bachelor-thesis/chap2/literature_review.tex

% MORE INFO
% WRUED: https://www.ccohs.ca/oshanswers/diseases/rmirsi.html

\section{Literature Review}
In this section, we gathered information about conventional methods, tools and
metrics that are used in the domain of keyboard related research. We were
particularly interested in how keyboard usage is related to health risks, the
status quo of keyboard design, which metrics to use when conducting a typing
related experiment and previous studies that evaluated the impact of key
actuation force on those metrics. The information collected during the
literature review ensured a profound understanding of those topics, which helped
us during the design phase of our two preliminary studies and the main
study. Additionally, we partially established our hypotheses based on the
results of previous research.

\subsection{Work Related Upper Extremity Disorders}
\label{sec:wrued}
\Gls{WRUED} is a term to describe a group of medical conditions related to
muscles, tendons and nerves in shoulder, arm, elbow, forearm or hand, such as
\gls{CTS}, \gls{RSI}, Tendinitis, \gls{TNS}, etc. Symptoms of
\gls{WRUED} are aching, tiredness and fatigue of affected regions that either
occur while working or even extend to phases of relaxation. A common way to
treat \gls{WRUED} is to avoid the potentially harmful activities that cause
discomfort in affected areas \cite{ccfohas_wrued}. Pascarelli and Hsu reported,
that out of 485 patients with \gls{WRUED} 17\,\% were computer users
\cite{pascarelli_wrued}. Since computers have become an essential part of many
jobs in almost any sector of employment, restrictions of computer related
activities would result in either reduced productivity or the complete inability
to fulfill required tasks, which in the worst case could require a change of
profession \cite{ccfohas_wrued}. Potential problems with current keyboard
designs and possible solutions are discussed in the following sections.

\begin{phga_sum*}
  \gls{WRUED} are a serious problem of modern society and since there is evidence
  pointing towards computer related work to be a possible factor for these
  diseases, it is likely that especially keyboards, as the main input device, are
  responsible for a portion of people affected by \gls{WRUED}.
\end{phga_sum*}

\subsection{Keyboards and Keyswitches}
\label{sec:kb_ks}
\subsubsection{Keyboard Models and Layouts}
\label{sec:kb_layout}

\begin{figure}[ht]
  \centering
  \includegraphics[width=1.0\textwidth]{images/keyboard_models.jpg}
  \caption{Different keyboards, including alternative split models, smaller form
    factors and traditional layouts such as ISO/IEC 9995 \cite{iso9995-2} and
    ANSI-INCITS 154-1988 \cite{ansi-incits-154-1988}}
  \label{fig:keyboard_models}
\end{figure}

Keyboards are well known input devices used to operate a computer. There are a
variety of keyboard types and models available on the market, some of which can
be seen in Figure \ref{fig:keyboard_models}. The obvious difference between
those keyboards in Figure \ref{fig:keyboard_models} is their general
appearance. The keyboards feature different enclosures and keycaps. Keycaps are
the rectangular pieces of plastic on top of the actual keyswitches that
sometimes indicate what letter, number or symbol, also known as characters, a
keypress should send to the computer. These keycaps are mainly made out of the
two plastics \gls{ABS} and \gls{PBT} which primarily differ in feel, durability,
cost and sound \parencite[8]{bassett_keycap}.

\begin{figure}[ht]
  \centering
  \includegraphics[width=1.0\textwidth]{images/keyboard_layouts.png}
  \caption{The three major physical keyboard layouts all labeled with the
    alphanumeric characters of the most popular layout―\gls{QWERTY}
    \cite{wiki_kb_layouts}}
  \label{fig:keyboard_layouts}
\end{figure}

Nowadays, there are three main standards that define the physical layout of a
keyboard―ISO/IEC 9995 \cite{iso9995-2}, ANSI-INCITS 154-1988
\cite{ansi-incits-154-1988} and JIS X 6002-1980 \cite{jis-x-6002-1980}. These
layouts propose slightly different arrangements of the keys and some even alter
the shape of a few keys entirely. Figure \ref{fig:keyboard_layouts} shows the
layouts defined by the three standards mentioned and shows the main
differences. In addition to the physical layout, there are also various layouts
concerning the mapping of the physical key to a character that is displayed by
the computer. Most of the time, the mapping happens on the computer via software
and therefore the choice of layout is not necessarily restricted by the physical
layout of the keyboard but rather a personal preference. As seen in Figure
\ref{fig:keyboard_models}, there are also non standard physical layouts
available which are often designed to improve the posture of the upper extremity
while typing to reduce the risk of injury or even assist in recovering from
previous \gls{WRUED} \cite{ripat_ergo, tittiranonda_ergo}. Those designs often
split the keyboard in two halves to reduce ulnar deviation and some designs also
allow tenting of the halves or provide a fixed tent which also reduces forearm
pronation \cite{baker_ergo, rempel_ergo}.

\subsubsection{Membrane Keyswitch}
\label{sec:mem_switch}

Besides the exterior design of the keyboard, there is another part of
interest—the keyswitch. This component of a keyboard actually sends the signal
that a key is pressed. There are different types of keyswitches available to
date. The most commonly used ones are scissor switches and rubber dome switches
which are both subsets of the membrane switch family. Scissor switches are often
found in keyboards that are integrated into notebooks while rubber dome switches
are mostly used in workplace keyboards. Both variants use a rubber membrane with
small domes underneath each key. When a key is pressed, the corresponding dome
collapses and because the dome's inner wall is coated with a conductive
material, closes an electrical circuit \cite{ergopedia_keyswitch,
  peery_3d_keyswitch}.

\subsubsection{Mechanical Keyswitch}
\label{sec:mech_switch}
Another type of switches are mechanical keyswitches. These switches are
frequently used in gaming and high quality workplace keyboards as well as by
enthusiast along with prosumers which build and then sell custom made keyboards
to the latter audience \cite{bassett_keycap, ergopedia_keyswitch}. These
keyswitches are composed of several mechanical parts which can be examined in
Figure \ref{fig:mech_keyswitches_dissas}. The housing is made up of two parts,
the bottom and top shell. The actual mechanism consists of two conductive
plates, which when connected trigger a keypress, a stainless steel spring which
defines how much force has to be applied to the switch to activate it and a stem
which sits on top of the spring and separates the two plates. The shape of the
stem, represented by the enlarged red lines in Figure
\ref{fig:mech_keyswitches_dissas}, defines the haptic feedback produced by the
keyswitch. When pressure is applied to the keycap, which is connected to the
stem, the spring gets contracted and the stem moves downwards. Thereby, it stops
separating the two plates, which closes the electrical circuit and sends a
keypress to the computer. After the key is released, the spring pushes the stem
back to its original position \cite{bassett_keycap, peery_3d_keyswitch,
  ergopedia_keyswitch, chen_mech_switch}. Usually, mechanical keyswitches are
directly soldered onto the \gls{PCB} of the keyboard. However, there are also
keyboards with \gls{PCB}s that feature special sockets where the keyswitches can
be \gls{swapped} without soldering at all \cite{gmmk_hot_swap}. It is also
possible to equip an already existing \gls{PCB} with sockets to make it
hot-swappable \cite{te_connect}.

\begin{figure}[ht]
  \centering
  \includegraphics[width=1.0\textwidth]{images/mech_keyswitches_dissas.jpg}
  \caption{Disassembled tactile, clicky and linear mechanical keyswitchs. The
    red lines resemble the shape of the stem which is responsible for the haptic
    feedback and thus, in combination with the strength of the spring, the
    required actuation force}
  \label{fig:mech_keyswitches_dissas}
\end{figure}

Mechanical keyswitches also have three main subcategories. Those categories
primarily define if and how feedback for a keypress is realised:
\begin{enumerate}
  \item \textbf{Tactile Switches} utilize a small bump on the stem to slightly
  increase and then instantly collapse the force required immediately before the
  actual actuation happens \cite{cherry_mx_brown}. This provides the typist with
  a short noticeable haptic feedback, which should encourage a premature
  release of the key. An early study by Brunner and Richardson suggested, that
  this feedback leads to faster typing speed and a lower error rate in both
  experienced and casual typists (n=24) \cite{brunner_keyswitch}. Contrary, a
  study by Akagi yielded no significant differences in terms of speed nor error
  rate between tactile and linear keyswitches and links the variation found in
  error rates to differences in actuation force (n=24)
  \cite{akagi_keyswitch}. Tactile feedback could still assist the typist to
  prevent \gls{bottoming}.
  \item \textbf{Tactile and audible Switches (Clicky)} separate the stem into
  two parts. The lower part also features a small bump to provide tactile
  feedback and is also responsible for a distinct click sound when the actuation
  happens \cite{cherry_mx_blue}. Gerard et al. noted, that in their study
  (n=24) keyboards with audible feedback increased typing speed and decreased
  typing force. This improvement could have been due to the previous experience
  of participants with keyboards of similar model and keyswitch characteristic
  \cite{gerard_keyswitch}.
  \item \textbf{Linear Switches} do not offer a distinct feedback for the
  typist. The activation of the keyswitch just happens after approximately half
  the total travel distance \cite{cherry_mx_red}. The only tactile feedback that
  could happen is the impact of \gls{bottoming}. However, with enough practice
  typist can develop a lighter touch which reduces overall typing force and
  therefore reduces the risk of \gls{WRUED} \cite{gerard_keyswitch,
    peery_3d_keyswitch, fagarasanu_force_training}.

\end{enumerate}
The corresponding force-displacement curves for one exemplary keyswitch of each
category by the manufacturer Cherry are shown in Figure
\ref{fig:ks_fd_curves}. The Operational position indicates the activation of the
keyswitch.

\begin{figure}[ht]
  \centering
  \includegraphics[width=1.0\textwidth]{images/ks_fd_curves.jpg}
  \caption{Actuation graphs for Cherry MX BROWN \cite{cherry_mx_brown} | BLUE
    \cite{cherry_mx_blue} | RED \cite{cherry_mx_red} switches. Tactile position marks the point where a haptic feedback happens, operational position marks the activation of the keyswitch and reset position is the point where the keyswitch deactivates again}
  \label{fig:ks_fd_curves}
\end{figure}


All types of keyswitches mentioned so far are available in a myriad of actuation
forces. Actuation force, also sometimes referred to as make force, is the force
required to activate the keyswitch \cite{radwin_keyswitch,
  ergopedia_keyswitch}. That means depending on the mechanism used, activation
describes the closing of an electrical circuit which forwards a signal, that is
then processed by a controller inside of the keyboard and finally sent to the
computer. The computer then selects the corresponding character depending on the
layout used by the user. Previous studies have shown, that actuation force has
an impact on error rate, subjective discomfort, muscle activity and force
applied by the typist \cite{akagi_keyswitch, gerard_keyswitch,
  hoffmann_typeright} and as suggested by Loricchio, has a moderate impact on
typing speed, which could be more significant with greater variation of
actuation force across tested keyboards \cite{loricchio_force_speed}.

\begin{phga_sum*}
Since this thesis is focused on keyboards and especially the relation
between the actuation force of the keyswitch and efficiency (speed, error rate)
and also the differences in satisfaction while using keyswitches with varying
actuation forces, it was important to evaluate different options of keyswitches
that could be used to equip the keyboards used in the experiment. The literature
suggested, that the most common switch types used in the broader population are
rubber dome and scissor switches \cite{ergopedia_keyswitch,
  peery_3d_keyswitch}. Naturally, those keyswitches should also be used in the
study, but one major problem due to the design of those keyswitches arises. It
is not easily possible to alter the actuation force of individual keyswitches
\cite{peery_3d_keyswitch}. This will be necessary to create a keyboard where
each key should have an adjusted actuation force depending on the finger that
normally operates it. It should be mentioned, that it is theoretically possible
to exchange individual rubber dome switches on some keyboards, e.g. keyboards
with \gls{Topre} switches, but the lacking availability of compatible keyboards
and especially the limited selection of actuation forces (30\,g to 55\,g for
\gls{Topre} \cite{realforce_topre}) makes this not a viable option for this
thesis \cite{keychatter_topre}. Therefore, we decided to use mechanical
keyswitches for our experiment. These keyswitches are broadly available
in a variety of actuation forces and because the spring which mainly defines the
actuation force can be easily replaced with any other compatible spring on the
market, the selection of actuation forces is much more appropriate for our use
case (30\,g to 150\,g) \cite{peery_3d_keyswitch}. We also decided to use linear
switches because they closest resemble the feedback of the more wide spread
rubber dome switches. Further, linear switches do not introduce additional
factors beside the actuation force to the experiment. In addition, based on the
previous research we settled on using a keyboard model with hot-swapping
capabilities for our experiment to reduce the effort required to equip each
keyboard with the required keyswitches and in case a keyswitch fails during
the experiment, decrease the time required to replace the faulty switch.
\end{phga_sum*}

\subsection{Measurement of Typing Related Metrics}
\label{sec:metrics}
Nowadays, a common way to compare different methods concerning alphanumeric
input in terms of efficiency is to use one of many typing test or word
processing applications which are commercially available. Depending on the
software used and the experimental setup, users have to input different kinds of
text, either for a predefined time or the time is measured till the whole text
is transcribed \cite{chen_typing_test, hoffmann_typeright,
  fagarasanu_force_training, akagi_keyswitch, kim_typingforces,
  pereira_typing_test, baker_ergo}.

\subsubsection{Readability of Text}
\label{sec:meas_fre}

The texts used should be easy to read for typists participating in studies that
evaluate their performance and are therefore chosen based on a metric called the
\gls{FRE} which indicates the understandability of text
\cite{fagarasanu_force_training, kim_typingforces, flesch_fre}. The score ranges
from 0 which implies very poor reading ease to 100 suggesting that the style of
writing used causes the text to be very easy to comprehend
\cite{flesch_fre}. Immel proposed an adjusted formula of the \gls{FRE} that is
suitable for German text \cite{immel_fre} and can be seen in
(\ref{eq:fre_german}). This formula was necessary, because all participants were
Germans.

\begin{equation}\label{eq:fre_german}
  FRE_{deutsch} = 180 - \underbrace{ASL}_{\mathclap{\text{Average Sentence Length}}} - (58,5 * \overbrace{ASW}^{\mathclap{\text{Average Syllables per Word}}})
\end{equation}

According to Flesch, the values retrieved by applying the formula to text can be
classified according to the ranges given in Table \ref{tbl:fre_ranges} \cite{flesch_fre}.
\begin{table}
  \centering
  \small
  \ra{1.3}
  \begin{tabular}{?l^c}
    \toprule
    \multicolumn{1}{c}{\emph{FRE}}& \emph{Understandability}  \\
    \midrule
    \multicolumn{1}{c}{0 - 30}    & Very difficult     \\
    30 - 50                       & Difficult          \\
    50 - 60                       & Fairly difficult   \\
    60 - 70                       & Standard           \\
    70 - 80                       & Fairly easy        \\
    80 - 90                       & Easy               \\
    \multicolumn{1}{r}{90 - 100}  & Very easy          \\
    \bottomrule
  \end{tabular}
  \caption{Categories for different FRE scores to classify the understandability
    of text \cite{flesch_fre}}
  \label{tbl:fre_ranges}
\end{table}

\subsubsection{Performance Metrics}
\label{sec:meas_perf}

There are several metrics to measure the performance of typists. Typical methods
to measure speed are
\begin{enumerate}
  \item \textbf{\Gls{WPM}}
  \begin{equation}\label{eq:wpm}
    WPM = \frac{|T| - 1}{S} * 60 * \frac{1}{5}
  \end{equation}
  In Eq. (\ref{eq:wpm}), $|T|$ is the length of the transcribed text, $S$ the
  time in seconds taken to transcribe $T$, $\frac{1}{5}$ the average word length
  and $60$ the conversion to minutes. $|T| - 1$ counteracts the first input
  which starts the timer in many typing tests \cite{mackenzie_metrics}.
  \item \textbf{\Gls{AdjWPM}} is especially useful if participants are allowed
  to make mistakes and at the same time not obligated to correct them. This
  method adds an adjustable factor to lower the \gls{WPM} proportionally to the
  uncorrected error rate $UER := [0;1]$ defined in Eq. (\ref{eq:uer}). The
  exponent $a$ in Eq. (\ref{eq:cwpm}) can be chosen depending on the desired
  degree of correction \cite{mackenzie_metrics}.
  \begin{equation}\label{eq:cwpm}
    AdjWPM = WPM * (1 - UER)^{a}
  \end{equation}
  \item \textbf{\Gls{KSPS}} measures the raw input rate of a typist by capturing
  the whole input stream including backspaces and deleted characters ($IS$)
  \cite{mackenzie_metrics}.

  \begin{equation}\label{eq:ksps}
    KSPS = \frac{|IS| - 1}{S}
  \end{equation}
\end{enumerate}

In addition to speed, the error rate of typists can be measured with the
following two methods

\begin{enumerate}
  \item \textbf{\gls{CER}} is the ratio of erroneous input that got fixed
  ($IF$) to any character typed during transcription, including $IF$
  \cite{soukoreff_metrics}.
  \begin{equation}\label{eq:cer}
    CER = \frac{|IF|}{|T| + |IF|}
  \end{equation}

  \item \textbf{\gls{UER}} is the ratio of erroneous input that was \textbf{not}
  fixed ($INF$) to any character typed during transcription, including $IF$
  \cite{soukoreff_metrics}.
  \begin{equation}\label{eq:uer}
    UER = \frac{|INF|}{|T| + |IF|}
  \end{equation}
\end{enumerate}

\subsubsection{Electromyography}
\label{sec:meas_emg}

In several other studies, in addition to the metrics mentioned so far, \gls{EMG}
data was captured to evaluate the muscle activity or applied force while typing
on completely different or modified hardware \cite{kim_typingforces,
  fagarasanu_force_training, gerard_audio_force, gerard_keyswitch, martin_force,
  rose_force, rempel_ergo, pereira_typing_test}. \gls{EMG} signals are captured
with the help of specialized equipment that utilize electrodes which are either
placed onto the skin above the muscles of interest (non-invasive) or inserted
directly into the muscle (invasive). The disadvantage of non-invasive surface
level electrodes is the lacking capability to capture the distinct signal of one
isolated muscle \cite{reaz_emg}. Nevertheless, because this type of electrode is
more likely to find acceptance among participants and is also easier to apply by
non-medically trained researchers, it finds wide adoption among studies
\cite{takala_emg}. To make \gls{EMG} data comparable across subjects, it is
necessary to conduct initial measurements where each individual participant is
instructed to first completely relax and then fully contract (\gls{MVC}) the
muscles of interest. These values are used to normalize further data obtained in
an experimental setting. The mean signal yielded by complete relaxation is
subtracted to reduce noise and the \gls{MVC} is used to obtain the individuals
percentage of muscle activity (\%MVC or EMG\%) during tests \cite{halaki_emg,
  takala_emg, rempel_ergo}. Muscles typically measured during typing exercises
are the \gls{FDS}, \gls{FDP} and \gls{ED}. The main function of the \gls{FDS}
and \gls{FDP} is the flexion of the medial four digits, while the \gls{ED}
mainly extends the medial four digits. Therefore, these muscles are primarily
involved in the finger movements required for typing on a keyboard
\cite{netter_anatomy, kim_typingforces, gerard_keyswitch, gerard_audio_force}.
A method frequently used to measure applied force is to place one or multiple
load cells under the keyboard \cite{gerard_keyswitch, rempel_ergo,
  bufton_typingforces}. Load cells are electronic components that are able to
convert applied force to an electrical signal. This signal usually gets
amplified by specialized circuits and then further processed by a micro
controller, computer or other hardware \cite{johnson_loadcell}.

\subsubsection{Subjective Metrics}
\label{sec:meas_sub}
Lastly, subjective metrics e.g., comfort, usability, user experience, fatigue
and satisfaction, are evaluated based on survey data collected after
participants used different input methods \cite{kim_typingforces,
  bell_pauseboard, bufton_typingforces, pereira_typing_test, iso9241-411}. In
their study, Kim et al. used a modified version of the \gls{KCQ} provided by the
\gls{ISO}, which is specifically designed to evaluate different keyboards in
terms of user satisfaction, comfort and usability \cite{kim_typingforces,
  iso9241-411}. This survey poses a total of twelve questions concerning e.g.,
fatigue of specific regions of the upper extremity, general satisfaction with
the keyboard, perceived precision and uniformity while typing, etc., which are
presented on a seven-point Likert scale \cite{iso9241-411}. Further, studies
concerning the usability and user experience of different text entry methods
used the \gls{UEQ} or \gls{UEQ-S} to evaluate the differences in those
categories \cite{nguyen_ueq, olshevsky_ueq, gkoumas_ueq}. While the full
\gls{UEQ} provides a total of 26 questions divided into six scales
(attractiveness, perspicuity, efficiency, dependability, stimulation and
novelty), the \gls{UEQ-S} only features 8 questions and two scales (pragmatic
and hedonic quality). Because of the limited explanatory power of the
\gls{UEQ-S}, it is recommended to only use it if there is not enough time to
complete the full \gls{UEQ} or if the participants of a study are required to
rate several products in one session \cite{schrepp_ueq_handbook}.

\begin{phga_sum*}
Measuring metrics related to data entry tasks can be performed with the help
several commercially available tools and equipment. Especially muscle activity
has to be measured with appropriate tools and accurate placement of the
electrodes is important to ensure quality results \cite{takala_emg, halaki_emg,
  kim_typingforces, gerard_keyswitch}. Metrics related to performance such as
\gls{WPM}, \gls{CER} and \gls{UER} are well defined and can be applied in almost
any experimental setup concerning the transcription of text
\cite{soukoreff_metrics, mackenzie_metrics}. In addition to the measured data,
questionnaires can help to gather subjective feedback about the keyboards and
thereby reveal differences that cannot be easily acquired by a device or formula
\cite{rowley_surveys}.
\end{phga_sum*}
\pagebreak
\subsection{Observer Bias and a Possible Solution}
\label{sec:bias}
As already discussed in Section \ref{sec:metrics}, it is common practice in
research related to typing to present a text that has to be transcribed by the
participant. Usually, the text was chosen by the researcher or already available
through the used typing test software. If the understandability of text is of
concern, the binary choice of―is understandable or not―made by the researcher
could lead to a phenomenon called the observer bias \cite{hrob_observer,
  berger_observer, angrosino_observer}. Thus, the text could potentially be too
difficult to understand for the participants if not evaluated with e.g. the
\gls{FRE} or other adequate formulas. Further, if there is previous knowledge
about the requested participants, the researcher could subconsciously select
text that is familiar to, or well received by some of the subjects and could
thereby conceivably influence the outcome of the study\cite{hrob_observer,
  berger_observer}. The same problem arises, if the typing test software already
provides such texts but the researcher has to select some of them for the
experiment. Furthermore, the difficulty of the provided texts should be verified
to ensure accurate results across multiple treatments. A possible solution to
this problem is crowdsourcing. Howe describes crowdsourcing as the act of
outsourcing a problem to a group of individuals that are voluntarily working
together to solve it \parencite[1-11]{howe_crowd_book} \&
\cite{howe_crowdsource, schenk_crowdsource}.

Observer bias can also occur while conducting the experiment when the researcher
has to give instructions to the subject. Therefore, it is important to treat
every participant equally by following a predefined procedure and minimize
unnecessary interaction where possible to further minimize the risk of bias
\parencite[674]{angrosino_observer}.

\begin{phga_sum*}
Summarizing, even seemingly arbitrary decisions or actions can have a potential
undesirable impact on the results of a study. If it is possible to implement
automated checks for the suitability of text e.g., a platform that verifies
submitted text based on \gls{FRE} scores, crowdsourcing could be used to
completely exclude the researcher from the text selection process and therefore
mitigate the risk of unwanted bias. In addition, the aspect of time in the
preparation phase of a study could be another factor to consider crowdsourcing
to acquire larger amounts of text with equal difficulty.
\end{phga_sum*}

\subsection{Influence of Actuation Force on Keyboard Use}
\label{sec:finger_force}
Section \ref{sec:kb_ks} discussed the differences of various keyswitch
models. One difference was the applied force, a keyswitch required to
activate. A study by Akagi tested the differences in performance and preference
across four visually identical keyboards with different keyswitches. The
keyswitches differed in actuation force and type. Two keyboards used tactile
keyswitches with 70.9\,g (\gls{KB} A) and 32.5\,g (\gls{KB} C) the other two
linear switches with 70.9\,g (\gls{KB} D) and 42.5\,g (\gls{KB} B). The (n=24)
subjects were required to type on each keyboard for 7 to 8 minutes where speed
and errors were recorded. The results showed, that \gls{KB} D (linear, 70.9\,g)
produced the lowest error rate followed by \gls{KB} A (tactile, 70.9\,g),
\gls{KB} C (linear, 42.5\,g) and \gls{KB} B (tactile, 35.5\,g). Further, the
difference in typing speed between the slowest (tactile, 70.9\,g) and fastest
(linear, 42.5\,g) keyboard was only 2.61\,\% and according to Akagi too small to be
significant in practical use. The study also revealed, that the preference for
neither of the four keyboards was significantly different
\cite{akagi_keyswitch}. A follow up survey by Akagi concerning the model of
keyboard typists would prefer to use in the future revealed, that 69\,\% of the 81
participating decided for a newly proposed keyboard with 56.7\,g resistance and
light tactile feedback \cite{akagi_keyswitch}. Further, a study by Loricchio,
where (n=16) participants typed on two identical keyboard models that only
differed in actuation force (58\,g and 74\,g), also yielded moderate differences in
typing speed. The keyboard with lower actuation force was 8.25\,\% faster and
preferred by 15 out of the 16 subjects compared to the keyboard featuring
keyswitches with higher actuation force \cite{loricchio_force_speed}. A study by
Hoffmann et al. even designed a keyboard that utilized small
electromagnets―instead of the typically used spring―to dynamically alter the
resistance of keys to prevent erroneous input by increasing the force required
to press keys that do not make sense in the current context of a word. This
design reduced the number of required corrections by 46\,\% and overall lowered
typos by 87\,\% compared to when the force feedback was turned off (n=12)
\cite{hoffmann_typeright}.

\begin{phga_sum*}
So far, studies concerning keyboards with uniform actuation force yielded
different results pertaining speed, but agreed that actuation force influences
the error rate during typing related tasks. To our best knowledge, there are no
studies that evaluated the effect of non-uniformly distributed actuation forces
across one keyboard on speed, accuracy, error rate or preference. This is why we
want to reevaluate the influence of actuation force on speed and determine if
keyboards with non-uniform actuation forces have a positive impact on all
metrics mentioned so far. The next section gives insights into why such
keyboards could make sense.
\end{phga_sum*}

\subsection{Strength of Individual Fingers}
As already mentioned in Section \ref{sec:mech_switch}, the force applied to a
keyswitch is the concern of multiple studies that evaluate the relation between
keyboarding and \gls{WRUED}. Further, multiple studies came to the conclusion
that there is a significant discrepancy in strength between individual fingers
\cite{bretz_finger, martin_force, baker_kinematics, dickson_finger}. Bretz et
al. found, that when participants squeezed an object between thumb and finger,
differences in applicable force between different fingers ranged from 1.6
\gls{N} up to 25.9 \gls{N} (n=16) \cite{bretz_finger}. Dickson and Nicolle
observed the effects of surgery on patients with rheumatoid hands. The pre and
post surgery force of finger flexion was recorded and the post surgery results
yielded a difference in flexion force, which is similar to the force required to
actuate a keyswitch, that ranged from 1 \gls{N} to 4 \gls{N}
\cite{dickson_finger}. Martin et al. measured applied average and peak force of
individual digits while typing on a keyboard (n=10). The measured differences
ranged from 0.10 \gls{N} to 1.49 \gls{N} for peak force and 0.01 \gls{N} to 0.08
\gls{N} for mean force \cite{martin_force}.

\begin{phga_sum*}
The goal of this thesis is to evaluate the possible advantages of keyboards with
non-uniform actuation forces. The fairly small difference of only 0.08 \gls{N} in mean
force applied to keyboards recorded by Martin et al. \cite{martin_force} but
rather big difference in finger strength measured by Bretz et
al. \cite{bretz_finger} could indicate that albeit the difference in strength,
all fingers have to apply equal force to generate a keypress because of the
uniform actuation force used in commercially available keyboards.
\end{phga_sum*}

\subsection{Summary}
\label{sec:lr_sum}
Keyboards are still the most commonly used input method for data entry to date
and so far the majority of keyboard users still operates non-alternative
keyboard designs. Thus, modifications that ideally could be implemented into
manufacturing processes of existing designs have to be explored, to ensure
availability and therefore adaption, which could help to reduce the risks of
\gls{WRUED}. One factor related to \gls{WRUED} is the actuation force of the
keyswitches \cite{bufton_typingforces, rempel_ergo, rempel_force,
  gerard_keyswitch}. Especially higher actuation forces have shown to be the
reason for discomfort in the upper extremity. On the other hand, higher
actuation forces also led to lower error rates while typing and therefore
enhance user satisfaction and performance \cite{gerard_keyswitch}. Therefore, a
desirable input method should offer enough resistance to prevent accidental key
presses but also reduce the stress induced on weaker fingers. With the help of
several methods to measure typing relate metrics such as muscle activity
(\gls{EMG}), error rates (\gls{CER} and \gls{UER}), typing speed (\gls{WPM}),
text readability (\gls{FRE}) and user satisfaction (\gls{UEQ} and \gls{KCQ}) it
is feasible to evaluate possible alternative input methods to the more
traditional keyboard. The availability of affordable surface level \gls{EMG}
measurement devices makes it possible for researchers that are not medically
trained to conduct non-invasive muscle activity measurements \cite{takala_emg}
In addition, load cells in combination with micro controllers are a reliable,
low-cost solution to visualize the strength of different fingers and monitor
applied forces while typing \cite{gerard_keyswitch, rempel_ergo,
  bufton_typingforces}. Although, the strength of individual fingers has already
been measured in different studies \cite{bretz_finger, martin_force,
  baker_kinematics, dickson_finger}, to our best knowledge, there are no
measurements concerning the maximum force each individual finger can apply in
different positions related to a key on the keyboard. Further, during our
research we only found one manufacturer of keyboards
(Realforce\footnote{\url{https://www.realforce.co.jp/en/products/}}), that
already offers models with variable actuation force. These keyboards feature two
types of keys and require less force towards the edges and more force towards
the middle \cite{realforce_topre}. We therefore try to provide a sensible
distribution of actuation forces across a non-uniformly equipped keyboard and
evaluate the possible advantages and disadvantages of such a design to encourage
other manufacturers to produce similar alternative keyboard designs.