Using opinion dynamics models with real-world data
Background
Opinion dynamics models have a huge potential for studying and solving some of the most pressing world problems. For example, we could use these models to understand which countries are more at risk of developing a strong anti-vaccine community.
Unfortunately, these models are often not tested against real-world data. Because of that, it is necessary to study how can we actually use these models with real data
Main findings
While modellers usually suppose uniform opinion distribution, real-world data may have quite complex structures [1]. Furthermore, while in physics we can use units of measurements, nothing similar exists for opinions. This means that opinion data will be more complex to use.
Up to now, models have been analysed almost in isolation. One of the main findings shows that this should be avoided. Indeed, these models make sense only when a precise way to measure opinions is specified [2].
To stress this point we showed that:
- Two people using two measurements of opinions may predict opposite outcomes. This even if we used the same model and dataset.
- Different models are actually the same under a measurement transformation. Meaning that model 1 paired with measurement 1 is the same as model 2 paired with measurement 2.
Output
- Research articles and preprints
- Preprint – A new degree of freedom for opinion dynamics models: the arbitrariness of scales
- Preprint – The sensitivity of the Deffuant model to measurement error
- Article – Deriving an Opinion Dynamics Model from Experimental Data
- Article – The psychometric house-of-mirrors: the effect of measurement distortions on agent-based models’ predictions
- Article – Propagation of measurement error in opinion dynamics models: The case of the Deffuant model
- [incoming] Article – How polarization extends to new topics: an agent-based model derived from experimental data
- Other output