Goleman’s 1995 book made a specific claim about leadership: that emotional intelligence accounts for approximately 90% of the difference between average and outstanding leaders. The claim has been repeated in corporate training programmes, executive coaching practices, and leadership development curricula ever since. In the peer-reviewed literature, there is no study that produces this finding. The number does not appear in the research. The evidence that does exist shows that EI contributes meaningfully to leadership effectiveness in specific conditions — but that its contribution sits alongside cognitive ability, personality, situational factors, and leadership behaviour in ways that assign it nothing close to the dominant role the popular account describes.

The gap between what the EI research actually shows and what the EI industry claims it shows matters practically, because it determines where the investment in EI development is most justified and where it is likely to produce disappointing returns.

What the meta-analytic evidence establishes

Dana Joseph and Daniel Newman’s 2010 meta-analysis in the Journal of Applied Psychology is the most methodologically rigorous available assessment of where EI predicts job performance and where it does not. Their cascading model established the finding that popular EI accounts consistently omit: EI predicts job performance most strongly in jobs with high emotional labour demands, those requiring sustained emotional management in direct interpersonal interaction. For jobs without high emotional labour demands, the incremental validity of EI over general cognitive ability and Big Five personality traits is weak to non-significant.

The practical consequence of this finding is substantial. Most of the people who go through corporate EI training programmes are in jobs where emotional labour is a moderate rather than dominant component of performance. For those people, the research predicts that EI training will produce modest returns on the performance outcomes the training is positioned as improving. For professionals in genuine high-emotional-labour roles, the prediction is stronger and the investment is more justified.

Van Rooy and Viswesvaran’s 2004 meta-analysis established the incremental validity picture more broadly: EI shows significant zero-order correlations with job performance, which is the basis for the popular claim that EI matters for outcomes. Once general cognitive ability and Big Five personality are accounted for, the additional variance that EI explains is real but substantially smaller than popular accounts suggest. The correlations are genuine; the increment over what standard predictors already capture is modest.

The measurement problem

The most consequential unresolved issue in EI research is the distinction between ability-measured and self-report-measured EI, and the dramatically different findings the two approaches produce. The Mayer-Salovey-Caruso Emotional Intelligence Test measures EI through performance tasks with objectively scorable answers — the ability model that the first article in this series described. Self-report EI measures ask people to rate their own emotional competencies on questionnaire items. Research consistently documents that self-report EI shows substantial overlap with Big Five personality traits, particularly emotional stability and agreeableness, while adding relatively little specific variance. Ability-measured EI shows more specific and less personality-confounded predictions, particularly for social relationship quality and certain interpersonal outcomes.

The implication is that a large portion of what the popular EI research literature reports as EI effects may be personality effects wearing EI labels. The studies use self-report measures that are substantially measuring personality, the personality variables predict the outcomes, and the result is reported as evidence that EI predicts those outcomes. This is not fraud — the researchers are using the measures as designed. The measurement tool’s overlap with personality is the problem, and it is a problem the field has documented clearly without it having much effect on how EI is sold or used commercially.

What EI genuinely predicts

The research does not suggest that EI is unimportant. The specific domains where the evidence is robust are worth identifying precisely, because they are the domains where EI development investment is most justified.

Quality of social relationships is the most consistently replicated EI prediction in the ability model research. Individuals who score higher on the MSCEIT report higher relationship satisfaction, are rated by relationship partners as more effective in interpersonal conflict, and show better outcomes in collaborative contexts that require emotional attunement. The effect is specific to the ability model and weaker or absent in self-report EI research, which suggests it is genuine cognitive ability rather than personality that is driving it.

Emotional perception accuracy, the first branch of the ability model, predicts negotiation outcomes in interpersonally demanding contexts. The mechanism is the informational one: the negotiator who accurately reads the counterpart’s emotional state is making decisions with better data about what the counterpart actually needs and how they are responding to proposals. This prediction holds in experimental studies and in research on professional negotiators.

Emotion regulation under evaluative threat — the high-stakes performance conditions this series has examined — shows robust prediction from ability EI. The entrepreneur in the investor meeting, the leader in the crisis, the professional in the evaluation conversation: these are exactly the conditions where the ability model’s prediction is most specific and most useful.

The cultural specificity limit

David Matsumoto’s cross-cultural research on emotional expression establishes a limit that the EI field rarely addresses. Most EI frameworks were developed with Western populations and embed Western norms about which emotional expressions are appropriate, which require regulation, and which constitute emotionally intelligent behaviour. The display rules — the cultural norms governing when and how emotions are expressed — vary substantially across cultural contexts. The emotion that constitutes appropriate assertion in one professional culture constitutes aggression in another. The deference that reads as emotional maturity in one context reads as disengagement in another. EI frameworks that do not account for this variation are applying one culture’s emotional norms as universal standards.

For the entrepreneur operating across cultures, or managing teams with significant cultural diversity, this limit is commercially consequential.

The 90% and what to do with it

Goleman’s 90% figure illustrates the mechanism through which the EI field’s popular claims became so much stronger than the research. The claim is memorable, dramatic, and actionable — it tells organisations that EI development is the primary lever for leadership performance improvement. The actual research literature, which shows EI contributing meaningfully alongside other factors in specific conditions, is considerably harder to turn into a training programme that justifies its cost. The population-level finding from a narrow domain became a universal principle through a combination of motivated reasoning, vivid anecdote, and the genuine desire to believe that the capacities the EI framework addresses are as powerful as the evidence shows they sometimes are.

The capacities are genuinely important. The evidence for them is genuinely interesting. The popular account of that evidence has consistently outrun what the studies actually show.

Books worth reading on this

Science Fictions by Stuart Ritchie. Ritchie’s account of how scientific findings are systematically inflated in the journey from peer-reviewed result to public claim covers exactly the pathway through which the EI research became the EI industry. His analysis of publication bias, outcome switching, and the specific incentives that produce overclaimed effect sizes in applied psychology research provides the methodological context for understanding why the self-report EI literature shows stronger and less conditional effects than the ability model research, and why the meta-analytic corrections to individual studies consistently produce smaller effects than the original findings reported. Reading Robson and Ritchie together gives the entrepreneur a framework for evaluating any psychological framework’s evidence base critically, which is a more durable and commercially useful skill than any specific finding about emotional intelligence.

If the dynamics described here are significantly affecting your wellbeing, speaking with a psychologist is the right next step. UK: Samaritans (116 123, free, 24/7). Mind (0300 123 3393). BACP: bacp.co.uk/search/Therapists. Crisis Text Line — text HOME to 741741 (US, UK, Canada, Ireland). International: internationaltherapistdirectory.com.

This article is for educational and informational purposes only. Sources: Joseph, D.L. & Newman, D.A. (2010), Emotional Intelligence: An Integrative Meta-Analysis and Cascading Model, Journal of Applied Psychology, 95(1), 54-78. Van Rooy, D.L. & Viswesvaran, C. (2004), Emotional Intelligence: A Meta-Analytic Investigation of Predictive Validity and Nomological Net, Journal of Vocational Behavior, 65(1), 71-95. Mayer, J.D., Caruso, D.R. & Salovey, P. (2016), The Ability Model of Emotional Intelligence: Principles and Updates, Emotion Review, 8(4), 290-300. Goleman, D. (1995), Emotional Intelligence, Bantam Books. Antonakis, J., Ashkanasy, N.M. & Dasborough, M.T. (2009), Does Leadership Need Emotional Intelligence?, The Leadership Quarterly, 20(2), 247-261. Matsumoto, D., Yoo, S.H. & Nakagawa, S. (2008), Culture, Emotion Regulation, and Adjustment, Journal of Personality and Social Psychology, 94(6), 925-937. Robson, D. (2019), The Intelligence Trap, Hodder & Stoughton. Ritchie, S. (2020), Science Fictions, Bodley Head.