MEASURING RESPONSE TIME WITH AN ERROR OF 1% OR LESS AND RELIABILITY OF 0.96:
PREPARING FOR THE 2002 BLUEBERRY HEALTH STUDY
R. Martin*, R. Coppings, K.E. Gerstmann, W. Holmes, A. Kokesh, D. Mathew, C. Pappas, A. Pruchnicki, B. Sachs, R. Schnoll, A. Wetherell
MMT Corp., Sherman, CT 06784, Lane College, Jackson, TN 38301, NY, NY 10014, Bethel, CT 06801, Charleston, WV 25312, New Fairfield, CT 06812, New Fairfield, CT 06812, Mount Sinai Medical Center, NY, NY 10029, HR Herbs, Sherman, CT 06784, Brooklyn College, Brooklyn, NY 11210, and Defence Science & Technology Laboratory, UK.
High measurement precision is critical for studies of aging and decline-reversal since change over one year may be 1% or less. Test-retest reliabilities for 77 performance tests listed in recent reviews averaged only 0.63, indicating that performance test reliability is often too low for measurement of age-related cognitive decline unless several hundred individuals are tested. Last year, we described methods for four-choice response time measurement that produced an overall test-retest reliability of 0.991 during a pilot study to determine whether consumption of flash-frozen blueberries could improve cognitive performance and overall health in people with multiple sclerosis. This high reliability was based on the average response time from a three-week measurement period and the corresponding average from the subsequent three-week period, for a total of six weeks of measurement. In this report, we describe methods that produced a within-session test-retest reliability of 0.96 for participants age 60 and over during weekly baseline monitoring prior to a larger, follow-up blueberry study. Participants in the 2002 study completed all measurements over the Internet, received no training other than web site instructions, and measured 2-choice decision speed by pressing either the 1 key when number 1 appeared in the number box, or the 2 key when the number 2 appeared. During each measurement session, 125 correct responses were obtained during 5 separate series of 25 responses. Participants were asked to proceed quickly enough to commit between 3 and 6 errors during each series of 25. Immediately before and after each response, the time-measurement accuracy of each computer was measured so that data obtained during periods of transient interference could be discarded. Within-session reliability was based on a comparison between the average of responses 26-75 and 76-125. Relative percent standard deviation between average response times for series 2 to 5 was 0.83% and between series 3 to 5 was 0.38% for 132 measurement sessions during the second and third weeks of baseline measurement, suggesting that measurement error may be 1% or less. We are grateful to I. Sullivan and J. Tasman for helping us to select a two-choice response time measurement as the most appropriate for this study.
Key words:
Response, reaction, time, measurement, reliability
Problems or questions regarding this site should be directed to
webmaster@americanaging.org