When it comes to surveys, significance and bias are separate issues

Today UNISON released a survey of nurses and midwives, which found that 65% reported not having an adequate amount of time with patients.

A UNISON representative, Christina McAnea, went on the Today programme to discuss the survey. Presenter Sarah Montague questioned how much it could really tell us about nurses in the NHS:

Montague: I was curious about this in terms of how much can we trust these figures. Because it's a survey of 3,000 nurses, but was it those who chose to respond, i.e. those who are perhaps most angry and unhappy?

McAnea: Well there will be an element of that.

Montague: Because it's self-selecting.

McAnea: It's self-selecting yes. But it's still statistically significant when you have a number like 3,000, but yes we totally accept that.

Statistical significance tests look at the likelihood of study results occurring due to chance.  If you choose a sufficiently large number of nurses at random to respond to your questions, this should reflect the views of all nurses fairly well.

But as Sarah Montague notes, UNISON's findings are based on people who've put themselves forward for the survey, so the results may not represent the view of nurses overall.

As the British Polling Council says in its guide to polls:

"A biased sample is a biased sample, however large it is."