2016/01/08 by R. Harald Baayen, Jacolien van Rij, Baayen, R. Harald +5
Computer Science · Decision Sciences · #Applications (stat.AP) #Bayesian Modeling and Causal Inference #FOS: Computer and information sciences #Multi-Criteria Decision Making #Rough Sets and Fuzzy Logic
paper · pdf · doi:10.48550/arxiv.1601.02043
openalex publication_date 2016/01/08 · openalex created_date 2022/08/15 · openalex updated_date 2026/07/28
A problem that tends to be ignored in the statistical analysis of\nexperimental data in the language sciences is that responses often constitute\ntime series, which raises the problem of autocorrelated errors. If the errors\nindeed show autocorrelational structure, evaluation of the significance of\npredictors in the model becomes problematic due to potential anti-conservatism\nof p-values. This paper illustrates two tools offered by Generalized Additive\nMixed Models (GAMMs) (Lin and Zhang, 1999; Wood, 2006, 2011, 2013) for dealing\nwith autocorrelated errors, as implemented in the current version of the fourth\nauthor's mgcv package (1.8.9): the possibility to specify an ar(1) error model\nfor Gaussian models, and the possibility of using factor smooths for\nrandom-effect factors such as subject and item. These factor smooths are set up\nto have the same smoothing parameters, and are penalized to yield the\nnon-linear equivalent of random intercepts and random slopes in the classical\nlinear framework. Three case studies illustrate these issues.\n