The Timing Accuracy of General Purpose Computers for Experimentation and Measurements in Psychology and the Life Sciences

Abstract

General purpose computers find increased use within behavioural, psychological, and neuroscientific experimentation, which raises concern for the timing accuracy that can be obtained with such systems. Here, we assessed the timing accuracy of such machines, considering both differences between different hardware and different versions of the Windows™ operating system (OS); Windows XP, Vista and Windows 7. The variability varied widely across machines and OS versions. The indeterministic variability within each OS and computer combination was mostly within +/-30 ms, and had a non-normal distribution with many small deviations and few large deviations. These large deviations are a characteristic feature that seems to constitute occasional additional delays up to about 150 ms. Thus, although measurements recorded from a general purpose PC running Windows should have an accuracy of -30 to +50 ms, occasionally larger variations suggest that experiments need a large test base to avoid significant distortions of the results.

Effects of tempo, swing density, and listener's drumming experience, on swing detection thresholds for drum rhythms

“…To avoid problematic timing inaccuracies during stimulus presentation (Madison and Wallace, 2012), drum patterns were not generated in real time. Instead, prior to each trial, audio data were rendered for all possible versions of the given drum pattern (i.e., for all onset displacements from 0 to nearly m, in 0.5 ms increments), using a 44.1 kHz sample rate.…”

Correction and Validation of Time-Critical Behavioral Measurements over the Internet in the Stage Twin Cohort with More Than 7000 Participants

“…The ISIP and ISMS data were thus described by the mean (M) and standard deviation (SD) and the reaction time data by the median (Md) and the inter-quartile range (IQR). Using non-parametric moments eliminates the influence of possible occasional long delays that typically appear in multi-tasking computer systems [16]. The ISIP/ISMS means were not considered, because they typically reflect the stimulus IOIs accurately, and therefore do not provide information relevant for timing problems incurred by the computer system.…”

“…These large differences across studies highlight the influence of different methods. Dodonova and Dodonov [72] listed a range of factors that may differ across reaction time studies, and that entail RT differences on the order of 10-30 ms, but did not acknowledge that errors related to general purpose computer systems can be considerably larger [16], and thus typically overshadow the former. It is therefore disconcerting that the delays and variability in instruments for measuring time-critical behavior are seldom reported [73][74][75].…”

“…Even the first personal computer, the IBM XT, with a clock frequency of 4.77 MHz, would entail a temporal variability of less than 1 ms, assuming around 50 basic processor operations per measurement cycle. It is apparently not the hardware that constitutes the bottleneck for precise timing, but the way that the software is designed for multi-purpose computers, including the operating systems themselves [16]. This points to the importance of tracking the software versions in users' computer when conducting time-critical online measurements.…”

“…Another reason is that the ways in which visual and auditory stimulation varies across platforms may be critical for test performance, whereas this is of no consequence for corresponding in text in a questionnaire. A third reason is that many tests rely on temporal accuracy in stimulus presentation or response measurement, which is known to be poor and to vary across general-purpose computer systems [9][10][11]13,16]. Both temporal delay and variability that vary across platforms decrease reliability and may even obliterate systematic participant effects.…”