CPT response time variation gives an uncertain attention forecast gain in 23 Penn LEAD test participants with fixed ridge models and known visit intervals

We tested whether baseline task performance improved a forecast of later self-reported attention ability. We used Penn LEAD, OpenNeuro ds007116, version 1.0.6. The task measure was the coefficient of variation of true-positive response times on the Penn Continuous Performance Test (CPT). We selected 57 participants with complete reports at two visits and eligible baseline task data. A fixed split assigned 34 participants to training and 23 to evaluation. The basic ridge model used the baseline report, age, sex, study group, and visit interval. The added model also used the task measure. Both models used a fixed penalty of 10. The forecast assumed that study group and visit interval were known. Mean squared error (MSE) was 1.30583 for the basic model and 1.29967 for the added model. The gain was 0.00617 mean item points squared. Its approximate 95% bootstrap interval was -0.18818 to 0.20510. All seven planned sensitivity intervals also included zero. These data leave the forecast gain uncertain. They do not show that the task measure has no useful predictive value. The package preserves all source files used, the in-run plan, code, predictions, bootstrap draws, and a figure. A new source download and two numerical replays passed the saved checks.

Penn LEAD · attention · response time variability · continuous performance test · forecasting · ridge regression · longitudinal study · bootstrap interval