Statistical information is not real information. A statistical description – this includes models – is, like a painting, only a representation of reality, although, as with a portrait, some have more ‘likeness’ than others.
Statistics apply to populations, not individuals.
Post-selection randomisation is not the same as selecting a random sample from the population you wish to study.
Increasing the size of a biassed sample reduces sampling error, but not bias. So the ratio of bias to sampling error increases.
If P is set at 5% then expect 1 in 20 of your tests to give ‘significant’ results even when H0 is true
An insignificant result is God’s way of telling you your sample is too small
