User`s guide

E-Prime User’s Guide
Chapter 3: Critical Timing
Page 75
to read a microsecond precision clock resulted in variance in the range of 200ms
2
, with clock miss
rates of 58% on a Pentium 120MHz computer and 47% on a Pentium 500MHz. When running
similar tests under E-Prime on the same machines, the clock miss rate was reduced to less than
0.1%.
We encourage the reader to perform a simple test to determine the accuracy of the timing of any
experiment presentation software that claims to provide millisecond accuracy. Create an
experiment using the software to display 10 sets of 10 bitmap images, displaying a single image
every 100ms. This program should take 10x10x100ms, or 10 seconds. Use the second hand of a
watch or a stopwatch to measure the total time, and look at the images to see if there are any
detectable pauses in the image presentation sequence. If there is any error, check the output of
the program to see if it reported the error. Much more extensive tests are required to verify
timing, but this is a simple check that many experimental control programs fail. When the timing
techniques presented in the following sections are utilized during experiment development, E-
Prime will present this sequence of displays accurately.
Scientists have an obligation to correctly report their data. No scientist can afford to publish
flawed data. However, accurate reporting of timing is difficult because many sources of computer
reporting of timing data are invalid, or too crude to be useful. Specifically, there are four major
problems related to maintaining and reporting timing precision in computerized behavioral
research. Below is a table listing these problems, and each will be discussed in turn.
Problems Reporting Timing Accuracy in Computerized Behavioral Research
1. Computer operating systems can falsely report timing data.
2. Actual durations can be significantly longer and deviate from the intended durations specified.
3. Displays are accomplished via refresh cycles. Refresh cycles are rarely multiples of the intended
display duration and cannot be restarted during an experiment.
4. Accurately measuring and reporting timing, then debugging an experiment can be difficult.
3.2.1 Problem 1: Computer operating systems can
falsely report timing data
Let us illustrate this with a simple example. Consider how the refresh rate (i.e., how often the
display is redrawn per second) is reported. This is a variable that should always be reported in
the Methods section of a computerized experiment that specifies short duration displays, since it
critically influences the stimulus quality and duration (see section 3.2.3). A researcher might
check the monitor refresh rate by checking the Windows desktop properties and use that value to
report the refresh rate
3
. The table below reports values for one card
4
tested, showing a
substantial mismatch between the “reported” refresh rate and the measured refresh rate.
Display
Resolution
Refresh Rate Reported
by Windows (Hz)
Measured Refresh
rate (Hz)
% Error
640 x 480 60 59.7 0.50%
800 x 600 60 60.3 0.50%
1024 x 780 60 60.2 0.33%
1280 x 1024 60 75.3 20.32%
3
The location of the refresh rate parameter varies because of dependence on the monitor virtual device
driver and because some cards do not report the refresh rate. Typically this information can be obtained by
clicking the right button on the Windows desktop, selecting Properties, then selecting Settings and selecting
the Advanced button. Finally, select the Adapter tab and check the Refresh Rate.
4
This example was done on a Gateway G6-266 with an STB Nitro 3D(S3 ViRGE-DX/GX 375/385 on
Windows 98.