A little off-topic for the forum, but wanted the cognitive science folks here to see this pre-print:
There was panic in the spring after a researcher published in PNAS a re-analysis of a bunch of his data and found that the test-retest reliability was really really low. Another paper came out in the fall showing a similar result. I have big issues with even the concept of reliability (which is an odd confounding of measurement noise and inter-individual variability), but those concerns aside, the above pre-print shows how the low-reliability result is an artifact of failing to account for the hierarchical structure of the data.