Autism is a highly heritable disorder, which is strongly influenced by genes. But it is not a genetic condition in the same way as Huntington's disease or cystic fibrosis, in which mutated DNA is the one and only cause. It is influenced by environmental factors, but what they are we do not yet know.

This uncertainty about environmental triggers has itself become a trigger for wild speculation. All sorts of hypothetical causes have been advanced and found wanting, from cold-hearted “refrigerator mothers” to vaccines.

This week a new contender has emerged. In a paper published in Archives of Paediatrics & Adolescent Medicine, Michael Waldman, of Cornell University, has suggested a link to rainfall. He noted that autism prevalence seemed to be lower in the south and west of the United States than in northern states, and wondered whether climate might be responsible. When he studied counties in Oregon, California and Washington, a trend emerged. Autism was more likely to be diagnosed in children living in wetter counties.

This supposed link has been seized on by those who claim, against the evidence, that mercury poisoning causes autism. Rain, they suggest, might be the conduit by which mercury pollution from industry is reaching the ground environment.

That, to be fair, is not Waldman's interpretation. He thinks it more likely that wet weather means children spend more time indoors, which may have knock-on effects. They may watch more TV. They get less sunlight and thus produce less vitamin D. They may be more exposed to household chemicals.

These suggestions are just about plausible - though it beats me how TV might affect a disorder for which warning signs often appear in the first year of life. But before British parents start to worry about our famously soggy climate, it is worth remembering that these are speculative ideas, with little or no evidence to recommend them. The link they purport to explain is in itself rather questionable.

An analysis of the paper by the Respectful Insolence blog (www.scienceblogs.com/ insolence) makes this plain. First, Waldman did not actually find a link between raw precipitation data and autism - rainfall figures had to be adjusted for a trend to emerge. This doesn't necessarily invalidate the claim, but it advises caution. Try enough adjustments, and associations will sometimes emerge through chance alone.

Rainfall, too, is variable from year to year, but the trend of rising autism diagnoses goes in one direction. Prevalence is also similar in many countries with different climates. These inconvenient truths compromise the credibility of the link, yet they pass unexplained. There is every chance that this link is a statistical artifact - and one found in a single region. There is no reason to think it can be generalised beyond the US West Coast.

These criticisms highlight a simple reason why one must always approach statistical studies with great care: correlation is not the same thing as causation. Causation can sometimes be inferred from association, but exacting conditions must be met.

An effect must be seen in multiple independent samples, and there should ideally be a relationship between dose and response. It should also be explicable by biological means, preferably by a mechanism backed by animal or human laboratory research.

In this case, none of these criteria has been fulfilled, casting doubt on whether the paper deserved to be published and reported. As Respectful Insolence says, studies like this are “custom-made for cranks”, as they are impossible to interpret and easily misrepresented to support maverick claims. Rarely do they turn out to be preliminary expositions of important links that have not yet been appreciated. Far more often they are diversions that alarm the public needlessly.

Mark Henderson is science editor of The Times

Explore Health

• Expert Advice

• Health Features

• Mental Health

• Alternative Medicine

• Child Health

• Health Club