Evaluation of Attribution Bias in Retrieval-Augmented Large Language Models