Pickup Usability Dominates - Human-Computer Interaction and Innovation in Handheld, Mobile and Wearable Technologies - page 54

User Studies Stuerzlinger, 2004), predicts a text entry time

for a 10 word phrase at 31.2 seconds, equating
to a speed of 19.3wpm—matching closely with
experienced user experiments with T9 of 20.4
wpm (James & Reischel, 2001).
In this work we modelled keystroke speed at
0.28s based on a fixed figure from Card et al.'s
work that is equivalent to “an average non-secre-
tary typist” on a full QWERTY keypad. This gives
fairly accurate predictions but cannot take into
account fine grained keyboard design elements
that can have a considerable impact on typing
speed in practice: for example different keyboard
layouts clearly affect the average time it takes a
user to move his/her fingers to the correct keys.
Mackenzie's group have conducted considerable
work using Fitt's law (Fitts, 1954) to calculate
the limit of performance given distance between
keys (e.g. (Silfverberg, MacKenzie, & Korhonen,
2000)). The basic form of their distance-based
modelling predicts 40.6 wpm for thumb-based
predictive input—assuming no next key operations
(essentially equivalent to no thinking or homing
times in equation 1). Later work modifies the Fitt's
distance models to take into account two inaccura-
cies that can noticeably affect predictions: repeated
letters on the same key (Soukoreff & MacKenzie,
2002) and parallel finger movements where users
move one finger at the same time as pressing with
another (MacKenzie & Soukoreff, 2002).
These models are useful in predicting perfor-
mance but focus on expert error-free performance.
More complex modelling approaches have been
researched to support novices to model more com-
plete interaction, and to model error behaviour (e.g.
(How & Kan, 2005; Pavlovych & Stuerzlinger,
2004; Sandnes, 2005)). Although users studies
are the acid test for any interactive system, these
models are valuable either in the early stages
of design or to understand methods where user
experiments are difficult, e.g. by being biased by
users' prior experience of current technologies.

Models that predict text entry performance only
give us part of the picture, proper user studies often
give a truer indication of how text entry methods
perform in reality. While there are many param-
eters that can affect the design of user studies, the
two prominent issues for text entry experiments
are the environment in which the experiments are
conducted and the phrases that users enter.
Most user studies into text entry have been
conducted in laboratories. A laboratory is a con-
trolled environment that leads to a more consistent
user experience than the real world and, thus,
considerably easier statistical analysis as there are
fewer confounding variables from the environment
to interfere with measurements taken. However,
conducting experiments on people entering text
on mobile phones in quiet office-like settings
where they can focus exclusively on the text entry
tasks is arguably not representative of normal use!
There is a growing debate in mobile HCI research
on the validity of laboratory experiments with
some researchers arguing that, while the focus
of most common errors is different in the real
world, laboratory experiments do not miss errors
that are found in real-world experiments (Kaik-
konen, Kekäläinen, Cankar, Kallio, & Kankainen,
2005) while others claim a wider range of errors
were found in the real-world than in laboratories
(Duh, Tan, & Chen, 2006). (Kjeldskov & Graham,
2003) report that “71% [of studied evaluations
were] done through laboratory experiments, 19%
through field experiments and the remaining 10%
through surveys”. As a specific example, (Brew-
ster, 2002) showed usability and text entry rates
were significantly reduced for users performing an
outdoor walking circuit, while entering on a soft
numeric key-pad, than those conducting the same
experiment in a traditional laboratory. Whereas
(Mizobuchi, Chignell, & Newton, 2005) showed
that, while walking was slowed down when using
a device, it did not impinge upon the text entry
rate or accuracy. Some researchers have tried to