There is a joke by the redoubtable Variety Show duo, Flanders and Swann that goes something like this:
"I hear that travelling by air is much less dangerous nowadays than crossing the road"
"Yes, but watch out for the airport buses. The drivers have orders to keep up those odds."
Performativity in philosophy of science concerns how a theory about practice can itself influence practice. For example theories of economics that posit selfish maximising agents may produce an economy made up of businesspeople trying to do just that.
Take cigarettes. As everyone knows, these are bad for you. Epidemiologists don't do causation, but they do say that smoking is a factor very highly correlated with the incidence of various life-threatening diseases. Governments around the world are therefore introducing smoking bans in public spaces to reduce smoking levels and change habits. But these will not necessarily reduce death rates among smokers. In fact they could raise them through both statistical and causal effects
1. By increasing the hassle and discomfort of smoking, these bans may be successful in persuading less addicted smokers to quit or reduce their smoking as the nuisance level surpasses their marginal pleasure from a cigarette. Therefore there should be less tobacco smoked and less deaths in total. Nevertheless lots of smokers have a differently shaped demand curve for tobacco and will continue to smoke a lot. But these are the very people - the heaviest smokers - for whom smoking is a very significant health risk. Taking away the light or 'social' smokers therefore leaves behind a concentrated population with a higher risk profile than before. Thus, death-rates among smokers will increase.
2. But the kind of people who smoke are different in other ways too. Epidemiologists identify a number of other risk factors for the chronic illnesses associated with smoking, including lifestyle (e.g. diet, exercise habits) predispositions (e.g. genes, childhood and prenatal nutrition), and social (e.g. income, marital status, class). In theory these are independent factors, but in practise of course they are often closely correlated with each other and with smoking itself. Much of epidemiology is concerned with statistically stripping out these interaction effects from the data in order to focus on the statistical relationships between single variables. Nevertheless government policies do not focus on single variables and surely influence the interaction of these risk factors. In particular, smoking bans have had some success in increasing the social stigma associated with smoking. It is therefore increasingly rare to find middle-class young people smoking. But smoking has been more resilient among working class people. These are the people who tend to have multiple other risk factors (such as diet, exercise and stress) that not only directly reduce their life-expectancy, but also increase the harm that smoking itself is likely to cause them (in the UK this combination of factors produces something like a 10 year life-expectancy gap between the poorest and richest boroughs). Again, by taking higher status, generally healthier smokers out of the risk pool one may be increasing the average risk profile of those remaining even though nothing about these people or the harm that smoking does them will have changed.
Smoking bans may also cause some increase in the risk of an early death by increasing the harm that smoking does people, though obviously it is difficult to assess how much.
3. Bans may increase risk by requiring smokers to expose themselves to more non-tobacco related dangers in order to smoke. i.e. they must leave warm secure indoor areas and go outside. That can be bad for your health if the environment is hostile, such as a Canadian winter. It may put your security at risk. For example in the UK pub smoking bans were associated with an increase in street aggression as smoking drinkers congregated on narrow pavements. It can be bad for people who are already in poor health to have to travel outside. Outside hospitals one can sometimes see very elderly or infirm smokers with walking frames who have to hobble in and out for their smoking fix. Obviously all those causal mechanisms are contingent ones that depend on the interaction of the ban with personal and environmental features.
4. There can also be more subtle indirect causal mechanisms at work. One striking one concerns whether the increase of social stigma associated with successful government policies may cause smoking to become a more dangerous 'life-style'. Firstly, as noted above, lower social status is itself a well-known risk-factor for the diseases of interest. Secondly, if people who smoke are considered disgusting drug addicts they may actually internalise and act on that view. Cigarettes are presently understood as a 'gateway drug' to harder and more dangerous substance abuse, but this was not the case 40 years ago when smoking was still cool.
In conclusion, smoking bans based on epidemiological evidence that smokers die earlier may have something of the character of a self-fulfilling prophecy. For purely statistical reasons relating to the population of smokers, the bans will themselves produce ever stronger correlations between smoking and early death. They may also directly harm smokers either through inducing riskier behaviour or through making them into social outcasts, with further possible (though disputable) effects on the risk profile associated with smoking.
None of these points concern the truth or accuracy of the original statistical analysis, and none would be news to epidemiologists. But the public is constantly presented with policy debates on areas as diverse as the benefits of university education or breast cancer screening, based on statistical evidence that is true as far as it goes. Our public discourse would benefit from a more sophisticated grasp of the underlying basis and limitations of such statistics about correlations when considering policies about causes.