Gravar-mail: Influence of proxy respondents in children's health interview surveys