Figure 1
Schematic representation of the affective bias task set up and scoring procedure. A) In the learning phase (20 trials), participants learned associations between stimuli (left-and right-tilted lines), responses (“z” or “m” respectively) and reward outcomes (counter-balanced across participants) through trial-and-error. Here we depicted one counterbalanced reward-contingency condition. Participants who pressed “z” when seeing a left-slanted line received 1 virtual pound. When seeing a right-slanted line, an “m” response received 4 virtual pounds. B) The testing phase consisted of three equally likely appearing stimuli: the left-slanted line, the right-slanted line, and a horizontal line. The horizontal line is called “ambiguous stimulus” due to its intermediate representation and was used to calculate the affective bias score. C) A schematic representation of ambiguous trials and rewards. On half of the trials, the ambiguous stimulus belonged to the left-tilt category, and on the other half to the right-tilt category, unbeknownst to the participant. If participants responded with the correct category-response association, they received the reward associated with that category.
Figure 2
A schematic of the scoring of correctness of the dynamic reward task. A) Participants moved a blue arrow using a slider. This corresponded to their task response. Once they let go of the slider, a second arrow was superimposed. B) On half of the trials, the second arrow is from category purple (P). The tip of the purple arrow must fall left of the blue arrow tip to be scored as correct. C) On trials with superimposed category orange (O) arrows, the tip of the orange arrow must fall right of the blue tip to be scored as correct. Not all correct trials are rewarded. Instead, category reward was yoked and changed throughout the experiment, necessitating a continuous adjusting of response bias to gain maximal rewards.
Figure 3
An example of a mixed-gamble and a gain-gamble trial. For 100 trials, participants made a choice between two cards, showing a safe bet (left option) or a 50/50 gamble (right option). They could be presented with either with a mixed-gamble (A) or a gain-gamble (B). Each safe bet, gamble combination was shown twice. Participants did not receive feedback. A) When safe bets gave 0 points, participants could either gain or lose a reward on the gamble. Here, there was a 50% chance to win £50 or loose £20. B) When safe bets gave a reward, the gamble consisted of winning nothing 50% of the time or winning a reward 50% of the time, here £60.
Table 1
The descriptive statistics of the affective bias score and the nine parameters used in the logistic regression model to predict the affective bias score. The outcome of the logistic regression, p(mid as high), represents the affective bias sore. The demographic predictor variables included were age and sex. The predictor variables from the dynamic SDT model were the learning rate (α), the reward sensitivity (i.e., gain, ln|G|), the bias parameter (b), and the setting noise parameter (σ). The predictor variables from the prospect theory model were the subjective value of reward (i.e., risk aversion, ρ), loss aversion (δ) and inverse temperature (τ).
| VARIABLE | VARIABLE SYMBOL | TASK | MEAN | SE | MEDIAN | IQR | MINIMUM | MAXIMUM |
|---|---|---|---|---|---|---|---|---|
| Negative Affective Bias Measure | p (mid as high) | Affective Bias Task | 0.57 | 0.02 | 0.55 | 0.25 | 0 | 1 |
| Age | N/A | N/A | 39.23 | 0.99 | 38 | 17 | 19 | 78 |
| Sex | N/A | N/A | 0.66 | N/A | N/A | N/A | N/A | N/A |
| Learning Rate | α | Dynamic Reward Task | 0.23 | 0.01 | 0.23 | 0.17 | 0 | 0.66 |
| Reward Sensitivity i.e., Gain | ln(|G|) | Dynamic Reward Task | –1.29 | 0.11 | –1.27 | 1.58 | –5.99 | 1.29 |
| Bias | b | Dynamic Reward Task | –0.09 | 0.5 | 0.09 | 4.05 | –31.63 | 32.73 |
| Setting Noise | σ | Dynamic Reward Task | 14.26 | 0.44 | 15.42 | 9.62 | 0.42 | 19.96 |
| Subjective Value Sensitivity i.e., Risk Aversion | ρ | Gambling Task | 0.68 | 0.02 | 0.69 | 0.41 | 0.22 | 1.3 |
| Loss Aversion | δ | Gambling Task | 2.31 | 0.1 | 2 | 1.57 | 0.4 | 7.27 |
| Inverse Temperature | τ | Gambling Task | 0.62 | 0.1 | 0.11 | 0.43 | 0 | 7.21 |
Figure 4
Box plot of the affective bias score with a density distribution plot and individual data points. The plot represents the affective bias score data i.e., p(mid as high) data of the full sample of 148 participants.
Figure 5
The logistic regression predicting p(mid as high) performed on the full sample (N = 148). The predictor variables were sex, age, δ = loss aversion, ρ = risk aversion conceptualising subjective value sensitivity, τ = inverse temperature, b = bias, σ = setting noise, ln(|G|) = gain parameter conceptualising reward sensitivity, α = learning rate.
Figure 6
Pairwise linear correlations between reward sensitivity, affective bias and subjective value sensitivity A) The raw correlation between the affective bias score and reward sensitivity. There is a weak positive, non-significant relationship. The dashed horizontal line indicates no preference between the low and high reward. The dashed vertical line represents a neutral gain parameter, which neither inflates nor shrinks optimal criterion estimates. Turquoise shaded region indicates participants are conservative on the learning task, the yellow shaded region shows that participants are liberal on the learning task. B) The raw correlation between reward sensitivity and subjective value sensitivity. The dotted horizontal and vertical lines indicate no bias. The yellow shaded area encompasses participants who are liberal in their criterion placements and risk seeking. The upper green shaded background indicates conservatism in the gain parameter but risk seeking behaviour in the gambling task. The right hand green shaded area indicates risk aversion but inflation of reward. The turquoise shaded region indicates conservatism on both the risk aversion and gain parameter.
Figure 7
Pairwise linear correlations between setting noise, affective bias and inverse temperature A) Showing the raw correlation between setting noise and the affective bias score in the full sample. B) Showing the raw correlation between setting noise and the affective bias score in the reduced sample C) showing the raw correlation between setting noise and inverse temperature in the full sample.