals. Experimental counterbalancing can separate prior presentation of particular numbers from prior exposure to particular physical intensities. Proposed ratio axioms should be tested within participants, alongside estimation and production methods, and reported with the frequency and magnitude of violations. A combination of axiomatic formalization and modern computational work makes clear predictions about which aspects of behavior would change if an observer relied primarily on multiplicative sensation, comparative anchoring, or mixed strategies Narens (1996) Kaspersen et al. (2026).
A third research priority is transportability. Scaling estimates drawn from narrow laboratory intensity ranges or highly practiced university samples may fail under different apparatus, visual environments, auditory conditions, ages, and language practices. Collaborative studies should publish calibration information, uncertainty intervals, stimulus sequences, and code for fitting competing models, making it possible to reproduce the inferential chain. New mathematical models can be assessed against independently replicated behavioral data and against simpler alternatives that include response context. The broad theoretical aim is a framework in which measurement level, neural coding hypotheses, behavioral decisions, and the organization of the environment are explicit, separately testable elements Laming (1997) Prat-Carrabin & Gershman (2025).
Conclusion
Psychophysical scaling and magnitude estimation comprise a mature experimental program for connecting physical stimulation with structured judgments. Its historical foundations include Weberian discrimination, Fechnerian logarithmic constructions, Stevens's direct estimation and power functions, and representational accounts of measurement. Each contributes a distinctive research tool: local thresholds summarize performance at discriminating changes; direct estimation yields numerical response functions; production and cross-modal matching test relations across operations; and formal measurement theory specifies conditions under which a ratio interpretation is justified Fechner (1860) Stevens (1957) Narens (1996). The scientific object is the entire stimulus–observer–judgment relation under a documented protocol.
The combined evidence supports the use of well-defined psychophysical functions while placing strong conditions on their interpretation. Context manipulations, category-frequency effects, numerical anchoring, sequential dependencies, and heterogeneous performance show that a reported magnitude reflects multiple contributing processes Parducci & Perrett (1971) Poulton (1968) Petrov & Anderson (2005). A fitted exponent does not alone identify a sensory mechanism or validate the claim that a particular numeral expresses a literal ratio of subjective experience. The appropriate standard is convergence among tasks that place different demands on the observer, together with explicit statistical uncertainty and prospectively tested theoretical predictions. Contradictory findings refine this standard by identifying the assumptions under which a proposed generalization survives.
Contemporary progress makes an integrated quantitative research agenda possible. Models relating perceived magnitude to discriminability, comparative and memory-based accounts of numerical estimation, and axiomatic analyses of production offer complementary ways to test explanations that once remained verbal or implicit Zhou et al. (2024) Kaspersen et al. (2026) Luce (2002). Future advances depend on reproducible experimental designs, careful calibration, population-aware analysis, and independent evidence across multiple modalities. Within psychology and across engineered systems, the lasting contribution of psychophysics is an exact methodological commitment: define the observed relation, specify its measurement assumptions, and determine which features remain invariant when the task, context, and observer are systematically changed.
References
- Borg, I., & Groenen, P. J. F. (2005). Modern multidimensional scaling: Theory and applications (2nd ed.). Springer. https://doi.org/10.1007/0-387-28981-X
- Ellermeier, W., & Faulhammer, G. (2000). Empirical evaluation of axioms fundamental to Stevens's ratio-scaling approach: I. Loudness production. Perception & Psychophysics, 62(8), 1505–1511. https://doi.org/10.3758/BF03212151
- Fechner, G. T. (1860). Elemente der Psychophysik (Vols. 1–2). Breitkopf und Härtel. https://doi.org/10.3931/e-rara-10879
- Gescheider, G. A. (1997). Psychophysics: The fundamentals (3rd ed.). Lawrence Erlbaum Associates. https://www.routledge.com/Psychophysics-The-Fundamentals/Gescheider/p/book/9780203774458
- Green, D. M., & Swets, J. A. (1966). Signal detection theory and psychophysics. Wiley. https://ntrs.nasa.gov/citations/19670040997
- Kaspersen, E., Solstad, T., Sabathiel, S., & Hodgen, J. (2026). Quantity estimation by repeated comparative judgments. Cognitive Science, 50(8), Article e70256. https://doi.org/10.1111/cogs.70256
- Kingdom, F. A. A., & Prins, N. (2016). Psychophysics: A practical introduction (2nd ed.). Academic Press. https://shop.elsevier.com/books/psychophysics/kingdom/978-0-12-407156-8
- Kohler, D., & Heller, J. (2026). On cross-modal commutativity in magnitude production for loudness and brightness. Perception. Advance online publication. https://doi.org/10.1177/03010066261470260
- Krueger, L. E. (1989). Reconciling Fechner and Stevens: Toward a unified psychophysical law. Behavioral and Brain Sciences, 12(2), 251–267. https://doi.org/10.1017/S0140525X0004855X
- Laming, D. (1997). The measurement of sensation. Oxford University Press. https://doi.org/10.1093/acprof:oso/9780198523420.001.0001
- Luce, R. D. (2002). A psychophysical theory of intensity proportions, joint presentations, and matches. Psychological Review, 109(3), 520–532. https://doi.org/10.1037/0033-295X.109.3.520
- Maloney, L. T., & Yang, J. N. (2003). Maximum likelihood difference scaling. Journal of Vision, 3(8), 573–585. https://doi.org/10.1167/3.8.5
- Narens, L. (1996). A theory of ratio magnitude estimation. Journal of Mathematical Psychology, 40(2), 109–129. https://doi.org/10.1006/jmps.1996.0011
- Parducci, A. (1965). Category judgment: A range-frequency model. Psychological Review, 72(6), 407–418. https://doi.org/10.1037/h0022602
- Parducci, A., & Perrett, L. F. (1971). Category rating scales: Effects of relative spacing and frequency of stimulus values. Journal of Experimental Psychology, 89(2), 427–452. https://doi.org/10.1037/h0031258
- Petrov, A. A., & Anderson, J. R. (2005). The dynamics of scaling: A memory-based anchor model of category rating and absolute identification. Psychological Review, 112(2), 383–416. https://doi.org/10.1037/0033-295X.112.2.383
- Poulton, E. C. (1968). The new psychophysics: Six models for magnitude estimation. Psychological Bulletin, 69(1), 1–19. https://doi.org/10.1037/h0025267
- Prat-Carrabin, A., & Gershman, S. J. (2025). Bayesian estimation yields anti-Weber variability. PNAS Nexus, 4(9), Article pgaf275. https://doi.org/10.1093/pnasnexus/pgaf275
- Stevens, S. S. (1946). On the theory of scales of measurement. Science, 103(2684), 677–680. https://doi.org/10.1126/science.103.2684.677
- Stevens, S. S. (1956). The direct estimation of sensory magnitudes: Loudness. The American Journal of Psychology, 69(1), 1–25. https://doi.org/10.2307/1418112
- Stevens, S. S. (1957). On the psychophysical law. Psychological Review, 64(3), 153–181. https://doi.org/10.1037/h0046162
- Stevens, S. S. (1975). Psychophysics: Introduction to its perceptual, neural, and social prospects (G. Stevens, Ed.). Wiley. https://openlibrary.org/books/OL5051914M/Psychophysics_introduction_to_its_perceptual_neural_and_social_prospects.
- Stevens, S. S., & Galanter, E. H. (1957). Ratio scales and category scales for a dozen perceptual continua. Journal of Experimental Psychology, 54(6), 377–411. https://doi.org/10.1037/h0043680
- Stevens, S. S., & Poulton, E. C. (1956). The estimation of loudness by unpracticed observers. Journal of Experimental Psychology, 51(1), 71–78. https://doi.org/10.1037/h0041089
- Teghtsoonian, R. (1971). On the exponents in Stevens' law and the constant in Ekman's law. Psychological Review, 78(1), 71–80. https://doi.org/10.1037/h0030300
- Treisman, M. (1964). Noise and Weber's law: The discrimination of brightness and other dimensions. Psychological Review, 71(4), 314–330. https://doi.org/10.1037/h0042445
- Zhou, J., Duong, L. R., & Simoncelli, E. P. (2024). A unified framework for perceived magnitude and discriminability of sensory stimuli. Proceedings of the National Academy of Sciences, 121(25), Article e2312293121. https://doi.org/10.1073/pnas.2312293121