4.3.1 Null hypothesis testing

Null hypothesis testing is a formal statistical approach to evaluating evidence for two different interpretations of a relationship between two variables, the null hypothesis and alternative hypothesis.

The null hypothesis (H₀) is generally that there is no difference between two groups. For example, the prevalence of resistant isolates in the group which received antimicrobial treatments is the same as the prevalence in the group which did not. The alternative hypothesis (H_a) is that there is a difference, that is, in this example that the prevalence of resistant isolates differs significantly between these two groups.

To reject or retain the null hypothesis, we need to assume a statistical model which represents our scientific hypothesis and then collect the data. Then, we calculate the probability of observing data as extreme as the data in hand, should the null hypothesis be true; this is the p-value. Typically, if the p-value is below a defined cut-off, then the null hypothesis is rejected. By convention, the cut-off is often set at 0.05, that is there is a less than 5% chances of observing data as extreme as the one we have by chance, however, this is an arbitrary value.