Ego States

The problem it solves

Feedback that is accurate, proportionate and kindly delivered will still, fairly often, be received as a telling-off. Neither person raises their voice and neither says anything unreasonable, and both come away describing the other one's tone, which is usually the first sign that the content was not the thing being argued about.

The smaller version of this happens in most teams several times a day. A plain factual question about when the window closes or who owns something now comes back as an answer about fairness, and a conversation about a schedule has turned into a conversation about something else entirely.

What the two have in common is that the words were fine. Whatever went wrong sat underneath them where neither person can point at it, and that's why these exchanges are so hard to talk about afterwards.

The idea

Eric Berne, a psychiatrist working in California in the 1950s, noticed that patients visibly changed part way through a session, and that what changed was their whole manner rather than their mood. Vocabulary, posture, pace and gesture would shift together, and shift back, and the person would be recognisably operating from somewhere else.

He proposed that behaviour comes from three distinct sources, and that you tell them apart by where the content came from rather than by what is said.

Enlarge: The three states arranged by where their content comes from, which is the test that holds up.

Three horizontal bands, one per ego state, arranged down the page by where the behaviour originates. The top band is Parent, borrowed whole from a caregiver or authority. The middle band is Adult, built now from the situation in front of you. The bottom band is Child, replayed from your own earlier life. A line down the left marks the middle band as the only one sourced from the present.

The three states arranged by where their content comes from, which is the test that holds up.Close

The three states arranged by where their content comes from, which is the test that holds up.

Parent is behaviour copied whole from a caregiver or an authority figure and replayed without ever being re-derived. It arrives as rules, standards, shoulds and judgements that went unargued because to the person holding them they were not in question, and it isn't always stern, since the reassuring and protective register is Parent as well.

Adult is assembled from what is actually in front of you now, which shows up as estimating, checking, and asking how many and by when. Of the three it's the one answering this situation rather than a stored one.

Child is feelings and strategies replayed from your own earlier life. It carries the compliance, the sulk and the flash of unfairness, and it carries curiosity, play, and most of the good ideas anybody has ever had.

Three things about this are routinely lost, and each of them matters more than the model itself.

It is not about age, and a forty-five-year-old operating from Child is as ordinary a sight as a nine-year-old operating from Adult.

Neither is it a personality type. Most people use all three on an ordinary day, often inside a single conversation, and there is no such thing as a Parent person.

And Adult is not the good one. A team working purely from Adult would be accurate, efficient, and would build very little that anybody wanted, so the aim is knowing which state you're in rather than living in one of them.

How to use it

Start with the shift rather than the label, because what you're listening for is the moment somebody's register changes, and that shift is the observable part while the categories are just names for where it went.

The cues are real but weak, and they work as hints rather than definitions. Absolutes and evaluations (always, should, frankly, obviously) lean Parent, while fairness, permission and defeat (it's not fair, am I allowed to, I can't do this) lean Child, and quantities, dates and conditions (how many, by when, what would it take) lean Adult. None of it is diagnostic on its own, and the principal engineer saying "we shouldn't ship without a rollback" may be repeating a rule or may have read the last incident report at midnight, since the sentence is identical either way.

Then use the move that reliably works, which is changing the question rather than naming the state. A question that needs a number, a date or a specific fact to answer it is very difficult to answer from anywhere but Adult. "Are we going to be okay for the board date" invites reassurance and will get it, whereas "what is left, and how long is each piece" mostly can't be answered with reassurance at all. If you keep receiving defensiveness in response to what you take to be a neutral question, the question is probably not arriving where you sent it, and rewording it is free.

Do all of this on yourself first, and for the most part on yourself alone. The useful version of the model is noticing that your own hands are shaking as you type a factual, well-evidenced, utterly reasonable reply, and then asking whether that reply is meant to inform somebody or to win something.

Where it breaks down

Every concept here has one. It is the section most summaries leave out.

Three bins will hold anything, which makes the model impossible to get wrong. Any exchange can be assigned states after the event, the assignment will feel convincing, and that feeling is not evidence of anything. A framework that explains every outcome is not tested by any outcome, and the confidence a neat retrospective mapping produces is the least trustworthy thing about it.

The evidence base is a good deal thinner than the popularity implies. Fifty years on there is still no dependable instrument that measures the three states, and a 2021 review that assessed fifty-six transactional analysis instruments across 263 studies against a standard quality framework found that a handful of them cleared it. A 2022 meta-analysis of forty-one trials did find moderate to large effects for therapy built on this theory, which tells you about the therapy rather than about the three states, because outcome research can't isolate which component of a method did the work.

Used outward it is a weapon. Give a manager this vocabulary and "you're being very Child about this" becomes available, which is a personal insult with a diagram behind it, unanswerable because any objection to it looks like confirmation. Berne built the model for a clinician to observe themselves and their patients under supervision, and there is no equivalent safeguard anywhere in an office.

The names carry a judgement the theory denies. Berne was explicit that Child is not immaturity and Adult is not virtue, while the words themselves argue the opposite continuously, and in practice the words win. Almost all the misuse of this model traces back to two badly chosen labels.

It came from one place and one period. The cues were drawn from mid-century American clinical practice, so directness that reads as Parent in one workplace is unremarkable in another, and deference that reads as Child in one culture is ordinary courtesy somewhere else. The model is silent on all of it, and its users rarely notice.

In one line

Notice which of three places your own next sentence is coming from, and change the question rather than naming anybody's state.