Table 1. Choice predictors.
| P (B chosen) | Two groups |
Continuous w |
||
|---|---|---|---|---|
| Two or more gazes | Exactly two gazes | Two or more gazes | Exactly two gazes | |
| Intercept | −2.49*** | −0.06 | −2.43*** | 0.01 |
| (0.29) | (0.44) | (0.30) | (0.45) | |
| QB−QA | 3.72*** | 3.92*** | 3.66*** | 3.89*** |
| (0.41) | (0.48) | (0.40) | (0.51) | |
| B chosen at t−1 | 1.68*** | 1.85*** | 1.69*** | 1.85*** |
| (0.27) | (0.33) | (0.26) | (0.33) | |
| First gaze on B | 1.20*** | 1.20*** | ||
| (0.21) | (0.21) | |||
| Last gaze on B | 2.06*** | −1.21 | 2.00*** | −1.21 |
| (0.36) | (0.72) | (0.38) | (0.76) | |
| Last gaze dwell time | −0.39*** | −0.84*** | −0.38*** | −0.83*** |
| (0.09) | (0.16) | (0.09) | (0.17) | |
| Model-based/w | 0.84*** | 1.58*** | −0.75 | −2.04** |
| (0.14) | (0.27) | (0.46) | (0.76) | |
| Last gaze time × last gaze on B | −0.50 | −1.45** | 0.83*** | 1.56*** |
| (0.35) | (0.54) | (0.14) | (0.29) | |
| Model-based/w × last gaze on B | 1.15* | 2.83** | 1.53* | 3.54* |
| (0.50) | (0.99) | (0.68) | (1.39) | |
| Model-based/w × last gaze time | 0.18 | 0.45* | 0.19 | 0.55* |
| (0.12) | (0.19) | (0.16) | (0.27) | |
| Model-based/w × last gaze on B × last gaze time | −0.39* | −0.92** | −0.44† | −1.09* |
| (0.17) | (0.34) | (0.25) | (0.49) | |
| AIC | 2027.81 | 1324.87 | 2028.99 | 1311.58 |
| BIC | 2267.11 | 1541.19 | 2268.30 | 1488.05 |
| Log likelihood | −974.90 | −624.44 | −975.50 | −624.79 |
| Number of trials | 3415 | 2192 | 3415 | 2192 |
| Number of subjects | 43 | 42 | 43 | 42 |
***P<0.001, **P<0.01, *P<0.05, †P<0.1. Standard errors in parentheses.
Fixed effects coefficient estimates of first-stage choice regressions in part 1 of the experiment (symbols are arbitrarily labelled A and B) using mixed-effects logistic models. Columns 1 and 2 display results that used a group dummy variable to compare model-based and model-free subjects coefficients; columns 3 and 4 present the results using a continuous value of w. Model-based behaviour is more affected by the gaze location, while model-free behaviour is more influenced by dwell time. The binary dependent variable was equal to 1 if symbol ‘B' was chosen; QB−QA is the hybrid Q-value difference estimated from the computational model; ‘B chosen at t−1', ‘Last gaze on B', and ‘First gaze on B' are binary variables; ‘Last gaze time' is duration of the last gaze on the particular trial; ‘model-based' was equal to 1 if the subject was in the model-based group (w<0.3). The coefficients of primary interest are shown in bold. AIC, Akaike information criterion; BIC, Bayesian information criterion.