The consequences of selective participation on behavioral-genetic findings: evidence from simulated and real data.

Nonresponse occurs when individuals either have no chance of being included in a study (noncoverage), refuse to take part (unit nonresponse), or fail to give complete information (item nonresponse). The purpose of this article is to test the possible biasing effects of nonresponse on the results of behavioral-genetic studies. Simulations and a real data 'natural' experiment were used to determine the impact of nonresponse on estimates of additive genetic and environmental effects. The simulations used realistic twin-pair correlations and models of nonresponse derived from prior research. The real data 'natural experiment' used data from a nationally representative birth-cohort twin study (E-Risk Study) and compared model results from families who had responded to a mail survey to those from all study cases. Results showed that the primary influence of nonresponse was to attenuate the effect of the shared environment and to inflate estimates of nonshared environment and additive genetic effects. At high levels of nonresponse a spurious nonadditive genetic effect (suggesting genetic dominance) was also found. Study nonresponse was shown to have the potential to bias the findings of behavioral-genetic research. Design and analysis methods that can be used to alleviate this potentially important biasing effect in behavioral-genetic studies are discussed in light of these findings.