A form at the clinic asks for a number between one and seven: how satisfied are you with your life these days? The same week, a friend rings and asks how things are going. A person can put down a four and tell the friend they’re fine, mostly, better than last month, and be answering honestly both times.
Where the treadmill idea came from
The belief that we return to a fixed baseline no matter what happens has a name and a founding anecdote. Philip Brickman and Donald Campbell called it the hedonic treadmill in 1971. Seven years later, Brickman with Dan Coates and Ronnie Janoff-Bulman compared the average well-being of recent lottery winners and of people paralysed in accidents against a control group, found the differences smaller than they expected, and concluded that both groups had adapted completely.
That study became the thing everybody knows, and the reading it acquired is not the reading its numbers support. Maike Luhmann and her co-authors point out that the standardised difference between the paralysed group and the controls was later put at 0.75 of a standard deviation, which is substantial, and that the 1978 paper is nonetheless routinely cited as evidence that life events leave no lasting mark. Earlier studies often reported significance tests without standardised effect sizes, so the size of what they had found was easy to lose. That account of the 1978 study comes from Luhmann and colleagues rather than from the original.
The correction arrived slowly, through panel data. Over the 2000s, Richard Lucas and colleagues went at the same question using the German Socio-Economic Panel and the British Household Panel Study, and reported that the effects of major events could persist for years, at rates that differed sharply from one event to the next. Lucas is one of the four authors of the meta-analysis that followed.
Bereavement, a birth, a job that ended
That 2012 meta-analysis pooled 188 publications, 313 samples and 65,911 people, all of it longitudinal, and split subjective well-being into its two standard halves. Affective well-being is the friend’s question: how much pleasant and unpleasant feeling is in the days. Cognitive well-being is the form’s question, a considered judgement of a life overall or of one part of it, such as a job or a marriage.
Both were followed through eight family and work transitions: marriage, divorce, bereavement, the birth of a child, unemployment, reemployment, retirement, and relocation or migration. (One sentence of the paper’s general discussion says ten. Its abstract enumerates four family and four work events, its summary table lists eight rows, and its constraints section says eight outright, which is what all eight event-by-event analyses cover. This is the accepted author manuscript, which the publisher notes has not been through final copy-editing.)
Separating the two questions changed the picture event by event. After bereavement, the considered judgement fell by almost half a standard deviation at the time of the loss, which the authors read as a worse initial hit than the one to day-to-day feeling, and both rose from there. After the birth of a child, life satisfaction rose at first and then slid; the pooled estimate for relationship satisfaction showed almost no immediate change and kept declining, ending up, in the authors’ words, permanently below its pre-birth level, which is a curve through sample averages rather than a forecast for any couple. Day-to-day affect after a birth came out slightly positive overall, though the paper describes that trajectory two different ways, calling the effects on affect small but clearly positive in one place and the initial reaction negative-then-rising in another.
Unemployment split the two on timing rather than on impact. The initial fall was not measurably different between the two: the difference did not reach significance, on an interval wide enough to leave room for day-to-day feeling being hit rather less hard. The considered judgement then climbed back, but from so low a floor that the authors put the point at which the pooled average returned to its pre-unemployment level at roughly three years, an average across samples rather than a schedule anyone is on, while day-to-day feeling did not change significantly over time at all.
The individual study estimates for day-to-day feeling disagreed sharply with each other, far more than the estimates for the considered judgement did. In the bereavement studies, the most negative estimate for day-to-day affect landed half a month after the death, and one of the most positive arrived five months later. The authors are careful about why: it may be that samples differ in personality, coping or support, or simply that they used different instruments. Sixty percent of the measures of day-to-day feeling were depression scales, which they say are presumably sharper at the low end of well-being than the high.
Welcome and unwelcome did not sort the events
The result they cannot organise is a negative one. Sorting the events by how welcome they are does not explain the pattern, and the authors put it as a hedge: these effects, they write, are “not a function of the alleged desirability of events.”
The initial reaction to divorce, in these studies, was relatively mild, and well-being rose afterwards, though that estimate comes from the stripped-down model described below rather than from the same specification as the others. The initial reaction to retirement was more negative than that. So was the initial reaction to getting a job again after unemployment. Across these eight events, filing them as good or bad did not predict the shape of the curve, and the rate at which well-being declined after the supposedly good ones was not systematically steeper than the rate at which it recovered after the bad ones.
Their reading of the divorce finding runs two ways. One is the plain one: “the legal act of divorce itself (though not necessarily the whole process) may actually be beneficial for peoples’ SWB, at least for those who perceive it as a relief from a bad marriage,” they write. The other is that the months before a divorce are not a neutral baseline, and neither are the months before a wedding. For divorce specifically they could not test the second, because the baseline estimate they needed was not available for the event.
They flag the wider limit themselves: the desirability finding rests on a very small sample of life events, and they say it needs replicating on other positive and negative ones. Their own checks for publication bias found effect size varying with sample size for prospective marriage studies, prospective childbirth studies, post-hoc unemployment studies, prospective studies of other occupational transitions and post-hoc relocation and migration studies, where they warn the estimates may be skewed. And the prediction they went in with did not generally hold. They expected feeling to adapt faster than judgement, and found what they called partial confirmation for marriage, bereavement, reemployment and retirement — on the ground that those events hit day-to-day feeling less hard in the first place, rather than that feeling recovered faster — with unemployment and child birth running the other way.
They could not pin down the ordinary baseline
The divorce analysis rests on eight prospective samples and a stripped-down model. The usual one would not converge, so the authors dropped its affect terms; the four post-hoc samples they found were too thin to model at all. Three quarters of the divorce studies used data that had been collected for something else entirely.
That thinness matters more than it looks, because the question underneath all of this is a question about baselines. Well-being appears to move in anticipation of an event, which means a study’s first measurement is often taken from an already-displaced position, and the paper’s estimate of where people ordinarily sit differed from the pre-event scores for almost every event.
So the paper can say how fast the considered judgement climbed back toward where it sat shortly before the event, and does. What it cannot say is when anyone got back to their ordinary baseline, because it could not pin down where that baseline sat. The question people actually want answered, the one about how long, is the one this evidence declines to settle.
Almost all of the underlying studies used self-report. The effect sizes were not corrected for measurement error, which the authors say may make their numbers somewhat understate the real ones, though they argue this leaves a clear picture of what adaptation research has actually observed. Where studies had not reported the correlation needed to weight them, the median value across the other studies was substituted. And the median study here was published in 2002; the unemployment work was older still.
This is a description of averages across populations, and it is not a diagnosis or a treatment for anyone in particular. Persistent low mood after a loss, or a job ending that has not lifted, is something a GP or a bereavement service can work on. The findings that touch individual difference are all of the shakiest kind — samples with more men in them adapted more slowly after bereavement, which is a fact about samples and not a fact about any man.
The phrase this evidence complicates is “you’ll get over it.” It bundles two promises, one about the texture of a person’s days and one about the verdict they would give on their life, and across most of the events here those two things came loose from each other and travelled at their own speeds.
