Eighteen statements, three minutes, no email. Scored on the two behaviours the demand–withdraw research actually measures — how much you press and how much you retreat — plus a separate score for how you argue, which turns out to matter more than whether you do.
Answer for how you usually are in a close relationship, not for how you would like to be. If you are thinking of one particular argument, that is fine — but note which one, because the result changes with the topic and that is the point.
Scored in your browser. Nothing you answer is sent anywhere or stored.
The four patterns this test can return are Steady, Pursuer, Withdrawer, Pursue-then-withdraw, each reported alongside a separate score for how harshly you argue.
Most conflict quizzes you will find are adapted from workplace instruments, where the dimensions are assertiveness and cooperativeness and the output is a negotiating style. That framework was built for colleagues splitting a budget. It transfers badly to two people arguing at eleven at night about whether one of them was dismissive at dinner.
The relationship literature converged somewhere else. When Andrew Christensen and Christopher Heavey filmed 31 couples in 1990 and had the arguments rated by both partners and by outside observers, the pattern that came through was sequential: one person pressing the issue — requests, complaints, criticism — and the other pulling back into defensiveness and passive inaction. Demand and withdraw. Heavey, Layne and Christensen replicated it in 1993 with a second sample and added a longitudinal edge: the direction mattered, and one version of the pattern predicted a decline in satisfaction a year later while the other did not.
The pattern has held up at scale. A 2014 meta-analysis by Schrodt, Witt and Shimkowski pooled 74 studies and roughly 14,255 participants and found demand–withdraw associated with worse individual, relational and communicative outcomes across the board — around r = .36 overall and about r = .42 for relational outcomes, and larger in distressed samples than in non-distressed ones. That is unusually consistent for this field. The honest limit on it: the pooled data are largely cross-sectional self-report, so they cannot tell you whether the pattern produces distress or distress produces the pattern. Both are plausible and both probably happen.
The reason this test scores pressure and withdrawal as two separate numbers rather than as the ends of one slider is that a person can be high on both. Pressing hard is expensive, and when it runs out the only remaining move is to stop. An argument that goes from intense to silent with nothing in between is one person doing both things twenty minutes apart.
The popular retelling of this research is “women demand, men withdraw”, and it is close enough to what people already believe that it rarely gets checked. The papers say something more specific and more useful.
Christensen and Heavey ran each couple through two conversations: one about a change the husband wanted, one about a change the wife wanted. Both partners were more demanding when the topic was their own request, and more withdrawing when it was their partner’s. There was a gender effect on top of that — men were more withdrawn overall — but, in their words, women were not overall more demanding. The role tracked who wanted something to change more than it tracked who was in the room.
This has a practical consequence for any test like this one, including this one. A single score for “your conflict style” is an average across topics, and the average hides the mechanism. If you want the useful reading, take it twice: once with a disagreement where you are asking for the change, once with one where your partner is. The gap between the two results is more informative than either number on its own.
It also means the label is not an identity. “I’m a withdrawer” is a sentence that turns a situation into a personality, which is the same move that makes jealousy harder to work with than it needs to be.
Whether you argue is a much weaker signal than how you argue. Gottman and Levenson’s 1992 paper followed 73 couples over four years and found that what separated the couples heading for dissolution was not the presence of conflict but its texture: more negative emotional expression, less positive expression, more stubbornness and withdrawal, more defensiveness. The popular packaging of that work is the four horsemen — criticism, contempt, defensiveness, stonewalling.
This test scores the first three together, as harshness, and treats the fourth as the withdrawal dimension, because stonewalling iswithdrawal and counting it twice would inflate the result. The distinction the harshness items are chasing is the one between a complaint and a criticism: a complaint names an event (“you were on your phone the whole meal”), a criticism names a character (“you never care about anything I say”). Same grievance, different sentence, and the second one is much harder to answer without a counter-attack.
Here is where we part company with most of the internet on this subject. The behaviours are well documented; the predictionclaims built on them are not. Divorce prediction papers report accuracies in the 67–95% range, and those figures come from statistical models fitted to the very couples they are then scored against — reconstruction rather than prediction. Heyman and Smith Slep demonstrated the cost in 2001: they built such an equation on half a sample, where it classified 90% of couples correctly and was right 65% of the time when it flagged a couple as heading for divorce, then applied it unchanged to the other half. On the fresh half, flagging a couple as heading for divorce was correct 29% of the time. Same equation, same population, one honest test.
So: contempt is worth removing from how you argue because it is corrosive to be on the receiving end of, and because the observational work keeps finding it clustered in distressed couples. Not because a questionnaire can tell you the odds your relationship ends. Nothing on this page can do that, and the places that imply otherwise are quoting the impressive half of a two-part result.
The move that changes things is not adopting the label, it is catching the loop one step earlier. Pressure and retreat feed each other: the more one person presses, the more leaving looks like the only way to stop it, and the more the other leaves, the more pressing looks like the only way to be heard. Neither person is reacting to the other unreasonably. They are reacting to a machine they built together.
If you scored high on withdrawal, the single highest-value habit is announcing the break and then keeping it. “I need twenty minutes, I’m coming back” is a different act from walking out, and the difference is entirely in whether the other person knows what is happening. If you scored high on pressure, the equivalent is saying the request once and stopping — the second and third versions add urgency but no information.
What happens after the argument matters at least as much as what happens during it. What the repair research supports doing after a fight covers the hour that most people spend re-litigating, and the habits of couples who stay in love covers the ordinary maintenance that makes the arguments smaller in the first place. Getting a difficult thing said at all is its own skill, and how people actually get someone to open up is about the version of that which does not require an argument as the trigger. More in Mindset.
There is also a good chance your conflict pattern is downstream of something else. Anxious attachment tends to look like pressure — the alarm goes off and staying quiet feels impossible — and avoidant attachment tends to look like withdrawal. If the description of your result felt familiar in a way that goes back further than this relationship, the attachment style test scores those two dimensions directly, and telling love apart from anxious attachment is the piece that connects them. For the other direction — what it looks like when someone handles this well — there is the habits of emotionally mature people, and for whether any of it is a good sign, what actually predicts that a new relationship has potential.
If arguing this way has been going on long enough that the question has moved from “how do we argue” to “is this worth it”, that is a different measurement. The relationship satisfaction test scores how the relationship is going and, separately, the two things that hold people in relationships regardless of how they are going.
It is the most studied pattern in relationship conflict: one person presses for the issue to be discussed and resolved, and the other retreats into silence, agreement or physical absence. Christensen and Heavey described it in 1990. A meta-analysis of 74 studies covering 14,255 people found it consistently associated with lower relationship satisfaction. It is a loop rather than two separate habits — pressure produces retreat, and retreat produces more pressure.
Less than the popular version claims. In Christensen and Heavey's own data both partners were more demanding when the conversation was about a change they wanted, and more withdrawing when it was about a change their partner wanted. The role followed the topic. There was an overall gender effect — men were more withdrawn on average — but women were not overall more demanding, which is the half of the finding that usually gets dropped.
Criticism, contempt, defensiveness and stonewalling — four behaviours John Gottman's team coded in recorded arguments and found more often in couples who later separated. This test scores the first three as one 'harshness' dimension and treats the fourth, stonewalling, as the withdrawal score, because that is what it is. Counting it twice would inflate your result.
No, and be suspicious of anything that says it can. The famous divorce-prediction accuracy figures come from equations fitted to the same couples they were then tested on. When Heyman and Smith Slep built such an equation and tested it on a fresh sample, correctly identifying a couple who would separate fell from 65% to 29%. The behaviours are real and worth knowing about. The prediction claims built on top of them are the part that has not held up.
No. The CPQ, by Christensen and Sullaway, is a validated instrument that asks about the couple — what each of you does in sequence — rather than about one person alone. This is an 18-item educational adaptation that follows the same conceptual structure from one side only. It will point you somewhere useful and it is not a validated scale.
No. Every question is scored in your browser and nothing is transmitted or saved. There is no account, no email field, and no results URL to share by accident. Close the tab and the answers are gone.
How we choose and read sources is set out in our editorial policy. If something here is wrong, tell us and it goes on the corrections page.