Regression-to-mean awareness in JOLs

Discount your highest-confidence JOLs most — they are the most likely to be inflated.

Why it works

JOL accuracy is not uniform across confidence levels — high JOLs are more likely to be overestimates than moderate JOLs, because the same fluency signals that produce high confidence (easy processing, recent exposure) are the ones most likely to inflate actual retention estimates. Knowing that the highest-confidence items are the most suspect allows calibrated discounting that counteracts the systematic bias.

How to do it

1. After any study session, look at the items you rated with 90%+ confidence.
2. Treat these as the highest-priority items to retest on your next session — not to skip.
3. If they pass delayed retrieval, then confirm the high JOL; if not, recalibrate.
4. Over sessions, track whether your 90%+ items are actually recalling at 90%+ — if not, lower your threshold.

Evidence

Overconfidence bias is more pronounced at the high end of confidence distributions. Research on the hard-easy effect in calibration shows overconfidence is concentrated on items rated as easy, paralleling the inflated-JOL phenomenon. (observational)

This practice is derived from calibration research rather than JOL-specific protocols; the specific instruction to focus on high-JOL items for early retesting is a practitioner application of the underlying finding.

Common mistake

Prioritizing restudy on low-JOL items (which feel hard) and skipping high-JOL items (which feel done), when the high-JOL items are precisely the ones most likely to produce a rude surprise at test time.