ABSTRACT Radio pulsar glitches probe far-from-equilibrium processes involving stress accumulation and relaxation in neutron star interiors. Previous studies of glitch rates have focused on individual pulsars with as many recorded glitches as possible. In this work, we analyse glitch rates using all available data including objects that have glitched never or once. We assume the glitch rate follows a homogeneous Poisson process, and therefore exclude pulsars that exhibit quasiperiodic glitching behaviour. Calculating relevant Bayes factors shows that a model in which the glitch rate λ scales as a power of the characteristic age τ is preferred over models that depend arbitrarily on powers of the spin frequency ν and/or its time derivative $\dot{\nu }$. For λ = A(τ/τref)−γ, where τref = 1 yr is a reference time, the posterior distributions are unimodal with $A=0.0066_{-0.002}^{+0.003}\ \rm {yr}^{-1}$ and $\gamma =0.27_{-0.03}^{+0.03}$. Importantly, the data exclude with 99 per cent confidence the possibility γ = 1 canvassed in the literature. When objects with zero-recorded glitches are included, the age-based rate law is still preferred and the posteriors change to give $A=0.0099_{-0.003}^{+0.004}\ \rm {yr}^{-1}$ and $\gamma =0.31_{-0.03}^{+0.03}$. The updated estimates still support increased glitch activity for younger pulsars, while demonstrating that the large number of objects with zero glitches contain important statistical information about the rate, provided that they are part of the same population as opposed to a disjoint population which never glitches for some unknown physical reason.