We developed a method to map and investigate temporal changes in the b-value using the signed Akaike information criterion (AIC) value and seismicity rate before the 2003 September 26-27 Northern Miyagi earthquake sequence. We studied the seismicity within a radius of 30 km around the epicenter of the largest shock, which was M 6.4, by three approaches: an analysis of all the events listed in the catalog; an analysis of only background events, but on data extracted by two different methods. We found that the change in the b-value of the clustered activities and background activities are different. The b-value of all events decreased from 1.2 to 0.7 and that of background event decreased even more. We empirically modeled the precursory temporal variation of the b-value and searched similar patterns among 971 areas chosen sequentially from inland Japan. The precursory pattern was both rare and significant; the AIC value was the second most significant factor among the more than 3000 tests conducted on 971 areas, each with four time windows. If the catalog was not declustered, the AIC of the precursory pattern became lower.