Article Figures & Data
Figures
- Figure 1.
Experimental design. Top: Study 1. Trial structure of the dual- and single-alternative experiment. The block part consisted of the context-setting presentation of the risky alternative and was identical for both experiments. By contrast, the trial part differed between experiments in that both the risky and the safe alternative were shown in the trials of the dual-alternative experiment, whereas only the safe alternative was shown in the trials of the single-alternative experiment. In the control part, participants indicated which risky alternative was active during the preceding block by pressing the corresponding number key on the keyboard. Bottom: Study 2. Trial structure was as in Study 1, but each block was preceded by a secondary working memory part, represented by a context-independent sequence of numbers and letters, that had to be maintained in working memory and recalled from longer-term memory at the end of the block.
- Figure 2.
fNIRS channel positions. The channel setup covered parts of the lateral and medial PFC. For final analysis, we averaged over all channels. The Matlab toolbox NFRI (Singh et al., 2005) was used to estimate the Montreal Neurologic Institute coordinates corresponding to the international EEG 10–20 positions. Channel positions were visualized using BrainNet Viewer (Xia et al., 2013).
- Figure 3.
Behavior and basic neural responses in Study 1. Top: RT slopes. Plots illustrate the effects of risk context and working memory load on RT, separately for the dual- and single-alternative experiments. (Middle) Choice slopes. Plots illustrate the effects of context and working memory load on choices of risky options, separately for the dual- and single-alternative experiments. Bottom: Δ[tHb] slopes. Plots illustrate the context- and experiment-dependence of Δ[tHb] response slopes averaged over all channels, separately for the dual- and single-alternative experiments. Slopes are shown for the peak response (time interval 4–7 s after alternatives onset). Note that all slopes are shown in terms of the coefficients for the second polynomial that best fitted the data (in a least-squares sense). Error bars represent SEM. See Table 1 for statistics.
- Figure 4.
Adaptive coding predicted by different adaptive coding models. Shown are predictions of neural activation in the dual-alternative experiment, based on theoretical parameter values as adapted from the models described by Louie et al. (2011), i.e., the difference model (model 1), the basic divisive normalization model (model 2), the enhanced divisive normalization model (model 3), and the full divisive normalization model (model 4).
- Figure 5.
Model comparison. Plots illustrating the time-resolved AIC-differences (ΔAIC) averaged over all channels between models 1–4, separately for the two experiments. Positive values correspond to a relatively better fit of the last-named model. Model 4 (full divisive normalization model) showed a relatively better fit compared with the other models, i.e., the difference model (model 1), the basic divisive normalization model (model 2), and the enhanced divisive normalization model (model 3). Moreover, model 4 provided a better overall fit around the canonical hemodynamic response peak (4 s) in the dual- compared with the single-alternative experiment.
- Figure 6.
Correlation between models and observed data. Kernel smoothing function estimate (KS-density) of the correlation coefficients between the four predicted models and the observed data in the two experiments. Significant differences between the dual- and single-alternative experiments were assessed using two-sample Kolmogorov–Smirnov test (KS-statistic, p value), considering all 40 sliding time intervals (left) and only the peak response (right).
- Figure 7.
Model parameters. We estimated the parameters for model 4, where Amax was the maximum activity, σ was the response slope, and β was the baseline parameter. The ratio (Amax * β/σ) of the three parameters can be summarized as the predicted background activity (ABackground). See Table 2 for statistical analysis using ANOVA.
- Figure 8.
Slopes in Study 2. Top: RT slopes. Plots illustrate the addition of context-independent working memory load to the risk context and decision-related working memory dependence of RT, separately for the dual- and single-alternative experiments. Middle: Choice slopes. Plots illustrate the addition of context-independent working memory load to the risk context and decision-related working memory dependence of risky choice, separately for the dual- and single-alternative experiments. Bottom: Δ[tHb] slopes. Plots illustrate the addition of context-independent working memory load to the risk context and decision-related working memory dependence of Δ[tHb] response slopes averaged over all channels, separately for the dual- and single-alternative experiments. Note that all slopes are shown in terms of the coefficients for the second polynomial that best fitted the data (in a least-squares sense). Error bars represent SEM. See Tables 3 and 4 for statistics.
Tables
In this issue
