Data were collected via postal questionnaire. The first wave of data was weighted (at the family level) to match the distribution of educational qualifications observed in a nationally representative sample (Meltzer et al. 2000; for full details of the first wave weighting see Rowe et al. 2006). The second wave was weighted (at the family level) according to predictors of attrition, using the inverse of the predicted probability of families remaining in the study at wave 2. Predictors were parental education, housing tenure, and child sex (girls being most likely to respond). This response weight was multiplied by the wave 1 sampling weight to provide a single weighting variable. In the present study substantively identical results were obtained when analyses were run with and without this weight.
