Modeling and predicting pointing errors in two dimensions

Abstract

Recently, Wobbrock et al. (2008) derived a predictive model of pointing accuracy to complement Fitts' law's predictive model of pointing speed. However, their model was based on one-dimensional (1-D) horizontal movement, while applications of such a model require two dimensions (2-D). In this paper, the pointing error model is investigated for 2-D pointing in a study of 21 participants performing a time-matching task on the ISO 9241-9 ring-of-circles layout. Results show that the pointing error model holds well in 2-D. If univariate endpoint deviation (SD x ) is used, regressing on N=72 observed vs. predicted error rate points yields R 2 =.953. If bivariate endpoint deviation (SD x,y ) is used, regression yields R 2 =.936. For both univariate and bivariate models, the magnitudes of observed and predicted error rates are comparable.

The effect of time-based cost of error in target-directed pointing tasks

“…Error models based on Fitts' law [23,34,35] allow for predicting error rates based on parameters of a Fitts' law task. Meyer et al [23] formulated an error model which relates the user's error rate and the model's predicted movement time.…”

“…Meyer et al [23] formulated an error model which relates the user's error rate and the model's predicted movement time. Similarly, Wobbrock et al proposed a 1-dimensional [34] and 2-dimensional [35] target-directed pointing error models that relate the user's effective movement time with the user's error rate. The explicit term for movement time in those two models [34,35] …”

“…In order to calculate P(E) we use Wobbrock et al's [35] error model for pointing in 2 dimensions, which is derived from the Fitts' law equation and predicts the error rate given the user's movement time:…”

“…Our model is not intended to replace Fitts' law. Instead, it builds on top of Fitts' law [9,20] in conjunction with Wobbrock et al's error model [34,35], to calculate an optimal movement time the user should attempt to acquire a target at, taking into account the time-based penalty associated should the user miss.…”

The effects of task dimensionality, endpoint deviation, throughput calculation, and experiment design on pointing measures and models

“…FittsStudy logs all pointing trials in a session in XML, from which it builds spreadsheets for statistical analysis. Although Fitts' law evaluation tools have been described in the literature [5,22,25], FittsStudy offers more features than previous tools, providing, among other things, trial browsing and zooming tools, submovement profile graphs, results from both endpoint deviation and throughput calculations, path analysis measures [17], a metronome for time-matching temporal-precision studies (e.g., [23]), and models of movement times and error rates [28,29]. Future work in FittsStudy includes offering not just the vertical ribbon and ring-of-circles tasks, but also random target fields (e.g., [6]), and crossing and steering tasks [1].…”