Do low survey response rates bias results? Evidence from Japan