Understanding predictability with LMM and cluster-based permutation test: a new approach in multiple comparison issues on LMM

BRUNO BIANCHI; DIEGO E. SHALOM; JUAN E. KAMIENKOWSKI

Buenos Aires

Congreso; LABMAN; 2017

As in almost every daily visual task, the brain generates a prediction on the forthcoming stimuli. In reading, this prediction is usually operationalized as the Predictability, the probability of knowing a future word before reading it. However, this predictions could be built on different factors depending on the stimuli, such as semantic relations with the context or memory retrieval of known sentences. In the the present study, we aimed to separate these contributions using proverbs and common sentences.Classically, EEG data is analyze by averaging trials within conditions, on a set of chosen electrodes and time samples, losing information about the differences between stimulus and trial execution. This issue was overcome by designing experiments carefully, keeping the possible values of the variables or conditions low (i.e. increasing the number of trials to be averaged). Conversely, behavioral experiments is moving towards more natural and complex scenarios. Particularly, in the field of reading and eye movements (EM), new statistical methods, such as Linear Mixed Models (LMM), allow researchers to explore the several variables usually involved in these natural stimuli. However, the EEG poses a new issue: while EM is usually one-dimensional measure (fixation time), EEG data is multidimensional (spatial-temporal), giving rise to a multiple comparisons problem. The cluster-based permutations test is a powerful and widely accepted method to face this kind of multiple comparison problem.We combined LMM for each electrode and time-point and a cluster-based permutation approach to correct for multiple comparisons. Here, we compare two different resampling methods with a classical correction criteria (Bonferroni). Using the former procedures we observed the classical predictability effect (N400), a conspicuous late effect of the word position in sentence and a sparse effect on the sentence type (proverb vs non-proverb), which is in line with our hypothesis of different components of predictability.