The seventh lecture in the Teaching Statistics Trust Lecture series is being hosted by the Royal Statistical Society South West and the Teaching Statistics Trust at the University of Plymouth.
This year's lecture will be given by Dr Peter Martin from University College London.
Abstract
Statistical hypothesis tests and p-values are poorly understood by many students, researchers, and even teachers of statistics. Wrong application and misinterpretation of statistical tests are common in several scientific fields, and over-reliance on the ‘p < 0.05’ threshold to claim ‘statistical significance’ often masks limitations of the data or weaknesses in the statistical methods. This has contributed to a crisis of replication: many published research findings, although claimed to be supported by strong statistical evidence, turn out to be false.
One root of these problems is the way that statistical hypothesis testing is frequently taught. Many excellent, thoughtful explanations and intuitive visualisations have been published, but in themselves these do not show how to use statistical tests in practice. It is one thing to understand the technical meaning of a p-value, and another to plan an investigation involving a test that can yield a valid measure of the strength of the statistical evidence for a scientific theory. Statistics teaching needs to emphasise how statistical inference is conducted in real research, demonstrate good practice, and show what can go wrong when p-values are misunderstood, misinterpreted, or intentionally gamed (p-hacking). The interpretation of a specific statistical result must take account of the context within which it was obtained. If this context is lost, misinterpretations are inevitable.
This lecture proposes an approach to teaching statistical tests that combines general principles of good statistics education – teaching statistics as a process of investigation in the context of real-world problems – with principles of good statistical inference: ‘Be thoughtful, open, and modest (Wasserstein et al 2019). The lecture will give specific narrative examples of how misconceptions about statistical tests can be challenged in a statistics class, and how teachers can exemplify good practice. This way, students and teachers can make tracks towards the ‘world beyond p < 0.05’.