Abstract We evaluate four hypotheses regarding b‐values in large earthquake sequences using the Japanese earthquake catalog: H1 ${mathcal{H} }{1}$: Around large earthquakes, b‐values vary more than expected at random. H2 ${mathcal{H} }{2}$: b‐values decrease prior to large earthquakes. H3 ${mathcal{H} }{3}$: After large earthquakes, b‐values increase. H4 ${mathcal{H} }{4}$: Large earthquakes localize in regions of lower b‐values. We show that H1 ${mathcal{H} }{1}$ is supported for spatial and, to a lesser extent, temporal variation. We find no evidence for a pre‐event drop H2, $left({mathcal{H} }{2}right),$ nor for post‐event increase H3. $left({mathcal{H} }{3}right).$ Large earthquakes occur preferentially in low b‐value regions H4, $left({mathcal{H} }{4}right),$ although the effect is small compared to the b‐value range across sequences. Together, the evidence suggests b‐values are mainly location‐controlled, which we verify with an independent data set. Our results support the view that b‐values depend primarily on rock properties and stress regime. This study targets large‐scale effects; highly localized shifts in b‐values remain to be investigated.