Estimating and controlling for the effects of volunteer bias with pairs of relatives.

If pairs of relatives correlate in their liability to participate in a research project, it is possible to test for the effects of volunteering on the criterion variable of interest. Much of the information for this test comes from a difference in criterion variable mean between individuals with and those without a cooperative relative. Also, if data are available from more than one class of relative, it may be possible to discriminate between (i) volunteering that occurs as a consequence of the criterion variable and (ii) volunteering as a cause of the criterion. Likelihood formulae are presented that permit quantification and significance testing of volunteer bias. If data are collected from a genetically informative design such as a twin study, it is possible to estimate genetic and environmental parameters independent of the contaminating effects of such bias. We describe some methods of reducing the computational burden of multidimensional integration to allow extension to multivariate data. Implications for research design and management are discussed.