Code and data for replicating the analysis reported in **"Structured speaker variability in Japanese stops: relationships within versus across cues to stop voicing"**, currently under review. **Structure** `aux` contains auxiliary files from the CSJ which include speaker metadata and lexical (frequency) information. `code` contains the Praat, Python and R scripts necessary to collect the data, format it, fit the statistical models, and generate predictions from the models. The data generation scripts assume access to the *Corpus of Spontaneous Japanese* (CSJ, Maekawa et al. 2000). `data` contains pre-generated data used in the original analysis. This includes the data file originally created from the CSJ, as well as the data files following exclusions and acoustic measurement of VDC. `models` contains the multivariate mixed-model reported in the analysis, as well as the same model fit with a weaker correlation prior. `predictions` contains pre-generated prediction files of both the model intercepts and voicing effects, and can be used to re-generate the paper plots and reported correlations.