The scaling of event sizes is one of the critical parameters controlling seismic hazard and risk. For induced seismicity, the non-stationary, heterogeneous character of subsurface stress perturbations can be a source of spatiotemporal variations in the scaling of event sizes. We analyse the spatiotemporal variations in the earthquake-size distribution of the Groningen induced seismicity catalogue using a method that circumvents arbitrary choices requiring a priori knowledge of these variations while systematically exploring the effect of possible bias in the derivation. Our results show that the b-value is spatially variable with indications of a temporal decrease. This spatiotemporal pattern may be explained by a stress-dependent b-value. At the same time, we postulate that the event-size distribution is tapered at a single, regional corner magnitude. Our results imply that the current risk assessment models overestimate the probability of larger magnitude events (M≥3.5) in the Groningen gas field and thus the risk posed.