Comparing Performance Models for Bivariate Pointing Through a Crowdsourced Experiment.

Evaluation of a novel user-performance model’s fitness requires comparison with baseline models, yet it is often time consuming and involves much effort by researchers to collect data from many participants. Crowdsourcing has recently been used for evaluating novel interaction techniques, but its potential for model comparison studies has not been investigated in detail. In this study, we evaluated four existing Fitts’ law models for rectangular targets, as though one of them was a proposed novel model. We recruited 210 crowd workers, who performed 94,080 clicks in total, and confirmed that the result for the best-fit model was consistent with previous studies. We also analyzed whether this conclusion would change depending on the sample size, but even when we randomly sampled data from five workers for 10,000 iterations, the best-fit model changed only once (0.01%). We have thus demonstrated a case in which crowdsourcing is beneficial for comparing performance models.