The goodness of the model fit (
R2) differed significantly among individual datasets, with especially low values for subject S3, who occasionally showed some rather unsystematic data trends (particularly in the BY color condition). In contrast, the
R2 values for subject S6 are extraordinarily high (
R2 > 0.81), which is at least to some extent due to the fact that this subject only participated in
Experiment 1. Generally, for currently unknown reasons, model predictions were better for the data obtained in
Experiment 1 than for the data of
Experiment 2. This can be seen in
Figure 11, where the mean luster settings are plotted against corresponding conflict measures
C (data points belonging to
Experiment 1 are shown as red and those belonging to
Experiment 2 are shown as green dots in the diagrams). For the color condition BY, the relationship between the mean luster settings and the conflict measures calculated by our model can be described by a decelerating power function (with exponent
c < 1.0), which is also in agreement with our previous findings (
Wendt & Faul, 2022a). However, this relationship is nearly linear for the other color condition, RG (
Figure 11). But, also regarding the shape of the fitting curve, there were large differences among subjects. For the pooled data, we also tested a version of our model with a linear transducer for the global conflict measure
C in the form
a*C (i.e., without an additive constant). As can be seen in the corresponding rows of
Table 3 (S0 linear), this has some impact on the model parameters. However, predictability was not greatly reduced (with
R2 values of 0.6197 and 0.5595 for the RG and the BY color conditions, respectively), suggesting that a linear transducer would also be generally compatible with our chromatic model. Note that better predictions (with
R2 = 0.6222 in the RG condition and
R2 = 0.6143 in the BY condition) were obtained when a linear model with an additive constant was used. We already found a strong linear relationship between luster judgments and model predictions with our original model for achromatic stimuli based only on conflicts due to interocular ON–OFF pairings (with
R2 values up to 0.84) (
Wendt & Faul, 2020). However, the positive
y-intercept of the linear regression was difficult to interpret in terms of an interocular conflict mechanism, as it meant that luster was perceived to a significant degree even in stimuli for which the model predicted no conflict at all. The zero point is therefore an essential component of our model, which eventually motivated our improved version with additional parameters and the use of a nonlinear transducer function.