Advanced modeling of selection and steering data: beyond Fitts' law. Issue 94 (October 2016)

• Record Type: Journal Article
• Authors: Nieuwenhuizen, Karin; Martens, Jean-Bernard
• Is Part Of: International journal of human-computer studies. Issue 94(2016)
• Page Start: 35
• Page End: 52
• Publication Date: 2016-10
• Subjects: Fitts' law -- Power-law models -- Throughput -- Multi-model selection -- Multi-level models -- Maximum likelihood; Human-machine systems -- Periodicals; Systems engineering -- Periodicals; Human engineering -- Periodicals; Human engineering; Human-machine systems; Systems engineering; Periodicals; Electronic journals; 004.019
• DOI: 10.1016/j.ijhcs.2016.03.009
• Languages: English
• ISSNs: 1071-5819

Abstract

Abstract: Human performance in selection tasks is frequently described by Fitts' law, which states that the average time needed to move to a target and select it is linearly related to the index of difficulty for the task, which is the logarithm of the task characteristic C = A / W, where A is the target distance and W is the target width. The coefficients in this linear relationship can vary across interaction conditions, such as when using distinct interaction devices, and can for instance be used to define throughput, which is a standardized measure to compare interaction conditions. Although Fitts' law has proven very useful in the field of human–computer interaction (HCI) over the past 50 years, there are several issues with Fitts' law that argue in favor of more advanced statistical modeling of experimental data. More specifically, we propose two generalizations of Fitts' law. The first generalization is not to limit the data analysis to average movement times, but to consider the distributions of the observed times instead. The second generalization is to extend Fitts' law to a more general relationship between task characteristics and index of difficulty and to use this generalized model to come up with additional measures that can be used alongside throughput. We use data from an experiment with both selection and tracing tasks to illustrate the proposed analysis method. The primary goal of the experiment is to compare the task performance in four experimental conditions, corresponding to all combinations of interaction device (mouse or pen) and target orientation (horzizontal/vertical versus oblique movements). First, we show that non-linear transformations on the measured task completion times are indeed advised to resolve problems with the normality and homoscedasticity of the data, especially in case of the tracing task. Second, we show that in case of the selection task the data supports a linear relationship between the logarithm of the task characteristic C and the logarithm of the movement time, which corresponds to a power-law between movement time and task characteristic, an alternative to Fitts' law that has previously been proposed by several authors. In case of the tracing task, the data supports a power-law function in between a linear and a logarithmic function. We conclude by demonstrating how multiple performance measures can be used simultaneously when comparing interaction conditions. Abstract : Highlights: Two generalizations of Fitts' law: modeling data distributions and modified index of difficulty Multi-model selection is used to compare alternative models for the observed data. Advanced data modeling accomplishes closer data approximation and higher contrasts between conditions. Performance rate measure includes the form of speed–accuracy relationship.