The null hypothesis is the statement or question upon which a hypothesis is based. e.g. “All apples are red”, “Eating pigs causes chubby faces”, “The average salary of astronauts is $120,000 per annum”.
Without evidence, the hypothesis is “null”: by default, it is neither true nor false. To prove our hypothesis we must test it.
Some hypotheses are true until proven false, and others are false until proven true. For example, we can say that all apples are red until we prove a non-red apple is found. On the other hand, eating pigs doesn’t cause chubby faces until we prove a correlation, and the average salary of astronauts is not $120,000 per annum until we find data to support the fact.
It’s impossible to prove a true hypothesis (e.g. “all apples are red”), because we can’t guarantee that an alternative hypothesis won’t come along to disprove it in the future. All it takes is one non-red apple to disprove our hypothesis, and we can’t possibly test every single apple in the world, past, present and future.
Hence, a true hypothesis makes for very good null hypothesis. A researcher will seek to disprove the null hypothesis rather than to prove it (i.e. “Innocent until proven guilty”). If they are able to provide evidence to prove that a green apple exists, they can claim a result: not all apples are red! Similarly, a medical scientist will seek to prove that eating pigs does not cause chubby faces, and an analyst will seek to prove that the average salary of astronauts is not $120,000.
But what happens if the researcher is unable to disprove the null hypothesis? They cannot conclusively say that “all apples are red” just because they couldn’t find a green apple. This is why experiments seek to prove statistical significance. They will attempt to prove that all the apples they found were red, and they couldn’t prove otherwise; that the majority of people who ate pigs have chubby faces; and that from the (large) set of data they collected about astronauts, the average salary was indeed $120,000.
Their conclusions can never be truly conclusive: the door is always open for another researcher to come along and disprove their theory. As such, a good piece of research will seek to both prove statistical significance and disprove the null hypothesis. If the researcher demonstrates that they took all reasonable means to disprove their own hypothesis and failed, it strengthens their argument.
What this means is that a huge amount of trust is placed on scientists to disprove their own theories. It’s easy to make a poor effort and then claim they couldn’t disprove themselves, and therefore their hypothesis is correct. Hence why reproducible research is so important.