Hadley Wickham
Rice University

Graphical inference

Visualisation is a critical part of any exploratory analysis, but how can we be sure that we are not seeing patterns in randomness? In this talk, I'll introduce the protocols of graphical inference that provide formal statistical inference for visual findings. Graphical inference uses much of the same infrastructure as classical hypothesis testing but replaces the test statistic with a human observer and the null distribution with null plots, visualisations of samples from the null hypothesis. With some care, the line up protocol allows us produce valid p-values for graphical findings, and provide an important tool for the exploratory analyst.