# How to Know When to End a Relationship
> The question is usually asked wrong. What the research says about ambivalence, about why you are a biased judge of it, and what really predicts an ending.
Written by Ahmet Abiç (Founding editor). Reviewed by Noah Cortez (Stanford Journalism Program). Published August 18, 2026.
Source: https://gritwright.com/social/when-to-end-a-relationship/
---
import Figure from '../../../components/Figure.astro';
import ResearchNote from '../../../components/ResearchNote.astro';
import PracticeNote from '../../../components/PracticeNote.astro';
import stayThinking from '../../../assets/guides/social/stay-thinking.webp';
import stayDistance from '../../../assets/guides/social/stay-distance.webp';
You have been having this argument with yourself for a while. Some weeks it is obvious you should leave and some weeks it is obvious you should not, and the fact that you cannot settle it feels like evidence of something.
It is evidence of something, but not of what you think. There are specific reasons this particular decision resists being made, and most of them are documented. Knowing them will not decide it for you. It will stop you spending another year on a process that was never going to converge on its own.
## The short answer
The question people ask themselves is the wrong question, and asking it better is most of the work.
| What you are asking | The better version |
| --- | --- |
| The wrong question | How happy am I, out of ten |
| The better one | How good is the good, and how bad is the bad, scored separately |
| Why you cannot decide | Staying needs no decision. Leaving needs one |
| What staying costs | Measurable, in self-esteem, health and life satisfaction |
| What predicts an ending | Withdrawal and defensiveness, more than conflict |
| What resolves it | A deadline you set in advance, not a feeling you wait for |
## The question is badly formed
Almost everyone tries to answer this on a single scale running from good to bad, lands in the middle, and concludes that the middle means they have not thought hard enough. [A theoretical review published this year](https://pmc.ncbi.nlm.nih.gov/articles/PMC12754029/) argues that the scale itself is the problem.
The authors argue against the standard one-dimensional picture of relationship evaluation, running from positive to negative, on the grounds that **positive and negative evaluations vary independently.**
That yields **four quadrants**: mostly positive, mostly negative, **high positive and high negative at the same time (ambivalence)**, and **low positive and low negative (indifference).**
The practical consequence they draw out: for someone experiencing ambivalence, filling in a normal satisfaction scale is **problematic, because they feel satisfied and dissatisfied at once**. They will report the middle of the scale, or **the extremes depending on which is salient at that moment**, satisfied one week and miserable the next.
Someone experiencing indifference **also lands in the middle**, for the opposite reason. The two states are qualitatively different and a single scale cannot tell them apart.
That explains something people find genuinely disorienting about their own answer: it moves, sharply, depending on the day and on what happened last. That is not indecision, it is what ambivalence does on a scale that was not built to hold it.
It also explains why "how do you feel about it out of ten" produces nothing usable. A five means two completely different things, and the two need opposite responses.
## A test that works better
Because the two dimensions are independent, the way to get an answer out of yourself is to score them separately, which is how the research measures it.
Ask two questions instead of one, over a period rather than in one sitting. **How strong are the positive feelings and thoughts you have about this person?** And separately, **how strong are the negative ones?**
Four outcomes, and each means something different.
**High positive, low negative.** Nothing to decide. If you got here, you were probably answering a question about something else.
**Low positive, high negative.** Also nothing to decide, though what follows is difficult. This is not ambivalence and the doubt you are experiencing is usually about the leaving rather than about the relationship.
**High on both.** Ambivalence. Real feeling on both sides, and the swing you have noticed in your own reading is the signature of it. This is the state where more thinking produces no convergence, because both inputs are true. It needs an external structure, which is what the last sections of this guide are about.
**Low on both.** Indifference, and it is the one people most often misread as being fine. There is no crisis and nothing much on either side. This tends to get called a comfortable relationship. Whether it is the one you want is a question about your life rather than about them.
## You are not a neutral judge of this
Before doing anything with that reading, it is worth knowing which direction your thumb is on the scale. [The evidence](https://pmc.ncbi.nlm.nih.gov/articles/PMC8597186/) says it is on the same side for almost everyone.
The review challenges the assumption that dating works as a test phase in which people evaluate each other and reject poor fits, arguing the evidence shows **a progression bias**: a systematic tilt towards **initiating, advancing and maintaining** relationships.
Their framing of what it explains: **why getting into a relationship is often easier than getting out of one, and why being in a less desirable relationship is often preferred to being in no relationship at all.**
Among the catalogued mechanisms: **relationship investments create barriers to dissolution; concern for the partner's feelings creates barriers to dissolution; fear of being single creates barriers to dissolution**; and, cognitively, **breakup decisions tend to be deliberative and effortful** while continuing requires nothing.
They also note that many people **struggle to leave even unhealthy relationships**, and that dissolved relationships frequently reconcile.
Read that as a correction factor rather than as an instruction. It does not mean you should leave. It means that when you find yourself constructing a case for staying, some part of that case is being generated by a documented bias, and you should discount accordingly.
The specific one to watch for is the second mechanism. Staying because ending it would hurt them feels like decency and is listed here as a barrier to a decision you have already privately made. It also does not work: nobody is spared by a long ambiguous ending, they are just given it in instalments.
## What staying in a bad one actually costs
The strongest evidence on the cost of staying comes from [a long study that followed the same people for over a decade](https://www.psu.edu/news/research/story/unhappy-marriages-detrimental-self-esteem-and-health), which is rare in this field.
**1,150 participants**, drawn from a national study of married Americans, interviewed at **four points across twelve years.** To count as unhappily married, a person had to score consistently below the study average on marital happiness.
Those who remained in long-term low-quality marriages showed **lower self-esteem, lower overall health, lower overall happiness and lower life satisfaction, with elevated psychological distress**, compared with people in long-term happy marriages.
Co-author Alan Booth's summary: unhappily married people may have greater odds of improving their well-being by dissolving those unions, as **there is no evidence that they are better off in any aspect of overall well-being than those who divorce.**
Two honest caveats, because this is exactly the kind of finding that gets stripped of them.
The first is that this is a study of marriages, not of eighteen-month relationships in your twenties, and the transfer is an inference. The mechanism is plausible and the population is different.
The second is bigger. Comparing people who stayed with people who left is not a controlled experiment, because who ends up in each group is not random. Other work has found that unhappily married people who stayed were no less happy five years later than those who divorced, and the honest reading of the two results together is that the difference is largely about which kinds of people, and which kinds of unhappiness, end up on each side. Anyone quoting either result without that caveat is selling something.
What survives both readings is narrower and still useful: there is no good evidence that staying in a durably bad relationship protects your well-being, and the intuition that it does is doing a lot of work in a lot of people's decisions.
## What actually predicts an ending
If you want to know whether this is heading somewhere, [the behavioural literature](https://bpl.studentorg.berkeley.edu/docs/41-Marital%20Processes92.pdf) points at something other than how much you argue.
**Seventy-three married couples**, studied in **1983 and again in 1987.**
Compared with couples whose relationships stayed regulated, those heading towards dissolution showed **more negative emotional expression, less positive emotional expression, more stubbornness and withdrawal from interaction, and greater defensiveness.**
The same couples also rated their problems as more severe, had lower satisfaction at both time points, and reported **poorer health** four years later.
Seventy-three couples is a small study, and it is one of the most cited in the field precisely because so few people have run a better one. Treat it as a well-described pattern rather than a law.
The item to take seriously is withdrawal, because it is the one that looks like peace. A relationship with arguments in it is a relationship where both people still think the outcome is worth the trouble. A relationship where nobody bothers to raise it any more is further along, and it very often gets described by the people in it as having got better lately.
If what you recognise is the arguing rather than the withdrawal, [the guide on stopping the fighting](/social/how-to-stop-fighting/) is a more useful place to be than this one.
## What is fixable and what is not
The other question folded into this one is whether the bad parts are the kind that respond to effort. [The largest analysis of relationship quality](https://sites.lsa.umich.edu/whirl/wp-content/uploads/sites/792/2020/08/Joel-et-al-2020-PNAS.pdf) is helpful here in a slightly indirect way.
Machine learning across **43 longitudinal datasets covering 11,196 couples**, comparing hundreds of variables against one another rather than one at a time.
**Relationship-specific perceptions explained up to 45% of the variance** in satisfaction. The strongest were **perceived-partner commitment, appreciation, sexual satisfaction, perceived-partner satisfaction and conflict.**
**Individual differences explained up to 21%**, led by life satisfaction, negative affect, depression, attachment avoidance and attachment anxiety.
The useful implication for this decision is what is missing from those lists. What predicts satisfaction is not a set of traits either of you has, it is a set of things about how the relationship is going, and those are the kind of thing that can move.
Which cuts both ways, and the direction depends on one honest question: have you actually tried, in a way the other person knows about. A relationship where neither person has ever said plainly what they need is not a relationship that has been tested. A relationship where that has happened repeatedly and changed nothing has been.
## Where this is not a judgment call
Everything above assumes a relationship that is unsatisfying rather than unsafe, and [those are different questions](https://www.who.int/news-room/fact-sheets/detail/violence-against-women) with different answers.
WHO estimates published in 2023 indicate that **about one in three women worldwide (31.6%, or 840 million) have been subjected to physical or sexual intimate partner violence, or non-partner sexual violence, in their lifetime.**
Around **a quarter of women aged 15 to 49 who have been in a relationship have experienced physical or sexual violence from a partner** at least once.
WHO's definition of intimate partner violence covers more than physical harm: **physical aggression, sexual coercion, psychological abuse and controlling behaviours.**
That last clause is the part worth reading twice, in either direction. Controlling behaviour and psychological abuse are named alongside physical violence, not below it. If what you are weighing includes being monitored, isolated from people, threatened, or systematically undermined, this is not a decision about compatibility and the progression bias research above describes exactly how such situations get rationalised into staying.
Men are on both sides of this and are less likely to name it. If any of it is recognisable, the next step is a specialist service in your country rather than a relationship guide.
## What produces an answer
Ambivalence does not resolve by being thought about, which is the single most useful thing in this article, because it means the strategy almost everyone is using is the one that cannot work.
**Set a date before you start.** Three months, six, whatever is honest, decided now and written down. Open-ended deliberation is the condition the bias operates in, and a deadline is the only thing that reliably converts a mood into a decision.
**Say the thing out loud, once, properly.** Not a hint, not a bad-mood version. The specific thing that is wrong, to them, in a way they could act on. A great many relationships people are quietly leaving have never had this conversation, and it is not possible to know whether something is fixable until somebody has tried to fix it.
**Watch behaviour for a set period after that, not words.** The response to being told is more informative than anything either of you says about the relationship in the abstract.
**Ask what you would tell a friend describing this.** Crude, and it works, because the bias is about your own investments and it does not transfer to somebody else's story.
**Notice which future you check for.** People are usually clearer about what they want than they admit. The imagined version of next year that you keep returning to, rather than the one you argue for, is data.
## When one of you has already left
There is a specific and common situation the general advice handles badly: one person is leaning out, the other is leaning in, and every conversation is really a negotiation about that. [A named protocol](https://assets.noviams.com/novi-file-uploads/oci/Discernment_Counseling_handout-240847f1.pdf) exists for exactly this.
Doherty developed **Discernment Counseling** through the Couples on the Brink project, after observing there was **no effective protocol for mixed-agenda couples**, where one partner is leaning towards ending it and the other wants to work on it.
It is short, typically **one to five sessions**, and its stated goal is **not to solve the problems but to establish whether they could be solved.** Three outcomes are on the table from the start: **carry on as things are, end it, or commit to six months of couples therapy with separation off the table.**
What it explicitly does not do is attempt to fix the presenting problems. The named target is the **leaning-in and leaning-out dynamic itself**, on the grounds that it must be addressed before any ordinary couples work can function.
This is a clinical practice rather than a controlled trial, and it is presented as that. The reason it belongs here is the design principle, which transfers whether or not you ever sit in the room.
If one of you has one foot out, every conversation about the relationship is secretly a conversation about that, and no amount of work on the actual issues will land while it is true. The pressure the leaning-in person applies is the thing keeping the leaning-out person leaning out. Naming the dynamic, rather than arguing the contents, is what unsticks it.
## The reasons that are not reasons
Whatever you decide, it is worth auditing the arguments you have been making to yourself, because several of the most persuasive ones appear in the research as barriers to a decision rather than as considerations.
**How long it has been.** Time invested is the textbook case of a cost you cannot recover by spending more. The progression bias review lists relationship investments as a mechanism that creates barriers to leaving, which is a polite way of saying that the longer it goes on, the more the length itself becomes the argument.
**That it would hurt them.** Also on the list, and dealt with above. Worth adding that this reasoning is unfalsifiable: it will be true next year too.
**That you might not find anyone else.** Fear of being single is named explicitly in the same review as a barrier to dissolution. It is worth noticing that this is a claim about you rather than about the relationship, and that it is doing its work by changing the subject.
**That it was better before.** Sometimes true and rarely decisive. The question is not whether a better version existed, it is whether it is available now, and the way to find out is the conversation rather than the memory.
**That everyone has doubts.** Everyone does, and the day-to-day swing in how satisfied people feel is measurable even in settled couples. That is an argument against acting on a bad Tuesday. It is not an argument against acting on a bad year.
**That the good parts are really good.** They may well be, and high positives are exactly what defines the ambivalent quadrant rather than the good one. Strong positives do not cancel strong negatives, which is the entire point of measuring them separately.
None of these mean stay is the wrong answer. They mean that if these are the load-bearing reasons, the decision has not actually been made yet.
## If you decide to stay
Deciding to stay is a real choice and it is different from not leaving, which is what most people who stay actually do.
The difference is that a decision comes with terms. What specifically has to change, by when, and what you are yourself going to do differently. Written down somewhere, even if only for you, because a resolution with no content dissolves back into the previous arrangement within about a fortnight.
The PNAS analysis is encouraging here in a specific way: the things that predicted satisfaction most strongly were relationship-level rather than trait-level, led by perceived-partner commitment and appreciation. Those are not fixed properties of either person. Whether your partner feels appreciated by you is something you can affect this week.
And if the thing you keep circling is one particular argument you have had forty times, it may not need this guide at all. Most recurring conflict is [not a solvable problem](/social/the-same-argument-again/) and the target is a version you can both survive having again, not a resolution.
## If you decide to go
Deciding to end it is the harder half to execute, and the two failures are opposite.
One is the slow ending: months of being half out, which the discernment work describes as the leaning-out position and which is the most corrosive state available to both people. It reads as kindness and functions as an extended, deniable rejection.
The other is the impulsive ending, delivered in an argument, which produces a version neither of you can hear and which the research suggests has a decent chance of reversing. Dissolved relationships reconcile frequently, and a breakup delivered at maximum heat is a strong candidate.
The workable version is neither: a decision made cold, said once, clearly, in person, without a list of grievances attached, and then held. Explaining at length is for your comfort rather than theirs. [Saying no without starting a fight](/social/how-to-say-no/) is the same skill applied to a smaller thing, and it is worth having practised.
## What this does not fix
Deciding is not the same as being right, and it is worth being clear about what a decision here does and does not buy you.
It does not tell you whether you will regret it. Nobody has that, in either direction, and waiting for it is another version of waiting for certainty.
It does not make the ending easy if that is the answer. The research notes that breakup recovery is difficult for the person who ends it as well as the person who is left, which is worth knowing in advance so it does not read as evidence you chose wrong.
It does not settle whether you will find something better, and the honest answer there is that [the research on predicting compatibility](/social/first-dates/) says nobody can forecast that from where you are standing.
What it does buy is the end of the deliberation, which has its own running cost, and which almost nobody counts while they are paying it.