Analyzing data from the Puerto Rican English in Philadelphia (PREP) corpus, we investigate participation in TH-stopping, a socially stigmatized yet stable variable documented in Philadelphia. While previous studies have been impressionistic and have considered voiced and voiceless tokens to pattern together, this work validates novel, acoustically based stopping indices: mean harmonics-to-noise ratio for voiced tokens and skewness for voiceless tokens. We apply these indices to the corpus data and analyze stopping under a Bayesian framework, and we compare results from a model built from impressionistic coding of a subset of the same data. We find convergent evidence that TH-stopping is a stable variable in the Puerto Rican English data as well. Findings are compared with those of existing studies, noting future directions for research on the variable and underscoring the importance of establishing demographically representative baselines for linguistic research in diverse urban centers.
