This question already has an answer here:
I'm currently reading An Introduction to Statistical Learning with Applications in R by Hastie & Tibshirani. In their discussion on the hierarchical clustering algorithm, they note that the notion of 'dissimilarity' between pairs of observations can be extended to pairs of groups of observations using the concept of 'linkage', then briefly describe four common types of linkage (complete, single, average, centroid).
Which linkage should I use when? I'd like to see examples to build up some intuition. ISLR doesn't provide them; they merely throw out some assertions (e.g. "average and complete linkage tend to yield more balanced dendrograms than single linkage"; "centroid linkage can lead to inversions"), which is understandable given the level of audience they're pitching at.
If it helps, my background is in physics, I have no formal statistical training.