This idea is the cornerstone to
Statistics.
When people measure things they are often trying to prove a theory,
e.g. 9/10 Cats prefer tinned food
To do the test, you need a theory to compare it against. This theory you compare it against should be the simplest theory you can come up with that fits the known facts. This theory is known as the Normal Hypothesis. (See Ockham's Razor)
e.g. Cats in the wild eat fish or mice, so they prefer to eat that
You then construct the experiment to try prove that they like tinned food more than fish or mice. You grab a bunch of cats and you give them some food and try to persuade them to eat it. (Good luck!)
Even if the evidence suggests that they like tinned food more- that's not really good enough if the result is within the normal variation of the experiment, or the normal variation in the tastes of cats; (and let's face it cats aren't noted for consistency ;-) you really need the result to be a 1000:1 chance thing, or some other pre-agreed standard (the estimate of the chances that we got a particular result based on luck with the normal hypothesis is called the p value).