[open] xpyriment back-end and response accuracy
I am working on an implicit associations test, and I selected xpyriment back-end to improve timing accuracy. However, when I look at my data, I see that accuracy is relatively low, and variable keyboard_backend shows that my keyboard is using legacy and not xpyriment. Does anyone know why this is the case and how I can switch my keyboard to xpyriment.
Thanks
Comments
Selecting the expyriment back-end does not immediately give you 'higher accuracy', and this certainly isn't measurably by looking at keyboard response data. If this is not what you're looking at, then what kind of benchmark did you use?
The temporal accuracy is about the screen refresh: both psychopy and expyriment provide you with the time of the screen refresh actually starting, whereas pygame (legacy) only gives you an indication (which, on a 60 Hz monitor, can be off by 0 tot 16.7 ms).
The keyboard timing accuracy is based on the polling rate of your keyboard, which using the standard Windows drivers is 8 ms. This means that the expected value of the random noise would be 4 ms, where you are likely to have reaction times of over a couple hundred milliseconds. So on a single trial basis, it's not that big a deal. Also note that you average out over all trials, which should theoretically leave you with equal amounts of noise in all conditions.
On top of this, human responses are intrinsically quite variable. There's actually research showing that in most cases, you needn't worry about the variability that is added by your keyboard.
Finally, to answer your question: the expyriment back-end is for the display; the back-end used for the keyboard, mouse, and sound is still PyGame. But because of what I went on about before, you needn't worry about this. And if you decide that the keyboard response timing is important to you, you need to look for a hardware solution (faster keyboard!) or different drivers than the default ones (polling more often than once every 8 ms). If you do not, you can use whatever software you want, but you won't get any less variant timing.
P.S.: see the Documentation for more info on how OpenSesame handles the display refresh, and what 'millisecond accurate timing' actually is.