Error-rate Prediction for Mouse-based Rectangular-target Pointing with no Knowledge of Movement Angles

TL;DR

This paper develops models to predict error rates in rectangular-target pointing tasks based solely on target dimensions, without prior knowledge of movement angles, validated through crowdsourced experiments.

Contribution

It introduces models that accurately predict error rates using only target size, enabling better UI design without knowing movement angle distributions.

Findings

Models achieve R^2 > 0.81 in prediction accuracy.

Prediction errors are less than 1.3% MAE.

Models generalize well to untested target sizes.

Abstract

In rectangular-target pointing, movement angles towards targets are known to affect error rates. When designers determine target sizes, however, they would not know the frequencies of cursor-approaching directions for each target. Thus, assuming that there are unbiasedly various angles, we derived models to predict error rates depending only on the target width and height. We conducted two crowdsourced experiments: a cyclic pointing task with a predefined movement angle and a multi-directional pointing task. The shuffle-split cross-validation with 60% training data showed R^2 > 0.81, MAE < 1.3%, and RMSE < 2.1%, suggesting good prediction accuracy even for predicting untested target sizes when designers newly set UI elements.