🧩 What This Introduces:
A Bayesian model for detecting changepoints in time series of overdispersed counts (for example, candidate contributions over a campaign or counts of terrorist violence). The model is built to identify salient structural breaks and to support formal inference on how many breaks appear in a given series.
🛠️ How Changepoints Are Identified:
The approach avoids specifying the number of changepoints ex ante by incorporating a hierarchical Dirichlet process prior. This lets the model:
- Estimate the number of changepoints from the data
- Estimate the location of each changepoint
- Accommodate overdispersed count observations without forcing simplistic variance assumptions
📊 Real-World Demonstrations:
Applications illustrate the model’s practical value:
- Campaign contributions during the 2012 U.S. Republican presidential primary
- Incidences of global terrorism from 1970 to 2015
These cases show how the method discovers structural breaks and supports inference about both the presence and number of changepoints.
🔎 Why It Matters:
Provides researchers and analysts a flexible, data-driven tool for detecting and quantifying regime shifts in count-based time series where variance exceeds simple Poisson assumptions. Especially relevant for empirical work on campaign dynamics, political violence, and other phenomena expressed as overdispersed counts.
