Bias resulting from missing information: some epidemiological findings.

Abstract

sumUmARY The biases resulting from missing information were examined in three psychiatric epidemiological studies. In each study, cases with missing information could be compared with the main sample because data were available from several sources or at several points in time through a longitudinal study. In almost all instances, cases with missing data differed systematically in terms of variables crucial to the questions being studied. In general, they tended to include a higher proportion with problems of various kinds-such as, behavioural deviance, reading backwardness, child or adult psychiatric disorder, and marital discord. The characteristics or circumstances of those giving information were generally more strongly associated with co-operation in testing or interviewing than the characteristics of those about whom information was sought. In some situations, the nature and degree of distortion resulting from missing information could lead to biased results.

The Nature of Covariation Between Autistic Traits and Clumsiness: A Twin Study in a General Population Sample

“…Although epidemiological samples have the advantage of limiting referral biases, our response rate was less than optimal, partially reflecting cultural attitudes towards issues of mental health in children. Some authors have suggested that a low response rate (Cox et al 1977) may reflect exclusion of the most severe cases; thus, the results of analyses addressing the psychometric and causal architecture of mental disorders in nonclinical samples do not necessarily apply to clinical populations. Moreover, since this is a general population sample, it could be argued that the information yielded by this design does not necessarily apply to the more severe cases of children with clumsiness, or autistic traits.…”

Aetiology and Risk Indicators of Hearing Impairments in a One-Year Birth Cohort for 1985-86 in Northern Finland

“…As Cox et al (1977) point out, the missing cases should also be assessed with regard to the variables central to the item of interest and not only in terms of demographic background. In this series, the missing subjects did not differ from the attendants in terms of such determinants as malformations or Apgar scores at 5 or 15 min, but there was a slight overrepresentation of children reported as having had episodes of otitis media among the attendants.…”

Noncompliance and dropouts as a problem of longitudinal studies in child psychology

“…Cox et al (1977) found that at a rate of noncompliance of 8.4%, children of nonrespondent parents show more psychiatric disorders, but do not differ from the participants as regards IQ. A possible explanation for this discrepancy are the different rates of noncompliance and the fact our data suggests that the noncompliants have a tendency to dissimulate (a higher share of pupils repeating a class in the noncompliant group and a tendency to show more symptoms).…”