Type II error is the complement of type I error (q.v.) in the NHST (Null Hypothesis Significance Testing) paradigm, see null hypothesis and inferential testing for more on that generally. It describes the error of deciding on the basis of a result telling you you can’t reject the null hypothesis that your finding was “non-significant” when in fact there is a non-null effect in the population.
Probability being what it is: probability not certainty, there will always be errors when we use statistical methods. When they give binary answers, as the NHST paradigm does, it is easy to categorise them as type I and type II.
For any one NHST there is only ever type I error risk: the value you chose when you planned the NHST (we hope), usually .05. By contrast there are an infinite number of type II error risks depending on how large the effect is in the population. That takes us to statistical power and to estimation and confidence intervals as alternatives to the NHST paradigm.
Try also #
Confidence intervals (CIs)
Type I error
Touched on in Chapter 5 but not expanded there to avoid making the book too “techy”.
Online resources #
First created 20.viii.23.