26
beyond this critical point, the effect of view size vanishes. This ceiling appeared unexpectedly
early in the curve, i.e. when the view’s half-diagonal had hardly reached 50 pixels, which
corresponds to a view of 80x60 pixels. This finding presumably reflects the non-linear impact
of the current panning error on target approach while zooming-in. Since each subsequent
magnification step multiplies all visualized distances by a constant zoom factor, the current
error increases exponentially, making it risky for the user to tolerate excessive deviations
from the beacon. Thus it seems that only the few pixels that surround the target beacon count
when controlling the zooming-in phase.
6. CONCLUSIONS AND DIRECTIONS FOR FUTURE
WORK
This article extends Fitts’ (1954) classic pointing paradigm to include target acquisition in
multiscale electronic worlds. This requires introducing the scale variable and defining the
notion of multiscale pointing. We show that, in the context of zoomable interfaces, multiscale
pointing required view navigation. Our theoretical analysis shows that Fitts’ law still applies
to multiscale pointing, even though the types of movement involved differs markedly from
classic pointing. In addition, the derivation explicitly takes into account the size of the view,
which had previously been considered constant. Our two studies provide empirical support for
the new theoretical model.
The interaction between view size and task difficulty is of particular interest. For small views,
the effect of task difficulty, i.e., the slope of Fitts’ law, depends on view size: the smaller the
view, the steeper the slope, that is, bandwidth is proportional to view size. This has important
consequences for the design of user interfaces with small displays, such as PDAs, cell phones
and wrist-watch computers. The smaller the display, the greater the need for multiscale
navigation but at the same time the lower the navigation bandwidth. Therefore reducing
display size risks decreasing the usability of multiscale interfaces. The appearance in our data
of a ceiling effect at a rather low level of view size may mitigate this problem, but further
tests with additional navigation tasks are necessary to confirm it. This ceiling effect also
suggests that display magnification far above the standard size, despite its merits in other
respects (see Hascoёt, 2003, for an example), should have no noticeable effect on the
bandwidth of multiscale view pointing.
The fact that current multiscale pointing techniques have dramatically increased the range of
IDs that computer users can handle has allowed us to inquire about the validity of Fitts’ law
over a range of task difficulty far wider than the classic 2-10 bit range: our experiments
involved IDs up to 30 bits. However, multiscale navigation takes time. In Experiment 1, our
participants took up to 10 seconds to reach an extremely difficult target using pan-and-zoom,
currently the best multiscale technique. This suggests that further facilitating target
acquisition for very difficult cases will require the design of novel interaction techniques.
Over the last few years a number of Fitts’ law based techniques aimed at artificially
facilitating pointing have been described in the HCI literature (see Balakrishnan, this issue).
Many of these solutions are effective, but apply only to fixed-scale GUIs. The present
theoretical framework should help design new pointing-facilitation techniques for multiscale
interfaces and improve our understanding of multiscale navigation.