User`s guide

E-Prime User’s Guide
Chapter 3: Critical Timing
Page 74
Modern multi-tasking operating systems (e.g., Windows, Mac OS, Unix) will not allow exclusive
and continuous execution of any process, because the operating system must take cycles to
perform critical functions (e.g., virtual memory management). No program running on a common
desktop operating system can deliver accurate measurements at every millisecond (i.e., a 0%
miss tick rate). For example, Windows will indiscriminately remove parts of a program from
memory to test to see if a program really needs them in order to reduce the applications working
memory set. This operation, itself, will produce random timing delays
1
. With E-Prime, we attempt
to manage or minimize all interactions with the operating system. Although we cannot deliver
measurements every millisecond, our methods make delays and/or missed reads of the clock tick
very rare, typically delivering 99.95% of the times precisely, and make sure that the results are
statistically valid and accurate.
Operationally, E-Prime defines the interpretation of millisecond precision as the ability to
report any reaction time or timing duration such that:
1. a measured and reported timing precision standard deviation is less than half a
millisecond.
2. recordings are within a millisecond from the time they are available to computer
hardware. When errors occur, they should be detectable, should not increase
measurement variance by more than 1ms
2
. The investigator should have a method
available to identify and filter out timing data that is in error.
3. screen refreshes (displays of the full screen) can be tracked without misses 99.9% of
the time, and misses can be detected and reported.
With such accuracy, researchers can be confident that they are reporting valid results of human
behavior, and that those results will replicate between laboratories. Remember that a 1ms
2
timing
variance is negligible to human timing variability. Statistically, the variances of two independent
processes add in the following manner for measuring human data:
Computer Recorded Variance = Human Variance + Measurement Variance
Let us take a concrete example. A typical human reaction time of 600ms response time would
have a standard deviation of 150ms and a variance of 22500ms
2
. With E-Prime on a 120 MHz
computer, with most hardware configurations and recommended software configurations (see
section 3.4), you would expect to have a measurement variance of 1ms
2
. This would increase
measurement variance to a negligible 0.00444%. To put it another way, to achieve equivalent
statistical power, the increase of 1ms
2
measurement error requires running an extra 0.00444%
trials
2
. Running one extra trial per 22500 trials is a small price to pay for the control, precision,
and cost savings offered by modern computerized research.
We caution you to note that the techniques commonly employed by most computer programs do
not achieve the 1ms
2
timing error variance. In our testing, using standard operating system calls
1
The Windows NT documentation on the virtual-memory manager states The working set manager (part of
operating system) periodically tests the quota by stealing valid pages of memory from a process… This
process is performed indiscriminately to all processes in the system (p. 12, The Virtual-Memory Manager in
Windows NT, by Randy Kath, Microsoft Developer Network Technology Group, December 21, 1992).
2
A similar variance argument can be made for display timing. The display timing variance of an experiment
that varied display durations of 1, 2 or 3 refresh cycles of 13, 26, or 39ms produces a timing manipulation
variance of 112.7ms
2
. If timing delays or computer hardware caused one extra refresh (13ms delay) in 1000
refresh detections, the added error variance would be 0.14ms
2
, or 0.13% of the manipulation variance. This
would require running one extra trial per thousand to compensate for the measurement error.