\section{Methodology and Structure of the Research Process}
\label{sec:methodology}
\subsection{Research Approach}
Because of the controversial findings about the impact of key actuation forces
on speed \cite{akagi_keyswitch, loricchio_force_speed} and the fact that
keyboard related work can increase the risk for \gls{WRUED} \cite{ccfohas_wrued,
  pascarelli_wrued}, we decided to further investigate possible effects of
different actuation forces and even a keyboard equipped with non-uniform
actuation forces on speed, error rate and satisfaction. To our best knowledge,
to this date, there is no published work about the influence of a keyboard with
non-uniform actuation forces on these metrics. Therefore, we first asked
seventeen people about their preferences, experiences and habits related to
keyboards to get a better understanding on what people might prefer as a
baseline for the design of the adjusted keyboard (keyboard with non-uniform
actuation forces) as well as to complement the findings obtained through our
literature review. Further, we collected information about available mechanical
keyswitches on the market. Additionally, we conducted a small preliminary
experiment with 6 subjects, where we measured the peak forces each individual
finger of the right hand was able to apply to distinct keys in different
locations. We then created the design for the adjusted keyboard based on those
measurements. Lastly, an experiment with twenty-four participants was conducted,
where we compared the performance and user satisfaction while using four
different keyboards, including our adjusted keyboard. Figure \ref{fig:s4_flow}
presents a brief overview of the consecutive sections.

\begin{figure}[H]
  \centering
  \includegraphics[width=1.0\textwidth]{images/section_4_flow}
  \caption{Overview of the topics covered in the following sections}
  \label{fig:s4_flow}
\end{figure}

\pagebreak

\subsection{Preliminary Telephone Interview}
\label{sec:telephone_interview}
Some of the studies we found that researched implications of actuation force on
speed, preference or other metrics were published between 1984 and 2010. That is
why we wanted to ascertain if and how, with the advance of technology in recent
years and especially the capabilities modern smartphones offer, keyboard usage
has changed. Further, we wanted to gather information about the preference of
key resistance, keyswitch type and experiences with \gls{WRUED}. Therefore, we
conducted a structured interview with seventeen volunteers (59\,\% females) via
telephone, from which the most important results are presented in Figure
\ref{fig:res_tel}. The age of the subjects ranged between 22 and 52 with a mean
age of 29 years. The professions of subjects were distributed among medical
workers, students, office employees, computer engineers and community
workers. The first question we asked was \textit{``Which keyboard in terms of
  actuation force would be the most satisfying for you to use in the long
  run?''}. Thirteen (76\,\%) out of the seventeen subjects mentioned, that they
would prefer a keyboard with light actuation force over a keyboard with higher
resistance. The next question \textit{``Have you ever had pain when using a
  keyboard and if so, where did you have pain?''} yielded, that 41\,\% of those
polled experienced pain at least once while using a keyboard. The areas affected
described by the seven who already experienced pain were the wrist
\underline{and} forearm (3 out of 7), wrist only (2 out of 7), fingers (1 out of
7) and forearm only (1 out of 7). The results for the third question
\textit{``Which keyboard are you currently using and for how many hours a day on
  average?''} were in line with the statements we found during our literature
review \cite{ergopedia_keyswitch, peery_3d_keyswitch}. Nine answered that they
use a notebook (scissor-switches, membrane), six stated that they use an
external keyboard with rubber dome switches and only two responded that they use
a keyboard featuring mechanical keyswitches. The average―self-reported―usage
ranged between half an hour and 10 hours with a mean of 4.71 hours. It is
important to note, that a study by Mikkelsen et al. found, that self-reported
durations related to computer work can be inaccurate
\cite{mikkelsen_duration}. The last question \textit{``Which tasks do you still
  prefer to perform with a keyboard rather than your mobile phone?''} revealed,
that all of the subjects preferred to use a keyboard when entering greater
amounts of data (emails, applications, presentations, calculations, research),
but also surprisingly 41\,\% preferred to use a keyboard to write instant messages
(chatting via Whatsapp Web\footnote{\url{https://web.whatsapp.com/}}, Signal
Desktop\footnote{\url{https://signal.org/download/}}, Telegram
Desktop\footnote{\url{https://desktop.telegram.org/}}).
\begin{figure}[H]
  \centering
  \includegraphics[width=1.0\textwidth]{images/res_telephone_interview}
  \caption{Most important results from the preliminary telephone interview}
  \label{fig:res_tel}
\end{figure}
\pagebreak
\subsection{Market Analysis of Available Mechanical Keyswitches}
\label{sec:market_forces}
To gather information about available actuation forces, we collected the product
lines of keyswitches for all well known manufacturers, namely
Cherry\footnote{\url{https://www.cherrymx.de/en/mx-original/mx-red.html}},
Kailh\footnote{\url{https://www.kailhswitch.com/mechanical-keyboard-switches/}},
Gateron\footnote{\url{http://www.gateron.com/col/58459?lang=en}},
Matias\footnote{\url{http://matias.ca/switches/}},
Razer\footnote{\url{https://www.razer.com/razer-mechanical-switches}} and
Logitech\footnote{\url{https://www.logitechg.com/en-us/innovation/mechanical-switches.html}}. Since
some of the key actuation forces listed on the manufacturers or resellers
websites were given in \gls{cN} and most of them in gram or gram-force, the values
were adjusted to gram to reflect a trend that is within a margin of ± 2\,g of
accuracy. The results shown in Figure \ref{fig:keyswitches_brands} are used to
determine the minimum, maximum and most common actuation force for broadly
available keyswitches. According to our findings, the lowest commercially
available actuation force is 35\,g ($\approx$ 0.34 \gls{N}) the most common one
is 50\,g ($\approx$ 0.49 \gls{N}) and the highest resistance available is 80\,g
($\approx$ 0.78 \gls{N}).

\begin{figure}[H]
  \centering
  \includegraphics[width=0.9\textwidth]{images/keyswitches_brands}
  \caption{Available actuation forces for keyswitches of major keyswitch manufacturers}
  \label{fig:keyswitches_brands}
\end{figure}

\subsection{Preliminary Study of Finger Strength}
\label{sec:meth_force}
To evaluate the impact of an adjusted keyboard\footnote{keyboard with
  non-uniform actuation forces} on performance and satisfaction we first needed
to get an understanding on how to distribute keyswitches with different
actuation forces across a keyboard. Our first idea was to use a similar approach
to the keyboard we described in Section \ref{sec:lr_sum}, where the force
required to activate the keys decreased towards the left and right ends of the
keyboard. This rather simple approach only accounts for the differences in
finger strength when all fingers are in the same position, but omits possible
differences in applicable force depending on the position a finger has to enter
to press a certain key.  To detect possible differences in peak force depending
on the position of the fingers, we conducted an experiment with six volunteers
(50\,\% females). Subject's ages ranged from 20 to 26 with a mean age of 24
years. The subjects were all personal contacts. Subjects professions were
distributed as follows: computer science students (3/6), physiotherapist (1/6),
user experience consultant (1/6) and retail (1/6). All Participants were given
instructions to exert maximum force for approximately one second onto the key
mounted to the measuring device described in Section
\ref{sec:force_meas_dev}. We also used a timer to announce when to press and
when to stop. We provided a keyboard to every participant, which was used as a
reference for the finger position before every measurement. To reduce order
effects, we used a balanced latin square to specify the sequence of rows (top,
home, bottom) in which the participants had to press the keys
\cite{bradley_latin_square}. Additionally, because there were only six people
available, we alternated the direction from which participants had to start in
such a way that every second subject started with the little finger instead of
the index finger. An example of four different positions of the finger while
performing the measurements for the keys \textit{Shift, L, I} and \textit{Z} can
be observed in Figure \ref{fig:FM_example}.

\begin{figure}[H]
  \centering
  \includegraphics[width=1.0\textwidth]{images/FM_example}
  \caption{Prototype of the force measuring device used to gather data about the
    maximum applicable force to a key with different finger positions. The
    positions for certain keys are simulated by aligning the wrist pad (left
    picture) to the scale of the device. The four different positions for the
    keys \textit{Shift, L, I, Z} (right pictures) are color coded according to
    the keys on the scale}
  \label{fig:FM_example}
\end{figure}

The results of the measurements are given in Table \ref{tbl:finger_force}. The
median of the means (15.47\,N) of all measurements was used to calculate the
actuation forces in gram for the keyswitches later incorporated in the layout
for the adjusted keyboard. We used Eq. (\ref{eq:N_to_g}) and
Eq. (\ref{eq:actuation_forces}) to calculate the theoretical gram values for
each measured keyswitch.

\begin{equation}
  \label{eq:N_to_g}
  GFR = \frac{50\,g}{M_{maf}} = \frac{50\,g}{14.47\,N} = 3.23 \frac{g}{N}
\end{equation}

\begin{equation}
  \label{eq:actuation_forces}
  AF_{key} = GFR * MAF_{key}
\end{equation}

With $M_{maf}$ the median of the means of applicable forces, $50\,g$ the most
commonly found actuation force on the market (Section \ref{sec:market_forces}),
$GFR_{key}$ the gram to force ratio, $MAF_{key}$ the median of applicable force
for a specific key and $AF_{key}$ the actuation force for that specific key in
grams.

An example where we calculated the theoretical actuation force for the \textit{P}
key can be seen in Eq. (\ref{eq:force_example}).

\begin{equation}
  \label{eq:force_example}
  AF_{P} = GFR * MAF_{P} = 3.23 \frac{g}{N} * 10.45\,N \approx 33.75\,g
\end{equation}

We then assigned each theoretical actuation force to a group that resembles
a spring resistance which is available on the market or can be adjusted to that
value. We matched the results from Table \ref{tbl:finger_force} to the groups
representing the best fit shown in Table \ref{tbl:force_groups}.


% Custom spring stiffness
% https://www.engineersedge.com/spring_comp_calc_k.htm
% https://www.eng-tips.com/viewthread.cfm?qid=198360


\begin{table}[H]
  \centering
  \ra{1.3}
  \begin{tabular}{?l^l^l^l^l^l^l^l^l^l^l}
    \toprule
    \textbf{Bottom Row} & \multicolumn{2}{c}{\emph{F5}} & \phantom{.} & \multicolumn{1}{c}{\emph{F4}} & \phantom{.} & \multicolumn{1}{c}{\emph{F3}} &  \phantom{.} &\multicolumn{3}{c}{\emph{F2}}\\
    \cmidrule{2-3}\cmidrule{5-5}\cmidrule{7-7}\cmidrule{9-11}
    \rowstyle{\itshape}
    Key                        & ↑     & -     && :     && ;     && M     & N     & B    \\
    \midrule
    \emph{Mean Force (N)}      & 11.23 & 10.84 && 14.22 && 15.34 && 16.38 & 15.60 & 14.36\\
    \emph{Actuation Force (g)} & 36.27 & 35.01 && 45.93 && 49.55 && 52.91 & 50.39 & 46.38\\
  \end{tabular}
  \begin{tabular}{?l^l^l^l^l^l^l^l^l^l^l}
    \\
    \textbf{Home Row} & \multicolumn{2}{c}{\emph{F5}} & \phantom{.} & \multicolumn{1}{c}{\emph{F4}} & \phantom{.} & \multicolumn{1}{c}{\emph{F3}} &  \phantom{.} &\multicolumn{2}{c}{\emph{F2}}\\
    \cmidrule{2-3}\cmidrule{5-5}\cmidrule{7-7}\cmidrule{9-10}
    \rowstyle{\itshape}
    Key                        & Ä     & Ö     && L     && K     && J     & H     &\\
    \midrule
    \emph{Mean Force (N)}      & 11.88 & 12.27 && 15.84 && 18.56 && 17.78 & 18.43 & \phantom{69.69}\\
    \emph{Actuation Force (g)} & 38.37 & 39.63 && 51.16 && 59.95 && 57.43 & 59.53 &\\
  \end{tabular}
  \begin{tabular}{?l^l^l^l^l^l^l^l^l^l^l}
    \\
    \textbf{Top Row} & \multicolumn{3}{c}{\emph{F5}} & \phantom{.} & \multicolumn{1}{c}{\emph{F4}} & \phantom{.} & \multicolumn{1}{c}{\emph{F3}} &  \phantom{.} &\multicolumn{2}{c}{\emph{F2}}\\
    \cmidrule{2-4}\cmidrule{6-6}\cmidrule{8-8}\cmidrule{10-11}
    \rowstyle{\itshape}
    Key                        & +     & Ü     & P     && O     && I     && U     & Z    \\
    \midrule
    \emph{Mean Force (N)}      & 10.80 & 10.70 & 10.45 && 14.34 && 17.95 && 17.00 & 16.80 \\
    \emph{Actuation Force (g)} & 34.88 & 34.56 & 33.75 && 46.32 && 57.98 && 54.91 & 54.26\\
    \bottomrule
  \end{tabular}
  \caption{Maximum force measurements for all digits of the right hand in
    different positions. The mean force of six participants is shown in the
    first row of each table and the resulting actuation force for the
    corresponding keyswitch in the following row. The columns indicate the label
    of the scale on the measuring device which can be seen in Figure
    \ref{fig:FM_example}. \textit{↑} stands for the shift key. \textit{F5} :=
    little finger, \textit{F4} := ring finger, \textit{F3} := middle finger,
    \textit{F2} := index finger}
  \label{tbl:finger_force}
\end{table}

\begin{table}[H]
  \centering
  \ra{1.3}
  \begin{tabular}{?l^c^c^c^c^c^c^c}
    \toprule
    \rowstyle{\itshape}
    \textbf{Spring Stiffness:} &              35\,g & 40\,g  & 45\,g  & 50\,g  & 55\,g  & 60\,g \\
    \midrule
    \emph{\textbf{F5:} Key (g)} & \centered{P&(33.75)\\Ü&(34.56)\\+&(34.56)\\-&(35.01)\\↑&(36.27)}& \centered{Ä&(38.37)\\Ö&(39.63)}&&&&&\\
    \midrule
    \emph{\textbf{F4:} Key (g)} &&& \centered{:&(45.93)\\O&(46.32)} &\centered{L&(51.16)}&&\\
    \midrule
    \emph{\textbf{F3:} Key (g)} &&&&\centered{;&(49.55)}&&\centered{I&(57.98)\\K&(59.95)}\\
    \midrule
    \emph{\textbf{F2:} Key (g)} &&&\centered{B&(46.38)}&\centered{N&(50.39)\\M&(52.91)}&\centered{Z&(54.26)\\U&(54.91)\\J&(57.43)}&\centered{H&(59.53)}\\
    \bottomrule
  \end{tabular}
  \caption{Categorization of theoretical actuation forces acquired with
    Eq. (\ref{eq:actuation_forces}), into groups of more commonly available
    stiffnesses of springs. The rows indicate which finger is used to press the
    key. \textit{F5} := little finger, \textit{F4} := ring finger, \textit{F3}
    := middle finger, \textit{F2} := index finger}
  \label{tbl:force_groups}
\end{table}

We simply mirrored the results of the right hand for keys operated by the left
hand and copied the values to keys which are out of reach without lifting the
hand. Finally, we created the adjusted keyboard layout that can be examined in
Figure \ref{fig:adjusted_layout}. This layout was used in our main experiment
where we compared it to four different keyboards with uniform actuation forces
which is discussed in more detail in the following section.

\begin{figure}[H]
  \centering
  \includegraphics[width=1.0\textwidth]{images/adjusted_layout}
  \caption{Adjusted keyboard layout based on the measurements conducted in this section}
  \label{fig:adjusted_layout}
\end{figure}

\subsection{Main User Study}
\label{sec:main_study_meth}
\subsubsection{Hypotheses}
\label{sec:main_hypotheses}
Based on the literature review and preliminary telephone interviews, we derived
the following hypotheses concerning the impact of actuation force on different
metrics related to performance and user experience to ultimately answer our
research question―\textit{``Does an adjusted actuation force per key have a positive
impact on efficiency and overall satisfaction while using a mechanical
keyboard?.''}

\begin{longtable}{p{0.3cm} p{0.5cm} p{13cm} p{0.5cm}}
  & \textbf{H1}	& Actuation force has an impact on typing speed (efficiency - speed). & \\
  \\
  & \textbf{H2}	& Higher key actuation force decreases typing errors compared to lower key actuation force (efficiency - error rate). & \\
  \\
  & \textbf{H3}	& Keys with lower actuation force are perceived as more satisfactory to type with than keys with higher actuation force. & \\
  \\
  & \textbf{H4}	& Differences in actuation force influence muscle activity while typing. & \\
  \\
  & \textbf{H5}	& An adjusted keyboard (non-uniform actuation forces) improves typing speed compared to standard keyboards (uniform actuation forces) (efficiency - speed).& \\
  \\
  & \textbf{H6}	& An adjusted keyboard decreases typing errors compared to standard keyboards (efficiency - error rate).& \\
  \\
  & \textbf{H7}	& An adjusted keyboard is perceived as more satisfactory to type with compared to standard keyboards. & \\
\end{longtable}

\subsubsection{Method}
\label{sec:main_method}
In our laboratory study, twenty-four participants were required to perform two
typing tests with each of the four keyboards provided by us and two extra typing
tests with their own keyboards as a reference. The four keyboards differed only
in actuation force and were the independent variable. The dependent variable
were, typing speed (\gls{WPM} and \gls{KSPS}), error rate (\gls{CER},
\gls{TER}), forearm muscle activity measured via \gls{EMG} and satisfaction
(preference, usability, comfort, post experiment semi structured interview and
ux-curves).

\subsubsection{Participants}
\label{sec:main_participants}
There were no specific eligibility criteria for participants (n=24) of this
study besides the ability to type on a keyboard for longer durations and with
all ten fingers. The style used to type was explicitly not restricted to
schoolbook touch typing to also evaluate possible effects of the adjusted
keyboard on untrained typists. All participants recruited were personal
contacts. 54\,\% of subjects were females. Participant's ages ranged from 20 to
58 years with a mean age of 29. Sixteen out of the twenty-four subjects (67\,\%)
reported that they were touch typists. Subjects reported the following keyboard
types as their daily driver, notebook keyboard (12, 50\,\%), external keyboard
(11, 46\,\%) and split keyboard (1, 4\,\%). The keyswitch types of those
keyboards were distributed as follows: scissor-switch (13, 54\,\%), rubber dome
(8, 33\,\%) and mechanical keyswitches (3, 13\,\%). We measured the actuation
force of each participants own keyboard. The resulting distribution of actuation
forces can be observed in Figure \ref{fig:main_actuation_force}. The
self-reported average daily usage of a keyboard ranged from 1 hour to 13 hours,
with a mean of 6.69 hours. As already mentioned in Section
\ref{sec:telephone_interview} it is important to note, that a study by Mikkelsen
et al. found, that self-reported durations related to computer work can be
inaccurate \cite{mikkelsen_duration}. All participants used the \gls{QWERTZ}
layout and therefore were already used to the layout used throughout the
experiment.

\begin{figure}[H]
  \centering
  \includegraphics[width=0.79\textwidth]{images/main_actuation_force}
  \caption{Distribution of actuation forces from participant's own
    keyboards. The colors represent the type of keyboard. \textit{EXT:} external
    keyboard, \textit{NOTE:} notebook, \textit{SPLIT}, split keyboard}
  \label{fig:main_actuation_force}
\end{figure}

\subsubsection{Experimental Environment}
\label{sec:main_environment}
All the experiments took place in a room normally used as an office. Chair, and
table were both height adjustable. The armrests of the chair were also
adjustable in height and horizontal position. The computer used for all
measurements featured an Intel i7-5820K (12) @ 3.600\,GHz processor, 16\,gB RAM
and a NVIDIA GeForce GTX 980 Ti graphics card. The operating system on test
machine was running \textit{Arch Linux}\footnote{\url{https://archlinux.org/}}
(GNU/Linux, Linux kernel version: 5.11.16). The setup utilized two 1080p (Full
HD, Resolution: 1920x1080, Refresh-rate: 144Hz) monitors were participants were
allowed to adjust the angle, height and brightness prior to the start of the
experiment. The only two applications that were used during the experiment were
the typing test application described in Section \ref{sec:gott} inside of the
\textit{Chromium}\footnote{\url{http://www.chromium.org/Home}} browser (Version:
v90.0.4430.93-r857950) and \textit{FlexVolt
  Viewer}\footnote{\url{https://www.flexvoltbiosensor.com/software/}} (Version:
0.2.15, Chrome App). The FlexVolt Viewer app was used to collect \gls{EMG} data
via a bluetooth dongle (\textit{Plugable USB 2.0 Bluetooth®
  Adapter}\footnote{\url{https://plugable.com/products/usb-bt4le/}}) from the
\textit{FlexVolt 8-Channel Bluetooth Sensor}. Because of the ongoing COVID-19
pandemic\footnote{\url{https://www.who.int/emergencies/diseases/novel-coronavirus-2019}},
we ensured proper ventilation of the room and all participants including the
researchers were tested with antigen tests prior to every appointment.

\subsubsection{Independent Variable: Keyboards}
\label{sec:main_keyboards}
Alongside the reference tests conducted with the participant's own keyboards, we
provided four keyboards which only differed in terms of actuation force
(Appendix \ref{app:equipment}). We decided to assign pseudonyms in the form of
Greek goddesses to the keyboards to make fast differentiation during the
sessions easier and reduce ambiguity. The pseudonyms for each keyboard and the
corresponding actuation force can be found in Table \ref{tbl:kb_pseudo}.

\begin{table}[H]
  \centering
  \ra{1.3}
  \begin{tabular}{?l^l^l^l}
    \toprule
    \rowstyle{\itshape}
    Pseudonym & Actuation Force && Description\\
    \midrule
    \textbf{Own}       & 35\,g - 65\,g & $\approx$ 0.34\,N - 0.64\,N & Participant's own keyboard (Figure \ref{fig:main_actuation_force})\\
    \textbf{Nyx}       & 35\,g & $\approx$ 0.34\,N & Uniform\\
    \textbf{Aphrodite} & 50\,g & $\approx$ 0.49\,N & Uniform\\
    \textbf{Athena}    & 80\,g & $\approx$ 0.78\,N & Uniform\\
    \textbf{Hera}      & 35\,g - 60\,g & $\approx$ 0.34\,N - 0.59\,N & Non-uniform / Adjusted (Figure \ref{fig:adjusted_layout})\\
    \bottomrule
  \end{tabular}
  \caption{Pseudonyms used for the keyboards throughout the experiment.}
  \label{tbl:kb_pseudo}
\end{table}

All keyboards used the standard ISO/IEC 9995 \cite{iso9995-2} physical layout
and provided keycaps representing the German \gls{QWERTZ} layout, which all
participants were already familiar with. All four keyboards used in the
experiment were
\textit{\gls{GMMK}}\footnote{\url{https://www.pcgamingrace.com/products/gmmk-full-brown-switch}}
equipped with \textit{Gateron} mechanical
keyswitches\footnote{\url{http://www.gateron.com/col/58459?lang=en}}. The order
in which participants would use the four keyboards during the experiment was
defined by a balanced latin square to reduce order effects. Additionally, the
mentioned reference tests with \textit{Own} were conducted at the start and end
of each session to detect possible differences in performance due to
exhaustion. The resulting groups used during the whole experiment were as
follows:

\begin{itemize}
  \item \textbf{Group 1:} \textit{Own $\rightarrow$ Hera $\rightarrow$ Athena $\rightarrow$ Nyx
    $\rightarrow$ Aphrodite $\rightarrow$ Own}
  \item \textbf{Group 2:} \textit{Own $\rightarrow$ Athena $\rightarrow$ Aphrodite $\rightarrow$ Hera
    $\rightarrow$ Nyx $\rightarrow$ Own}
  \item \textbf{Group 3:} \textit{Own $\rightarrow$ Aphrodite $\rightarrow$ Nyx $\rightarrow$ Athena
    $\rightarrow$ Hera $\rightarrow$ Own}
  \item \textbf{Group 4:} \textit{Own $\rightarrow$ Nyx $\rightarrow$ Hera $\rightarrow$ Aphrodite
    $\rightarrow$ Athena $\rightarrow$ Own}
\end{itemize}


\subsubsection{Experimental Design}
\label{sec:main_design}
\textbf{Preparation and Demographics}

The whole laboratory experiment was conducted over a total time span of three
weeks. Participants were tested one at a time in sessions that took in total
$\approx$ 120 minutes. Prior to the evaluation of the different keyboards, the
participant was instructed to read the terms of participation which included
information about privacy, the \gls{EMG} measurements and questionnaires used
during the experiment. Next, the participants filled out a pre-experiment
questionnaire to gather demographic and other relevant information e.g., touch
typist, average \gls{KB} usage per day, predominantly used keyboard type,
previous medical conditions affecting the result of the study e.g.,
\glsfirst{RSI}, \glsfirst{CTS}, etc. The full questionnaire can be observed in
Appendix \ref{app:gott}. Further, participants could adjust the chair, table and
monitor to a comfortable position.

\textbf{\gls{EMG} Measurements}

Since we measured muscle activity during all typing tests, electrodes were
placed on the \glsfirst{FDS}/\glsfirst{FDP} and \glsfirst{ED} of both
forearms. As already discussed in Section \ref{sec:meas_emg}, 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[650-653]{netter_anatomy}. We used ECG-Electrodes (Ag/AgCI/Solid Adhesive,
Pregelled, Size: 43mm) from TIGA-MED Deutschland
GmbH\footnote{\url{https://www.tiga-med.de/Diagnostik-Geraete/EKG-Elektroden-Zubehoer/EKG-Klebeelektrode-Festgel-50-Stueck-Pack}}.
To identify the correct locations for the electrodes, participants were
instructed to wiggle their fingers till contractions of the \gls{FDS}, \gls{FDP}
or \gls{ED} could be felt \cite{kim_typingforces}. A reference electrode was
placed next to the pisiform bone onto the dorsal side of the wrist. The
locations were then shaved and subsequently cleaned with alcohol before applying
the electrode. The distance between electrodes was 20mm. The correct placement
was then confirmed, by observing the data received by the \textit{FlexVolt
  8-Channel Bluetooth Sensor} in the \textit{FlexVolt Viewer} application while
the participant performed flexion and extension of the wrist. The
\textit{FlexVolt 8-Channel Bluetooth Sensor} used following hardware settings to
record the data: 8-Bit sensor resolution, 32ms \gls{RMS} window size and
Hardware smoothing filter turned off. To gather reference values
(100\,\%\gls{MVC} and 0\,\%\gls{MVC}), which are used later to calculate the
percentage of muscle activity for each test, we performed three
measurements. First, participants were instructed to fully relax the \gls{FDS},
\gls{FDP} and \gls{ED} by completely resting their forearms on the
table. Second, participants exerted maximum possible force with their fingers
(volar) against the top of the table (\gls{MVC} - flexion). Lastly, participants
applied maximum possible force with their fingers (dorsal) to the bottom of the
table while resting their forearms on their thighs (\gls{MVC} - extension). We
decided to also measure 0\,\%\gls{MVC} before and after each typing test and
used these values to normalize the final data instead of the 0\,\%\gls{MVC} we
retrieved from the initial \gls{MVC} measurements. A picture of all participants
with the attached electrodes can be observed in Appendix \ref{app:emg}.

\textbf{Familiarization with \glsfirst{GoTT} and the Keyboards}

Participants could familiarize themselves with the typing test application
(\gls{GoTT}) for up to five minutes with a keyboard that was not used during the
experiment. Further, representative of the other keyboard models used in the
experiment (\gls{GMMK}), participants could familiarize themselves with
Aphrodite (50\,g). Additionally, because of a possible height difference between
\gls{GMMK} compared to notebook or other keyboards, participants were given the
choice to use wrist rests of adequate height in combination with all four
keyboards during the experiment. If during this process participants reported
that an electrode is uncomfortable and that it would influence the following
typing test, this electrode was relocated and the procedure in the last
paragraph\footnote{\gls{EMG} Measurements} was repeated\footnote{Happened one
  time during the whole experiment}.

\textbf{Texts Used for Typing Tests}

As described in Section \ref{sec:gott}, we acquired ten, non-overlapping, texts
so that every keyboard could be tested twice. The texts were labeled T0\_1,
T0\_2, T1\_1, ..., T4\_1, T4\_2 and could be selected before each typing
test. The order of the texts did not change during the experiment. All texts had
almost identical \gls{FRE} scores (mean = 80.10, SD = 0.48).

\textbf{Questionnaires}

To receive feedback about several aspects that define a satisfactory user
experience while using a keyboard, we decided to incorporate two questionnaires
into our experiment. The first questionnaire was the \glsfirst{KCQ} provided by
\cite[56]{iso9241-411} and was filled out after each individual typing test
(\glsfirst{PTTQ}). The second survey, that was filled out every time the
keyboard was changed, was the \glsfirst{UEQ-S} \cite{schrepp_ueq_handbook} with
an additional question―\textit{``How satisfied have you been with this
  keyboard?''}―that could be answered with the help of a \gls{VAS} ranging from
0 to 100 (\glsfirst{PKQ})\cite{lewis_vas}. Due to the limited time participants
had to fill out the questionnaires in between typing tests (2 - 3 minutes) and
also because participants had to rate multiple keyboards in one session, the
short version of the \gls{UEQ} was selected \cite{schrepp_ueq_handbook}.

\textbf{Post Experiment Interview \& \Gls{UX Curve}s}

To give participants the chance to recapitulate their experience during the
whole experiment, we conducted a semi-structured interview, after all typing
tests were completed. We recorded audio and video for the whole duration of the
interviews and afterwards categorized common statements about each
keyboard.

Further, we prepared two different graphs were participants had to draw \Gls{UX
  Curve}s related to subjectively perceived typing speed and subjectively
perceived fatigue for every keyboard and corresponding typing test. The graphs
always reflected the order of keyboards for the group the current participant
was part of. Furthermore, before the interview started, participants were given
a brief introduction on how to draw \Gls{UX Curve}s and, that it is desirable to
explain the thought process while drawing each curve \cite{kujala_ux_curve}. An
example of the empty graph for perceived fatigue (group 1) can be seen in Figure
\ref{fig:empty_ux_g1}.

\begin{figure}[H]
  \centering
  \includegraphics[width=1.0\textwidth]{images/empty_ux_g1}
  \caption{Empty graph for participants of group 1 to draw an \gls{UX Curve} related
    to perceived fatigue during the experiment}
  \label{fig:empty_ux_g1}
\end{figure}

\textbf{Main Part of the Experiment}

Each subject had to take two, 5-minute-typing-tests per keyboard, with a total
of 5 keyboards, namely \textit{Own (participant's own keyboard)}, \textit{Nyx
  (35\,g, uniform), Aphrodite (50\,g, uniform), Athena (80\,g uniform)} and
\textit{Hera (35\,g - 60\,g, adjusted)} (Table \ref{tbl:kb_pseudo}). As
described in Section \ref{sec:main_keyboards}, the order of the keyboards
\textit{Nyx, Aphrodite, Athena} and \textit{Hera} was counterbalanced with the
help of a balanced latin square to reduce order effects. The keyboard
\textit{Own} was used to gather reference values for all measured metrics. Thus,
typing tests with \textit{Own} were conducted before (one test) and after (one
test) all other keyboards, to also capture possible variations in performance
due to fatigue. Participants were allowed, but not obligated to, correct
mistakes during the typing tests. The typing test application allowed no
shortcuts to delete or insert multiple characters and correction was only
possible by hitting the \textit{Backspace} key on the keyboard. The
\textit{Capslock} key was disabled during all typing tests, because there was
only visual feedback in form of coloring of correct and incorrect input and no
direct representation of entered characters (Figure \ref{fig:gott_colorblind}),
which could have led to confusion when the \textit{Capslock} key was activated
by accident.

\subsection{Summary}
\label{sec:meth_summary}
The preliminary telephone interview and the market analysis of available
mechanical keyswitches allowed us to gather profound information concerning
user's preferences and availability of hardware components. Additionally, the
preliminary study, where we measured the maximum applicable force onto a
keyswitch for each finger of the right hand in different positions, yielded
necessary data for the design of the adjusted keyboard layout. Throughout the
main user study, where we compared five different keyboards, we were able to
obtain various qualitative and quantitative data regarding performance and
satisfaction. The statistical evaluation of this data will be presented in the
next sections.