This document discusses the relationship between diversity in classification ensembles and single-class performance measures for class imbalance problems. It first analyzes when and why ensemble diversity, as measured by Q-statistic, can improve overall accuracy based on classification patterns. It then extends this analysis to consider diversity's impact on single-class performance measures like recall, precision and F-measure. Theoretical analysis identifies six situations of how diversity may affect these measures. Extensive experiments on artificial and real-world datasets with skewed class distributions find strong correlations between diversity and the discussed performance measures. Diversity generally has a positive impact on the minority class and is beneficial to overall performance in terms of AUC and G-mean.