Shades of Neutrality: Detecting Media Slant in a Political Monopoly Using Quotation Accuracy

A natural starting point in the study of political media bias has been to examine differences in coverage intensity. This study develops a notion of quotation accuracy as a dimension of media slant. Using Natural Language Processing (NLP), I operationalize three measures of quotation accuracy—substring alignment, bag-of-words overlap, and a two-stage transformer-based semantic measure. I combine these measures with applied econometrics to analyze coverage of parliamentary speeches in Singapore’s flagship daily newspaper, a setting where coverage intensity shows no discernible slant. Across all three measures, opposition speeches are quoted less accurately than those of the ruling party. NLP-based features and bounding analyses rule out competing explanations, such as topical composition, ignorable tokens, or linguistic coherence. Finally, I contextualize the findings in a government-media machinery where ruling-party politicians pre-circulate transcripts to the media, but the opposition does not, which could explain accuracy gaps. I theorize that this difference in media engagement embeds private information about slant.