Determining the onset of experimental effects in time series data, such as those obtained in eye-tracking and EEG studies, is critically important in psycholinguistics because it allows evaluating accounts of when different types of information become available during language processing. In this talk, I examine methods for characterising the timecourse of experimental effects, with a focus on data from the visual world paradigm.

One increasingly used approach is the bootstrap-based onset detection method proposed by Stone et al. (2021), but its statistical properties have not yet been formally evaluated. In two simulation studies, we found that this method has poor coverage and exhibits inflated Type I error rates. To address these shortcomings, we propose a novel method based on generalised additive mixed models (GAMMs) that seamlessly integrates onset detection with timecourse modelling.

We illustrate the new method by reanalysing two visual world datasets representing common experimental designs. In our simulations, the GAMM-based approach yielded estimates with low bias and well-calibrated confidence intervals.

Our method can also be extended to estimate onset timing for individual participants, opening the way to the study of individual differences in language processing. In principle, it could also be adapted to other techniques that yield time series data, such as EEG or pupillometry. We believe this method is a valuable addition to the psycholinguistic analytical toolkit and provide a user-friendly R package to facilitate its use.