User`s guide

E-Prime User’s Guide
Chapter 3: Critical Timing
Page 83
Recall that Figure 2 illustrated a small cyclic variation of the stimulus duration. Figure 6 shows an
enlarged view of the intended and actual time measured in E-Prime.
E-Prime Stimulus Display Duration Options
180
185
190
195
200
205
210
1 3 5 7 9
11 13 15
Image Number
Display Duration (ms)
Intended Time
E-Prime 200 ms
E-Prime 15 Refreshes
E-Prime 14 Refreshes
Figure 6. Intended and measured display time for three approximations to 200ms display.
The intended duration was 200ms. Assuming a monitor refresh rate of 73.1Hz (13.67 refresh
duration), the nearest integer refresh cycle onsets are 191.3ms and 205.0ms (for 14 or 15
refreshes respectively). The display hardware limits the display duration to only the values of
191.3ms or 205.0ms. Because of this, the experimenter has only 3 options: 1) make all the
displays 191.3ms, 2) make all the displays 205.0ms, or 3) make the displays a mixture of 37%
191.3ms and 63% 205.0ms to average 200ms. When choosing the average duration option, you
can export the data from E-Prime (using the E-DataAid utility) into a spreadsheet such as
Microsoft Excel and calculate the mean and standard deviation of the actual display duration
(e.g., mean display time of 200.057ms with 6.61ms SD). See section 3.4.4 for more information.
3.2.4 Problem 4: Accurately measuring and reporting
timing, then debugging an experiment
You want to be sure that the experimental procedure that was intended was the procedure that
actually ran on the computer and presented to the subject. This is difficult for several reasons.
First, programming experimental specifications is inherently error-prone due to the large number
of details that must be handled correctly
6
. Second, testing humans often occurs at the limits of
human perception. Hence, errors may not be accurately detected by simple observation (e.g.,
adding 50ms to a 600ms display may not be noticed without checking special hardware or
recording routines). Third, equipment designed for external monitoring is generally not available
due to both costs and operating complexity. Fourth, access to critical variables (e.g., when the
refresh actually did occur) may not be available. Fifth, the insertion of timing code may actually
alter the timing and, additionally, output of appropriate information may not be possible while
critical timing events are occurring.
6
In computer programmer studies, estimates of the typical number of bugs are often reported to be in the
range of one bug per ten lines of code.