25
Exp. 1
Figure 23. Bandwidth for a broad range of view sizes. Error bars represent α=.05 confidence limits based
on between-participant variability. Note that the sign of the y variable (the slope of the IDt vs. time
function) has been inverted.
5.3.2. Experiment 2b: Lower Range of View Size
Using a new sample of volunteers, we replicated Experiment 2a with a lower range of view
sizes to check whether the variations of V below 40 pixels does indeed influence bandwidth
and, if yes, to evaluate whether or not the proportionality relation stated by Equation 6 holds.
The task, the levels of ID (14 and 17 bits), and the design were the same. The only change
was that the levels of V were now 5, 7, 10, 15, 20, and 40 pixels.
Results and Discussion
Exp. 2
Bandwidth
(bit/s)
Figure 24. Bandwidth for a lower selection of view sizes.
The results, illustrated in Figure 24, confirmed our suspicions. Not only did the bandwidth
decrease dramatically as V was reduced below 20-40 pixels but the shape of the curve was
consistent with our hypothesis that the relationship starts as a proportionality. Evaluated over
the leftmost three and four points of the plot, the r2 for fitting a straight line was fairly high in
all six participants (mean±sd = .988±.009 and .913±.046, respectively) and the intercept small
enough to give credit to the proportionality hypothesis (-0.82±0.32 and 0.03±0.34,
respectively).
Summary
Experiment 2 supports the hypothesis that in multiscale view pointing (1) the bandwidth of
the interaction varies proportionally with view size up to a certain critical point and (2)