A novel method for detecting the onset of experimental effects in visual world eye-tracking

Determining the onset of experimental effects in timeseries data is central to psycholinguistics, because onset timing constrains accounts of the timecourse of different types of information during language processing. Several onset detection methods have been proposed for the visual world paradigm. An increasingly adopted approach is the bootstrap-based method of Stone et al. (2021), but its statistical properties have not yet been formally evaluated. We address this gap with two simulation studies, and show that the bootstrap-based method produces delayed onset estimates, has poor confidence interval coverage, and can inflate Type I error rates. We propose a novel onset detection method based on generalised additive mixed models (GAMMs) and posterior simulation. This method seamlessly integrates timecourse modelling with onset estimation, and supports the direct comparison of onset times across experimental conditions and participant groups. Our simulations show that the GAMM-based method has low bias and yields well-calibrated confidence intervals. We demonstrate its practical use by reanalysing two published visual world datasets representing common experimental designs, and provide a user-friendly R package to facilitate adoption. By enabling more precise measurement of effect timing in the visual world paradigm, our method can contribute to advancing theories of prediction and language comprehension.