Building your experiments with Paradigm means millisecond accuracy without the headaches
All Around Accurate:
Visual stimuli presentation and vertical blank detection using DirectX.
Audio presentation on Windows 7/8/8.1 using ASIO or OpenAL.
EEG, Eye-Tracker and fMRI synchronized onset and response triggers using Paradigm Elements.
Reaction time measurement using WM_INPUT and DirectInput.
How We Measure Timing Accuracy
All of our timing measures are taken using the Black Box Toolkit.
We take two accuracy measures each for visual and audio
stimuli. The first is a relative onset error measure which benchmarks the duration error between stimuli from onset to onset. The second is an offset duration measure where we calculate the difference between when Paradigm
"thinks" the stimulus was presented and when it actually appeared on the screen or was rendered by the speakers.
This is a measure of the total latency in the system (software and hardware) and is typically what most researchers care about. These benchmarks were done on a 32-bit Windows 7, Dell Precision T3500 Xeon w/ 2 GB RAM, a NVIDIA Quadro 6000
graphics card and a Creative XtremeGamer sound card.
Image Stimulus Timing Accuracy
Paradigm presented a set of images 400 times, each for a duration of 50ms. Relative onset errors show an average error of 0.32ms (SD 0.13). Offset duration measures show an average difference of 1.88ms (SD 0.02) meaning
that there was a 1.88ms difference between when Paradigm "thinks" the image was display and when it actually appeared on
the screen.
Average Offset Duration: 1.88ms (SD 0.02)
Average Relative Onset Error: 0.32ms (SD 0.13)
Audio Stimulus Timing Accuracy
Paradigm presented a single audio file 195 times, each for a duration of 15ms. Measurements were taken using a microphone placed in front of the speaker so these results are for the whole audio system not just when Paradigm started the playback buffer. Relative onset error measures show an average error of 0.08ms (SD 0.12). The offset duration measure
shows an average difference of 0.71ms (SD 0.05) meaning that there was virtual no difference difference between when
Paradigm "thinks" the sound was played and when it actually was heard. Note: achieving latencies this low requires you
use a sound card that supports ASIO or OpenAL with "bit-matched" playback enabled in the device's properties.
Average Offset Duration: 0.71ms (SD 0.05)
Average Relative Onset Error: 0.08ms (SD 0.12)
How Much Will Your Experiment's Timing Differ From Ours?
Honestly, it's almost impossible to tell. There are many sub-optimal combinations of hardware and software that could
negatively affect your experiment's timing. We recommend you use a system similar to the one we've tested here which will
give you similar results. We also recommend you read the following reference that explains all the issues involved when
trying to achieve good timing:
Millisecond Precision psychological research in a world of commodity computers: New hardware, new problems? Richard R Plant and Gary Turner. Behavior Research Methods, 2009, Volume 41, Number 3, Pages 598-614
Average Offset Duration: 1.88ms (SD 0.02)
Average Relative Onset Error: 0.32ms (SD 0.13)
Average Offset Duration: 0.71ms (SD 0.05)
Average Relative Onset Error: 0.08ms (SD 0.12)
