Natural Sentences as Valid Units for Coded Political Texts

摘要: All methods for analyzing text require the identification of a fundamental unit analysis. In expert-coded content analysis schemes such as Comparative Manifesto Project, this is ‘quasi-sentence’: natural sentence or part judged by coder to have an independent component meaning. Because they are subjective constructs identified individual coders, however, quasi-sentences make fundamentally unreliable. The justification supposed gain in coding validity. We show that unfounded: using does not produce valuable additional information characterizing substantive political content. Using sentences units, contrast, delivers perfectly reliable unitization with no measurable loss validity resulting estimates.