Where in the RT distribution do cues act? A distributional (ex-Gaussian) reanalysis of arrow, gaze and colour cueing in Schmitz et al. (2024), Experiments 1 and 2

Abstract

Secondary analysis of two public cueing datasets (Schmitz et al., 2024). Experiment 1 uses non-predictive arrow and face cues. Experiment 2 uses 100% counter-predictive arrow, face and colour cues. The question is where in the RT distribution cues act: a shift of the whole distribution (the ex-Gaussian location mu), a change in spread (sigma), or a change in the slow tail (nu). We also ask whether arrow and face cues differ, whether uncued targets show a cost, and whether cueing effects differ for errors. All models are distributional GAMLSS regressions on trial-level RT with accuracy as a predictor and full by-participant random slopes. The registration fixes the fitting ladder (independent then compound-symmetry random effects, then dropping accuracy if error trials are too sparse), the equivalence bounds, the adequacy check, and a shifted-Wald fallback. Expected outcomes: in Exp 1 a mu shift for cued targets with no tail change; in Exp 2 a smaller benefit for faces than for arrows, which we do not expect to be specific to the tail. Exp 1 is analysed blind apart from published statistics. Exp 2 is registered after viewing diffusion-model fit summaries, which are disclosed. Narrative summary (contents and difference from prior registrations): This registration contains the full analysis plan for ex-Gaussian distributional models of Schmitz et al. (2024) Experiments 1 and 2: hypotheses D1–D6 with blind or informed tags, the dependent variables, exclusions, the model formulas, the fitting ladder, inference rules, and a decision table (G1–G5) covering convergence, non-convergence, adequacy failure, recoverability and participant exclusions. The attached files are the analysis code, the patched shifted-Wald family, and this plan as a PDF. It differs from the earlier registration (https://osf.io/9qn8p), which covered only Experiment 2 and a diffusion-model family. That family failed its registered adequacy check, so it was abandoned. The earlier registration is not edited and its outcome is noted here. This registration adds Experiment 1, switches to a distributional model of correct and error trials with accuracy as a predictor, and discloses prior knowledge of the Exp 2 diffusion fits.