The effect of survey attrition in longitudinal surveys: evidence from Peru, Côte d'Ivoire and Vietnam