Table 3.
Results of All Simulations
| Sim | Model | SSE | BIC | AIC | Scaling | k1 | k2 | Outcome Salience | Ctx Salience | Cue1 Salience | Cue 2 Salience | Cue 3 Salience |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | RWM | 46.01 | 28.33 | 23.62 | 12.56 | 1.00 | NA | 0.21 | NA | 0.17 | NA | NA |
| WCM | 36.71 | 29.27 | 22.20 | 12.72 | 1.00 | −0.39* | 0.16 | NA | 0.22 | NA | NA | |
| CEM | 25.79 | 20.79 | 13.73 | 8.99 | 0.05 | 0.37* | 0.99 | NA | 0.99 | NA | NA | |
| 2 | RWM | 6913.70 | 102.83 | 99.99 | 82.22 | 0.69 | NA | 0.97 | NA | 0.97 | NA | NA |
| WCM | 3375.40 | 94.78 | 91.24 | 80.93 | 0.53 | 1.00 | 1.00 | NA | 0.70 | NA | NA | |
| CEM | 3963.00 | 97.19 | 93.65 | 82.21 | 0.23 | 1.00 | 1.00 | NA | 0.77 | NA | NA | |
| 3 | RWM | 0.99 | −17.47 | −19.89 | 1.64 | 0.65 | NA | 0.59 | 0.54 | 0.58 | NA | NA |
| WCM | 0.98 | −15.12 | −18.03 | 1.19 | 0.33 | −0.99 | 0.69 | 0.99 | 0.99 | NA | NA | |
| CEM | 0.35 | −27.50 | −30.41 | 2.03 | 0.42 | 0.99 | 0.44 | 0.07 | 0.53 | NA | NA | |
| 4 | RWM | 0.20 | −7.82 | −6.36 | 21.07 | 0.04 | NA | 0.56 | 0.77 | 0.06 | 0.41 | 0.09 |
| WCM | 0.06 | −13.74 | −12.07 | 2.83 | 0.78 | −1.00 | 0.99 | 1.00 | 0.33 | 0.03 | 1.00 | |
| CEM | 0.09 | −10.26 | −8.59 | 1.82 | 0.93 | 0.45 | 0.99 | 1.00 | 1.00 | 0.05 | 0.13 | |
| 5 | RWM | 1.14 | −1.04 | −1.60 | 1.16 | 0.83 | NA | 0.89 | 0.73 | 0.64 | 0.77 | 0.76 |
| WCM | 1.04 | 0.33 | −0.31 | 1.65 | 0.99 | −0.81 | 0.22 | 0.96 | 0.53 | 0.99 | 0.99 | |
| CEM | 0.06 | −22.88 | −23.51 | 21.89 | 0.10 | 0.99 | 0.11 | 0.60 | 1.00 | 0.04 | 0.22 |
Note: k1 represents reduced activation of an absent outcome. RWM = Rescorla-Wagner (1972) model. WCM = Within Compound Model (as described in Witnauer & Miller, 2011). CEM = Conjoint Error Model. For WCM and CEM k2 controls activation and eligibility, respectively, of an absent cue. In principle, simulations of the CEM could use k1 for both cues and outcomes. In practice, the addition of a second parameter for cues (k2) often allowed the model to provide a better fit based on BIC and AIC statistics. This reflects the trial-wise nature of the model being a poor match to the serial nature of a typical cue-outcome or CS-US pairing.
k2 values for Simulation 2 represent the best-fitting values for the k2 parameter in Group Uninformed (see Table 3 for details). The best-fitting values for Group Informed were 1.00 for both WCM and CEM, which captures the psychological intuition that informing subjects about the absent target would increase activation (WCM) or error (CEM). Notably in Simulations 1 and 2, SSE was based on only one set of predictions for both groups because the models anticipated no difference between the two groups prior to any experimental treatment (see Table 4). In Simulation 4, Cue 1 salience was used for both Cue A and Cue D, Cue 2 Salience was used for both Cue B and Cue C, and Cue 3 salience was used for X. In Simulation 4, Ctx Salience was used to represent the salience of two different contexts (inhibition training and target training), Cue 1 Salience was used for both A and C, Cue 2 Salience was used for B, and Cue 3 salience was used for X and Y.