It is an open question whether glitch activity in individual pulsars varies on decadal time-scales. The Crab pulsar has experienced 23 spin-up glitches in the last 36 years, interrupting an otherwise monotonic deceleration. A homogeneous Poisson process, i.e. a process with constant rate, is not sufficient to describe the time-ordered distribution of glitch epochs in the Crab pulsar. There are signs of clustering at the $2\sigma$ level when testing with Ripley's $K$ function. Two alternative, inhomogeneous models with one and two step-wise rate changes are found to have higher relative evidence (Bayes factors of 1.74 and 2.86 respectively) than the homogeneous Poisson process. The distinction between clustering, where events are correlated, and rate variation is discussed. The implications for glitch microphysics, in particular trigger mechanisms based on avalanche processes, are briefly discussed.