Deriving a zero-truncated modelling methodology to analyse capture–recapture data from self-reported social networks

Capture–recapture (CRC) is widely used to estimate the size (N) of hidden human populations (e.g., the homeless) from the overlap of sample units between two or more repeated samples or lists (a.k.a., capture occasions). There is growing interest in deriving CRC data from social-network data. The current paper hence explored if self-reported social networks (lists of social ties) submitted by participants from the target population could function as distinct capture occasions. We particularly considered the application of zero-truncated count distribution modelling to this type of data. A case study and follow-up simulation study focused on two methodological issues: (1) that a participant cannot be named in their own self-reported social network and hence cannot be named as many times as non-participants; and (2) positive dependence between being a participant and being named by (a social tie of) other participants. Regarding the latter, a further motivation of the simulation study was to consider the impact of using respondent-driven sampling to select participants, because all non-seed RDS participants are recruited as a social tie of another participant. Exponential random graph modelling was used to generate the simulation study’s target populations. Early comparison was also made to estimates of N from Successive Sampling.