Applications

Question 1

How would your models work on health or election data?

Answer 1

• Health: tracking disease counts across regions and time. The GDM-HMM allows for detecting shifts and uncertainty in disease dominance.
• Elections: modelling vote share compositions across areas or over time. Spatial or temporal models could uncover political clusters or sudden shifts in voter behaviour

Question 2

What would be the challenges of applying your framework to financial data?

Answer 2

• often have heavy tails and non-count-based proportions (e.g. weights)
• temporal dynamics may involve high-frequency volatility, requiring model adaptation
• real-time constraints in financial decisions may require faster inference

However, the frameworks could be adapted to fit this application.

Question 3

Can your approach help in ecological monitoring?

Answer 3

• animal species counts over space and time
• biodiversity tracking, especially with rare or zero-inflated species
• help identify hotspots or gradual ecological transitions

Question 4

How do your models support decision-making under uncertainty?

Answer 4

• provide posterior distributions, not just point estimates, enabling risk-aware decisions
• offer probabilistic forecasts, helping policymakers or forensic analysts understand confidence and uncertainty in their inferences
• use calibration tools like ECE and Brier Score to ensure that probabilities are reliable and interpretable

Question 5

How would a practitioner without a Bayesian background apply your model?

Answer 5

• framework is accessible as it was implemented in an user-friendly platform like NIMBLE
• using weakly informative priors it allows practitioner with limited information to ensure the posterior is driven by the data rather than prior knowledge
• packaged model versions with example templates or wrappers could further lower the barrier to use.