We used sum-of-sine data to fit the parameters of a modified McRuer Crossover model (McRuer and Krendel, 1974). (A) In the most flexible setting, there are a total of eight free parameters, allowing for patients and controls to have completely independent model parameters. (B) In the model fitting and model selection process, we tested all possible variations of the model structure, for a total of 2667 model configurations. For example, some model structures yoke parameters together or eliminate parameters, and some yoke variables together across patient and control models. Each row illustrates an individual model configuration. Variables with the same color within that row are yoked together for that model configuration and white indicates that the specified variable was removed from the model. The table includes the following: the best model structures for all possible numbers of free parameters (Best 1–8); the hypothesized model structure (Hypothesis, shown in C and D); two special pure delay (D1-2) model structures. Best 1 is a pure gain model structure. Best 2 is a pure delay (D3) model structure. The model names are shown in the left column. Lowest Err means lowest Leave-One-Out error. Lowest Var means lowest model inconsistency. (C) Results of model selection procedure in frequency domain for the models in B. The models in red dots and blue asterisk, highlighted in the red circle, provide the best trade-off between minimizing model inconsistency (i.e. minimizing parameter variability and over-fitting) and minimizing error (improving model fit). The top four models are labeled 4 Lowest Err, 4 Lowest Var, 5 Lowest Err and 5 Lowest Var. These top models had nearly equivalent performance according to the model selection criteria. (D) Time domain validation. As in (C), we plotted error versus model inconsistency, but here we used the time-domain error (see Materials and methods). Critically, the time-domain data were not used directly for model fitting, and yet the top models in left bottom corner of the frequency domain are also the top models in the time domain, validating that our frequency-domain model selection process selected the models that were also best in the best time-domain.