bachelor-thesis/chap6/discussion.tex

\section{Discussion}
\label{sec:discussion}
In the following sections, we reiterate on our findings presented in the last
section and try to derive answers regarding our seven hypotheses and research
question \textit{``Does an adjusted actuation force per key have a positive
  impact on efficiency and overall satisfaction while using a mechanical
  keyboard?''}.

\subsection{Impact of Actuation Force on Typing Speed}
\label{sec:dis_speed}

Our main experiment yielded that there are differences in typing speed for both
metrics related to transcribed text we measured―namely \glsfirst{WPM} and
\glsfirst{AdjWPM}. Especially the keyboard with the lowest uniform actuation
force of 35\,g―\textit{Nyx}―performed worse than all other keyboards. In terms
of \gls{WPM}, \textit{Nyx (35\,g)} was on average 4.1\,\% slower than
\textit{Athena (80\,g)} and \textit{Aphrodite (50\,g)} and 4.8\,\% slower than
the adjusted keyboard \textit{Hera (35 - 60\,g)}. Similarly, for \gls{AdjWPM},
\textit{Nyx} was 4.3\,\% slower than \textit{Athena} and \textit{Aphrodite} and
4.9\,\% slower than \textit{Hera}. The 4\,\% to 5\,\% difference in \gls{WPM}
and \gls{AdjWPM} in our sample account for approximately two words per
minute. When extrapolated with the mean daily keyboard usage of 6.69 hours
reported by our participants, this difference would be as big as 803 words,
which when put into perspective, is equivalent to roughly two full pages of only
written content (11pt font size). Although, this specific example would assume
constant typing for 6.69 hours, it is still a useful estimate of the loss in
productivity under normal working conditions over the course of several
days. These differences in \gls{WPM} and \gls{AdjWPM} could be explained by the
higher error rates and thereby the loss of ``typing flow'' we discuss in the
next section. \gls{KSPS} reflects the raw input speed by including backspaces
and previously deleted characters. The reason we included \gls{KSPS} in our
analysis was to reveal possible differences in the physical speed participants
type on a keyboard and not to further asses speed in the sense of
productivity. We could not find any statistically significant differences in
\gls{KSPS} but saw a trend, indicating that subjects typed a bit slower (<
3\,\%) on \textit{Athena (80\,g)} compared to \textit{Aphrodite (50\,g)} and
\textit{Hera (35 - 60\,g)}. With the differences in metrics that are commonly
used to measure typing speed more closely related to productivity (\gls{WPM},
\gls{AdjWPM}) and the trends that indicate a slight difference in operating
speed we could have accepted our hypothesis. However, with the relation between
error rate and typing speed described in the next section and the thereby rather
indirect effect of the actuation force, we can only partially accept our
hypothesis that a difference solely in actuation force, has an impact on typing
speed.

\begin{phga_hyp*}[1 $\rightarrow$ \cmark\kern-1.1ex\raisebox{.7ex}{\rotatebox[origin=c]{125}{--}}]
  Actuation force has an impact on typing speed (efficiency - speed).
\end{phga_hyp*}

% During our telephone interviews 76\,\% of respondents would have preferred a
% keyboard with lighter actuation force.

% Our study tried to present the participant with a typing scenario that is as
% close to a typical text input situation as possible, by allowing but not
% enforcing the correction of erroneous input.

\subsection{Impact of Actuation Force on Error Rate}
\label{sec:dis_error}

As already briefly mentioned in Section \ref{sec:dis_speed}, measured error
rates like \glsfirst{UER}, \glsfirst{CER} and \glsfirst{TER} differed especially
between \textit{Nyx (35\,g)} and the other test keyboards. The statistical
analyses further revealed that \textit{Athena}, the keyboard with the highest
actuation force of 80\,g, produced on average 1\,\% less \gls{TER} than
\textit{Hera (35 - 60\,g)} and \textit{Aphrodite (50\,g)} and 3\,\% less than
\textit{Nyx (35\,g)}. Furthermore, \textit{Hera} and \textit{Aphrodite} both had
a 2\,\% lower \gls{TER} than \textit{Nyx}. Additionally to the quantitative
results, fourteen of the twenty-four participants also reported, that
\textit{Nyx's} light actuation force was the reason for many accidental key
presses. It further stood out that, as shown in Figure \ref{fig:max_opc_ter},
\textit{Athena} was the most accurate keyboard for 58\,\% of participants and
also more accurate than keyboard \textit{Own} for eleven of the
subjects. Overall, this concludes that a higher actuation force has a positive
impact on error rate.

\begin{phga_hyp*}[2 $\rightarrow$ \cmark]
  Higher key actuation force decreases typing errors compared to lower key
  actuation force (efficiency - error rate).
\end{phga_hyp*}

\textbf{Impact of \gls{TER} on \gls{WPM}}

The higher error rates and the possibility to correct erroneous input could have
also been a factor that led to lower \textit{WPM}. To evaluate the likelihood of
this additional relation we conducted a \gls{LRT} of fixed effects for our
linear mixed-effects model with two random effects (participant and first/second
typing test), fixed effect \gls{TER} and response variable \gls{WPM}. The
results of the \gls{LRT} ($\chi^2(1)$ = 110.44, p = 0.00000000000000022)
together with the trends of lower \gls{WPM} with increasing \gls{TER}, visible
in Figure \ref{fig:reg_ter_wpm}, suggest that the \gls{TER} indeed had an
impact on \gls{WPM}. This could have been because every time an error was made,
almost all participants decided to correct it right away. With a higher error
rate, this obviously leads to many short interruptions and an increased number
of characters that are not taken into account when computing the \gls{WPM}
metric.

\begin{figure}[H]
  \centering
  \includegraphics[width=1.0\textwidth]{images/reg_ter_wpm}
  \caption{Regression lines for the relation between \gls{TER} and \gls{WPM}.
    The trends indicate a decrease in \gls{WPM} with rising \gls{TER} and
    therefore the existence of a relation between the two metrics}
  \label{fig:reg_ter_wpm}
\end{figure}

\subsection{Impact of Actuation Force on Satisfaction}
\label{sec:dis_sati}

We tried to narrow down the rather broad term ``satisfaction'' to individual
categories that we, with the information gathered through our literature review
and telephone interviews, defined as necessary for a positive user experience
while using a keyboard \cite{giese_sati}. We decided for the following metrics
to evaluate whether or not a user experience with a keyboard that features
lighter actuation forces was more satisfactory:

\begin{table}[H]
  \centering
  \ra{1.0}
  \small
  \begin{tabular}{l}
  $\rightarrow$ Pragmatic scale from the \glsfirst{UEQ-S} \\
  $\rightarrow$ Score of the additional question \textit{``How satisfied have you been with this keyboard?''}\\
  $\rightarrow$ Results of the \glsfirst{KCQ}\\
  $\rightarrow$ Ranking of the keyboards during semi-structured interview\\
  $\rightarrow$ Ratio of positive and negative feedback for each keyboard during semi-structured interview\\
  \end{tabular}
\end{table}

\textbf{[\xmark] Pragmatic Scale (\gls{UEQ-S})}

As described in Section \ref{sec:res_ueqs}, we could not find statistically
significant differences for any of the test keyboards regarding the pragmatic
scale of the \gls{UEQ-S}. From visual assessment of the graph shown in Figure
\ref{fig:ueq_tkbs_res} we could conclude that there is a slight trend towards a
more positive rating for keyboards that utilized keyswitches with higher
actuation forces than \textit{Nyx (35\,g)}. This trend in the opposite direction
of our hypothesized outcome that lighter actuation force leads to more user
satisfaction, could be due to the longer familiarization time required for
keyboards with very light actuation force \cite{gerard_keyswitch}.

\textbf{[\xmark] Additional Question of Satisfaction with Keyboard}

The results deduced from the additional question \textit{``How satisfied have
  you been with this keyboard?''}, which were answered on a \glsfirst{VAS}
from 0 to 100 after both tying tests with a keyboard, suggested that \textit{Nyx
  (35\,g)}, the keyboard with the lightest actuation force and also
\textit{Athena (80\,g)} the keyboard with the highest actuation force, were rated
significantly worse than \textit{Aphrodite (50\,g)}. Additionally, \textit{Hera
  (35 - 60\,g)}, the adjusted keyboard, showed a trend towards a significantly
better rating than \textit{Nyx}. These results indicate that neither of the
keyboards with extreme actuation forces were perceived as a overwhelmingly
pleasant keyboard to use during our typing tests. This is further supported by
the visualization of the mean ratings in Figure \ref{fig:res_tkbs_sati} where
the average ratings for \textit{Aphrodite} and \textit{Hera} were approximately
10 points higher than those for \textit{Nyx} and \textit{Athena}.

\textbf{[\xmark] Keyboard Comfort Questionnaire (\gls{KCQ})}

For the \gls{KCQ} we found several statistically significant differences. For
questions related to effort or fatigue while operating a keyboard,
\textit{Athena (80\,g)} received significantly lower ratings than the other test
keyboards. Additionally to the measured differences in error rates discussed in
Section \ref{sec:dis_error}, we discovered that participants also perceived the
accuracy of \textit{Athena (80\,g)} and \textit{Aphrodite (50\,g)} higher
compared to \textit{Nyx (35\,g)}. Similarly to the results discussed in the last
paragraph, the scores of the two keyboards with extreme actuation forces,
\textit{Nyx (35\,g)} and \textit{Athena (80\,g)}, fluctuated quite a bit and, on
average both keyboards scored lower than \textit{Aphrodite (50\,g)} or
\textit{Hera (35 - 60\,g)} (Figure \ref{fig:kcq_tkbs_res}). Thereby, these
results do not indicate a clear trend towards enhanced user experience when
using keyboards with lower actuation forces.

\textbf{[\xmark] Post Experiment Ranking of All Keyboards}

The ranks in terms of favored test keyboard, provided by all twenty-four
participants during the post-experiment semi-structured interview, can be
observed in Figure \ref{fig:tkbs_ranking}. The results further support the
tendencies towards keyboards with medium actuation forces, that we already
observed in the last couple paragraphs.

\begin{figure}[H]
  \centering
  \includegraphics[width=0.8\textwidth]{images/tkbs_ranking}
  \caption{Rankings for only the test keyboards, gathered during the
    post-experiment interview. It was possible to rank two or more keyboards the
    same}
  \label{fig:tkbs_ranking}
\end{figure}

\textbf{[\xmark] Ratio of Positive and Negative Feedback}

Lastly, we analysed all recordings of the post-experiment interviews and
categorized the feedback given for each keyboard into positive and negative
responses. We then calculated a ratio of these responses, which can be seen in
Figure \ref{fig:ratio_interview}, to evaluate preferences towards specific
keyboards that could not be expressed by our participants through any other
supplied method during the experiment. Like all other factors we identified as
reasonable indicators for satisfaction, these ratios yielded, that neither
\textit{Athena (80\,g)} nor \textit{Nyx (35\,g)} received more positive than
negative feedback. It should be noted, that previous research has shown that
people tend to remember and process bad experiences more thorough than good
ones, which could be a reason for the increased number of negative feedback for
\textit{Nyx} and \textit{Athena}, but would also indicate a worse experience
with those two keyboards \cite{baumeister_bad}.

\begin{figure}[H]
  \centering
  \includegraphics[width=0.80\textwidth]{images/ratio_interview}
  \caption{The ration of $\frac{Positive Responses}{Negative Responses}$ during
    the semi-structured interview for all test keyboards}
  \label{fig:ratio_interview}
\end{figure}

\textbf{Conclusion}

Contrary to the responses of our preliminary telephone interview, where 76\,\%
of attendees preferred a keyboard with light actuation force, none of the
factors we defined as relevant for user satisfaction suggests that keyboards
with lower actuation force are more satisfactory to use than keyboards with
higher actuation force. Therefore, we have to fully reject our hypothesis. We
can conclude thought that keyboards with actuation forces in between those two
extremes (35\,g / 80\,g), are persistently perceived as more pleasant to use and
that ratings of keyboards with extreme actuation forces are highly influenced by
personal preference. The latter is indicated by the high fluctuation of almost
all responses regarding our evaluated factors.

\begin{phga_hyp*}[3 $\rightarrow$ \xmark]
  Keys with lower actuation force are perceived as more satisfactory to type
  with than keys with higher actuation force.
\end{phga_hyp*}

\subsection{Impact of Actuation Force on Muscle Activity}
\label{sec:dis_emg}
In contrast to other studies that suggested that actuation force has an impact
on muscle activity, we could not identify significant differences in terms of \%
of \glsfirst{MVC} for any of our \gls{EMG} measurements. Only a slight trend
that \textit{Nyx (35\,g)} produced the highest flexor \%\gls{MVC} for only
14\,\% of participants, could be interpreted as anecdotal evidence towards our
hypothesis that actuation force has an impact on muscle activity. Therefore, we
have to reject our hypothesis.

\begin{phga_hyp*}[4 $\rightarrow$ \xmark]
  Differences in actuation force influence muscle activity while typing.
\end{phga_hyp*}

%\subsection{Impact of an Adjusted Keyboard on Typing Speed, Error Rate and
%  Satisfaction}

\subsection{Implications for the Adjusted Keyboard}
\label{sec:dis_hera}

As discussed in the previous sections, there were no statistically significant
differences in terms of satisfaction for any of the test keyboards, including
our adjusted keyboard \textit{Hera}. Still, the rather unconventional design
choice of non-uniform actuation forces across a keyboard did not negatively
influence the satisfaction compared to \textit{Aphrodite}, which was often
perceived as equivalent to the participant's own keyboard. In fact,
\textit{Hera} was the keyboard with the most occurrences in the top three, tied
first place with \textit{Aphrodite} and was never ranked 4th place during the
post-experiment interview (Figure \ref{fig:tkbs_ranking}). Since \textit{Hera},
among others, utilized keyswitches with light actuation force (35\,g), the
satisfaction could improve during prolonged usage, because of the longer
familiarization period required by keyboards with lighter actuation forces
\cite{gerard_keyswitch}. Interestingly, participant \textit{I3Z4XI7H} who
reported a currently present disease of the left arm and wrist (Syndrome Sudeck,
complex regional pain syndrome (CRPS)), ranked Hera 30 points higher than all
other keyboards. \textit{I3Z4XI7H} also reported in the post-experiment
interview that \textit{Hera} was surprisingly pleasant to use and that pain was
significantly lower than with all other keyboards including
\textit{Own}. However, because of the nearly identical scores to
\textit{Aphrodite} in almost all categories, we have to reject our hypothesis
that an adjusted keyboard is more satisfactory to use than standard keyboards.

\begin{phga_hyp*}[7 $\rightarrow$ \xmark]
  An adjusted keyboard is perceived as more satisfactory to type with compared
  to standard keyboards.
\end{phga_hyp*}

Similarly, the resulting error rates measured for \textit{Hera (35 - 60\,g)}
were close to equal to the results of \textit{Aphrodite (50\,g)} and for speed
related metrics between those two keyboards only slight improvements while using
\textit{Hera} in \gls{WPM} (0.8\,\%), \gls{AdjWPM} (0.6\,\%) and \gls{KSPS}
(1\,\%)― that were not statistically significant―were recorded during our
experiment. It was still interesting to see, that \textit{Hera}, out of all four
test keyboards, was the fastest for eleven (45\,\%) out of the twenty-four
subjects and that albeit the usage of 30\,\% keyswitches\footnote{That were
  actually pressed during our typing tests} that required 35 - 40\,g actuation
force, which is similar to the actuation force of \textit{Nyx (35\,g)}, we did
not see comparably high error rates. Because of the lacking evidence, that an
adjusted keyboard produces less errors or supports the typist in achieving
higher typing speeds, we have to reject our two hypotheses regarding those
improvements.

\begin{phga_hyp*}[5 $\rightarrow$ \xmark]
  An adjusted keyboard improves typing speed compared to standard keyboards
  (efficiency - speed).
\end{phga_hyp*}

\begin{phga_hyp*}[6 $\rightarrow$ \xmark]
  An adjusted keyboard is perceived as more satisfactory to type with compared
  to standard keyboards.
\end{phga_hyp*}

Our experiment basically revealed that keyboards which utilized keyswitches
with actuation forces that were neither too light (35\,g) nor too heavy (80\,g),
generally outperformed keyboards which featured those extreme actuation
forces. In the following section, we elaborate on possible limitations of our
experimental design and future research that could be reasonable to further
investigate advantages and disadvantages of adjusted keyboard designs.







% ---