Dyadic Relationships

Skills & Tools

• R

• Python

• Automatic Event Classification

• Dynamic Time Warping

• Pearsons and Spearman Correlation

Differences in Assessing Dyadic Relationships with Automatic and Manual Event Classification

Dyadic relationships are important to understanding international relations. Tracking how countries respond to each other and exogenous shocks is critical to understanding what drives their interactions. To make this analysis possible there needs to be a consistent and reliable source for measuring these relationships. Event databases seek to fill this role by classifying events and scoring them based on the conflictual or cooperative nature of each event. Historically these events were human coded but more recently they are automated by performing text analysis on news stories.

How accurately each of these methods classify events is an ongoing topic of research. As is defining the scales on which the intensity of the events is measured. This study though, does not seek to compare these databases to ground truth. Instead it examines the agreement between human coded and machine coded datasets in measuring the dyadic relationships of 6 countries from 1955 to 1978. This agreement is measured using the Pearson and Spearman correlations and the dynamic time warping distance to capture similarity between signals despite possible misalignment. Conducting this analysis at varying time scales offers insight to the effect of granularity.

Ideally, this research can be used to make more informed decisions about event database usage and understand the conditions for agreement between human and machine coded datasets.