What Is Emotional Intelligence? A Popular Idea With a Contested Basis

The term was introduced in a 1990 paper by Peter Salovey and John Mayer, who defined it as the ability to monitor one's own and others' feelings, to discriminate among them, and to use that information to guide thinking and action. Their framing was deliberately narrow: a set of mental abilities concerning emotional information, analogous to verbal or spatial ability and measurable in the same way, with right and wrong answers. Daniel Goleman, a science journalist, popularised the idea in 1995 in a book that broadened it enormously, folding in motivation, empathy, self-control, optimism and social skill, and asserting that this cluster accounted for the majority of success while conventional intelligence accounted for very little. The figures he gave had no empirical basis and he later distanced himself from them, but the claim is what sold the book and it is what most people still believe the research says.

Much of the confusion in the field comes from the word being used for three distinct constructs measured in incompatible ways:

• •Ability models, following Mayer and Salovey, which test emotional perception, understanding and regulation with performance tasks that have scored answers, most commonly the MSCEIT. These correlate moderately with general intelligence and behave like a cognitive ability
• •Trait models, which use self-report questionnaires asking people how emotionally skilled they consider themselves. These correlate strongly with established personality traits, particularly low neuroticism, extraversion and conscientiousness, and are better described as a personality profile than an intelligence
• •Mixed models, following Goleman, which combine abilities, traits, motivations and competencies into a single package designed for organisational use, and which are the basis of most commercial assessments
• •The three correlate only weakly with each other, which means studies reporting on emotional intelligence are frequently not measuring the same thing

Meta-analyses give a consistent and moderate picture. Emotional intelligence, on any measure, predicts job performance, but the effect is modest and shrinks considerably once general mental ability and the five major personality traits are controlled for, which is the crucial test of whether a new construct adds anything. The incremental contribution is real but small in most jobs and largest in work with heavy emotional demands, including nursing, teaching, sales, customer service and management, where regulating one's own feelings and reading other people's is a core task rather than a peripheral one. Beyond work, it predicts relationship satisfaction, mental health and lower burnout reasonably well, though the self-report measures are partly predicting themselves here, since a questionnaire about emotional wellbeing correlates with a questionnaire about emotional skill for uninteresting reasons. Academic performance shows a small effect. The claim that it matters more than conventional intelligence has no support in any of this literature.

Assessing an ability by asking people to rate it is the weakest method available in psychology, and it fails here for the predictable reason that the people worst at reading emotions are also worst placed to notice that they are. Self-report is also trivially easy to fake in a hiring context, where every candidate knows which answer sounds better. The performance-based alternative has its own difficulty: a test of emotional understanding needs a correct answer, and correctness is established either by expert consensus or by what the majority of respondents say, neither of which is obviously the truth, and both of which embed the culture the test was written in. A face judged as expressing contempt in one society may not be read that way in another. These problems are why the field has not converged on a single measure after three decades, and why commercially sold assessments should be treated with care.

The training industry built on the idea is large and its evidence is mixed but not empty. Reviews of workplace programmes find small to moderate improvements on the measures used, which is unsurprising when the measure is a questionnaire and the participants have just spent two days learning what the desirable answers are; changes in actual behaviour and performance are less consistently demonstrated and less often looked for. The better evidence comes from school-based social and emotional learning programmes, where large meta-analyses covering hundreds of studies have found improvements in behaviour, emotional adjustment and, notably, academic attainment, with effects persisting for months or years. The plausible summary is that specific component skills, recognising an emotion, naming it, pausing before responding, taking another person's perspective, can be taught and practised like other skills, while the broad trait-like version is close to personality and about as changeable.

Defined by Salovey and Mayer in 1990 as an ability to perceive, understand and use emotional information, and popularised in 1995 with the unsupported claim that it outweighs conventional intelligence, the idea is now measured in three incompatible ways: performance tests that behave like a cognitive ability, self-report questionnaires that largely measure personality, and commercial mixed models. It predicts job performance modestly, adds little once intelligence and personality are accounted for except in emotionally demanding work, and its component skills appear teachable even where the broad trait is not.