Fig 1.
Example of a self-referential rule block in the social evaluation learning task.
Each block contains a probabilistic learning phase where 32 word pairs and feedback are presented, and a global rating phase. There were 6 blocks in total, self-liked, self-neutral, self-disliked, other-liked, other-neutral, other-disliked. P[positive word correct] = 0.8, 0.5 and 0.2, for the liked, neutral and disliked rules respectively.
Fig 2.
Cumulative mean positive responses over the 32 trials during the learning phase.
Learning curves for high (n = 50) and low FNE (n = 50) individuals based on median split of test-day BFNE scores. The high and low FNE groups vary most over the initial trials where high FNE made fewer positive responses. After the initial trials the high and low groups behave similarly except in the neutral and dislike rules in the self-referential condition where the learning curves are clearly separated. The clear differentiation of the curves by rule indicates that individuals were adjusting their response in response to feedback.
Table 1.
Participant characteristics.
Table 2.
Mean positive response rate and global rating scores for referential condition and rule by screening group.
Fig 3.
Predicted values for learning phase and global interpretations from regression models testing for differential effects of condition and rule on FNE.
BFNE = Brief Fear of Negative Evaluation Scale. Rule contingencies like 80%, neutral 50%, dislike 20%. All regression models were a good fit for the data, each explaining around 64% of overall variance for the learning phase, and around 60% for the global ratings. We predicted the linear relationship between each rule and condition with social anxiety (coefficient and 95% confidence intervals) using the lincom command in Stata. These indicate that learning in the self-referential condition and in the neutral and disliked rules is most associated with FNE, and this holds for both the learning and global rating phases.
Table 3.
Regression coefficients (95% CI) from regression models testing for interactions of rule and self-referential condition with BFNE for the positive response proportion and global ratings in the social learning task.
Fig 4.
Correct-repeat behaviour during the learning phase.
Predicted values from the Poisson regression models testing for the referential-condition by FNE interaction on correct-repeat behaviour during the learning phase in separate models for negative (left) and positive (right) words. We used the lincom command in Stata to estimate the relationship (rate ratio and 95% confidence intervals) between correct-repeat behaviour and a one standard deviation increase in BFNE (1 s.d. corresponds to 11.5 BFNE points) by referential condition. FNE was selectively associated with self-referential negative-correct-repeat responses, with the coefficient indicating a 17% increase for each 11.5 point increase in BFNE, p < 0.001. There was no evidence to suggest that negative-correct-repeat behaviour in the other other-referential condition, or that positive-correct-repeat behaviour in either condition, varied with FNE, p’s > 0.2.