Selective attrition and bias in a longitudinal health survey among survivors of a disaster