Response bias in community surveys of mental health: systematic bias or random noise?