Validating models with real data

“Modeling event data allows for observations into the data generating process.” — David Masad

David Masad, a Ph.D student at George Mason University, led a Friday Seminar on Oct. 10, 2014, in the Computational Social Science department of the Krasnow Institute, presented, ‘Comparative Modeling of Conflict Decision Making.’

Masad’s research employs data from the Correlates of War for some of his research. Correlates of War is a project started in the ‘60s that focuses on the history of war, with a great set of resources.

To validate the data his models were producing he used a dataset called Behavioral Correlates of War (BCOW).

According to Masad, the BCOW data consists of hand-coded event data that includes date, actor, action (verbal or physical), categorization, and target.

Masad has built out two models that produce data. The models are a case study in the Honduras-Nicaragua conflict in 1957. The theories tested are — Bueno de Mesquita Expected Utility Model and Tit for Tat.

Afterwards, he compared the data produced by the models against the actual data to test different theories, which includes the evaluation of outcome validity vs process validity.

Masad’s research is still in progress, but it is a clear example of the testing process for agent based modeling.