This is also suggested by the fact that their analysis also amplifies the effect in the pre-article time window (a marginally significant effect at p=0.058), so it might just be ‘amplifying noise’ (by amplifying the between-item differences at the lower end of the cloze scale), and does not yield a significant effect when we used the original DUK baseline.
Crucially, the difference between significant and not-significant itself may not be significant ( Gelman and Stern, 2006 ), log-transformation does not yield higher t-values consistently across laboratories, does not necessarily improve model fit, and does not yield higher t-values or improve model fit in another large dataset (collapsed data from Ito et al., 2017a ; Nieuwland, 2016 ; Nieuwland and Martin, 2012 ; total N = 124).