Where does contrast collide?

In the analysis of variance, a contrast is a weighted combination of group averages whose weights add up to zero, built to test one specific planned comparison, for example whether the combined average of two treatment groups differs from a control. Because the weights sum to zero, the contrast isolates a single question about the pattern of means, and a set of contrasts can be made orthogonal so that each asks a non-overlapping question. Abelson and Prentice's 1997 Contrast tests of interaction hypothesis showed how such focused comparisons test a precise prediction rather than a vague overall effect. Here a contrast is an abstract number computed from data to answer a planned question, not a difference in brightness the eye detects, a difference in signal between tissues, or a difference between two speech sounds.[1]

Unit

A single weighted combination of group means, with coefficients summing to zero

Recovery means

Not applicable

Measured by

Computed from the group means and tested with a t or F statistic against its standard error

In vision science, contrast is the difference in luminance or colour that lets an object be told apart from its surroundings: a pattern with high contrast has a large gap between its light and dark parts, and the smallest gap a viewer can still detect defines the limit of their contrast sensitivity. It is scaled by formulas such as Michelson contrast, the difference between the lightest and darkest luminance divided by their sum. Campbell and Robson's 1968 Application of Fourier analysis to the visibility of gratings measured how this detectable contrast varies with the fineness of a striped pattern, mapping the eye's contrast sensitivity. Here contrast is a property of a visual scene and of the observer's ability to detect it, not a weighted number in a table, a signal gap in a medical scan, or a meaning-bearing difference between sounds.[2]

Unit

A luminance or colour difference, scaled between the light and dark regions of a scene

Recovery means

Not applicable

Measured by

Quantified as Michelson or Weber contrast, and probed by contrast thresholds for gratings

In radiology, contrast is the difference in signal, brightness, or density between two tissues in a medical image: a scan has good contrast when neighbouring structures, such as a tumour and the organ around it, produce clearly different values and can be told apart. When the natural difference is too small to read, it is often boosted by injecting a contrast agent, a substance that changes how a tissue absorbs X-rays or responds in magnetic resonance. Lohrke and colleagues' 2016 review 25 years of contrast-enhanced MRI traced how such agents widen the signal difference between tissues to make disease visible. Here contrast is a measurable property of an acquired image and the agents that enhance it, not a statistical weighting, the eye's own sensitivity to brightness, or a difference between speech sounds.[3]

Unit

A signal, brightness, or density difference between two tissues in an image

Recovery means

Not applicable

Measured by

Assessed as a contrast-to-noise ratio between regions, and enhanced by injected contrast agents

In phonology, a contrast is a difference between two speech sounds that can distinguish the meaning of words in a language: if swapping one sound for another turns one word into a different word, as the p and b do in pat and bat, the two sounds stand in contrast and count as separate phonemes. Such minimal pairs are the classic evidence that a difference is contrastive rather than a mere variation that carries no change in meaning. Swadesh's 1934 The phonemic principle set out how linguists identify these meaning-distinguishing units within a particular language. Here a contrast is a functional, meaning-bearing opposition between sounds, specific to one language, not a weighted number, a brightness difference the eye detects, or a signal gap in a medical scan.[4]

Unit

A meaning-distinguishing opposition between two sounds in a given language

Recovery means

Not applicable

Measured by

Established by minimal pairs and contrastive distribution within a language