Page 70 -

3.3 Investigating a single independent variable 55

control these negative impacts is counterbalancing the condition or treatment orders
through a Latin Square Design.
When the objective of the study is not initial interaction with the application, an
effective approach to reduce the impact of the learning effect is to provide sufficient
time for training. Research suggests that, for many types of tasks, the learning curve
tends to be steeper during the initial interaction stages and flatter after that stage
(see Figure 3.5). People achieve quicker progress in learning during initial stages,
followed by gradual lesser improvement with further practice. Therefore, providing
sufficient training time for users to get acquainted with the system or the task greatly
reduces the learning effect during the actual task sessions. Of course, training cannot
completely eliminate the learning effect. It only reduces its impact. This approach,
combined with the counterbalancing of task conditions, is widely adopted in HCI
studies to control the impact of learning.

Success

Trials
FIGURE 3.5
Typical learning curve.

To address the problem of fatigue caused by multiple experimental tasks, we
need to design experiment tasks frugally, reducing the required number of tasks and
shortening the experiment time whenever possible. It is generally suggested that the
appropriate length of a single experiment session should be 60 to 90 minutes or
shorter (Nielsen, 2005). When a session lasts longer than 90 minutes, the participant
may get tired or frustrated. It is strongly suggested that a single session should defi-
nitely not last longer than 2 hours. During the experiment, the participant should be
provided with opportunities to take breaks as needed. Interestingly, even when the
experimenter encourages the participants to take breaks, the participants may not
realize that they are getting tired and tend to ignore the suggestion to take a break.
Therefore, some researchers find it helpful to force the participants to take a break
during an experiment. For more discussion regarding the benefit of breaks in HCI
studies, please refer to Chapter 15.