Differential Threshold

Abstract

The differential threshold is the smallest change in a stimulus that an observer can reliably detect, the just-noticeable difference that marks the resolution of a sense. This article traces the construct from Weber's discovery that the just-noticeable difference is a constant fraction of the baseline stimulus, through Fechner's derivation of a logarithmic relation between stimulus and sensation by treating that difference as a unit, to the classical and adaptive methods that measure it and the psychometric function from which it is read. It sets out Stevens' power-law challenge to Fechner's scale and the signal-detection reinterpretation on which the threshold dissolves into a separation of sensitivity from decision criterion. Three demonstrations let the reader vary the Weber fraction, read a difference limen from a psychometric function, and compare Fechner's logarithmic scale against Stevens' power law.

Keywords: differential threshold, just-noticeable difference, weber's law, fechner's law, psychophysics

The differential threshold is the least difference between two stimuli that a perceiver can tell apart, the amount by which one must exceed the other before the two are reliably discriminated. It is measured as the just-noticeable difference (JND), also called the difference limen, and is conventionally fixed at the increment detected on some criterion proportion of trials, so that a smaller threshold means a finer sense (Gescheider, 1997). The construct is the discriminative counterpart of the absolute threshold, the least intensity that can be detected at all: where the absolute threshold asks whether a stimulus is present, the differential threshold asks whether two stimuli differ. Its study founded quantitative psychology, because the differential threshold gave the first stable law relating the physical world to the mental one — the just-noticeable difference varies with the stimulus in an orderly way — and the attempt to build a scale of sensation from that unit produced the field of psychophysics (Fechner, 1860).

What the Differential Threshold Is

The differential threshold names a limit on discrimination, not on detection. Two stimuli that differ physically are not always told apart: below some difference the perceiver responds as though they were the same, and only when the difference grows large enough does discrimination become reliable. Because that transition is gradual rather than abrupt, the threshold is a statistical construct fixed by convention — the difference discriminated on a criterion proportion of trials, classically the point discriminated 75% of the time in a two-alternative task, halfway between chance and certainty (Gescheider, 1997). The stimulus difference at that criterion is the difference limen or just-noticeable difference; a related quantity, the point of subjective equality, is the comparison value that is judged equal to the standard as often as not, and the difference limen is measured as the change from that point needed to reach the criterion.

The construct rests on a distinction between the physical stimulus and the sensation it produces. Weber and Fechner worked on the physical side, relating the just-noticeable difference to the measurable stimulus, and it was Fechner's ambition to cross from that measurable difference to a scale of unmeasurable sensation that turned a psychophysical regularity into a theory of mind (Fechner, 1860). Table 1 sets out the core quantities the rest of the article uses.

Weber's Law

The founding empirical regularity is Weber's. Testing weight discrimination by having observers judge which of two lifted weights was heavier, and touch by finding the least separation at which two points on the skin felt double, Ernst Heinrich Weber found that the just-noticeable difference is not a fixed amount but a fixed proportion of the standard: a weight of 100 units required an increment of about 2 units to feel heavier, a weight of 200 units required about 4, and the ratio between the increment and the baseline held roughly constant across the range (Ross & Murray, 1996). Fechner named this regularity Weber's law and wrote it as ΔI/I = k, where ΔI is the differential threshold, I the baseline intensity, and k the Weber fraction, a constant characteristic of the sense being tested (Fechner, 1860). The fraction is small for keen discriminations — on the order of 0.02 for lifted weight and 0.01 for line length — and larger for coarse ones such as taste concentration, and it gives a single number by which the acuity of different senses can be compared. The first demonstration makes the proportionality concrete: adjusting the baseline intensity rescales the just-noticeable difference in step, holding the fraction fixed.

Weber's law is an approximation, accurate across the middle of a sensory range but failing at its ends. Near the absolute threshold the Weber fraction rises sharply, because the observer's own baseline noise adds to the weak stimulus, and at very high intensities it may drift as well. For the discrimination of pure-tone intensity the fraction does not stay flat but declines slowly and systematically as intensity rises, a robust departure christened the near-miss to Weber's law and better described by making the exponent on intensity slightly less than one (McGill & Goldberg, 1968). Ekman noted a complementary regularity relating the Weber fraction to the growth of sensation, so that the just-noticeable difference corresponds to a constant increment on the sensation scale rather than the stimulus scale (Ekman, 1959). These qualifications refine Weber's law without overturning its central claim, that discrimination is governed by ratios rather than absolute differences.

Fechner's Logarithmic Law

Weber described how the threshold varies with the stimulus; Fechner used that description to build a scale of sensation. His reasoning began from an assumption of equal-appearing units: if every just-noticeable difference is subjectively the same size — one step of sensation, wherever it falls on the stimulus continuum — then a scale of sensation can be built by counting just-noticeable differences up from the absolute threshold (Fechner, 1860). Combining this assumption with Weber's law yields a differential equation, dS = c·(dI/I), whose integral is the logarithmic function now called Fechner's law, S = c·log(I/I₀), where S is sensation magnitude, I the stimulus intensity, I₀ the absolute threshold, and c a constant set by the Weber fraction (the Worked Example derives c = 1/log(1 + k), so that the scaling constant and the Weber fraction k stay distinct). The law states that sensation grows as the logarithm of intensity: equal ratios of stimulus produce equal increments of sensation, so that each successive doubling of a weight or a light adds the same amount to how intense it feels, and the physical world is compressed into the mental one.

This logarithmic compression is why a candle lit in a dark room is dramatic but the same candle added to a bright one is imperceptible, and why decibel and other logarithmic scales match perception so well. Fechner's achievement was methodological as much as theoretical: by making sensation a countable multiple of a measurable unit, he turned the study of mind into a quantitative science and gave it its first law (Link, 1994). The move was also his most contested, because the equal-units assumption — that every just-noticeable difference feels the same — cannot itself be observed directly, and the century of work that follows is in large part an argument over whether the scale it builds is the right one.

Measuring the Threshold

The differential threshold is not read off a single trial but estimated from many, and Fechner bequeathed three classical procedures for doing so (Fechner, 1966). In the method of limits the experimenter presents a comparison stimulus in ascending or descending series and records the value at which the observer's judgment changes, averaging the crossover points. In the method of constant stimuli a fixed set of comparison values is presented many times each in random order, and the proportion of trials on which each is judged greater than the standard is plotted against its value. In the method of adjustment the observer moves the comparison until it appears just different from the standard, and the setting is repeated. Table 2 sets the classical methods beside the adaptive procedures that now dominate the laboratory.

Whatever the method, the data converge on the psychometric function, the S-shaped curve relating the observer's judgment to the size of the stimulus difference, summarised by two numbers: a location, giving the threshold, and a slope, giving the precision of discrimination. Its exact shape depends on the response the observer makes, and two conventions must be kept apart (Gescheider, 1997). When the observer judges whether a comparison is greater than the standard, the function plots the proportion of greater judgments and runs from 0 to 1: its midpoint, where the comparison is called greater half the time, is the point of subjective equality, and the difference limen is computed as half the interval of uncertainty, the stimulus range between the values judged greater on 25% and 75% of trials. When instead the observer must say on each trial which of two intervals contained the larger stimulus, the function plots the proportion correct and runs from chance — 50% in a two-alternative task — up to certainty, and the threshold is conventionally the difference correct on 75% of trials, halfway between chance and certainty. Either way the differential threshold is the stimulus difference at which the function crosses the chosen criterion, and fitting the function properly — with attention to its slope, to lapses of attention that flatten its top, and to confidence intervals on the estimate — is a technical literature in its own right (Wichmann & Hill, 2001). The second demonstration lets the reader adjust the location and slope of a psychometric function and read the resulting difference limen directly from the curve.

Classical methods are wasteful, because they spend most trials on stimulus values far from the threshold that carry little information about it. Adaptive procedures concentrate trials where they are informative by letting each stimulus depend on the responses so far: a simple staircase steps the stimulus down after a correct response and up after an error, hovering around the threshold, while model-based methods place each trial at the stimulus value expected to be most informative and maintain a running estimate of the threshold (Treutwein, 1995). The Bayesian adaptive method QUEST, for instance, keeps a probability distribution over the threshold and tests at its current best estimate on every trial, converging in far fewer trials than the method of constant stimuli (Watson & Pelli, 1983). These procedures are the modern standard for measuring the differential threshold efficiently.

Stevens' Power Law

Fechner's logarithmic law dominated for a century, but its equal-units assumption drew a direct challenge from S. S. Stevens. Rather than building sensation up indirectly from just-noticeable differences, Stevens asked observers to assign numbers directly to their sensations in magnitude estimation, reporting that one light seemed twice as bright as another or one shock three times as strong (Stevens, 1957). Across many continua the numbers followed not a logarithmic function but a power function of stimulus intensity, S = c·Iᵃ, where the exponent a is characteristic of the sensory continuum. The exponent is less than one for compressive senses such as brightness (about 0.33), close to one for line length, and greater than one for expansive senses such as electric shock (about 3.5), so that doubling the physical intensity may less than double, roughly double, or far more than double the sensation depending on the modality (Stevens, 1961). The third demonstration plots the logarithmic and power laws together and lets the reader vary the exponent to see how a single family of curves captures compressive, linear, and expansive senses.

The power law and the logarithmic law make genuinely different predictions and cannot both be exactly right, and the dispute between them structured mid-century psychophysics. Stevens argued that the power law, resting on direct judgments of sensation, needed no unverifiable assumption that all just-noticeable differences feel alike, and moved to repeal Fechner's law outright (Stevens, 1961). Yet the two are not wholly unrelated: Teghtsoonian showed that if the range of sensation across a continuum is roughly constant, Stevens' exponents become predictable from the Weber fraction, tying the power-law exponent back to the discrimination data Fechner had used (Teghtsoonian, 1971). Later theorists argued that the logarithmic and power laws can be reconciled as limiting cases of a more general relation, or as reflecting different stages of the same process, so that the choice between them is less a contradiction than a question of what each measures (Krueger, 1989). What both share is the premise that discrimination data reveal a lawful, measurable relation between stimulus and sensation.

The Signal-Detection Critique

A deeper challenge questioned the threshold concept itself. Classical psychophysics treated the threshold as a fixed sensory barrier that a stimulus difference either does or does not cross, but the observer's judgment also depends on willingness to say different — a cautious observer reports fewer differences than a lax one presented with identical stimuli. Signal-detection theory separates these two influences. It models each stimulus as producing a noisy internal magnitude, and the observer as setting a decision criterion on that internal continuum, reporting a difference when the magnitude exceeds it; discrimination performance then decomposes into sensitivity (d′), the separation between the internal distributions for the two stimuli, and bias, the placement of the criterion (Green & Swets, 1966). On this account the classical threshold conflates a genuine limit on sensitivity with a movable decision policy, and Swets marshalled the evidence to argue that there is no fixed sensory threshold at all, only a continuum of sensory evidence read out through an adjustable criterion (Swets, 1961).

The reinterpretation does not abolish the differential threshold but relocates it. The just-noticeable difference remains a useful summary of discriminative acuity, but it is now understood as a point on a sensitivity function rather than a wall in the nervous system, and careful measurement uses detection-theoretic indices that are uncontaminated by response bias (Macmillan & Creelman, 2005). There is a historical irony in this resolution: the signal-detection analysis of discrimination echoes ideas latent in Fechner's own treatment of variability, so that the theory presented as overturning classical psychophysics in part rediscovers it (Link, 1994). Figure 1 shows how Weber's proportional threshold produces Fechner's logarithmic sensation scale, the relationship at the heart of the classical account.

Worked Example

The link from Weber's law to Fechner's law can be worked by hand, and reproduces the first and third demonstrations. Take the discrimination of lifted weights, whose Weber fraction is about k = 0.02. Weber's law, ΔI = k·I, fixes the just-noticeable difference at each baseline: at I = 100 g the increment needed is 0.02 × 100 = 2.0 g; at I = 300 g it is 0.02 × 300 = 6.0 g; at I = 500 g it is 0.02 × 500 = 10.0 g. The increment grows in proportion to the baseline, yet the fraction ΔI/I stays at 0.02 throughout — the invariant that is Weber's law and that the first demonstration holds fixed as the baseline slider moves.

Now build Fechner's scale by counting these just-noticeable differences up from an absolute threshold of I₀ = 50 g. Because each step multiplies the intensity by (1 + k), reaching intensity I takes N = log(I / I₀) / log(1 + k) steps. From 50 g to 100 g that is log(2) / log(1.02) = 35.0 just-noticeable differences; from 50 g to 200 g it is log(4) / log(1.02) = 70.0; from 50 g to 400 g it is log(8) / log(1.02) = 105.0. Each doubling of the weight — 50 to 100, 100 to 200, 200 to 400 — adds the same 35 steps of sensation. That constant increment per constant ratio is exactly a logarithm: counting Weber's proportional thresholds yields S = c·log(I / I₀) with c = 1 / log(1 + k), the sensation scale drawn in Figure 1. The same continuum measured by Stevens' magnitude estimation would instead be fitted by S = c·Iᵃ; for lifted weight the exponent is about a = 1.45, so that doubling the weight multiplies the reported sensation by 2^1.45 ≈ 2.7 rather than adding a constant — the divergence the third demonstration draws as the gap between the logarithmic and power curves.

Discussion

The differential threshold has proved more durable as a measurement than as a mechanism. Weber's law, that discrimination is governed by stimulus ratios, remains one of the most general regularities in perception, holding approximately across modalities and even extending to the discrimination of numerosity, where the ratio between two counts predicts how easily they are told apart (approximate number system). Its systematic failures — the rise near the absolute threshold, the near-miss for tone intensity — are themselves informative, pointing to the internal noise that limits discrimination (McGill & Goldberg, 1968). What has not survived unchallenged is the interpretation Fechner built on it: that the just-noticeable difference is a fixed unit of sensation from which a mental scale can be summed.

Two lines of critique bound the classical account. Stevens' power law showed that direct scaling of sensation gives a different function from the one Fechner derived indirectly, and though the two can be partly reconciled, the disagreement made clear that no single scale of sensation is forced by the data alone (Stevens, 1961; Krueger, 1989). Signal-detection theory cut deeper, dissolving the threshold into a sensitivity that reflects the nervous system and a criterion that reflects the observer's decision policy, and showing that much of what classical psychophysics measured as a sensory limit was in part a choice (Green & Swets, 1966; Swets, 1961). The modern practice keeps the differential threshold as an efficient, bias-corrected summary of discriminative acuity, measured by adaptive procedures and read from a carefully fitted psychometric function, while treating the sensory scale it once promised as a separate and harder question (Wichmann & Hill, 2001). The construct that founded quantitative psychology thus survives as its workhorse measure, even as the grander theory that motivated it has been revised around it.

Common Misconceptions

The differential threshold is a fixed amount of stimulus change.

It is a fixed proportion, not a fixed amount. Weber's law states that the just-noticeable difference grows with the baseline: the 2 g that is noticeable against 100 g is invisible against 1,000 g, where about 20 g is needed. Only the ratio ΔI/I stays roughly constant (Ross & Murray, 1996).

The just-noticeable difference is the smallest difference that can ever be detected.

Discrimination is probabilistic, not all-or-none. Smaller differences are still detected on some trials, and the threshold is a point on a continuous psychometric function fixed by convention — typically the difference detected 75% of the time — not a hard boundary below which nothing is seen (Gescheider, 1997).

A measured threshold reflects only the sensitivity of the senses.

A classical threshold also reflects the observer's decision criterion — how much evidence they require before reporting a difference. Signal-detection theory separates sensitivity from this response bias, showing that two observers with identical senses can yield different thresholds if they adopt different criteria (Green & Swets, 1966).

Glossary

Absolute threshold.

The least stimulus intensity that can be detected against no stimulus, the baseline I₀ from which Fechner's sensation scale is counted; the detection counterpart of the differential threshold.

Adaptive procedure.

A threshold method in which each stimulus depends on prior responses, concentrating trials near the threshold; includes staircases and Bayesian methods such as QUEST.

Decision criterion.

In signal-detection theory, the amount of internal evidence an observer requires before reporting a difference; its placement is bias, separable from sensitivity.

Difference limen.

A synonym for the differential threshold: the stimulus difference discriminated at the criterion proportion of trials.

Differential threshold.

The smallest detectable difference between two stimuli, measured as the just-noticeable difference and expressed as an increment ΔI on a baseline I.

Fechner's law.

The logarithmic relation S = c·log(I/I₀) between sensation and intensity, obtained by integrating Weber's law under the assumption that every just-noticeable difference is an equal unit of sensation.

Just-noticeable difference (JND).

The operational measure of the differential threshold: the stimulus increment that is noticed on a criterion proportion of trials.

Magnitude estimation.

Stevens' direct scaling method in which observers assign numbers proportional to the perceived intensity of stimuli, yielding the power law rather than the logarithmic law.

Method of constant stimuli.

A classical procedure presenting fixed comparison values in random order and fitting a psychometric function to the proportion of greater-than judgments.

Near-miss to Weber's law.

The systematic slow decline of the Weber fraction with intensity for pure-tone discrimination, described by an exponent on intensity slightly below one.

Point of subjective equality.

The comparison value judged equal to the standard as often as not, the centre of the psychometric function from which the difference limen is measured.

Power law (Stevens' law).

The relation S = c·Iᵃ between sensation and intensity, with an exponent a characteristic of the sensory continuum, obtained from direct magnitude estimation.

Psychometric function.

The curve relating the probability of a correct discrimination to the size of the stimulus difference; its location gives the threshold and its slope the precision.

Sensitivity (d′).

In signal-detection theory, the separation between the internal distributions evoked by two stimuli, a bias-free index of discriminability.

Signal-detection theory.

The framework that decomposes a discrimination judgment into sensitivity and decision criterion, reinterpreting the classical threshold as a movable criterion on a continuum of evidence.

Weber fraction.

The ratio k = ΔI/I of the just-noticeable difference to the baseline intensity, approximately constant within a modality and a compact index of discriminative acuity.

Weber's law.

The regularity that the just-noticeable difference is a constant fraction of the baseline stimulus, ΔI/I = k, the founding empirical law of psychophysics.

Key Researchers

Gustav Theodor Fechner (1801-1887). Physicist and philosopher at the University of Leipzig; in Elemente der Psychophysik (1860) he named Weber's law, treated the just-noticeable difference as a unit of sensation, and derived the logarithmic psychophysical law, founding psychophysics. Wikipedia - Wikidata

David M. Green (1932-2022). Auditory psychophysicist, latterly at the University of Florida; with John Swets he co-authored Signal Detection Theory and Psychophysics (1966), recasting the sensory threshold as a separation of sensitivity from decision criterion. Wikidata

Stanley Smith Stevens (1906-1973). Professor of psychophysics at Harvard University; he introduced magnitude estimation and the power law of sensation, arguing that sensation grows as a power function of intensity and moving to repeal Fechner's logarithmic law. Wikipedia - Wikidata

John A. Swets (1928-2016). Psychophysicist at Bolt Beranek and Newman and formerly MIT; a co-developer of signal-detection theory in psychology, he argued in Is there a sensory threshold? (1961) that classical threshold data reflect a movable decision criterion rather than a fixed limen. Wikipedia - Wikidata

Ernst Heinrich Weber (1795-1878). Physiologist at the University of Leipzig; from experiments on lifted weights and tactile discrimination he established that the differential threshold is a constant fraction of the baseline stimulus, the regularity Fechner would name Weber's law. Wikipedia - Wikidata

Felix A. Wichmann. Professor of Neural Information Processing at the University of Tübingen; his analysis of the psychometric function set a modern standard for fitting perceptual thresholds and their confidence intervals from psychophysical data. Faculty Page - Google Scholar - ORCID

Frequently Asked Questions

What is the differential threshold?
It is the smallest change in a stimulus that an observer can reliably detect, measured as the just-noticeable difference and fixed at a criterion proportion of correct discriminations, such as the difference detected on 75% of trials (Gescheider, 1997).

How does the differential threshold differ from the absolute threshold?
The absolute threshold is the least intensity detectable against no stimulus, asking whether something is present; the differential threshold is the least detectable difference between two stimuli, asking whether they differ (Fechner, 1860).

What is Weber's law?
Weber's law states that the just-noticeable difference is a constant fraction of the baseline stimulus, ΔI divided by I equals a constant k, so a proportionally larger change is needed to notice a difference against a stronger standard (Ross & Murray, 1996).

What is the Weber fraction?
It is the ratio of the just-noticeable difference to the baseline intensity, a single number that indexes the acuity of a sense; it is about 0.02 for lifted weight and larger for coarser discriminations (Ekman, 1959).

How is Fechner's law related to Weber's law?
Fechner integrated Weber's law, treating each just-noticeable difference as an equal step of sensation, to obtain a logarithmic relation between stimulus and sensation, S equals c times the logarithm of I over I₀ (Fechner, 1860).

How does Stevens' power law challenge Fechner's law?
Using direct magnitude estimation, Stevens found that sensation follows a power function of intensity rather than a logarithmic one, with an exponent that varies by sense, and he argued this needed no assumption that all just-noticeable differences feel alike (Stevens, 1957).

How is a differential threshold measured?
By classical methods of limits, constant stimuli, or adjustment, or by adaptive procedures that place trials near the threshold, with the result read from a fitted psychometric function relating the probability of discrimination to the difference (Wichmann & Hill, 2001).

Does signal-detection theory replace the differential threshold?
It relocates rather than abolishes it, separating a bias-free sensitivity from the decision criterion so that the threshold becomes a point on a sensitivity function rather than a fixed sensory barrier (Green & Swets, 1966).

References