The research behind this assessment
If you are about to answer forty questions about the most private part of your life, you are owed an account of where those questions came from. This page lists every concept the program uses, the studies behind each one, and — just as importantly — what this tool cannot tell you.
Three things up front
- Scoring is arithmetic. Fixed weights, no algorithm deciding anything about your relationship.
- No AI returns a verdict. AI is used only to reflect your own written entries back to you at the end.
- This is research-informed, not a validated psychometric instrument. The limitations are stated in full further down.
Why a single quiz cannot answer this
Ask someone on a Tuesday after a good weekend whether the relationship is worth saving and the answer is yes. Ask the same person on a Thursday after a fight and it is over. Both feel like clarity. Neither is evidence.
The mechanism has a name. Mood-congruent memory, documented across decades of experimental work, means the emotional state you are in when you recall something biases which memories become available and how they are weighted. Your current feeling quietly re-edits the past six months every time you review them. That is why the stay-or-go question can stay open for years: each week overwrites the last, and no single sitting can measure anything except the mood you sat down in.
The response is structural rather than clever. Rate the same eight dimensions each week, in writing, as things happen. After two months you are no longer arguing with yourself from memory — you are reading a trend line that your mood cannot rewrite.
Diagnostic questions, not pros and cons
The structural idea behind the question set comes from Mira Kirshenbaum's Too Good to Leave, Too Bad to Stay: replace the balance sheet with diagnostic questions that carry asymmetric predictive value. A pros-and-cons list weights whatever is loudest today and produces a different answer each time it is made honestly. A diagnostic question is one where a particular answer tells you far more than its opposite does — the way contempt tells you more about a relationship's future than its absence does.
The eight dimensions and their sources
Eight is a deliberate number. Fewer and you cannot see which part is failing. More and nobody answers honestly every week for two months.
1. Respect and contempt
What it measures. Whether basic warmth survives irritation, and whether either of you speaks to or about the other with mockery, sarcasm or dismissal.
The research. Gottman and Levenson's longitudinal studies of married couples found that contempt — eye-rolling, name-calling, talking down — is the single strongest behavioural predictor of separation, and part of a cluster (with criticism, defensiveness and stonewalling) that forecast divorce with high accuracy across multi-year follow-ups. Criticism is recoverable. Sustained contempt rarely resolves unless the person expressing it decides it is unacceptable.
How it becomes questions. Four items ask about tone rather than content — how you are spoken to when they are annoyed, whether you are talked down to, how you speak about them to others, whether your view carries weight in joint decisions. The weekly experiment asks you to count instances in both directions for seven days without raising them, because counting cuts through both "it's not that bad" and "it's constant".
2. Repair after conflict
What it measures. What happens in the hours and days after a fight, and who moves first.
The research. The same body of work found that stable and unstable couples argue at similar rates. The discriminating variable is repair: whether an attempt to de-escalate is made, and crucially whether it is accepted. Repair success predicts outcomes better than conflict frequency or intensity does.
How it becomes questions. Items ask whether one of you finds a way back within a day or two, whether arguments get buried instead of resolved, whether apologies are meant, and whether raising something hard turns into a fight about how you raised it. The experiment is to make one explicit repair attempt and record what is done with it — accepted, ignored, or used against you.
3. Trust and felt safety
What it measures. Whether you can relax in their company, and what you have stopped saying.
The research. Attachment research, from Bowlby's original formulation through Hazan and Shaver's extension to adult romantic bonds, treats felt security as the base layer of a relationship rather than one feature among many. When a partner is appraised as a source of threat rather than a safe base, self-protective strategies — withholding, monitoring, bracing — take over, and everything built above that layer becomes unstable.
How it becomes questions. Items measure bracing before speaking, reliability of stated intentions, and how much of your life you keep private to avoid a reaction. The experiment asks you to say one small true thing you would normally hold back, and to note your fear level beforehand against what actually happened.
4. Closeness and intimacy
What it measures. Emotional and physical connection, and — more importantly — whether it responds.
The research. Reis and Shaver's intimacy process model defines intimacy not as a quantity of affection but as perceived partner responsiveness: the sense that disclosure is met with understanding, validation and care. That reframing matters here because absolute levels fluctuate enormously with stress, children, illness and work, while responsiveness does not fluctuate in the same way.
How it becomes questions. Items ask about genuine closeness versus logistics, loneliness in their presence, mutual wantedness of physical affection, and curiosity about your inner life. The experiment is to initiate one moment of connection with no agenda and observe whether they meet you there. A low that moves when attended to is a different signal from a low that does not move at all.
5. Autonomy and sense of self
What it measures. How much of your own life, opinion and perimeter is still intact.
The research. Aron and Aron's self-expansion model describes close relationships as including the other in the self — healthy up to the point where the self stops being distinguishable. Differentiation-of-self work in the Bowen tradition makes the parallel argument: the capacity to hold a position while staying emotionally connected is what allows a bond to survive disagreement. Loss of self inside a relationship is associated with regret in both directions, staying and leaving alike.
How it becomes questions. Items ask about friendships and interests that are still yours, whether you have become a smaller version of yourself, whether you can disagree without managing the fallout for days, and whether you edit what you want to keep the peace. The experiment is to reclaim the smallest lost thing and note both their reaction and your own resistance.
6. Shared future and direction
What it measures. Whether you want the same life — children, location, money, faith, ambition.
The research. Value and life-direction mismatches behave differently from other relationship problems. Where communication and conflict patterns respond to skill and effort, a genuine disagreement about whether to have children or where to live does not average out over time; it resolves when one person changes their mind, or it does not resolve. Gottman's distinction between solvable and perpetual problems, and the wider literature on incompatibility-driven dissolution, both point the same way.
How it becomes questions. Items ask whether five years from now includes them and whether that feels good, whether you want fundamentally different things, whether the future can be discussed without tension, and whether you are waiting for them to change before committing. The experiment is to raise one concrete future question and record whether you got a position or a deflection.
7. Effort balance
What it measures. Who initiates the plans, the repairs, the difficult conversations, the affection.
The research. Equity theory in close relationships holds that perceived under-benefit is corrosive over time — distress tracks the perception of imbalance rather than the raw division of labour. Asymmetric effort is the most common pattern reported by people stuck in the stay-or-go question. The diagnostic signal in the research is not the imbalance itself but its responsiveness to being named.
How it becomes questions. Items ask whether the work is shared, whether anything would happen if you stopped, whether naming a problem clearly produces change, and whether they notice you struggling unprompted. The experiment is to name one imbalance once, as information rather than complaint, then step back and observe what fills the gap.
8. Staying out of fear
What it measures. Whether you are choosing them, or unable to leave.
The research. Rusbult's investment model shows that commitment is predicted by three separable factors: satisfaction, quality of alternatives, and investment size. People stay in relationships they are not satisfied with when alternatives look poor and investment is high — a mechanism that operates independently of relationship quality. Sunk-cost research explains why past investment feels like a reason to continue even when it carries no information about the future.
How it becomes questions. Items separate choice from constraint: whether you would still choose them if leaving carried no cost, how much of your staying is about what you have already invested, and whether fear of being alone is the main anchor. For the married track, items on shared money, property and children are added, because those are the constraints most often mistaken for reasons. The experiment is to take one practical step on the biggest fear, since solving a fear does not commit you to leaving — it just stops the fear making the decision.
Why each week ends in an experiment
Every week the program hands you one small, specific thing to try in the real world, aimed at whichever dimension is scoring lowest. The format is borrowed from behavioural experiments in cognitive therapy, where a belief is tested against reality rather than debated internally — you predict what will happen, do the thing, and record what actually happened.
It earns its place here because of one specific blind spot: a relationship that responds to effort and a relationship that absorbs effort without changing look identical from the inside. Self-rating cannot separate them. A test can.
Safety is screened, not scored
Before any question about relationship quality, the assessment screens for physical harm, threats, financial and movement control, surveillance, and sexual coercion. A positive answer stops the flow rather than lowering a score.
This follows Evan Stark's work on coercive control, which reframes abuse as a pattern of domination and liberty deprivation rather than a series of violent incidents on a severity scale. A controlling relationship is not a low-scoring relationship; it is a different category with different answers and different timelines. Eight weeks of observation is the wrong response to it. Support is here.
How the scoring actually works
Each item is answered on a five-point frequency scale from Never to Almost always. Items where agreement is a bad sign — "I brace myself before telling them something they won't like" — are reverse-scored, so a high answer lowers the dimension rather than raising it.
Within each dimension the answered items are averaged and mapped onto a 0–100 scale, so every dimension is comparable with every other regardless of how many items it carries. The overall figure is the plain mean of the eight dimension scores. Nothing is weighted by importance, because how much respect matters relative to intimacy is a judgement about your life rather than a fact we can supply.
Track selection changes which items appear. The married and long-term track adds items on shared property, children, and structured attempts at change; the dating track adds items on whether the relationship has a named direction and whether this doubt repeats a pattern from previous relationships. The eight dimensions stay the same either way, which is what keeps your weekly scores comparable.
Weekly modules are ordered weakest-first: the dimension scoring lowest at baseline is the one you work on first, because it is the one carrying most of the doubt.
Limitations, stated plainly
- Not a validated instrument. The dimensions are drawn from published research, but this particular question set has no published reliability or validity testing. It has not been normed against a population, and its scores are not clinical measurements.
- One side of the story. Everything here is self-report from one partner. It measures your experience of the relationship, which is the thing you have to decide about, but it is not an objective account of the relationship.
- Population-level findings. Research that predicts outcomes across thousands of couples does not predict yours. Contempt strongly predicts separation in aggregate; it does not mean your relationship is over.
- Not therapy. This is a structured way of gathering your own evidence. It is not counselling, not treatment, and not a substitute for either.
- No verdict, by design. The program will never tell you to stay or leave. It cannot know what you can live with, and that is the whole of the question.
Common questions
Is this relationship quiz actually accurate?
It is research-informed, not a validated psychometric instrument. Every dimension is drawn from published relationship science, but the specific wording has not been through published reliability or validity testing. Its accuracy comes mostly from repetition over weeks rather than from any single reading.
What research says whether a relationship will last?
The most durable findings are that contempt predicts separation more strongly than conflict frequency, that whether repair attempts land matters more than whether fights happen, and that investment and lack of alternatives keep people in relationships independently of how satisfied they are.
Can a quiz tell me whether to leave my marriage?
No, and this one does not try. Scoring here is arithmetic — no algorithm and no AI returns a verdict. What the program produces is your own evidence over time, so the decision is made on a pattern rather than on the week you happen to be having.
Why does the program take weeks instead of giving an answer now?
Memory is mood-congruent: your current feeling quietly re-edits your recollection of the past months. A single sitting measures your mood. Weekly ratings, written as things happen, measure the relationship.
What if I am afraid of my partner?
Then this is the wrong tool. Coercive control and violence are a category difference, not a low score, and the assessment screens for them at the start and routes to support instead of scoring.
References
- Gottman, J. M., & Levenson, R. W. (1992). Marital processes predictive of later dissolution: Behavior, physiology, and health. Journal of Personality and Social Psychology, 63(2), 221–233.
- Gottman, J. M., & Levenson, R. W. (2000). The timing of divorce: Predicting when a couple will divorce over a 14-year period. Journal of Marriage and Family, 62(3), 737–745.
- Bowlby, J. (1969/1982). Attachment and Loss, Vol. 1: Attachment. Basic Books.
- Hazan, C., & Shaver, P. (1987). Romantic love conceptualized as an attachment process. Journal of Personality and Social Psychology, 52(3), 511–524.
- Reis, H. T., & Shaver, P. (1988). Intimacy as an interpersonal process. In S. Duck (Ed.), Handbook of Personal Relationships (pp. 367–389). Wiley.
- Aron, A., & Aron, E. N. (1986). Love and the Expansion of Self: Understanding Attraction and Satisfaction. Hemisphere.
- Bowen, M. (1978). Family Therapy in Clinical Practice. Jason Aronson.
- Rusbult, C. E. (1980). Commitment and satisfaction in romantic associations: A test of the investment model. Journal of Experimental Social Psychology, 16(2), 172–186.
- Rusbult, C. E., Martz, J. M., & Agnew, C. R. (1998). The Investment Model Scale. Personal Relationships, 5(4), 357–387.
- Arkes, H. R., & Blumer, C. (1985). The psychology of sunk cost. Organizational Behavior and Human Decision Processes, 35(1), 124–140.
- Bower, G. H. (1981). Mood and memory. American Psychologist, 36(2), 129–148.
- Bennett-Levy, J., et al. (2004). Oxford Guide to Behavioural Experiments in Cognitive Therapy. Oxford University Press.
- Hatfield, E., Utne, M. K., & Traupmann, J. (1979). Equity theory and intimate relationships. In Social Exchange in Developing Relationships. Academic Press.
- Stark, E. (2007). Coercive Control: How Men Entrap Women in Personal Life. Oxford University Press.
- Kirshenbaum, M. (1996). Too Good to Leave, Too Bad to Stay. Plume.
See where you stand today
The free baseline scores all eight dimensions in about 25 minutes and shows you the three driving your doubt. No verdict, no pressure to continue.
Take the free assessmentRelated reading: how the program works, should I stay or leave? and is my marriage over?