Using GAMMs to model trial-by-trial fluctuations in experimental data: More risks but hardly any benefit