The effect of non-response bias on the results of two-stage screening surveys of psychiatric disorder

Summary The effects are examined of two varieties of non-response bias (that due to illness and that due to defensiveness, as defined in this paper) on two-stage screening surveys for psychiatric disorder. Equations to model these effects are first derived; these are then used, in conjunction with pre-existing data, to estimate the sixe of the non-response bias. It is concluded (a) that bias due to illness results in prevalence estimates being some 5% lower than the true prevalence; (b) that specificity is little affected by either variety of bias; but that (c) bias due to defensiveness results in the sensitivity being overestimated by about 6%.