BLIN and penalized logit results for environmental treaty ratification.

A Lasso regularization parameter which maximized AUCPR in 10-fold cross-validation was used for each model. The use of the BLIN model over penalized logit improves the predictive performance of the model: AUCPR increases by 0.6 when using BLIN over penalized logit. The substantive implications of these models differ in interpreting control variables. Hard law, legislative approval, log SO2, log GDP per capita, global mixed goods, and log economic openness are all regularized to zero when including network effects. Significance was determined using the selectiveInference package, which returns a p-value for each coefficient. We indicate significance of a given coefficient at the 0.05 level.