Using EQ data in manager development without a label

A promotion panel in Gurugram, last November. Twelve managers on the slate and one printed emotional intelligence report per person on the table. The HR business partner flipped to the third page and said the sentence that decided the case: "He is low empathy." Nobody asked what he had done. The report went back into the folder and the conversation moved to the next name.

The score did not end that promotion. The sentence did. A percentile is a fact about a comparison group. A trait word is a claim about a person. The distance between those two is one sentence long, and it is where most EQ work in organisations quietly stops being useful.

How a score turns into a label

The compression happens in three steps, and each one loses information.

First the raw score becomes a rank. A manager sits at the 34th percentile on empathy. That is a statement about where he stands relative to a norm group, nothing more.

Then the rank becomes an adjective. Low empathy. The norm group disappears from the sentence, and with it the only thing that made the number meaningful.

Then the adjective becomes the person. He is low empathy. That is not a measurement claim any more. It is a description of character, and it behaves like one: it feels permanent, it explains everything, and it cannot be tested.

I have watched this happen to good data in six or seven organisations now. The instrument is rarely the problem. The report is rarely the problem. The problem is that a human being has to turn a table of numbers into a sentence, under time pressure, in front of a committee, and the shortest sentence available is always a trait word.

Why the label kills the feedback

The research on feedback is blunt about this. Kluger and DeNisi pooled 607 effect sizes from 23,663 observations and found that feedback interventions improved performance by an average d of 0.41, but that over a third of them made performance worse. Their explanation holds up: feedback effectiveness falls as attention moves away from the task and toward the self.

Their own summary is the sentence to keep on the wall: "FI effectiveness decreases as attention moves up the hierarchy closer to the self and away from the task."

A label is pure self-level information. "You are low empathy" gives the manager nothing to do, only something to be. The two available responses are to accept it, which is demoralising, or to reject it, which is defensive. Neither one changes what happens in a Tuesday review meeting.

There is a second body of work that points the same way from the other direction. Mueller and Dweck found that children praised for intelligence cared more about performance goals than learning goals, gave up sooner after failure, and described intelligence as a fixed trait. Children praised for effort did the opposite. The studies were with fifth graders, so do not stretch them too far into a plant in Coimbatore. The part that transfers is the grammar. "You are good at this" and "you worked the problem that way" produce different responses, and the difference is that one describes a person while the other describes an action the person can repeat.

The order of work: behaviour before score

This is the rule that fixes most of the problem, and it is a scheduling decision rather than a writing one.

Before anyone opens the report, the manager writes down three recent situations. Not reflections, not ratings. Situations: what was happening, what he did, what followed. Twenty minutes, one page, no instrument in the room.

Then he reads the report. Then the reviewer reads the behaviour notes and the report together.

The reason is anchoring. If you read "34th percentile on empathy" first and then try to recall how the man handles a difficult conversation, you will find examples. You will find them because you were sent looking. The order protects both of you from that.

Step three is the one people skip. It is also the one that produces the finding. If the report says low empathy and the three situations show a man who noticed the problem and said nothing because the plant head always spoke first, you do not have an empathy problem. You have a meeting protocol problem, and the fix costs nothing.

Report the band, not the rank

Every score on a psychometric report carries measurement error. The standard error of measurement converts the instrument's reliability into the units of the score itself: SEM equals the standard deviation times the square root of one minus the reliability.

Take a scale with a standard deviation of 15 points and a reliability of 0.85. The SEM is 15 times the square root of 0.15, which is about 5.8 points. A band of one SEM either side of the score is the ordinary working range for a single result. Two managers whose scores sit eight points apart on that scale are not distinguishable from each other, and a seven point gap between two subscales for the same person is not a finding either.

So ask the vendor for the SEM of every subscale you plan to interpret. If it cannot be produced, you cannot report a band, and if you cannot report a band you should not report a rank either.

The expiry date matters more than it looks. An EQ report is a snapshot of how a person answered or performed on one day. Twelve months later it is a historical document, and teams keep quoting it for three years because nobody wrote an end date on the cover.

Write the sentence as a condition, not a trait

If the output of the review is a written sentence, then the sentence is the intervention. Get the grammar right and half the labelling problem disappears.

A usable sentence has three parts: the situation, the observable act, and the consequence. The test is simple. Could a colleague who was not in the room see it happen? If not, the sentence is still a trait word wearing a longer coat.

This is the same conversion that makes a competency framework usable, and it is worth doing once, properly, rather than rewriting it for every manager. The five domains become useful only when each one is turned into a cue a rater can actually see, and the same discipline applies to whatever subscales your instrument reports.

Keep development data out of administrative decisions

This is a policy question, and it needs a line in writing rather than good intentions.

The International Test Commission's guidelines on test use put it plainly: tests should be supported by evidence of reliability and validity for their intended purpose, and evidence should be provided to support the inferences drawn from the scores. The same document asks test users to explain and support the use of results that classify people into categories, which is exactly what a promotion decision does.

Now look at what the evidence supports for self-reported emotional intelligence. Joseph and colleagues, in a meta-analysis of mixed EI measures, found a correlation of 0.29 with supervisor-rated job performance. After controlling for cognitive ability, the Big Five, ability EI, self-efficacy and self-rated performance, the relationship became essentially nil, with a standardised coefficient of -0.02. The measures also overlapped heavily with those other constructs, with a multiple correlation of 0.79.

That is a perfectly good basis for a development conversation. It is not a basis for a promotion decision, and it is certainly not a basis for a hiring cut-off. If you would not put a personality score in a promotion note, do not put this one there either.

A worked example

The numbers below are illustrative. The pattern is not.

Twenty four managers, six subscales, a 100 point scale with a standard deviation of 15 and a reliability of 0.85, so an SEM of about 5.8 points and a band roughly 12 points wide.

Manager A scores 44 on empathy and 51 on self-regulation. The gap looks like a story: he understands people but cannot control himself. It is seven points, which is inside the band, so it is noise. The right action is to treat both as "around the middle" and go and look at the behaviour notes.

Manager B was labelled low empathy in the previous cycle on the strength of a 46. His three situations read differently. In each one he noticed the problem early, formed a view, and said nothing, because the plant head opened every meeting and the supervisors took their cue from him. The development action was not empathy training. It was a ten minute pre-meeting with the plant head before each review, and an agreement that B raises issues in the first ten minutes.

Six months later his score had barely moved. The number of issues raised by supervisors in those meetings had gone from two or three to eleven. That is the outcome the programme was actually bought for, and no subscale would have found it.

The honest limit

EQ data is a weak predictor and a decent conversation aid. The four rules above make it a better conversation aid. They do not make it a strong predictor, and nothing in this article should be read as saying the instrument will tell you who to promote.

Three more limits, stated plainly.

The framing only works if the manager trusts the source. If he thinks the questionnaire was nonsense, or that his raters had it in for him, a well written condition sentence will read as a verdict with better manners. Fix the trust problem first, or do not run the exercise.

A label free report does not fix a manager who is genuinely behaving badly. If he is shouting at people, that is a performance management issue with a process and a timeline. EQ data should not be used to soften it, and it should not be used to delay it while another cycle is run.

And the method costs reviewer time: about forty minutes per manager once you include the behaviour sample and the comparison. That is the real reason teams skip it. Budget the time before you announce the programme, because a process that only works when nobody is busy will be abandoned in the first quarter that gets busy.

What to do on Monday

Pull the last cycle's EQ reports and count how many contain a trait adjective about a person. In most organisations it is nearly all of them.

Rewrite the three worst sentences using the situation, act, consequence format. Add one line to the footer: development use only, not to be used for selection, promotion or pay decisions, review by this date twelve months from now. Then send the whole set back to the reviewers with the three rewrites attached and no commentary.

That is twenty minutes of work, and it changes what the next cycle produces. If you want the reporting side handled by the tool rather than by a reviewer with a template, the development report is built around bands and behaviours rather than percentiles. If you are still deciding whether a self-report instrument is the right one for this at all, the case against reading self-report scores as ability is worth an hour first.