Predicting nonresponse bias from teacher ratings of mental health problems in primary school children.

The impact of nonresponse on estimates of mental health problems was examined in a prospective teacher screen in a community survey of 9,155 7-9 year olds. For 6,611 of the children, parents consented to participation in the actual study (Responders), while for 2,544 children parental consent was not obtained (Nonresponders). The teacher screen involved assessment of a broad set of symptoms of mental health problems and functional impairment. Calculations of non-response coefficients, a function of effect sizes and non-response proportion, revealed only ignorable nonresponse bias for both mean scores and correlations. However, the results from binary logistic regressions revealed that children ascribed signs of mental health problems by their teachers were less likely to participate. This was most frequent among children with only moderate symptoms. However, it also involved children with high symptom scores related to inattention, hyperactivity, emotions and peer relationship problems. These findings suggest that measures based on effect size can underestimate the magnitude of non-response bias and that a logistic regression approach may be more appropriate for studies geared at estimating prevalence of mental health problems in children.