Hyperparameter Sensitivity in Political Text Scaling Algorithms

An analysis of 17 computational text-based ideal point estimation (CT-IPE) algorithms reveals that most methods function as highly configurable pipelines rather than fixed estimators. Through 5,537 experimental runs, the study found that for 13 of the 17 algorithms, hyperparameter profiles explain little residual variance, showing low ICC values below 0.10.