Age-related errors in the assessment of children
Abstract
Children's functioning can only reasonably be measured relative to that of other individuals of the same age. In practice, age ranges are usually used to group children for this purpose. Examples include school grades, age groups in sports, and age bands used in developmental assessments. Age grouping is associated with systematic errors, often known as relative age effects (RAEs): Within each age group, older children outperform younger ones. This type of assessment error may lead to opportunities and interventions being offered inefficiently or unfairly.
This thesis comprises 5 research projects that aim to clarify underlying causes of RAEs, examine their importance in different contexts, and develop analytic methods relevant to their study. I use data drawn from two studies: A prospective cohort study including athletic performance measures (the Physical Health Activity Study Team project) and a validation study undertaken to compare measures of child development (Psychometric Assessment of the Nipissing District Developmental Screener). I develop linear models to characterize age-related variation and then use results to draw conclusions, to inform other analyses, and to generate synthetic datasets.
Together, studies demonstrate a set of methods for the exploration and correction of RAEs. They also yield several concrete findings: (1) A simple mathematical interpretation of RAEs can fully explain the errors seen in real datasets, meaning that other explanations are, in at least some contexts, unnecessary. (2) RAEs have different effects in ranking and selection contexts, with ranking errors largest among average individuals but selection errors greatest when more extreme thresholds are used. (3) Age bands cause misclassification in measures of child development, and the error rate rises rapidly with the width of age bands used. (4) The use of different sets of age bands will prevent different assessments from agreeing closely. (5) Age grouping in developmental assessments will create an illusion of longitudinal instability. Finally, I demonstrate the use of alternative scoring approaches and discuss how these can reduce or eliminate errors related to RAEs.