Goleman's Emotional Intelligence: The Best-Selling Construct Academic Psychology Doesn't Recognize

Daniel Goleman’s 1995 book sold 5+ million copies and launched a multibillion-dollar EI coaching and assessment industry. The empirical reality: meta-analyses (Joseph & Newman 2010; O’Boyle 2011) show “mixed-model” EI is largely a repackaging of Big Five personality, ability EI adds tiny incremental validity over IQ, and the “EI matters more than IQ” claim has no empirical basis. Here is the honest story for CEOs evaluating EI assessments.

Picture the meeting. A leadership-development vendor is presenting to your executive team. The pitch deck features brain icons, the Goleman four-quadrant model (self-awareness, self-management, social awareness, relationship management), and a chart claiming that “emotional intelligence accounts for nearly 90% of the difference between star performers and average ones in senior leadership roles.” The proposed engagement: a year-long EI assessment-and-coaching program for forty managers, priced around $400,000. The assessment instrument is one of the brand-name commercial EI inventories — perhaps the EQ-i 2.0, the ESCI (Emotional and Social Competency Inventory), or a Goleman-licensed derivative. The vendor’s collateral cites a 2008 Harvard Business Review article and Goleman’s bestselling books. The line that closes the room: “EI predicts career success more reliably than IQ does.”

The empirical literature behind that pitch — the actual peer-reviewed meta-analytic work on EI and job performance, conducted by industrial/organizational psychologists with no commercial stake in EI assessment tools — tells a substantially different story. The headline finding, repeated across multiple large meta-analyses, is that the popular “mixed-model” version of EI (the one Goleman popularized, the one most commercial EI assessments measure) overlaps so heavily with already-established personality traits — primarily conscientiousness, extraversion, agreeableness, and emotional stability — that it is not clearly measuring a distinct cognitive capacity at all. The narrower, academically more defensible “ability-based” version of EI (the Mayer-Salovey-Caruso MSCEIT) does appear to measure something somewhat distinct, but its incremental validity over IQ and personality for predicting job performance is small — under one percentage point of explained variance in Joseph and Newman’s (2010) cascading-model meta-analysis. The claim that EI matters more than IQ for job success, which Goleman attributed (incorrectly, as it turns out) to corporate research, has no empirical basis.

This article walks through what Goleman actually claimed in 1995, what the academic EI construct Salovey and Mayer proposed in 1990 looked like before popularization, the critical distinction between ability-based and mixed-model EI, what the major meta-analyses (Joseph & Newman 2010, O’Boyle et al. 2011) actually found about incremental validity, why academic critics like John Antonakis have called the EI construct “fundamentally flawed” as commercially operationalized, and what is honest to say about emotion-related skills for hiring, leadership development, and HR investment. The goal is calibration, not takedown. There are real, trainable emotion-related skills. There is a real if narrow academic EI construct. What the empirical literature does not support is the popular framing that “emotional intelligence” is a unified, separately measurable capacity that predicts career success above and beyond personality and IQ — and that justifies the assessment and training industry built around it.

What Goleman 1995 Actually Claimed

The book that launched the popular EI industry is Daniel Goleman, Emotional Intelligence: Why It Can Matter More Than IQ (Bantam Books, 1995). Goleman was a science journalist for The New York Times at the time, with a PhD in clinical psychology from Harvard but without a research program in the EI field. The book synthesized work from neuroscientists (Joseph LeDoux on the amygdala), educational psychologists (Howard Gardner on multiple intelligences), and the small academic EI literature that existed at the time (Peter Salovey and John Mayer’s 1990 paper, primarily) into a sweeping argument about the importance of emotion regulation, self-awareness, empathy, and social skill for life success.

The book’s framing — captured in the subtitle, Why It Can Matter More Than IQ — is what made it a phenomenon. Goleman argued that IQ accounts for “at best about 20 percent of the factors that determine life success, which leaves 80 percent to other forces” (p. 34), and that emotional intelligence was a substantial portion of that remaining 80%. He repeatedly cited corporate-research claims that “emotional intelligence” predicted star performance in management roles two-to-one over technical skills and IQ. In his 1998 follow-up Working with Emotional Intelligence, he wrote: “for star performance in all jobs, in every field, emotional competence is twice as important as purely cognitive abilities.” This is the line that Joseph and Newman (2010, p. 69) singled out in their meta-analysis as empirically unsupported.

Goleman’s model — four quadrants, eighteen competencies, with quadrants for self-awareness, self-management, social awareness, and relationship management — was not derived from a factor-analytic study or a single research program. It was a synthetic framework assembled from multiple lines of work, packaged for a general business audience. The Hay Group (now Korn Ferry) subsequently developed the Emotional Competence Inventory (ECI, later ESCI) as a commercial assessment instrument operationalizing Goleman’s quadrants. The Six Seconds organization, the Bar-On model (EQ-i), and several other commercial assessment systems built on related “mixed model” foundations.

It is important to be fair to Goleman. He did not fabricate data, did not commit academic fraud, did not misrepresent himself as the inventor of EI (he credits Salovey and Mayer), and he did not personally build the entire commercial industry that grew up around the book. The original academic concept came from Salovey and Mayer; Goleman popularized it. The commercial assessment market emerged because there was massive demand from corporate HR for psychological constructs that sounded scientific and actionable. Goleman was the right messenger at the right cultural moment. The criticism the academic literature has leveled — and that this article is summarizing — is not that Goleman was dishonest, but that the empirical foundation for the strongest claims in the book (especially “EI matters more than IQ”) was always thinner than the popular framing suggested, and that the commercial assessment industry that followed has continued making those claims long after the meta-analytic evidence stopped supporting them.

What Salovey & Mayer 1990 Actually Proposed

The academic origin of the EI construct is Peter Salovey and John D. Mayer, “Emotional Intelligence,” published in Imagination, Cognition and Personality, volume 9, issue 3, pages 185–211, in 1990. Salovey (then at Yale, later its President) and Mayer (at the University of New Hampshire) proposed EI as a specific type of intelligence — that is, a cognitive ability — involving the capacity to perceive, appraise, and express emotion accurately; to use emotion to facilitate thought; to understand emotion and emotional knowledge; and to regulate emotion in oneself and others.

The 1990 paper is a constrained, modest academic conception. It does not claim EI matters more than IQ. It does not promise that EI predicts career success or stellar leadership performance. It does not claim EI is the secret to romantic success, parenting, or organizational effectiveness. It proposes EI as one specific kind of intelligence — analogous to spatial intelligence or verbal intelligence — that might be measurable through tasks with right and wrong answers, and that might add some incremental validity over standard cognitive ability tests for predicting outcomes in emotionally-laden domains.

Mayer and Salovey subsequently developed this into the four-branch ability model (Mayer & Salovey, 1997) and operationalized it through the Multifactor Emotional Intelligence Scale (MEIS, 1999) and then the Mayer-Salovey-Caruso Emotional Intelligence Test (MSCEIT, 2002). The MSCEIT scores respondents on tasks like identifying the emotion expressed by a face, predicting how emotions blend or evolve over time, and selecting effective emotion-regulation strategies for hypothetical scenarios. Scoring is done either by consensus (compared to a large normative sample) or by expert judgment (compared to a panel of emotion researchers). The two scoring methods correlate highly.

The academic ability-EI tradition, in other words, was always narrower and more cautious than the Goleman popularization. Mayer, Roberts, and Barsade’s 2008 Annual Review of Psychology summary (“Human Abilities: Emotional Intelligence,” volume 59, pp. 507–536, DOI: 10.1146/annurev.psych.59.103006.093646) explicitly distinguishes the “specific-ability and integrative-model approaches” that they consider scientifically defensible from the “mixed-model” approaches (Goleman, Bar-On, etc.) that they describe as inadequately specified. The same 2008 review notes that even within the defensible ability tradition, the incremental validity over established cognitive ability and personality measures is modest, and that the practical importance of EI has been overstated in popular treatments.

The Two Models — Ability-EI Versus Mixed-Model-EI

The single most important distinction for evaluating an EI assessment, training program, or research claim is the difference between ability-based EI and mixed-model EI. These are two genuinely different things often labeled with the same term, and conflating them is the source of most of the popular confusion in this area.

Ability-based EI treats emotional intelligence as a cognitive ability — analogous to verbal or spatial intelligence — measurable through performance-based tasks with right and wrong answers (scored by consensus or expert judgment). The flagship instrument is the MSCEIT (Mayer-Salovey-Caruso Emotional Intelligence Test, 2002), with 141 items across four branches: perceiving emotions, using emotions to facilitate thought, understanding emotions, and managing emotions. Ability EI correlates moderately with general cognitive ability (g) — Joseph and Newman (2010) found a corrected correlation of about ρ = .22 between performance-based ability EI and cognitive ability, consistent with the idea that ability EI is in fact a form of intelligence rather than a personality trait. Ability EI has modest correlations with Big Five personality traits (most below ρ = .30 in the Joseph and Newman meta-analysis, with the strongest being agreeableness at ρ ≈ .29). The MSCEIT has known limitations: modest test-retest reliability for some subscales, scoring controversies, and concerns about whether “correct” emotion-management responses can be validly identified by consensus.

Mixed-model EI treats emotional intelligence as a heterogeneous combination of personality traits, motivational tendencies, self-perceived competencies, and emotional styles. The Goleman model, the Bar-On Emotional Quotient Inventory (EQ-i 2.0), the Emotional and Social Competency Inventory (ESCI), and most other commercial EI assessments are mixed-model instruments. They are typically self-report (you rate your own emotional skills on a Likert scale) or 360-feedback (others rate you), not performance tests. Joseph and Newman (2010, Table 2) found mixed-model EI correlates substantially with Big Five personality traits — ρ = .53 with emotional stability, ρ = .46 with extraversion, ρ = .43 with agreeableness, ρ = .38 with conscientiousness, ρ = .29 with openness — and only ρ = .11 with cognitive ability. In other words, mixed-model EI scores look much more like a composite of Big Five personality traits than like a measure of cognitive ability.

This distinction matters enormously for what an EI assessment is actually measuring. If your vendor is using the EQ-i, the ESCI, a Goleman-derived instrument, or any other self-report “mixed model” tool, you are mostly measuring a person’s Big Five personality profile dressed up in EI language. If your vendor is using the MSCEIT, you are measuring something somewhat distinct — but with weaker predictive power, more measurement controversy, and less commercial uptake. The MSCEIT itself shows only ρ ≈ .12 correlation with self-report ability EI (Joseph & Newman 2010, Table 3) — these two “ability EI” methods barely converge, suggesting they may not even be measuring the same construct.

The Meta-Analytic Reckoning

The most rigorous meta-analytic test of EI’s incremental predictive validity over standard personality and cognitive ability measures is Dana L. Joseph and Daniel A. Newman, “Emotional Intelligence: An Integrative Meta-Analysis and Cascading Model,” published in the Journal of Applied Psychology, volume 95, issue 1, pages 54–78, in 2010 (DOI: 10.1037/a0017286). They synthesized meta-analytic correlations across all three construct-method pairings of EI (performance-based ability EI like the MSCEIT, self-report ability EI like the WLEIS, and self-report mixed EI like the EQ-i) with the Big Five personality traits, cognitive ability, and job performance.

The headline incremental-validity results from Joseph and Newman’s Table 5c — which shows the unique variance each EI type adds to job performance prediction after controlling for both Big Five personality and cognitive ability — are sobering for the popular EI narrative. Performance-based ability EI added a statistically significant but tiny 0.2% of incremental variance (ΔR² = .002, p < .05) over Big Five and cognitive ability for predicting overall job performance. Self-report ability EI added 2.3% (ΔR² = .023). Self-report mixed EI — the kind Goleman popularized and most commercial assessments measure — added 14.2% (ΔR² = .142) of incremental variance over Big Five and cognitive ability.

That mixed-model figure (14.2%) sounds impressive until you understand what it actually means. Joseph and Newman themselves are explicit about the interpretation: “this incremental validity is likely attributable to the nonability (motivational) content of mixed EI measures… that is not captured by ability EI measures” (p. 66). In other words, mixed-model EI predicts incrementally over Big Five and cognitive ability because mixed-model EI contains content beyond Big Five and cognitive ability — specifically, self-efficacy, motivational tendencies, and self-perceived competence. The “incremental validity” is not telling you that emotional intelligence is a powerful new construct; it is telling you that the mixed-model EI scale includes a bunch of additional self-report content that happens to predict job performance, and that content is not “EI” in any conceptually clean sense. It is mostly motivation and self-efficacy with a thin coating of emotion-language items.

The second major meta-analysis is Ernest H. O’Boyle, Ronald H. Humphrey, Jeffrey M. Pollack, Thomas H. Hawver, and Paul A. Story, “The Relation Between Emotional Intelligence and Job Performance: A Meta-Analysis,” published in the Journal of Organizational Behavior, volume 32, issue 5, pages 788–818, in 2011 (DOI: 10.1002/job.714). O’Boyle and colleagues built on Joseph and Newman’s framework, expanded the dataset (65% more studies, more than twice the sample size), and used dominance analysis to estimate relative importance.

The three EI streams in O’Boyle et al. showed corrected correlations with job performance of r_c = .24 (Stream 1, performance-based ability EI), r_c = .30 (Stream 2, self-report ability EI based on the four-branch model), and r_c = .28 (Stream 3, mixed-model EI). All three are statistically significant. All three are also modest in absolute terms — well below the corrected correlations between general cognitive ability and job performance (typically reported around r ≈ .50–.65 across the broad I/O literature) or between conscientiousness and job performance (typically r ≈ .25–.35). The streams correlate differently with cognitive ability and the Big Five, indicating that — as Joseph and Newman also showed — the three EI streams are not really measuring the same construct.

O’Boyle and colleagues did find that all three EI streams contributed some incremental relative importance beyond Big Five and cognitive ability in dominance analyses, and on this basis the authors describe their results as “supporting the overall validity of EI.” But the magnitudes are modest, the differences between streams are large (suggesting they are not the same construct), and the largest incremental validity figures come from mixed-model measures whose “EI” content is heavily contaminated with personality and self-efficacy content. The two meta-analyses are broadly consistent: ability-based EI adds little incremental prediction over IQ and Big Five; mixed-model EI adds more, but mostly because it includes content that is not really EI.

The “Jangle Fallacy” Problem

Psychologists have a term for the situation where two scientific constructs share a common name but measure different things — the “jingle fallacy” (different things, same name). They have a complementary term for the inverse — when the same thing gets measured under different names, creating the illusion of multiple discoveries — the “jangle fallacy” (same thing, different names). Mixed-model EI sits squarely in jangle-fallacy territory.

If you compute self-report mixed EI correlations with the Big Five personality traits from Joseph and Newman’s (2010) Table 2, the picture is unmistakable. Mixed-model EI correlates ρ = .53 with emotional stability, ρ = .46 with extraversion, ρ = .43 with agreeableness, ρ = .38 with conscientiousness, ρ = .29 with openness. The multiple correlation between mixed-model EI and the Big Five composite is in the range where one might reasonably describe mixed-model EI as a weighted composite of Big Five personality traits. Schulte, Ree, and Carretta (2004), in their pointedly titled paper “Emotional Intelligence: Not Much More Than g and Personality,” published in Personality and Individual Differences, volume 37, issue 5, pages 1059–1068 (DOI: 10.1016/j.paid.2003.11.014), showed that EI scores could be largely predicted from cognitive ability (g), the personality trait of agreeableness, and the respondent’s sex. The “EI” label, they argued, was largely a re-naming of constructs already well-established in personality and intelligence research.

This is not a fatal critique of EI as a research program — there are legitimate scientific reasons to study how Big Five traits and cognitive ability combine to produce competent emotion-related behavior in real-world settings. But it is a serious challenge to the commercial EI assessment industry, which sells the proposition that EI is a distinct measurable capacity that has been neglected by traditional psychometric approaches and that organizations need to measure and develop separately from personality and cognitive ability. The empirical reality is that the popular mixed-model EI assessments are mostly measuring Big Five personality (especially the emotional-stability-extraversion-agreeableness cluster) with some self-efficacy and motivational content layered on top. If your organization already has good Big Five personality data on candidates or employees — which is inexpensive, well-validated, and decades-mature as a psychometric technology — there is essentially no incremental information to be gained from adding a mixed-model EI assessment on top.

The Trainability Question

The corporate EI training industry rests on the claim that EI is malleable — that you can hire someone with average EI and through targeted training significantly improve their EI, with downstream improvements in leadership effectiveness, team performance, customer satisfaction, and so forth. The empirical evidence for this claim is weaker than the marketing collateral implies.

Meta-analyses of EI training interventions (for example, Mattingly and Kraiger’s 2019 meta-analysis in Human Resource Management Review) generally find statistically significant but small-to-moderate effect sizes (often around d = 0.40–0.50 on EI assessment outcomes immediately post-training). The problem is identifying what these effect sizes mean. Post-training increases on self-report EI scales — which are the typical outcome measure — could reflect genuine skill acquisition, but they could also reflect demand characteristics (participants know they are supposed to “improve” and answer the post-test accordingly), social desirability, or simply learning the language and frame of the training program. When studies use behavioral outcome measures (supervisor ratings of effectiveness, objective performance data, customer satisfaction scores), effect sizes tend to be smaller and harder to attribute specifically to “EI development” versus general communication-skills training or general leadership development.

A reasonable summary of the trainability literature, fair to both sides: there is some evidence that targeted training in specific emotion-related behaviors — active listening, perspective-taking, emotion-regulation strategies like cognitive reappraisal, conflict de-escalation techniques — can produce real behavioral change. There is much weaker evidence that the unified, abstract construct of “emotional intelligence” is being trained, as opposed to a collection of specific social-cognitive skills that have been studied under various other labels for decades. The cleanest, most actionable behavioral-change literature in this space is probably the cognitive-behavioral therapy and behavioral skills training literatures, not the EI training literature per se. If you want managers to handle difficult conversations better, train them in the specific conversational skills involved. The “EI” wrapper adds branding, not measurable incremental effectiveness.

What Antonakis 2009 Concluded

The sharpest published academic critique of EI as a leadership construct is John Antonakis, Neal M. Ashkanasy, and Marie T. Dasborough, “Does Leadership Need Emotional Intelligence?” published in The Leadership Quarterly, volume 20, issue 2, pages 247–261, in 2009 (DOI: 10.1016/j.leaqua.2009.01.006). Antonakis (University of Lausanne) is a leadership researcher with a longstanding skepticism of EI’s predictive validity for leadership outcomes; Ashkanasy and Dasborough have been more sympathetic to EI work. The paper is structured as a dialogue between the two camps and is a useful read for anyone trying to understand the scientific debate.

Antonakis’s substantive critique runs along several lines. First, the construct itself is poorly specified — different EI inventories purport to measure different things and converge poorly with each other (the Joseph and Newman 2010 figure of ρ ≈ .12 between performance-based and self-report ability EI is illustrative). Second, the incremental validity over established constructs (cognitive ability and Big Five personality) is small enough that it does not justify the commercial enthusiasm. Third, the studies showing strong EI-leadership relationships often fail to properly control for personality and cognitive ability, fail to use objective leadership outcomes, and rely on common-method variance (the same person rating both their own EI and their leadership effectiveness). Fourth, the body of work cited by EI advocates is heavily contaminated by publication bias and by researchers with commercial stakes in EI assessment instruments. Antonakis’s conclusion in the dialogue is that the EI-leadership case as commercially presented is built on a foundation of weak measurement, inadequate controls, and motivated reasoning by researchers and consultants who benefit from EI being real.

A 2022 follow-up — Antonakis, “Does Leadership Still Not Need Emotional Intelligence? Continuing ‘The Great EI Debate,’” published in The Leadership Quarterly, volume 33, issue 6 (DOI: 10.1016/j.leaqua.2021.101539) — argues that more than a decade of subsequent research has not substantially improved the empirical case. The mixed-model assessments still mostly measure personality. The ability assessments still have measurement problems. The incremental-validity picture has not changed in EI’s favor.

This is not a fringe academic critique. It is the consensus view of a substantial portion of the industrial/organizational psychology community on commercial mixed-model EI: the construct as marketed is built on weak measurement and oversold incremental validity, and the assessment-and-training industry around it persists primarily because of demand-side momentum (HR departments and L&D vendors find EI useful as a marketing concept) rather than because of strengthening empirical evidence.

What’s Honest To Say About Emotion-Related Skills

The case against popular EI is not a case against the importance of emotion-related skills in work and life. People who are skilled at perceiving others’ emotions, regulating their own emotional responses, communicating empathy, managing conflict, and reading social situations are genuinely more effective in most roles — especially roles involving leadership, sales, customer service, healthcare, and any work involving sustained interaction with other people. The question is not “do emotion-related skills matter?” (they do) but “is there a unified, separately measurable cognitive capacity called ‘emotional intelligence’ that captures these skills and that organizations should measure and develop as a distinct thing?”

The honest empirical answer to that question is: probably not, as currently operationalized. What there is, instead, is a collection of partially overlapping constructs — the Big Five personality traits (especially the emotional-stability/extraversion/agreeableness cluster), specific social-cognitive skills (active listening, perspective-taking, emotion regulation), domain-general cognitive ability, and accumulated emotion knowledge that people pick up through experience. These constructs are well-studied under their existing names. The “emotional intelligence” wrapper bundles them into a marketable package that organizations can buy assessment and training programs around. The bundling is commercially useful but scientifically arbitrary; you could re-bundle the same underlying constructs as “social effectiveness,” “interpersonal competence,” “emotional self-management,” or any other plausible-sounding name and you would get essentially the same construct with essentially the same predictive validity.

If you want to take emotion-related skill development seriously in your organization, the empirically defensible approach is to target specific behaviors with documented training programs: cognitive-behavioral techniques for emotion regulation, structured communication training (motivational interviewing, crucial conversations, nonviolent communication), conflict-resolution training with measurable behavioral outcomes, perspective-taking and mentalization exercises drawn from the social-cognition literature. These are specific, measurable, trainable. The “develop your EI” framing is a vaguer wrapper around the same underlying skills, and the meta-analytic incremental validity it adds over a well-designed program in the specific underlying skills is small.

What This Means For Hiring And Leadership Development

For senior leaders evaluating EI-based assessment or training investments, the practical implications follow from the empirical picture.

On using mixed-model EI for hiring decisions: the meta-analytic incremental validity over a well-administered Big Five personality assessment plus a cognitive ability test is small. Most of the predictive power in mixed-model EI assessments comes from personality content these instruments share with established personality inventories. If your existing assessment battery includes a validated personality measure and a cognitive ability test, adding a mixed-model EI assessment is unlikely to materially improve selection decisions. The legal-defensibility profile of mixed-model EI for hiring is also worth examining — these instruments are commercial products, not always validated for the specific roles where they are deployed, and the science is contested enough that an adverse-impact challenge could find expert witnesses on both sides.

On using EI for executive coaching and leadership development: the diagnostic value of an EI assessment (helping a leader become aware of how they come across to others, identifying specific behavioral patterns to work on) can be useful even when the underlying construct is contested. The risk is in over-interpreting EI scores as if they were measuring a fundamental cognitive capacity, rather than recognizing that mixed-model EI scores are largely personality scores with some self-efficacy content layered on. A coach who treats an “EI assessment” as a starting point for behavioral conversations is doing legitimate work; a coach who treats it as a measure of a fixed-but-trainable capacity that justifies a year-long $400,000 engagement is selling a stronger empirical story than the literature supports.

On vendor selection: if you are going to use an EI framework, the ability-based MSCEIT has the cleanest academic pedigree and the most defensible construct validity, but it has limitations (modest reliability, scoring controversies, and weaker correlations with leadership outcomes than the marketing-friendlier mixed-model instruments). Mixed-model instruments (EQ-i 2.0, ESCI, Goleman-derived assessments) are commercially polished, easier to debrief with executives, and more widely adopted — but you should know that you are essentially buying a Big Five personality assessment with EI branding and some additional self-efficacy items, and you should price the engagement accordingly.

On the “EI matters more than IQ” claim: when you hear this in a vendor pitch, recognize that it has no specific empirical basis in the meta-analytic literature. Goleman’s original claim was largely rhetorical; the academic EI researchers Goleman cited (Salovey, Mayer) have never endorsed the strong version of this claim; and the meta-analyses we have today (Joseph & Newman 2010, O’Boyle et al. 2011) do not support it. Cognitive ability remains the single best predictor of job performance across most occupational categories in the I/O psychology literature. EI, however measured, predicts substantially less than cognitive ability does, and the incremental contribution of EI over cognitive ability and personality is small for ability-based measures and modest-but-contaminated for mixed-model measures.

The honest pitch a thoughtful vendor could make, if they wanted to sell EI training that holds up to scrutiny, is something like: “We can help your managers develop specific emotion-related behavioral skills — active listening, conflict de-escalation, emotion regulation under stress, perspective-taking — that are useful in leadership roles. We use an EI framework as a pedagogical structure because it is familiar to participants and gives the training a coherent narrative. We do not claim that ‘EI’ is a magic distinct capacity that we are measuring and developing as a separate thing from personality and skill.” That is a defensible pitch. The current commercial standard pitch — “EI is the differentiator between average and star performers, our assessment measures it precisely, our program develops it reliably” — is not.

Sources

• Goleman, D. (1995). Emotional Intelligence: Why It Can Matter More Than IQ. New York: Bantam Books.
• Salovey, P., & Mayer, J. D. (1990). Emotional intelligence. Imagination, Cognition and Personality, 9(3), 185–211. DOI: 10.2190/DUGG-P24E-52WK-6CDG
• Mayer, J. D., Roberts, R. D., & Barsade, S. G. (2008). Human abilities: Emotional intelligence. Annual Review of Psychology, 59, 507–536. DOI: 10.1146/annurev.psych.59.103006.093646
• Joseph, D. L., & Newman, D. A. (2010). Emotional intelligence: An integrative meta-analysis and cascading model. Journal of Applied Psychology, 95(1), 54–78. DOI: 10.1037/a0017286
• O’Boyle, E. H., Humphrey, R. H., Pollack, J. M., Hawver, T. H., & Story, P. A. (2011). The relation between emotional intelligence and job performance: A meta-analysis. Journal of Organizational Behavior, 32(5), 788–818. DOI: 10.1002/job.714
• Antonakis, J., Ashkanasy, N. M., & Dasborough, M. T. (2009). Does leadership need emotional intelligence? The Leadership Quarterly, 20(2), 247–261. DOI: 10.1016/j.leaqua.2009.01.006
• Antonakis, J. (2022). Does leadership still not need emotional intelligence? Continuing “The Great EI Debate.” The Leadership Quarterly, 33(6), 101539. DOI: 10.1016/j.leaqua.2021.101539
• Schulte, M. J., Ree, M. J., & Carretta, T. R. (2004). Emotional intelligence: Not much more than g and personality. Personality and Individual Differences, 37(5), 1059–1068. DOI: 10.1016/j.paid.2003.11.014
• Mattingly, V., & Kraiger, K. (2019). Can emotional intelligence be trained? A meta-analytical investigation. Human Resource Management Review, 29(2), 140–155. DOI: 10.1016/j.hrmr.2018.03.002
• Mayer, J. D., & Salovey, P. (1997). What is emotional intelligence? In P. Salovey & D. Sluyter (Eds.), Emotional development and emotional intelligence: Educational implications (pp. 3–34). Basic Books.
• Mayer, J. D., Salovey, P., & Caruso, D. R. (2002). Mayer-Salovey-Caruso Emotional Intelligence Test (MSCEIT). Toronto: Multi-Health Systems.
• Goleman, D. (1998). Working with Emotional Intelligence. New York: Bantam Books.

Related Articles in This Hub

• Replication Crisis Hub — full index — start here for the broader landscape of contested behavioral-science claims that have entered corporate practice.
• Grit: Real, But Barely Distinguishable From Conscientiousness — the structural parallel to EI: a popular construct that meta-analysis shows is essentially a repackaging of an existing personality trait.
• Fredrickson’s 3:1 Positivity Ratio — another positive-psychology construct that achieved massive corporate adoption before the mathematical foundation was retracted.
• Growth Mindset — overlapping audience (school and corporate L&D buyers), similar pattern of construct becoming oversold relative to the meta-analytic evidence.
• Self-Esteem Movement — earlier example of psychology-to-public-policy export that outran its empirical foundation.
• 10,000 Hours Rule — same general phenomenon: a popularized scientific claim that the underlying researcher’s actual position is much more constrained than the bestseller version.

FAQ

Does this mean emotional intelligence is fake?

No. Specific emotion-related skills — perceiving emotions accurately in oneself and others, regulating one’s emotional responses under stress, communicating empathy, managing conflict productively — are real, measurable, and matter for leadership and interpersonal effectiveness. The critique is more specific: the popular Goleman/mixed-model EI construct that organizations buy assessments and training programs around is largely a repackaging of Big Five personality traits with some self-efficacy content, not a separately measurable cognitive capacity that adds substantial incremental information beyond personality and IQ. The narrower academic ability-EI construct (MSCEIT) is more defensible but has its own measurement problems and weaker predictive validity than the popular framing claims. Honest answer: emotion-related skills are real, the “EI” wrapper around them is more marketing than science.

Is the MSCEIT (ability-based EI assessment) better than the EQ-i or ESCI (mixed-model)?

For research purposes, yes — the MSCEIT has the cleanest academic pedigree, the strongest construct-validity case, and the most defensible measurement properties. For commercial-coaching purposes, the MSCEIT is less popular because it produces less actionable, less debrief-friendly output than the mixed-model instruments. The trade-off is honest: the MSCEIT measures something that is more likely to be “real EI” but its incremental predictive validity is small (about 0.2% incremental variance over Big Five and cognitive ability in Joseph & Newman 2010); the mixed-model instruments have larger incremental-validity numbers but mostly because they include personality and self-efficacy content that is not really “EI.” There is no clean answer; pick the instrument that fits your purpose and be honest with yourself about what it is measuring.

What about EI for leadership specifically? Doesn’t research show EI matters more for leaders?

The empirical picture for EI-and-leadership specifically is no stronger than the general EI-and-job-performance picture, and possibly weaker. Antonakis and colleagues (2009, 2022) have argued at length that the EI-leadership literature is contaminated by common-method variance (same rater scoring both EI and leadership), inadequate controls for personality and cognitive ability, and publication bias. When studies use objective leadership outcomes (financial performance, retention, follower satisfaction) and properly control for established predictors, EI’s incremental contribution is small. There is some evidence that EI matters more for jobs with high emotional-labor demands — customer service, healthcare, hospitality — which makes intuitive sense, but this is a moderation finding, not a vindication of the strong “EI predicts leadership effectiveness” claim.

Should I stop using EI assessments in hiring decisions?

For mixed-model EI assessments (the EQ-i, ESCI, Goleman-derived instruments), there is a strong case for not using them as primary selection screens, especially if you are already using a validated Big Five personality assessment and a cognitive ability test. The incremental information from a mixed-model EI assessment on top of those existing tools is small, and the legal-defensibility profile is contested. For ability-based EI (the MSCEIT), the empirical case for hiring use is even weaker — the incremental validity over cognitive ability is essentially nil. If you want to assess for emotion-related skills in hiring, structured behavioral interviews focused on specific past situations are probably your highest-validity option, with a validated personality assessment as a useful supplement. EI assessments add cost and complexity without proportionate selection-quality improvement.

What about teaching emotional skills to children in schools?

Social-emotional learning (SEL) programs in K-12 education are a related but distinct topic from corporate EI. Meta-analyses of well-designed SEL programs (the CASEL framework, for example) show genuine benefits for student outcomes — improved academic performance, reduced behavioral problems, better social adjustment — with effect sizes that are modest but meaningful. The honest interpretation is that targeted instruction in specific social-emotional skills (emotion regulation, perspective-taking, conflict resolution, self-management) works for kids; the broader “EI” framing is not what is doing the work. Schools considering SEL investment should evaluate specific evidence-based programs (Second Step, RULER, MindUP, etc.) on their own outcome research rather than buying into the broader EI marketing narrative.

How do I tell whether an EI vendor’s claims are scientifically defensible?

Useful diagnostic questions to ask a vendor pitching an EI program: (1) Are you using ability-based or mixed-model EI? — answer should not be “both” or evasive. (2) What is the incremental validity of your assessment over Big Five personality and cognitive ability for our specific use case? — answer should cite a specific number from a peer-reviewed study, not a marketing claim. (3) What independent (not vendor-funded, not authored by people with financial stake in the instrument) peer-reviewed studies validate this specific instrument for this specific use? — answer should be specific citations. (4) What is the test-retest reliability of your assessment over 3-6 months? — should be above .70 to support the “develop your EI” narrative. (5) For training programs: what behavioral outcomes (not self-report) does your training improve, and what are the effect sizes? — answer should be specific. Vendors who can answer these questions clearly have something defensible; vendors who hedge or pivot to testimonials are selling marketing, not science.

Is the “90% of star performance is EI” claim from Goleman accurate?

No. This claim appears in Goleman’s 1998 Working with Emotional Intelligence and has been widely repeated in the consulting and HR-vendor literature. It originated from Goleman’s analysis of internal competency models developed by various corporations, which is methodologically incommensurable with peer-reviewed psychometric research on EI’s incremental validity. Joseph and Newman (2010, p. 69) explicitly call out a similar Goleman claim (“emotional competence is twice as important as purely cognitive abilities for star performance”) as empirically unsupported by their meta-analytic data. When you hear the “90%” figure in a vendor pitch, you are hearing a marketing claim with no specific empirical basis in the peer-reviewed EI-validity literature. The most accurate honest summary the academic meta-analyses support is something like: “EI adds modest incremental prediction of job performance beyond cognitive ability and Big Five personality, with the largest incremental-validity figures coming from instruments whose ‘EI’ content is heavily contaminated with personality and self-efficacy content.”

If I’m a CEO and I just want to improve my managers’ people skills, what do I actually do?

The empirically defensible path looks something like: (1) Stop thinking about it as “developing EI” and start thinking about specific behavioral skills you want to see — handling difficult conversations, giving and receiving feedback, regulating emotions in conflict, recognizing and responding to direct-report distress, etc. (2) Invest in training programs that target those specific skills with documented behavioral outcomes — structured communication frameworks (motivational interviewing, crucial conversations), cognitive-behavioral skill training for emotion regulation, role-play-based conflict resolution practice. (3) Measure outcomes behaviorally — 360 feedback on specific behaviors, observable changes in meeting dynamics, conflict-resolution outcomes, employee retention in the manager’s team — not by re-administering the same self-report EI assessment that was used to justify the program. (4) Be skeptical of any program whose primary outcome measure is “increased EI scores on the same instrument used to diagnose the deficit.” This is the same epistemological problem as a weight-loss program whose only success metric is “participants report feeling thinner.” Real behavioral change in real outcomes is what matters; the EI scaffolding is dispensable.