Habits appear to spread between people, and proving it is very hard

The observation and the excitement it produced

Analyses of large social network data have reported that various behaviours and states cluster among connected people, and that a person is more likely to show a characteristic when their contacts do. Some of this work reported that the association extended beyond immediate friends to friends of friends, which was presented as evidence of behaviour propagating through a network like something transmitted.

The findings received enormous attention, and the metaphor of contagion is powerful enough that it has passed into ordinary conversation. It also raised an obvious practical hope: if behaviour spreads, then influencing a few well-connected people would be an unusually efficient way to change a population.

The confound that won’t go away

Three explanations produce the same clustering and they’re extremely difficult to separate with observational data. Influence, where one person’s behaviour changes another’s. Homophily, where people form connections with others who already resemble them. And shared environment, where connected people are exposed to the same conditions, prices, weather, workplaces and neighbourhoods.

Statistical controls help with the parts you can measure and do nothing about the parts you cannot, and the traits driving both friendship formation and behaviour are largely unmeasured. Methodologists have argued forcefully that the standard analyses can’t distinguish these mechanisms, that some of the specific results are sensitive to modelling choices, and that similar apparent contagion can be produced for characteristics nobody believes are transmissible. That critique has not been fully answered.

What would settle it, and why it is rarely available

The clean answer requires assignment. Where an intervention is randomly given to some people and the behaviour of their untreated contacts is measured, any change in the contacts is attributable to the treated person rather than to shared traits. Studies of this kind exist and are far more persuasive than network correlations, and they are expensive, difficult and limited to behaviours somebody could ethically assign.

Where such designs have been run, the general picture is that social influence on behaviour is real and considerably more modest than the contagion metaphor suggests. That is the pattern one usually sees when a striking observational finding is finally tested experimentally, and it is a reason to expect it in advance rather than a surprise.

The mechanisms that do not need a network

Some of what looks like transmission is better described in duller terms. If the people around you do something, the thing is more available, more convenient and more normal, and each of those changes your behaviour without anybody influencing anybody. A household where a particular food is bought is a household where you eat it, and no attitude was transmitted.

Perceived norms are a separate channel and they are frequently inaccurate. People estimate what others around them do from a biased sample — the visible, the vocal, the memorable — and adjust towards a norm that may not exist. That means an intervention correcting a mistaken belief about what everybody else does can change behaviour even where no real influence was flowing, which is a different mechanism with different limits.

What to take from an unresolved literature

The defensible summary is that who you spend time with affects what you do, that this is obvious and does not require network science to establish, and that the specific quantitative claims about how far behaviour spreads through a network are contested and rest on designs that cannot cleanly support them.

For a personal decision, the modest version is sufficient and does not depend on the disputed part. The people you see regularly determine what is available, what is convenient, what is normal and what gets suggested on a given evening. Those are environmental facts about your circumstances, which is the category of thing that reliably shapes behaviour, and they operate whether or not anything is being transmitted in the interesting sense.

A note on how this claim is used

Contagion language carries an implication worth noticing: that behaviour arrives from outside and that responsibility sits with whoever transmitted it. That framing is occasionally apt and frequently convenient, and it should be held to the same evidential standard as any other causal claim.

It also invites a policy conclusion — target the influential — that the underlying evidence does not currently support at the strength usually implied. Where interventions of that kind have been tested directly, results have been mixed. That is an argument for testing them rather than for assuming either that they work or that they cannot.