Can emotional intelligence be trained? Yes, but check what moved

A plant in Pune ran a two-day emotional intelligence workshop for forty supervisors. Six weeks later the L&D manager re-tested them on the same questionnaire and the average went up eight points. The business head saw the slide, approved a rollout to two hundred more supervisors, and asked one question at the end: what changed on the floor?

Nobody had an answer. The score had moved. The floor had not been looked at.

That is the usual shape of this problem, and it is worth being precise about it, because the research on whether emotional intelligence can be trained is more interesting than either the vendor pitch or the sceptics allow.

What the meta-analyses actually found

Three reviews are worth reading before you sign anything.

Mattingly and Kraiger pooled 58 published and unpublished studies of emotional intelligence training, using either pre-post or treatment-control designs. They found a moderate positive effect, and the effect held up across participant gender and across the type of measure used, whether an ability test or a mixed self-report instrument. Effects were larger in published studies than in dissertations, which is the first hint that the literature is friendlier than reality.

Mehler and colleagues went further and separated the designs. Their pre-post meta-analysis produced a standardised mean difference of 0.44 (95% CI 0.29 to 0.59). The controlled-trials-only analysis, 27 studies, produced 0.46 (95% CI 0.30 to 0.63). Both effects persisted more than three months after training ended, and all professional groups benefited equally. That is a real result and it is not small: roughly half a standard deviation, in line with what competency-based training programmes deliver generally.

Powell and colleagues looked at 17 longitudinal studies in healthcare. Every study showed an increase. They then said the thing that matters: methodological limitations in those studies may have led to an overestimate of the actual effects.

So the honest answer to the title question is yes. Training programmes reliably raise emotional intelligence scores by something close to half a standard deviation, and the gain survives a quarter. Anyone telling you the construct cannot be developed is overreading the criticism.

The next question is harder.

The score you moved is the thing you have to doubt

Emotional intelligence is measured two ways, and they are not interchangeable.

Self-report instruments, usually called mixed models, ask the person to rate agreement with statements about themselves. Ability instruments give the person problems to solve with right and wrong answers: which face shows fear, which emotion would help here, how does this feeling progress.

A meta-meta analysis pooled 484 effect sizes from 102,579 participants across nine meta-analyses and found an average correlation of 0.244 between emotional intelligence and the outcomes it is supposed to predict. Mixed models sat at 0.272. Ability models sat at 0.160. Effect sizes in the field declined over time, and the decline appeared only in the mixed models. The authors' reading is blunt: the mixed-model effects were probably overestimated in the first place.

Put that beside a training result and the mechanism becomes obvious. A self-report item asks whether you can tell when a colleague is upset. A two-day workshop teaches you the vocabulary of emotional intelligence, gives you a framework, and rewards you with attention for talking about feelings accurately. Six weeks later you answer the same item. You are not lying. You are better at the language the item is written in, and the item cannot tell the difference between that and skill.

The training effect in Mattingly and Kraiger held across measure types, which is the part most critics miss. Ability scores move too. But ability scores move less easily, cost more to administer, and are the ones that show a smaller, more durable relationship with actual outcomes. If you are spending money to change behaviour, the instrument you retest should be the one least able to flatter you. I have written before about how self-report EQ tests mostly measure self-confidence, and the training literature is where that problem does its damage.

Design the programme around one ability, not around the construct

Emotional intelligence is four abilities, not one. Training all four in a two-day block gives you a broad, shallow exposure and nothing you can check. The method below runs in the order the work actually happens.

Name the ability and the behaviour it should produce. Pick one. Then write down what a person doing it well looks like at work, and what you would count.

If you cannot fill the third column for the ability you chose, stop. You are buying a score, not a change.

Choose the measure that matches the ability. An ability task if the skill is a skill. A self-report instrument only as a secondary reading, and never as the headline. Third-party ratings of the behaviour are the only measure that cannot be answered by vocabulary alone. The table below is the trade-off.

Build the practice loop, not the content. The programmes inside those meta-analyses were multi-session, with repeated practice and feedback. I did not find a meta-analysis that isolates the single-day lecture as a design, because the trials are not built that way. If your plan is one facilitator, one hall, one afternoon of slides, you are outside the evidence base and should not expect the 0.46.

Measure at three points, and measure the behaviour separately. Before, at the end, and eight to twelve weeks later. The behaviour measure comes from someone other than the trainee. This is the same discipline as Kirkpatrick level 3 without a research team: pick one behaviour, one observer group, one counting rule, and accept a small sample over a large claim.

A worked example, with illustrative numbers

Forty supervisors at a distribution centre. The ability chosen was managing emotions, defined narrowly as staying in a difficult conversation rather than ending it. The behaviour counter was conversations escalated to HR or ended early, tracked by the ER log.

Baseline over the previous quarter: 9 such events across the group. Self-report score averaged 68 out of 100. An ability subtest on emotion management put the group at the 45th percentile of the vendor norm.

The programme ran six sessions of ninety minutes over six weeks, in pairs, with recorded role plays and structured peer feedback, plus one observed live conversation per supervisor with a coach. Total participant time: 40 people at 9 hours each, 360 hours. At a loaded cost of Rs 900 an hour that is Rs 3.24 lakh of staff time before the facilitator's fee. Label those figures as illustrative; the shape is what matters.

At the end: self-report 76, up eight points. Ability subtest up 0.3 standard deviations. At twelve weeks: 4 escalated or abandoned conversations in the quarter, down from 9. Peer ratings on "stays calm when challenged" moved from 3.1 to 3.6 on a five-point scale.

Now the decision. The self-report gain of eight points is the number that will look best on a slide and it is the least trustworthy. The behaviour count is the one that justifies the next batch, and it moved by more than half, on a measure no participant could fill in themselves. If the behaviour count had stayed at 9 while the score rose eight points, the correct conclusion would have been that the workshop taught vocabulary.

The honest limit

Almost every study in these reviews uses an emotional intelligence instrument as the outcome. Very few measure what the person does at work afterwards. That is not a small gap. It means the 0.46 is an effect on scores, and the effect on behaviour is largely unmeasured in the literature you would cite to defend the budget.

Two further reasons to discount it. Published studies showed larger effects than dissertations, which is publication bias in plain sight. And the field-level analysis found effect sizes falling over time, concentrated in the self-report models. The true score effect is probably below 0.46.

There is also a limit on speed. Ability-based emotional intelligence behaves like a skill: it improves with practice and it improves slowly. A six-week programme may move the ability subtest by a third of a standard deviation and no more. If your business case needs a large movement in eight weeks, you are buying the self-report gain, and you should say so out loud rather than let the rollout decision rest on it.

And none of this fixes a manager who is set up to fail. Emotional intelligence training does not compensate for a supervisor running 14 direct reports across three shifts with no cover. Train the skill, then look at the system before blaming the person for not using it.

What to do on Monday

Pick one ability. Write the behaviour it should produce in a sentence a supervisor would recognise. Write the counter, the person who will count it, and the date you will read it. Put that page in front of the facilitator before you brief them.

Then, when a vendor shows you a lift, ask two questions: which instrument produced that number, and who rated the behaviour. If the answer to the second is "the participants", you have bought a vocabulary lesson. That is not worthless. It is just not the thing you told the business head you were buying. If you want the assessment side built so the ability task and the behaviour rating sit in one place, that is what our assessment platform is for.