Preference-based serial decisions are counterintuitively influenced by emotion regulation and conscientiousness

Abstract

Our decisions have a temporally distributed order, and different choice orders (e.g., choosing preferred items first or last) can lead to vastly different experiences. We previously found two dominant strategies (favorite-first and favorite-last) in a preference-based serial choice setting (the 'sushi problem'). However, it remains unclear why these two opposite behavioral patterns arise: i.e., the mechanisms underlying them. Here we developed a novel serial-choice task, using pictures based on attractiveness, to test for a possible shared mechanism with delay discounting, the 'peak-end' bias (i.e., preference for experienced sequences that end well), or working-memory capacity. We also collected psychological and clinical metric data on personality, depression, anxiety, and emotion regulation. We again found the two dominant selection strategies. However, the results of the delay, peak-end bias, and memory capacity tasks were not related to serial choice, while two key psychological metrics were: emotion regulation and conscientiousness (with agreeableness also marginally related). Favorite-first strategists actually regulated emotions better, suggesting better tolerance of negative outcomes. Whereas participants with more varied strategies across trials were more conscientious (and perhaps agreeable), suggesting that they were less willing to settle for a single, simpler strategy. Our findings clarify mechanisms underlying serial choice and show that it may reflect a unique ability to organize choices into sequences of events.

Fig 1. Illustration of the picture-rating task procedure.

Fig 2. Illustration of the serial-choice task procedure.

Fig 3. Illustration of the sequence-rating task procedure.

Fig 4. Illustration of the delay-discounting task procedure.

Fig 5. Average number of (two-button-sequence) responses in the Effort Task (i.e., the average number of times the participants pressed the two-button-sequence to view the picture again for 0.5s within the 10-second trial duration) for each category.

The effect of attractiveness level was found to be significant, indicating that the classification process for picture sets worked reliably, providing a clear distinction between each category, and providing evidence that the pictures were treated as consummatory rewards (to be worked for). The error bar denotes standard error of the mean. (** p < .000 for all pairs).

Fig 6. Individual results of the Serial-Choice (SC) task.

The vertical axis represents the SC score, which is the slope of the best-fit line of the participant’s average choice of stars for each order position within the sequence. The horizontal axis represents each participant, aligned by the SC score. Higher (close to +1) SC scores represent the tendency toward favorite-last (i.e., 1-2-3-4) preference, while the opposite (close to -1) represent favorite-first (i.e., 4-3-2-1) behavior. The error bar is standard error of the mean. Favorite-last was the dominant strategy, while a distinct proportion of favorite-first behavior also existed.

Fig 7. Comparison of SC task results with a random sequence group.

The difference between the empirical SC Task data and its random pair (i.e., the average of 1,380 randomly generated four-length sequences). Panel A shows the average number of stars for each order position within the sequence. The average number of stars for 1^st choice was significantly lower and for 4^th significantly higher than its random pair (p = .000). Panel B shows the mean SC score of the participant group and its random pair. The slope from our empirical data was significantly higher than the slope of random sequences (p = .000). The error bar is the standard deviation. These results indicate that the favorite-last choice was the dominant strategy for our participants. (** p = .000).

Fig 8. Reaction times of serial-choice task.

The horizontal axis represents the order of choices and the vertical axis is mean reaction time. The reaction time of the first choice (M = 2.564) was significantly longer than the others (p = .000). This suggests that the participants planned the order of choices before their first choice. The error bar denotes standard error of the mean. (** p = .000).

Fig 9. Average number of stars of each choice for different SC scores.

The vertical axis represents the average number of stars, while the horizontal axis represents the order of choice in the SC task, with participants divided into seven groups (red to blue) along the SC score (-1 to +1). The first (red, far left for all four order positions) group exhibited strict favorite-first behavior while the other groups behaved oppositely (except the second group). The second group exhibited a unique behavioral pattern, which was U-shaped (e.g., 4-1-2-3), although their SC slope had a positive number since they also mostly picked their best option last. These results indicate that there were clear preferences among the participants for both favorite-first and favorite-last strategies, as well as mixed strategies, although favorite-last prevailed overall. The error bar denotes standard error of the mean.

Fig 10. Reaction-time of serial-choice task together with SC score.

The vertical axis represents the mean reaction time and the horizontal axis represents the order of choice in the SC task, with participants divided into seven groups (red to blue) along the SC score (-1 to +1). In the first position of the choice sequence, the middle (purple) group exhibited a significantly longer reaction time than the first (far left, red) and the last (far right, blue) groups. This suggests that the middle group in particular (and the other middle groups more generally) took more time to decide the serial-choice strategy. The error bar denotes standard error of the mean. (* p < .002).

Fig 11. Sequence-Rating task results.

Panel A shows the results for each participant, with the vertical axis the SR score (i.e., from the best-fit line of the sequence-rating dependent measure across the tested sequences). The horizontal axis represents each participant, aligned by the SR score (i.e., the same participant order as Fig 6). The higher (positive) SR score reflects a preference for ascending (e.g., 1-2-3-4) sequences, while the opposite (negative) reflects a descending (e.g., 4-3-2-1) preference. Preference for ascending sequences was dominant, while a distinct proportion preferred descending. Panel B shows the mean SR slope of all participants versus its random pair. The empirical data slope was significantly higher than from random sequences (p = .000). The error bar is the standard deviation. Both results show a prevailing preference for favorite items coming later in the sequence (i.e., rather than first), notwithstanding the distinct group who preferred a descending order. (** p = .000).

Fig 12. Working-memory performance for order of target picture in the sequence-rating task.

Working-memory performance was significantly higher when the target picture was located in the last position of a sequence (M = 0.90, SD = 0.10) than the 1st (M = 0.78, SD = 0.13), 2nd (M = 0.74, SD = 0.13), or 3rd (M = 0.67, SD = 0.13) positions. These results indicate both “recency” and “primacy” memory biases occurred with the sequential paradigm. The error bar denotes standard error of the mean. (** p = .000 for all pairs).

Fig 13. Delay-Discounting task results.

Panel A shows the mean k-value (delay-discounting parameter) for each attractiveness condition, with a larger k-value signifying greater discounting. Panel B shows the mean reaction time for each condition. The discounting parameter and reaction time both decreased when the amount of (delayed) reward was larger in both absolute (“2 vs 3” > “3 vs 4”) and relative (“2 vs 3” > “2 vs 4”) terms. These results indicate that participants discounted the delayed, higher-rated pictures properly, and were more willing to wait for the larger reward when it was more attractive. The results verify that the participants actually considered the picture stimuli rewarding according to the “star” categories, validating this critical component of the paradigm. The error bar denotes standard error of the mean. (** p = .000 for all pairs).

Fig 14. Individual results of the Sequence-Rating (SR) Task aligned with Serial-Choice (SC) score.

The vertical axis represents the SR slope, i.e., the retrospective sequential experience preference; while the horizontal axis represents each participant aligned by the SC score from the Serial-Choice Task (i.e., the same participant order in Fig 6). The higher (positive) SR score reflects a preference for ascending (e.g., 1-2-3-4) sequences, while the opposite (negative) reflects that for descending (e.g., 4-3-2-1) sequences. The SC (x-axis) and SR (y-axis) scores were not correlated (Pearson-r = .133, N = 69, p = .276), and thus preferences for ascending or descending sequences did not remain consistent for individuals across the two tasks.

Fig 15. Relationship between Serial-Choice and ERQ scores.

There was a negative correlation between the SC and ERQ scores, indicating that the favorite-first strategy also showed better emotion-regulation ability. (Pearson-r = -.254, N = 66, p = .040).

Fig 16. Relationship between standard deviation of Serial-Choice slope (SC SD) and BFI Conscientiousness and Agreeableness scores.

We found a significant positive correlation between the standard deviation of the SC slope and the BFI Conscientiousness (15A) (Pearson-r = .390, N = 69, p = .001) and Agreeableness (15B) (Pearson-r = .269, N = 69, p = .025) scores. These results indicate that the use of more varying strategies in serial choice—compared to strict favorite-first or favorite-last strategies—was linked to more self-disciplined and considerate personality traits.

Fig 17. Correlation of BDI score with both delay discounting and working-memory performance.

The relationships between BDI score and two behavioral task results: delay discounting parameter (A: Pearson-r = .440, N = 65, p = .000) and working-memory performance (B: Pearson-r = -.266, N = 69, p = .027). Consistent with previous studies, higher discounting of future reward and lower working-memory performance were linked to higher BDI scores, suggesting more symptoms of depression.