Divorce contagion
A Socratic walk-through of divorce contagion — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why does a close friend's separation raise the chance that a stable-looking marriage ends too?
Someone in the circle announces a separation, and within a year or two a second marriage ends — one nobody had worried about. The usual reading is that divorce spreads like a cold: an idea catches, and people who would otherwise have stayed are infected by it. But an idea is not a virus, and nobody wants a divorce because a friend had one. So what would actually have to pass between two households for the first ending to cause the second — and are we even sure anything passed?
Reasoning it through
REASONING #Start with the finding, since it is why the question exists. Working from the Framingham Heart Study network, researchers reported that people whose friend had divorced were about 75 per cent more likely to divorce themselves, roughly 33 per cent at two degrees of separation, and no more likely at three. Strong, weaker, gone: that is the shape transmission makes, and it was reported as transmission.
Before accepting it, ask what else makes that shape. Friends are not a random draw. People befriend people who married at a similar age, with similar schooling, similar religious commitment, similar money trouble; two couples alike in the things that predict divorce will divorce at similar times without either watching the other. Friends also share circumstances — a plant closes, a unit deploys, a neighbourhood's prices collapse. Neither story needs influence, and both decay with network distance, because similarity and shared circumstance decay with distance too.
This is not a quibble. Applying the same machinery to the same kind of data, critics recovered apparent "contagion" of acne, of headaches, and of height; and a formal result showed that homophily and influence are generically confounded in observational network data — from the network alone you cannot pull them apart. So the honest starting position is that the 75 per cent is a correlation of unknown and disputed causal share.
Now suppose part of it is real. What crosses? Not the wish to divorce. Ask instead what actually decides whether a marriage ends, and it helps to treat it as three quantities rather than one: how good the marriage is, how good the best alternative looks, and the barriers to leaving — money, children, housing, religion, shame, the fear that the far side is unliveable. A marriage ends not when quality falls but when the alternative plus the shrinking barriers clears the quality. It is a threshold, not a slide.
Watch what a friend's divorce does to those three. It does nothing to the first. It does a great deal to the other two. You now have a worked example in front of you: she kept the house, the children are all right, she is not destitute, and the circle did not drop her. The uncertainty you were charging against the outside option shrinks, and the social penalty turns out smaller than you had priced it. Two terms move, and the one that would have to move for the marriage to look troubled does not.
Which is exactly why the marriage that ends is the one nobody worried about. Outsiders can see conflict; they cannot see barriers. A marriage of low quality and high barriers is quiet — the arguing stopped years ago — and that is precisely the configuration a fallen barrier is decisive for. Marriages of real quality do not move when a barrier drops. Marriages with low barriers have already ended.
Can we separate this from plain norm-copying? One line of evidence is suggestive: Swedish workplace register studies found that separation risk rises with the share of recently separated colleagues of the opposite sex. If an attitude were spreading, the sex of the divorced peers should be irrelevant; if the visible supply of alternatives had changed, it should not be. That points at the alternatives channel rather than the normative one — one study family, not a verdict.
The analogy
THE ANALOGY #Think of an old dam nobody inspects. The water behind it has sat at the same level for decades and from downstream everything looks fine. Then a similar dam one valley over gives way, and what changes is not the water but the engineers' estimate of the wall. Inspections begin, the safety margin is rewritten, and a structure that had held for forty years is condemned. Nothing came over from the other valley except information about what a wall like this can bear.
a dam has one true strength that inspection merely reveals, whereas a marriage's barriers are partly made of what people believe about them — so a friend's divorce does not only show that the wall is thin, it genuinely thins it, by lowering the social cost of being next.
Clarifying the model
THE MODEL #Three refinements hold this together.
First, "contagion" is the wrong metaphor for a threshold effect, and the difference is practical. A virus adds risk to everyone it touches. A fallen barrier only matters to couples already sitting just above the line, so the same event does nothing at all in most marriages and ends a few — which is why the population-level effect stays modest even if the individual mechanism is strong.
Second, some clustering is bookkeeping. A divorce reorganises the network around it — couple-friendships fracture, each ex keeps some friends and loses others — so the friendship measured afterwards is not the one that existed before.
Third, the causal question is not settled and the field knows it. The strongest defence of influence is the decay with distance, and that effects were largest between mutual friends; the strongest objection is that mutual friends are also the most similar people in the network, so the same pattern falls out of selection. Both readings survive the data we have.
A picture of it
THE PICTURE #How to readRead across for how much holds a couple in place and up for how good the marriage is. The bottom-right quadrant is the one to watch: poor marriages held together by barriers alone, which look calm from outside because the arguing is long over. The two labelled points on that row are the same couple before and after a friend's separation — the marriage has not changed at all, only its position on the horizontal axis, and that horizontal move is the whole effect. The couple in the top half barely notices the same event, and the couple bottom-left was leaving anyway.
What became clearer
WHAT CLEARED #Divorce does not spread; barriers fall. Much of the measured clustering may be similarity and shared circumstance rather than influence at all, and the standard methods cannot tell the difference. Whatever influence remains works not by making a marriage worse but by making the exit cheaper and the far side legible — which is why it strikes marriages that looked fine, since to an outside eye a high-barrier marriage and a happy one are indistinguishable.
Where to go next
ONWARD #- Why divorce also runs in families across generations, and whether that is the same mechanism or a different one.
- How barriers themselves have fallen over a century — law, women's earnings, housing — and what that did to the baseline everyone's threshold sits against.
Key terms
TERMS #| Term | What it means |
|---|---|
| Homophily | the tendency to form ties with people already similar to you, which produces clustering with no influence at all. |
| Barriers to dissolution | the costs of leaving a relationship, distinct from how good the relationship is; the term that peer divorce moves. |
| Confounded identification | the situation where two explanations imply the same observable pattern, so no amount of that data decides between them. |
Every term the collection defines is gathered in the glossary.