Why the Honeymoon Phase Ends: What Is Measured and What Is Folklore

What the search engines told us when we asked this question

On 18 October 2026 we put this article's exact question to Google and to Brave and read what came back, because the answer is unusually revealing about where the standard explanation comes from.

Google's AI Overview opened with a mechanism: the honeymoon phase ends “because of natural shifts in neurochemistry, brain adaptation, and daily routine.” It then laid out three neurochemical steps — a surge of dopamine and norepinephrine, a tolerance-and-baseline-reset as the nervous system adapts, and a handover to oxytocin and vasopressin. Each step carried a source chip. The dopamine claim was attributed to a podcast site. The tolerance claim was attributed to The Gottman Institute. The claim that “subconscious patterns emerge” was attributed to TikTok.

Brave's AI answer said the same thing in the same order and added one more compound: the brain is “flooded with dopamine, oxytocin, and nerve growth factor,” and “over time, these chemicals naturally subside, replaced by vasopressin and oxytocin.”

Below those summaries, the first page on both engines contained Cleveland Clinic, Healthline, Verywell Mind, Psychology Today, Vogue, Bustle, Brides, several therapy practices, a handful of Reddit threads and five videos. We could not find a study cited on any of them.

So there is a mechanism in wide circulation, stated with confidence, assembled by search engines, and sourced — where it is sourced at all — to a podcast and a short-video platform. This page is an attempt to find out which parts of it have been measured.

The short answer, worked through below: the decline is real and it is one of the best-documented findings in relationship research. Almost nothing else in that paragraph survives contact with the papers. The neurochemical story rests on two brain-imaging studies of seventeen people each and one blood test — and the second of those brain studies found the opposite of what the story predicts.

Every empirical claim here carries a linked source, which is the promise in our editorial policy. Where the evidence is thin, this page says so instead of rounding up.

Three different questions are wearing the same words

Before any of the numbers, a note that belongs at the top rather than the bottom. If what prompted this search is that a partner's behaviour changed once the relationship became secure — they started deciding who you see, going through your phone, punishing you with silence, or escalating when you say no — that is not the honeymoon phase ending, and a later section explains why the two get confused and why the confusion is costly. Find A Helpline, in the sources below, lists free and confidential services by country. In the United States the National Domestic Violence Hotline is 1-800-799-7233, or text START to 88788, 24 hours a day.

With that said. “Why does the honeymoon phase end” is asked by three different people.

The first wants to know whether the drop is normal — whether what they are feeling is what everyone feels. That question has a good answer with real numbers behind it, and the numbers are smaller and slower than the word end suggests.

The second wants the mechanism. Why does it happen? That is where the chemical story lives, and it is the part of this subject with the widest gap between confidence and evidence.

The third is not really asking about honeymoons. They are asking whether the change they are watching means the relationship is in trouble. That question has been studied directly, by following couples for up to thirteen years, and the finding is sharper than anything the genre reports: it is not the average decline that distinguishes the couples who end up unhappy. It is the steepness, and the steep ones were identifiable at the start.

Those three get answered in that order. One framing point before starting, because it governs the whole page: almost all the trajectory data comes from married couples, usually measured from the wedding onwards. “The honeymoon phase” as the internet uses it — the first months of dating — is barely measured at all. That gap matters and the last section returns to it.

There is a measured honeymoon, and it is in the happiness data

The cleanest version of the honeymoon effect does not come from relationship research. It comes from the literature on whether people adapt to life events, where marriage is one of the events studied and the pattern has a name.

Richard Lucas, Andrew Clark, Yannis Georgellis and Ed Diener used a 15-year longitudinal study of over 24,000 individuals to look at what marital transitions do to life satisfaction. On average, people reacted to the event and then adapted back toward their baseline. That is the honeymoon, measured at population scale: a bump, then a return.

The part that almost never travels with the headline is the rest of their abstract. There were substantial individual differences in that tendency. People who initially reacted strongly were still far from baseline years later. And many people showed trajectories running in the opposite direction to the one adaptation theory predicts. The authors' own conclusion is that marital transitions can be associated with long-lasting changes in satisfaction, and that those changes get overlooked when only average trends are examined.

That sentence is worth keeping for the whole of this page. The average says it fades. The average is also hiding people it does not describe.

The wider meta-analysis makes the same point about what is actually being measured. Maike Luhmann, Wilhelm Hofmann, Michael Eid and Lucas pooled longitudinal data from 188 publications, 313 samples and 65,911 people across four family events — marriage, divorce, bereavement, childbirth — and four work events. Two results matter here. Life events affect affective well-being and cognitive well-being differently, and for most events the effects on cognitive well-being are stronger and more consistent. And the rate of adaptation is not a function of how desirable the event is supposed to be.

Translated: how you feel day to day and how you judge your life when you stop and assess it are two different measurements that move differently after the same event. A great deal of the confusion about whether the honeymoon “ends” comes from people comparing a memory of the first kind with a judgement of the second. It is the same measurement trap that makes people misread their own history with a partner in the signs you have actually moved on.

How much satisfaction actually drops, and when

The question “how much” has a single best answer, and it is recent and large.

Janina Bühler, Samantha Krauss and Ulrich Orth assembled every study they could find on how relationship satisfaction develops: 165 independent samples and 165,039 participants, with mean ages from 20 to 76 and mean relationship durations from three months to 46 years. They analysed it two ways, and the distinction between the two is the single most useful thing on this page.

The first was mean level — cross-sectional information, the score at a sample's first assessment. Plotted against age, satisfaction decreased from 20 to 40, reached a low point at 40, increased until 65, then plateaued. Plotted against relationship duration, it decreased over the first ten years, bottomed out at ten years, rose until twenty, and declined again.

The second was mean change — longitudinal information, how much scores moved per year inside a relationship. Here there was no rise. Relationship satisfaction decreased within a given relationship, with the largest declines in young adulthood and in the first years of a relationship.

Two things follow. The decline is front-loaded: whatever is happening, most of it happens early, which is exactly where the honeymoon-phase question is pointed. And the shape depends entirely on which of the two analyses you read — a problem the next section is about.

The moderator results are worth having as well, because they rule out most of the explanations people reach for. Only two things explained variance in the mean level: the presence of children, and which questionnaire was used. The pattern otherwise held across birth cohort, sample type, country, ethnicity, gender, whether the couple shared a household, marital status, relationship transitions and whether the data were dyadic.

That last list is doing a lot of quiet work. The decline is not a feature of one country, one generation, one sex, or married couples specifically. Whatever drives it is not cultural in any way this dataset can detect.

The famous U-shape is an artifact, and one paper contains both halves

For decades the standard picture of marital happiness was a U: high at the start, sagging in the middle while the children are at home, recovering once they leave. It is intuitive, it is widely repeated, and the best test of it says it is a measurement artifact.

Jody VanLaningham, David Johnson and Paul Amato used a national, 17-year, five-wave panel sample and did something careful. First they took the cross-sectional data from the first wave alone — one snapshot, people at different marriage durations — and replicated the U-shape. Then they analysed the same study as a panel, with a fixed-effects pooled time-series model that follows the same people across waves.

In the panel analysis they found declines in marital happiness at all marital durations and no support for an upturn in the later years. The relationship was slightly curvilinear, with the steepest declines occurring in the earliest and the latest years. The negative effect of duration survived controls for other life-course variables. Their conclusion is blunt: the U-shaped pattern is an artifact of cross-sectional research and is not typical of US marriages.

Why the snapshot lies is easy to state once you see it. A cross-section of long-married people is not a sample of how marriages go. It is a sample of the marriages that lasted — the unhappy ones have left — so the apparent recovery in the later years is partly the unhappy couples disappearing from the data rather than anyone's happiness returning.

Now put that next to the meta-analysis in the previous section, because the same split is sitting inside it. Bühler and colleagues describe their mean-level results as cross-sectional information and their mean-change results as longitudinal. The cross-sectional half shows the rise after the low point. The longitudinal half shows decline within relationships and no rise at all. Twenty years after VanLaningham's paper, the largest synthesis ever run on this question reproduces both the artifact and its correction, in the same tables, correctly labelled.

This matters practically. If someone tells you it gets better after year ten, ask which of those two numbers they are quoting.

Most couples do not fall off a cliff

The word phase implies an edge — a period that is on, and then is off. One study set out to test whether that edge exists as a distinct pattern, and it is the most directly relevant piece of research on this entire question.

Michael Lorber, Ariel Erlanger, Richard Heyman and Daniel O'Leary studied 395 community couples from one month before the wedding through two and a half years after it, and looked for what they called a honeymoon effect: initially high but rapidly decreasing marital satisfaction. They used trajectory modelling, so rather than averaging everyone they asked how many distinct patterns the data contain and who is in each.

The answer: a supermajority of couples showed initially high marital satisfaction that either slowly decreased — 86% of women — or remained steady — 78% of men. The rapidly-decreasing pattern, the one the phrase describes, was a minority class. The authors put the men at risk of it at around 14% and the women at around 10%.

James Lavner and Thomas Bradbury found the same shape with a different sample and a longer horizon: 464 newlywed spouses, eight reports of satisfaction over four years. Five trajectory groups emerged, and the list of them is the finding. Some had high intercepts and no declines at all. Some had moderate intercepts and minimal declines. Some had low intercepts and substantial declines. The groups differed systematically in their four- and ten-year divorce rates, and wives tended to follow more satisfying trajectories than their husbands.

So the honest answer to “does the honeymoon phase end for everyone” is no, not in the sense the question means. For most couples in these samples the first years are a gentle slope or a flat line, not a drop. A distinct rapid decline is real, it happens to a minority, and — as the next section shows — it is not a mystery.

Which also means the average decline quoted everywhere is a blend of several different things happening to several different groups. Averaging a flat line and a cliff produces a gradient that describes nobody. Our relationship satisfaction test exists for the same reason: a snapshot of where a relationship sits is more honest than a forecast of where it is going.

The steep declines were visible at the start, and one predictor keeps recurring

If the rapid-decline group is a minority, the obvious question is who is in it. Three studies have answered that, and they agree in a way that is uncomfortable for the standard explanation.

In Lorber and colleagues' trajectory analysis, membership in the less optimal classes was most consistently predicted by spouses' own intimate partner violence and depression, and by their partners' marital satisfaction, intimate partner violence and depression. The authors' framing is preventive: men at risk of the honeymoon effect can be identified in advance from such predictors, and so can the women at risk of decreasing satisfaction — though the women at risk may also show somewhat attenuated satisfaction as early as the engagement.

In Lavner and Bradbury's five-group study, the variables that discriminated among the trajectory groups were personality traits, stress, aggression and communication behaviours assessed shortly after marriage.

The third study is the one that should be better known. Lavner, Benjamin Karney and Bradbury asked 502 newlywed spouses in 251 marriages to predict how their feelings about the relationship would change over the following four years, then compared those forecasts with what actually happened. Nearly every spouse predicted their satisfaction would stay the same or improve. On average it declined. And the wives with the most optimistic forecasts showed the steepest declines — those wives also had lower self-esteem, higher stress and higher levels of physical aggression toward their partners at the start.

Three samples, three designs, and the same variable keeps appearing in the group with the cliff: aggression, present and measurable before the decline happened.

That is a different causal story from the one the search engines gave. In these data the steep version of “the honeymoon ended” is not what happens when novelty runs out. It is what was already visible at baseline in couples where something was wrong, and the optimism of the first months was covering it rather than protecting against it. The authors of the forecasting study say exactly that: believing your marriage will improve does not make it so, and may paradoxically mask risky relationships.

Disillusionment, not emerging conflict

There are two competing accounts of how a good relationship becomes a bad one, and they make opposite predictions about what you would see in the first years.

The emergent distress account says trouble arrives: negativity builds, conflict escalates, and a relationship that started fine deteriorates into one that is not. The disillusionment account says the trouble is the fading itself — that what distinguishes couples heading for separation is not new conflict but the loss of what was there at the beginning.

Ted Huston, John Caughlin, Renate Houts, Shanna Smith and Laura George tested the two directly, following newlywed couples and their first two years of marriage and then checking their marital fate after 13 years. The result provides, in their words, little support for the idea that the emergence of distress leads to marital failure. What distinguished couples headed for divorce from those who built a stable bond was disillusionment: an abatement of love, a decline in overt affection, a lessening of the conviction that one's spouse is responsive, and an increase in ambivalence.

There is a second finding in the same paper that cuts even closer to this article's question. Differences in the intensity of newlyweds' romance, and in how much negative feeling they expressed, predicted both whether couples who stayed married were happy thirteen years later and how long the marriages that ended lasted before separating. The differences were present at the start.

So the end of the honeymoon is not a neutral developmental stage that everyone passes through on the way to mature love. In the longest-running test of it, how much the early intensity abates is the variable that carries the signal — and it carries it over more than a decade.

The background to all of this is Karney and Bradbury's review of the field, which evaluated 115 longitudinal studies representing over 45,000 marriages and remains the standard map of what this literature can and cannot establish. It is worth knowing it exists, because the confident paragraph a search engine will hand you was not built from it.

What Huston's result does not license is reading a quiet patch as doom. It is a group-level contrast between couples who separated and couples who did not, measured over thirteen years. The behavioural side of what the stable couples were doing is in the habits of couples who stay together, and because the disillusionment account turns on responsiveness rather than on the volume of arguing, our conflict style test is built on the demand-withdraw literature rather than on how often a couple fights.

The dopamine story, and how much of it exists

Now the mechanism both AI summaries led with. The claim is that early love floods the brain with dopamine, that the brain adapts and the high cannot be sustained, and that the reward chemistry hands over to the bonding chemistry. Here is the evidence base for it, in full.

The founding study is by Arthur Aron, Helen Fisher, Debra Mashek, Greg Strong, Haifang Li and Lucy Brown. They scanned 10 women and 7 men who were intensely in love, looking alternately at a photograph of their beloved and a photograph of a familiar person. Activation specific to the beloved appeared in dopamine-rich areas associated with reward and motivation — the right ventral tegmental area and the right caudate nucleus. Activation in the right anteromedial caudate correlated with questionnaire scores for intensity of romantic passion.

That is a real result and it is the origin of everything downstream. Note what it is not. It is seventeen people, scanned once. The participants had been in love from one to seventeen months, and the study was not designed to track anyone over time. It contains no measurement of anything fading, because it did not measure anyone twice.

The follow-up is where the story breaks. Bianca Acevedo, Aron, Fisher and Brown ran the same paradigm on 10 women and 7 men married an average of 21.4 years who reported still being intensely in love. If the handover account is right, the dopamine-rich reward response should be gone, replaced by the quieter attachment circuitry.

It was not gone. Effects specific to the long-term partner appeared in the ventral tegmental area and dorsal striatum — the same dopamine-rich reward system — which the authors describe as consistent with results from early-stage romantic love studies. The attachment-related regions showed up as well, not instead. And the correlations inside the data split the constructs apart neatly: ventral tegmental and caudate activity tracked romantic love scores and felt closeness; globus pallidus activity tracked friendship-based love; and a separate set of regions tracked obsession. The authors' conclusion is that for some individuals the reward value associated with a long-term partner may be sustained, similar to new love.

So the two imaging studies that the chemical story rests on amount to 34 people from one research programme, and the one that actually looked at long relationships found the reward response still there after two decades. The honest reading is not that the chemistry cannot fade. It is that nobody has shown it must.

The one blood marker that was followed up, and what it can and cannot show

Brave's summary named a specific molecule — nerve growth factor — and there is exactly one study behind that, which makes it checkable.

Enzo Emanuele and colleagues measured plasma neurotrophins in 58 people who had recently fallen in love and compared them with people who were single and people already in long-lasting relationships. NGF was significantly higher in the group in love: 227 pg/ml, against 123 for those in long-lasting relationships and 149 for the single group, p < 0.001. NGF also correlated with scores on the Passionate Love Scale at r = 0.34, p = 0.007. The other three neurotrophins they measured showed no differences at all.

Then the part everyone quotes. 39 of the people in love were re-tested after 12 to 24 months. They had kept the same relationship but were, in the paper's phrasing, no longer in the same mental state they had described at the first visit. Their plasma NGF had decreased and become indistinguishable from the control groups.

That is a genuine longitudinal observation and it is the closest thing to a measured timeline this subject has. It is also narrower than it looks, in three ways worth stating plainly.

The follow-up group was selected on the outcome. These were the subjects who were no longer in the early state; the paper is reporting that a marker tracked a state that had already ended by self-report. It cannot tell you that the marker's decline caused the state to end, and it does not claim to.

It is a blood measure, not a brain measure. Plasma NGF is not a love chemical and nothing in the paper suggests it is a switch. The authors' own conclusion is carefully hedged: some behavioural or psychological features associated with falling in love could be related to raised NGF levels.

And the second-most-cited chemical finding is smaller still. Donatella Marazziti, Hagop Akiskal, Alessandra Rossi and Giovanni Cassano compared 20 people who had fallen in love within the previous six months, 20 unmedicated patients with obsessive-compulsive disorder, and 20 controls, and found the density of the platelet serotonin transporter significantly lower in the first two groups than in controls. That study — twenty people, a platelet measure — is the entire basis for the widely repeated claim that early love is chemically like OCD.

So how long does it last? Three measured answers, none of them the question

This is the most-asked version of the query, and it deserves a direct answer rather than a hedge. The direct answer is that nobody has measured it, and here is what exists instead.

The adaptation window. In the 24,000-person panel, people reacted to marriage and drifted back toward baseline over the following years — with the caveat from the same paper that strong reactors stayed far from baseline much longer and some never returned at all.

The 12-to-24-month window. That is the follow-up interval in the NGF study, and it is where the popular figure of “one to two years” almost certainly comes from. It describes 39 people who were asked again at some point in that range, not a measured endpoint. Nobody tested at six months, or at three years, so the interval is an artifact of the study design rather than a finding about when the state ends.

The satisfaction slope. The largest meta-analysis puts the steepest within-relationship declines in the first years, with the low point against relationship duration at around ten years.

None of those is the honeymoon phase as the question means it. To measure that you would need to follow new couples from the first weeks, with a defined construct and frequent measurement, and record when it ended for each person. That study has not been published. The nearest thing is the Passionate Love Scale, built by Elaine Hatfield and Susan Sprecher, which gives you an instrument — but the trajectory studies that use it start at the wedding, by which point the phase in question is often already over.

Which is why the ranges in circulation vary so much and cite nothing. The StyleCraze page on the first results page says six months to two years, with no source. The term limerence, often attached to a figure of eighteen months to three years, comes from a 1979 trade book; we searched Crossref and OpenAlex and could not find a peer-reviewed measurement of its duration. The most honest page we found was Cleveland Clinic's, which says the honeymoon phase can last “weeks, months or, in some cases, years” and leaves it there.

An unsourced number that everyone repeats is a pattern on this site rather than a one-off. It happened to the 7-38-55 figure, to Gottman's 69% of perpetual problems, to the three-day rule and to every other numbered dating rule we have traced. What is new, and what the first section of this page documents, is that the search engine now assembles the number and places it above the results rather than leaving it in somebody's blog post.

Passion, obsession and desire are three different measurements

Part of the confusion is that “the honeymoon feeling” is at least three constructs that researchers measure separately and that move differently. Separating them answers the question better than any single number.

Passionate love. Susan Sprecher and Pamela Regan studied 197 couples at different stages of courtship and the transition to marriage, testing three predictions about how passionate love differs from companionate love: that it is more sexualised, that it is more associated with intense positive and negative emotional experience, and that it declines with the passage of time. They found support for the first and the third. They found little support for the second — the idea that passionate love is the emotionally turbulent kind did not hold up. Both forms were associated with satisfaction and commitment, and companionate love was more strongly associated with satisfaction than passionate love was. The design compares couples at different stages rather than following the same couples, so read it as a gradient across stages, not a filmed decline.

Obsession. This is the component most people are actually describing when they say the spark went: the preoccupation, the checking, the inability to think about anything else. Acevedo and Aron found that romantic love and obsession load as independent factors, and that obsession correlates positively with satisfaction in short-term relationships and negatively in long-term ones. The thing that reliably fades is the thing whose persistence predicts worse outcomes. That result is unpacked at length in what actually builds attraction over time, and it is the single most useful idea in this subject.

Desire. Fritz Klusmann surveyed 1,865 German students aged 19 to 32 in steady heterosexual partnerships and found three things as relationship duration increased. Sexual activity and sexual satisfaction declined in both sexes. Sexual desire declined only in women. And desire for tenderness declined in men while rising in women. The author is careful about his own design: the data are cross-sectional, so a longitudinal reading is precarious, and differences in mating strategy could account for part of the pattern.

The practical consequence of holding those three apart: “the honeymoon ended” can mean the obsession lifted, which the data treat as good news in a long relationship, or it can mean desire and satisfaction fell, which is a different measurement with a different answer. And the two get confused in a specific, costly direction, because the preoccupied kind of intensity is the one that feels most like love from the inside — the distinction worked through in love or anxious attachment, and one of the reasons our attachment style test reports two continuous scores rather than handing out a label.

Children change the shape more than the total

Having a baby is the explanation most couples reach for, and it is the one place where the meta-analytic answer and the long-prospective answer appear to disagree. They are both worth having, because the disagreement is informative.

The meta-analysis says the effect is small and not unique to parenthood. Danielle Mitnick, Richard Heyman and Amy Smith Slep aggregated 37 studies tracking couples from pregnancy to after the birth of a first child, plus four studies tracking childless newlyweds. From pregnancy to eleven months after birth, both men and women showed significant but small declines; five studies that followed to 12–14 months found moderate declines. Then the result that reframes it: the four studies of newlyweds indicated that couples who did not become parents experienced a decrease in satisfaction similar to that of parents across a comparable span of time. Their conclusion is that the decrease may not indicate anything unique about the transition to parenthood.

The cross-sectional comparison gives the effect a size. Jean Twenge, W. Keith Campbell and Craig Foster found parents reporting lower marital satisfaction than non-parents at d = −.19, and a negative association with the number of children at d = −.13. The gap was most pronounced among mothers of infants: 38% of mothers of infants had high marital satisfaction, against 62% of childless women. The effect was more negative among higher socioeconomic groups, younger birth cohorts and in more recent years.

The eight-year prospective study says the timing is different even if the total is not. Brian Doss, Galena Rhoades, Scott Stanley and Howard Markman followed 218 couples across the first eight years of marriage, with observed as well as self-reported measures. Parents showed sudden deterioration following the birth, small to medium in size, which tended to persist for the rest of the study. Mothers and fathers changed by similar amounts. The couples who did not have children showed more gradual deterioration without the sudden change.

Put together, the honest summary is this: relationships decline whether or not a child arrives, by amounts that are not wildly different, but a birth concentrates the decline into a step rather than a slope. A step is much easier to notice, and much easier to attribute — which is probably why parenthood gets more of the blame than the data support.

The number the decline literature does not predict

Everything so far describes averages going down. So here is the result that the average conceals, and it is the most encouraging finding in this article precisely because nobody was looking for encouragement when they ran it.

Daniel O'Leary, Bianca Acevedo, Arthur Aron, Leonie Huddy and Debra Mashek took a random sample of 274 married Americans and asked a blunt question. Among those married more than ten years, 40% reported being “very intensely in love.” They then cross-validated in a second random sample — 322 people in New York married more than ten years — where the figure was 29%, and the predicted correlates held in both.

Those correlates are the useful part. Long-term intense love went with thinking positively about the partner, thinking about the partner when apart, affectionate behaviour and sexual intercourse, shared novel and challenging activities, and general life happiness.

The shared-novelty item is the one that connects to a model rather than sitting as a loose association. The self-expansion account — reviewed by Chelsea Branand, Debra Mashek and Aron — holds that relationships supply a sense of growth by letting another person's perspectives and resources be experienced as partly one's own, and that this requires new material to work with. That literature, including the seven-minute experiments and the one randomised field trial that tested exciting against merely pleasant activities, is covered in what actually builds attraction over time, along with the preregistered studies where the effect did not replicate cleanly. We are not going to restate it here; the point for this page is that the correlate turns up in a representative sample of people who are still in love after a decade. The same mechanism read from the far side — what happens when a separation stops supplying novelty and the couple finally lives in the same place — is in what distance actually does to attraction.

Two gendered details from the same study, and the second needs naming rather than reporting.

For women but not men, passion about things outside the relationship correlated with intense love — which is what the self-expansion account predicts and the opposite of the folk theory that devotion means contraction.

And for men but not women, wanting to know where the partner is at all times correlated significantly with intense love. That is a correlate in a dataset, and it is also, described plainly, a monitoring behaviour. The next section is about why that distinction is the most important one on this page.

When it is not the honeymoon phase ending

This section exists because the frame in the title has a failure mode, and the failure mode has a cost.

“The honeymoon phase is over” is a ready-made, reassuring, developmentally-normal explanation for a partner's behaviour changing after the relationship becomes secure. It is sometimes the right explanation. It is also the most available wrong one, and it is the explanation that keeps people explaining things away.

The distinction is not about intensity or timing. It is about what changed. Affection becoming less demonstrative, conversations becoming more practical, sex becoming less frequent, the other person being less fascinated by every detail of your day — those are the things the research above describes, and they happen on both sides at once.

What the research above does not describe is one person's behaviour acquiring new control over the other. Deciding who you may see. Needing to know where you are, as a requirement rather than as interest, and not reciprocally. Going through your phone. Money becoming conditional. Silence used as a sanction. Pressure continuing after you have said no. Anger that arrives when you do something independently. Those are not a phase ending, and they do not resolve by waiting or by trying harder to be interesting.

The trajectory studies make the point in their own language, which is why it belongs in an evidence article rather than a disclaimer. When Lorber and colleagues asked which couples showed the steep decline, the predictor that came back most consistently was intimate partner violence — present and measurable before the slope appeared. When Lavner and Bradbury asked what separated their five groups, aggression was on the list. When Lavner, Karney and Bradbury asked which wives were most optimistic about the future, those wives had higher levels of physical aggression at baseline. Three samples, and in each one the cliff and the aggression travel together.

So if what is happening feels less like a slope and more like a cliff, the literature's own answer to “what predicts this” is not novelty ran out.

If any of that is recognisable, the useful next step is a person rather than an article. Find A Helpline, in the sources below, lists free and confidential services by country. In the United States the National Domestic Violence Hotline is 1-800-799-7233, or text START to 88788, 24 hours a day. Where the feeling and the behaviour have to be told apart, the psychology of jealousy works through the distinction at length, and what builds attraction over time has a section on why a shrinking world is not always a shared choice.

What this evidence cannot tell you

The limits go in the article rather than in a footnote, because on this question they are a large part of the honest answer.

The phase itself has not been measured. Nearly all the trajectory data starts at a wedding. There is no published study that follows new couples from the first weeks, with a defined construct, measuring often enough to say when the early state ended for each person. Every duration in circulation — six months, two years, three years — is an estimate, a study's follow-up interval mistaken for a finding, or a figure from a trade book.

The neuroscience is 34 people from one research programme. Two fMRI studies of seventeen participants each, from the same lab, neither designed to track change over time. Seventeen people is a normal size for imaging work; it is not a basis for the confident causal mechanism search engines are assembling from it. And the long-term study found the reward response present, not absent.

The chemical follow-up selected on its outcome. The 39 people re-tested in the NGF study were those who reported no longer being in the early state. The design can show a marker tracking a state. It cannot show the marker ending it.

Averages hide the people who do not fit. This is the recurring lesson. Lucas and colleagues found strong reactors still far from baseline years later and some trajectories pointing the opposite way. Lorber's modelling found the rapid decline in roughly one couple in eight. Lavner and Bradbury found a group with no decline at all. A page that reports only the mean would be accurate and would describe almost nobody.

The samples are narrow. Overwhelmingly American and Western European, mostly mixed-sex, mostly married, heavily self-report. The desire data are German students in one cross-section. The same caution applies to anything you read about how much of a relationship pattern is fixed, which is why whether attachment styles change ends in a coefficient rather than a promise.

And the trajectory of a particular relationship is not predictable. The largest project ever run on relationship quality — 11,196 couples, 29 laboratories — could not account for more than about 5% of the variance in whether satisfaction rose or fell, from any combination of self-report variables. That result is set out in what actually builds attraction over time. It means nothing on this page is a forecast about you. It is a description of what moves the average, and the whole argument above is that averages are not people.

More of this under Mindset. Claims this site has had to retract are listed on the corrections page.