haines_2020_anxiety
All participants (N=982 across three samples: 810 online MTurk, 136 student replication, 36 substance-use-disorder/treatment-seeking patients; the paper analyzed 800, 132 and 35 of them after its survey-based exclusions) completed a delay-discounting task (DDT) in which, on each trial, they chose between a smaller-sooner reward (SS, immediate) and a larger-later reward (LL, delayed). Reward amounts (in dollars; the LL reward of the adaptive trials is $800) and delays (in weeks) were titrated trial-by-trial using an adaptive algorithm; responses are binary (0=SS, 1=LL). Each participant completed two adaptive (ADO) sessions separated by a 5-min break (42 trials per session in the lab samples, 20 for MTurk). Trait impulsivity (BIS-11, incl. non-planning subscale) and state anxiety (STAI-S) were measured via survey. The research question was how trait impulsivity and state anxiety interact to modulate delay discounting, tested with hierarchical Bayesian computational models estimating reward sensitivity (α) and discounting rate (k).
Likelihood of human responses
| model | PNLL | test NLL (avg) | test responses | train NLL (avg) | train responses |
|---|---|---|---|---|---|
| CORE | 2,191.9 | 0.3795 | 5,775 | 0.3513 | 60,678 |
| CORE (no prediction errors) best | 2,178.6 | 0.3773 | 5,775 | 0.3538 | 60,678 |
| CORE (no phonemes) | 2,183.6 | 0.3781 | 5,775 | 0.3382 | 60,678 |
| CORE (no forgetting) | 2,316.5 | 0.4011 | 5,775 | 0.3789 | 60,678 |
| CORE (no hierarchy) | 3,715.7 | 0.6434 | 5,775 | 0.6431 | 60,678 |
| Transformer | 2,763.8 | 0.4786 | 5,775 | 0.4615 | 60,678 |
| Centaur | 2,573.5 | 0.4456 | 5,775 | – | – |
Experiments
| exp | task class | participants | responses | conditions | simulated | effects |
|---|---|---|---|---|---|---|
| exp0 | DelayDiscountingTask | 982 | 66,453 | 1 | yes | delay_discounting_gradient |
Behavioral effects
| effect | experiment | paper | CORE | CORE (no prediction errors) | CORE (no phonemes) | CORE (no forgetting) | CORE (no hierarchy) |
|---|---|---|---|---|---|---|---|
delay_discounting_gradient | exp0 | -1.000 | ✓ yes -0.010 | ✓ yes -0.012 | ✓ yes -0.011 | ✓ yes -0.015 | ✗ no -0.000 |
participants discount delayed rewards — as the delay to the larger-later reward (rightTime) increases, and as the smaller-sooner amount (leftValue) increases, the probability of choosing the larger-later option (response=1) decreases. Test: pooled logistic regression of response on leftValue and rightTime; both coefficients expected negative and p < .05. | |||||||
Transcript
In this study you complete a series of choices about money. Each trial shows you two reward options side by side. In the main part of the task the left option is a smaller reward available immediately (now), and the right option is a larger reward available only after a delay (later). You press the key A to choose the option on the left, or the key B to choose the option on the right. There are no right or wrong answers; pick the option you would genuinely prefer. The full session also includes practice and staircase trials with the same key layout. You compare: you get $600 now or $800 in 4 months. You press B. You compare: you get $770 now or $800 in 10 years. You press A. You compare: you get $240 now or $800 in 3 years. You press B. You compare: you get $790 now or $800 in 10 years. You press A. You compare: you get $540 now or $800 in 2.5 months. You press B. You compare: you get $440 now or $800 in 10 years. You press A. You compare: you get $440 now or $800 in 2 years. You press B. You compare: you get $420 now or $800 in 5 years. You press B. You compare: you get $380 now or $800 in 10 years. You press A. You compare: you get $470 now or $800 in 5 years. You press A. You compare: you get $410 now or $800 in 5 years. You press B. You compare: you get $340 now or $800 in 10 years. You press A. You compare: you get $420 now or $800 in 5 years. You press B. You compare: you get $320 now or $800 in 5 years. You press B. You compare: you get $320 now or $800 in 5 years. You press B. You compare: you get $360 now or $800 in 5 years. You press B. You compare: you get $330 now or $800 in 5 years. You press B. You compare: you get $470 now or $800 in 5 years. You press B. You compare: you get $360 now or $800 in 5 years. You press B. You compare: you get $330 now or $800 in 5 years. You press B. You compare: you get $320 now or $800 in 5 years. You press B. You compare: you get $320 now or $800 in 5 years. You press B. You compare: you get $320 now or $800 in 5 years. You press B. You compare: you get $330 now or $800 in 5 years. You press B. You compare: you get $540 now or $800 in 2.5 months. You press B. You compare: you get $440 now or $800 in 10 years. You press A. You compare: you get $360 now or $800 in 3 years. You press B. You compare: you get $420 now or $800 in 5 years. You press B. You compare: you get $380 now or $800 in 10 years. You press A. You compare: you get $470 now or $800 in 5 years. You press B. You compare: you get $330 now or $800 in 10 years. You press A. You compare: you get $340 now or $800 in 10 years. You press A. You compare: you get $390 now or $800 in 5 years. You press B. You compare: you get $330 now or $800 in 5 years. You press B. You compare: you get $330 now or $800 in 5 years. You press B. You compare: you get $340 now or $800 in 5 years. You press B. You compare: you get $360 now or $800 in 5 years. You press B. You compare: you get $360 now or $800 in 5 years. You press B. You compare: you get $330 now or $800 in 5 years. You press B. You compare: you get $360 now or $800 in 5 years. You press B. You compare: you get $360 now or $800 in 5 years. You press B. You compare: you get $350 now or $800 in 5 years. You press B. You compare: you get $320 now or $800 in 5 years. You press B. You compare: you get $340 now or $800 in 5 years. You press B.