In this study, we collected mixed method social network data alongside a survey of people who inject drugs (PWID) to assess key assumptions of respondent-driven sampling (RDS). We found adherence to some assumptions and non-adherence to others. Specifically, sampling did occur through a connected network of PWID and the reported degree size of survey participants did correlate with network degree and likelihood of recruitment. However, recruitment was significantly more likely via a multiplex relationship and recruitmen dyads were significantly more likely to live in the same geographical area. In addition, people actively injecting drugs, those with 'two lives' and 'dealers', may have been isolated from the survey. We believe this is the first mixed-method social network study to report an assessment of the assumptions of RDS. We highlight the potential value of network data in understanding the validity of population-level estimates in surveys using RDS.