Answers are grounded in this paper and cite their sources.
Earthquake occurrence is a complex stochastic process characterized by temporal clustering, where periods of heightened seismic activity are frequently followed by aftershock sequences, leading to significant overdispersion in count data. This overdispersion — where the variance of earthquake counts substantially exceeds the mean — poses a challenge for standard Poisson-based modeling approaches, which assume equidispersion. Within this context, this study investigates the performance of Poisson Hidden Markov Models (PHMM) and Negative Binomial Hidden Markov Models (NBHMM) in capturing the latent regime-switching dynamics of earthquake frequency data in Türkiye, a region of high seismic activity situated at the intersection of major tectonic plates. The results show that the two-state PHMM provides the best overall representation of the annual earthquake counts. Both AIC and BIC select this model; AIC provides stronger support, whereas BIC indicates that the one-state Negative Binomial model remains a competitive alternative. The stronger overall performance of the PHMM supports persistent changes between two statistical activity levels. Nevertheless, the competitive Negative Binomial fit indicates that within-state overdispersion also explains part of the observed variability. Thus, the Negative Binomial model provides a useful alternative interpretation, but it does not outperform the selected two-state PHMM.
No takes yet. Share an insight, caveat, or question.
Hakan Karaca (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: