Statistically significant

Yesterday on the radio I heard the phrase 'statistically significant' twice (one on a discussion about crime and other about cancer). We all think we know what it is: If all the birds overhead poo on you simultaneously, you know, it cannot be a random event. It is surely a statistically signifcant event full of practical consequences ordered by an angry diety to dirty your atheistic bald head (alright, my atheistic bald head). Nevertheless, if you aren't a scientist or a statistician, a short readup may help, because, it ain't so simple. The wikipedia page is not laymanish (if you are an expert, why not create a page that's simpler). I liked this answer, which is a good starting point.

More like this

Nice links.

The most common misunderstanding, I think, is to believe that a "statistically significant" difference means "big difference." Size of difference matters to statistically significance, but so does both the actual variation involved in the measures and the degree of confidence in the estimate of the means and variance. So a very small difference could be statistically significant if the range of variation is narrow and the means are close.

In the examnple cited in your second link, the difference between 18 and 20 individuals is unlikely to be be statistically significant given the provided information, but that does not mean that the difference is not real. But a second experiment with 10,000 rats in each treatment (control and provided with the mediation) with the same result would likely be statistically significant. But not otherwise particularly significant.

Unless you are one of the two percent of rats that were saved by the drug!

@ Greg Laden

But a second experiment with 10,000 rats in each treatment (control and provided with the mediation) with the same result would likely be statistically significant.

Just to clarify, this is true as long as the "same result" means 18% versus 20% survival (not 18 versus 20 individual survivors). Obviously, that's what you meant, but I thought I'd make it explicit.

Selva, perhaps reading statistical significance along with p-value will give a better understanding as well as address the issues raised by the above two commenters

Thanks folks. I am muddling through. Interestingly enough, it seems impossible to be able to intuitively appreciate 'statistical significance', I'll lay the blame on our evolutionary predisposition for pattern spotting even when there is none.