Covariation bias in depression - a predictor of treatment response?