## Default lab data analysis choices Here's a page where we write down lab default data analysis choices. The goal is that this guides our practice and constitutes a "default registration." ### Data analysis + Reaction times are clipped at +/- 3 SDs, computed in log space + RTs are modeled in log space, not linear. + Mixed effects models always start with maximal random effects structure (e.g., from Barr et al.). + In cases of non-convergence, we prune interaction random slopes, first from items, then from subjects. Then we remove item random effects altogether. + We use the t = z approximation for computing p-values in mixed effects models. ### Children + We exclude children for bilingualism at < 75% reported English exposure on our form. + We also exclude children for parent/sibling interference, and for self-reported developmental disability (e.g., ID, SLI, ASD, ADHD). + We *don't* exclude premature infants in preschooler studies (e.g., age 2 and above). + When not otherwise specified, we use the 50/50 rule for excluding trials and children from eye-tracking studies: exclude the trial if > 50% of data is missing, exclude the child if > 50% of trials are missing.