Corrigendum: Reading Single Words Aloud With Monocular Presentation: The Effect of Word Frequency

Serje Robidoux, Derek Besner
2018 Frontiers in Communication  
In the original article, there was an error. An error in the script used to analyse the experiment in Robidoux and Besner (2018) led to an overstatement of the directional evidence against the Jainta et al. (2017) account for that data. When adjusting the general Bayes factor for the directional prediction made by Jainta et al.'s account, we stated that ". . . the evidence favoring the null model over the Jainta et al. account increases from 12.6 to 46.0 times, " where in fact the evidence
more » ... ases to 23.0 times. Reasoning and conclusions for the study are not affected by the correction. Revised analysis scripts have been uploaded to the original location (https://osf.io/t9dux/), and the original analysis script contains a prominent warning. A correction has been made to The Experiment, Results, Paragraph 1: The model predicted means are summarized (along with percentage errors) in Table 1 . The experiment provided strong evidence for the null model (no interaction) over the interactive alternative model (BF 01 = 12.6 ± 2.36%). That is, the evidence strongly favored the view that the word frequency effect was equivalent in both the monocular and binocular reading conditions. In fact, from Table 1 it is apparent that if there were an interaction, it would go in the wrong direction-monocular reading produced a slightly larger word frequency effect. The implication is that the evidence for the form of the interaction reported by Jainta and colleagues is specific to their eye movement paradigm rather than general to other measures (such as reading aloud, as here). When the Bayes factor is corrected for this directional prediction, the evidence favoring the null model over the Jainta et al. account increases from 12.6 to 23.0 times 1 . The authors apologize for this error and state that this does not change the scientific conclusions of the article in any way. The original article has been updated. 1 The BayesFactor package in R does not allow directly specifying a one-tailed prior. To correct for the direction of the predicted effects, we followed Richard Morey's advice here: http://bayesfactor.blogspot.com/2015/01/multiple-comparisonswith-bayesfactor-2.html and adjusted the BF 10 for the proportion of the posterior distribution that was consistent with the Jainta et al. data. See the scripts and data for details.
doi:10.3389/fcomm.2018.00033 fatcat:lrwztuevinb7tc7c5kqrx44p7e