User`s guide

E-Prime User’s Guide
Chapter 3: Critical Timing
Page 92
If Cumulative timing mode was in use, the duration for each display would be shortened so that
the end of the display time would always occur as close to 100ms boundaries as the hardware
allows. In Cumulative mode, the second display would start looking for the vertical blanking at
100ms (rather than at 106.83ms). In Cumulative mode, the average display duration would be
100ms with a standard deviation of 3.95ms. After 100 displays, the start of the next display would
occur at 10.000 seconds (plus the delay until the next refresh).
Event mode and Cumulative mode studies are used in two different classes of paradigms. If the
goal is to present a single stimulus, or to present short sequences of stimuli (e.g., fixation, probe,
mask) with an inter-trial interval that varies acceptably, then you should use Event mode. If the
goal is to maintain a constant rate of presentation (e.g., a memory task presents stimuli every 2
seconds without cumulative error or drift), you should use Cumulative mode. The default timing
mode for all objects in E-Prime is Event.
3.3.1.4 Technique 4: Logging the millisecond timestamp of
all actions
Due to the overall complexity of timing issues it is easy to make a mistake in the specification of
timing, or for some unexpected system timing delays to alter the timing of the experiment. These
mistakes can easily go undetected and unreported. In E-Prime, it is possible to log the times that
stimuli were presented to the subject and the times that responses were made. Therefore, it is
also possible to analyze the timing of the experiment with the same analysis tools used to
analyze the behavioral data. The accurate logging of timing data is the principal line of
defense against the operating system or hardware putting artifacts into data. As described
before, the makers of the operating system are willing to “indiscriminately” (see footnote, section
3.2) introduce timing errors into an experiment. These types of errors are rare and generally
small, but there is always a possibility that they may occur. By logging stimulus timing data you
can identify those errors, and when they occur you can choose to include or exclude the data
from trials on which they occur in your data analysis.
Each stimulus presentation object in E-Prime has a Data Logging property on it’s Duration/Input
tab. This property allows the settings of Standard, Time Audit Only, Response Only, and Custom
and controls which set of values the object will add to the data log file after the object runs. When
Data Logging is set to Standard for an event, the following variables are logged in addition to the
basic set of dependent measures (i.e., accuracy, response, response time, etc):
OnsetTime
Timestamp (in milliseconds from the start of the experiment) when the stimulus
presentation actually began.
OnsetDelay
Difference in milliseconds between the actual onset time and the expected or
“target” onset time.
DurationError
Difference in milliseconds between the actual duration and intended duration
before the command ended. If the event is terminated by a response, a value of –
99999 will appear in the column (i.e., since the duration error cannot be computed
in that situation).
In addition to the variables listed above, stimulus presentation objects can also log the Duration,
PreRelease, ActionTime, ActionDelay, StartTime, FinishTime, OffsetTime, OffsetDelay,
TargetOnsetTime, TargetOffsetTime and TimingMode (see Chapter 1-E-Studio in the
Reference Guide, and the Time Audit topic in the E-Basic Online Help for property descriptions).
The logging of timing related variables is accomplished in a way that does not significantly impact
the execution time of the experiment, and allows output to the data analysis stream to facilitate
rapid analysis. The only arguable downside to logging all of this timing data for each time critical
object in the experiment is that it can significantly increase the number of columns in the data file
(which can sometimes be confusing to examine). However, because this data can easily be