A Bayesian Model for Estimating Population Means Using a Link-Tracing Sampling Design
Article first published online: 13 JUN 2011
© 2011, The International Biometric Society
Volume 68, Issue 1, pages 165–173, March 2012
How to Cite
St. Clair, K. and O'Connell, D. (2012), A Bayesian Model for Estimating Population Means Using a Link-Tracing Sampling Design. Biometrics, 68: 165–173. doi: 10.1111/j.1541-0420.2011.01631.x
- Issue published online: 23 MAR 2012
- Article first published online: 13 JUN 2011
- Received September 2010. Revised April 2011. Accepted April 2011.
- Bayesian modeling;
- Exchangeable distributions;
- Link-tracing designs;
- Population networks;
- Snowball sampling
Summary Link-tracing sampling designs can be used to study human populations that contain “hidden” groups who tend to be linked together by a common social trait. These links can be used to increase the sampling intensity of a hidden domain by tracing links from individuals selected in an initial wave of sampling to additional domain members. Chow and Thompson (2003, Survey Methodology 29, 197–205) derived a Bayesian model to estimate the size or proportion of individuals in the hidden population for certain link-tracing designs. We propose an addition to their model that will allow for the modeling of a quantitative response. We assess properties of our model using a constructed population and a real population of at-risk individuals, both of which contain two domains of hidden and nonhidden individuals. Our results show that our model can produce good point and interval estimates of the population mean and domain means when our population assumptions are satisfied.