Where does sensitivity collide?

In diagnostic testing, sensitivity is the proportion of people who truly have a condition that a test correctly labels as positive: it is the true positive rate. It sits opposite specificity, the true negative rate, and the two are read together from the same table of true and false positives and negatives. Altman and Bland's 1994 note Diagnostic tests 1: sensitivity and specificity defined the pair and warned that neither number alone tells you how useful a test is. A test can look strong on one measure while failing badly on the other, and moving the decision threshold trades one against the other, which is what a ROC curve traces. This is a property of a classifier scored against a known truth, not a feature of a model, a sense organ, or a temperament.[1]

Unit

Proportion of true positives correctly identified, from 0 to 1

Recovery means

Not applicable

Measured by

Computed as true positives divided by all who truly have the condition, using a test scored against a gold standard, and plotted against the false positive rate as a ROC curve

In modeling and analysis, sensitivity analysis asks how much the output of a model changes when its inputs are varied: it tells you which assumptions the result really depends on. A local analysis nudges one input at a time around a point, while a global analysis varies all of them across their whole range and shares out the output variation among them, for example through Sobol variance indices. Saltelli's 2002 paper Sensitivity analysis for importance assessment set out this variance-based view and argued that ranking inputs by how much they drive the output is the honest way to judge their importance. Here sensitivity is a property of a model and its inputs, not a test score, a detection threshold, or a personality trait. A model can be highly sensitive to an input no one thought mattered, which is exactly what the analysis is meant to reveal.[2]

Unit

Share of output variation traced to a given input

Recovery means

Not applicable

Measured by

Estimated by varying inputs and recording how the output moves, from one-at-a-time local derivatives to global variance decompositions such as Sobol indices

In perception, sensitivity is how faint a stimulus can be while a person still detects it: a more sensitive observer notices weaker signals, so their absolute threshold is lower. Classical psychophysics tried to measure this threshold directly, but a faint signal is always mixed with noise, and how often someone says yes also depends on how willing they are to guess. Tanner and Swets's 1954 paper A decision-making theory of visual detection separated these two things, treating detection as a decision and measuring true sensitivity as d prime, the distance between the signal and the noise, apart from the observer's bias. So sensitivity here is a property of a sensory system and how well it tells signal from noise, not a test's true positive rate, a model's response to inputs, or a trait of temperament.[3]

Unit

The faintest stimulus a person can reliably detect

Recovery means

Not applicable

Measured by

Estimated in detection experiments as an absolute threshold, or as d prime in a signal detection task that separates true sensitivity from response bias

In personality research, sensory-processing sensitivity is a proposed temperament trait: some people process what they sense more deeply and react more strongly to their surroundings, whether pleasant or harsh. On this view a highly sensitive person notices subtle stimuli, is more easily overwhelmed by intense or busy settings, and is more affected by both supportive and stressful environments. Aron and Aron's 1997 paper Sensory-processing sensitivity and its relation to introversion and emotionality introduced the trait and a scale to measure it, and argued that it is distinct from introversion and from simple emotional reactivity. Here sensitivity names a stable difference between people, gauged by self-report, not a test statistic, a model property, or a bare detection threshold. It describes a temperament, and this page treats it as a research construct rather than a clinical diagnosis.[4]

Unit

Where a person falls on a trait of heightened sensitivity

Recovery means

Not applicable

Measured by

Assessed by self-report scales such as the Highly Sensitive Person scale introduced by Aron and Aron