null hypothesis
/ nul hy-POTH-uh-sis /
Think of a courtroom where the accused is presumed innocent until proven guilty. The trial does not start from guilt; it starts from the cautious assumption of innocence, and the evidence must work hard to overturn it. The null hypothesis plays exactly the role of that presumed innocence in a statistical test.
The null hypothesis is the default, deliberately boring claim that nothing interesting is going on: that two means are really equal, that a new method has no bias, that two methods are equally precise. A test then gathers the data and asks whether the evidence against this dull claim is strong enough to reject it. You either reject the null hypothesis or fail to reject it; you never prove it true.
It matters because framing the question this way forces you to demand real evidence before announcing a difference, guarding against seeing patterns in pure noise. The subtle caveat is that failing to reject the null hypothesis is not the same as confirming it — like a not-guilty verdict, it means the evidence was insufficient, not that innocence was proven.
To test a new analyzer against a certified standard, the null hypothesis is no bias: the analyzer's mean equals the certified value. A t-test then judges whether the measured difference is large enough to reject that assumption.
The cautious starting assumption a test tries to overturn.
Rejecting a true null hypothesis (a false alarm) is a type I error, set by the significance level; failing to reject a false one (a missed difference) is a type II error.