Clarifying selection bias in cluster randomized trials