Reliance on small samples, the wavy recency effect, and similarity-based learning.
Distribution of the number of citations over years.