Blas Kolic

Comparing Data Assimilation and Likelihood-Based Inference on Latent State Estimation in Agent-Based Models In this paper, we present the first systematic comparison of Data Assimilation (DA) and Likelihood-Based Inference (LBI) in the context of Agent-Based Models (ABMs). These models generate observable t...

While limited to one ABM, this work fills a critical gap in the ABM calibration literature, providing the first structured comparison of DA and LBI for latent state inference.

Kudos to Marco, Corrado, and Gianmarco for such a wonderful collaboration!

Hope you enjoy it
👉 arxiv.org/abs/2509.17625

⚖️ Essentially:
➖ DA: Great for macro-level patterns. Easy to apply, doesn’t need a formal likelihood.
➖LBI: Superior for micro-level accuracy, but needs explicit likelihoods (often hard to derive).

➡️ Trade-off between generality and precision.

📊 Main results:
➖ At the agent level, LBI outperforms DA in reconstructing latent opinions. LBI is more accurate and robust to model errors.
➖ At the aggregate level, both methods perform similarly well → DA remains competitive for forecasting population-level trends.

We test this using the Bounded-Confidence Model of opinion dynamics, where agents interact only if their opinions are sufficiently close, resulting in nonlinear updates.

⚙️ Scenarios:
➖ Observed: agent interactions
➖ Latent: agent opinions
➖ Noisy opinions
➖ Mis-specified model parameters

Can we recover the latent agent states (e.g., opinions) from observed data in an ABM?

🆎 First systematic comparison between:
➖ Data Assimilation (DA) → Approximate, model-agnostic
➖ Likelihood-Based Inference (LBI) → Precise, but model-specific

🚨 Fresh from ArXiv:
“Comparing Data Assimilation and Likelihood-Based Inference on Latent State Estimation in Agent-Based Models”
with @marcopangallo.bsky.social, @c0rrad0.bsky.social, & @gdfm.bsky.social
👉 arxiv.org/abs/2509.17625

Amazing collab with Fabián Aguirre-Lopez and the data science crew at Sinnia, Mexico.

📄 Journal: doi.org/10.1093/comn...

📝 ArXiv (OA): arxiv.org/abs/2206.14501

💻 Code & plots: github.com/blas-ko/Twit...

Last month, we published
"From chambers to echo chambers: quantifying polarization with a second-neighbor approach applied to Twitter’s climate discussion" 🌍🔥

We find stable climate echo chambers despite ~90% weekly user churn, and show how events like #FridaysForFuture can disrupt polarization.

very nasty, indeed

🌱