Gravar-mail: Accounting for dropout reason in longitudinal studies with nonignorable dropout