Stress and cognitive-effort avoidance
Motivation
When people are stressed, they often aren’t as good at cognitively demanding tasks. Psychologists have documented that learning, memory, and decision-making performance is impaired when people are acutely stressed, and have developed various theories to explain this. Roughly speaking, there are two competing explanations:
- When we are stressed, we reallocate cognitive effort from executive function (decision-making) to salience (controlling our attention and looking for important information), which impairs our executive function
- People who are stressed don’t want to exert cognitive effort, and choose to exert less, regardless of how much the stress has caused us to reallocate it
Option 2 should make sense to anyone who has chosen to watch Netflix on the sofa after a stressful workday.
To test which theory better explains behavior, we need an experiment where we can manipulate the stress level of the subjects, then see whether they choose to exert cognitive effort or not. Bogdanov et al. (2021) designed such an experiment.
In their experiment, subjects were stressed using the Trier social stress test (TSST), which has subjects prepare and give a 5-minute presentation in front of three impassive judges, then do a mental arithmetic task in front of them. This has been shown to reliably stress people out.1
Subjects then performed a demand-selection task (DST), a simple procedure where subjects click one of two options in a colorful diagram, and then must either judge the evenness/oddness of a number or judge whether it is larger or smaller than 5. The tasks are easy, but depending on the subject’s choice, they may have to switch between them more or less option: in one choice, they do the same task (evenness or magnitude) repeatedly with probability 0.9, in the other choice they usually have to switch between each trial (repeat probability 0.1). Constantly switching the task takes cognitive effort, so generally subjects prefer the low-effort option.
Thus, the researchers hypothesized:
Accordingly, we predicted that if stress engenders a withdrawal of effort—either by increasing the subjective costs or decreasing the perceived benefits of exerting cognitive effort—we should observe an even stronger demand-avoidance effect in the stress condition relative to the control condition.
Data
The study has 20 participants. Each participant did one trial in the stressed condition (with the TSST) and one in the control condition (no TSST), seven days apart. On the first day they also took various psychological questionnaires.
One data file (cognitive-effort-demographics.csv) gives the demographics and questionnaire results for each subject.
The other data file (cognitive-effort.csv) gives the result of every single DST trial for each subject. In each trial, the subject is presented with the choice, makes their choice, is presented with the number, and must perform the task. The time they take to choose and the time they take to do the task are recorded, as well as their choice, whether the task switched, and whether they did the task correctly. The trials are divided into blocks of roughly 70 individual trials; each subject did 4 blocks.
Data preview
cognitive-effort.csv
cognitive-effort-demographics.csv
Variable descriptions
For cognitive-effort.csv:
| Variable | Description |
|---|---|
| PID | Participant ID |
| condition | Experimental condition (control or stress) |
| block | Block of the DST (1 through 4) |
| effort_choice | 0 = chose high-effort cue, 1 = chose low-effort cue |
| effort_choice_RT | How quickly they selected high or low effort (seconds) |
| Switch | Whether current trial is a switch (true) or repetition trial (false) |
| TS_correct | Accuracy of response in the task-switching part of the DST (0 = incorrect, 1 = correct) |
| TS_RT | Reaction time in task switching (seconds) |
| Session | Session number |
For cognitive-effort-demographics.csv:
| Variable | Description |
|---|---|
| ID | Participant ID |
| Gender | Participant gender (0 = male, 1 = female) |
| Age | Participant age, years |
| BMI | Participant’s body mass index |
| GAD | Score on a generalized anxiety disorder scale (details not specified) |
| PHQ | Score on the patient health questionnaire (details not specified) |
| BAS_Drive | Behavioral Activation System subscale score for drive; higher = higher persistence toward goals |
| BAS_Fun_seeking | Behavioral Activation System subscale score for fun-seeking behavior; higher = higher desire for fun, novel things |
| BAS_Reward_responsiveness | Behavioral Activation System subscale score for reward responsiveness; higher = greater responsiveness to rewards |
| BAS_overall | Overall Behavioral Activation System score (sum of three subscale scores) |
| BISBAS_dummy_items | Score on the 4 filler questions on the BIS/BAS scales |
| BIS | Score on the Behavioral Inhibition System scale; higher = experience greater anxiety/inhibition about possible negative consequences of actions |
| DAS | Score on the Dimensional Apathy Scale; higher = more lack of motivation toward goal-directed behaviors |
| NFC | Score on the Need for Cognition Scale (higher = higher tendency to engage in and enjoy effortful cognitive endeavors) |
| PSS | Score on the Perceived Stress Scale (higher = respondent finds their life unpredictable, uncontrollable, or overloading) |
| SHAPS | Score on the Snaith-Hamilton Affective Pleasure Scale (measures ability to feel pleasure; higher = less ability) |
| UPPS_P_Negative_urgency | UPPS-P Impulsive Behavior Scale; subscale score for negative urgency, the tendency to act impulsively due to negative emotions |
| UPPS_P_Perseverance | UPPS-P Impulsive Behavior Scale; subscale score for lack of perseverance, the tendency to not complete tasks |
| UPPS_P_Positive_urgency | UPPS-P Impulsive Behavior Scale; subscale score for positive urgency, the tendency to act impulsively due to positive emotions |
| UPPS_P_Premediation | UPPS-P Impulsive Behavior Scale; subscale score for lack of premeditation, the tendency to act without forethought |
| UPPS_P_Sensation_seeking | UPPS-P Impulsive Behavior Scale; subscale score for sensation-seeking, the tendency to pursue novel or thrilling experiences |
| UPPS_P_overall | Overall UPPS-P Impulsive Behavior Scale score |
| systolic_[condition]_x | Systolic blood pressure at time point x in this condition (treatment or control); e.g., systolic_control_2 is systolic blood pressure in the control condition at time 2 |
| diastolic_[condition]_x | Diastolic blood pressure at time point x in this condition |
| heart_rate_[condition]_x | Heart rate at time point x in this condition |
| PANAS_pos_[condition]_x | Score for the positive scale of the Positive and Negative Affect Schedule at time point x in this condition |
| PANAS_neg_[condition]_x | Score for the positive scale of the Positive and Negative Affect Schedule at time point x in this condition |
| cort_[condition]_x | Salivary cortisol level at time point x in this condition |
| TSST_difficult_[condition] | Participant’s assessment of difficulty for the stress manipulation in this condition, on a scale from 0 (not at all) to 100 (very much) |
| TSST_unpleasant_[condition] | Participant’s assessment of unpleasantness for the stress manipulation in this condition, on a scale from 0 (not at all) to 100 (very much) |
| TSST_stressful_[condition] | Participant’s assessment of stressfulness for the stress manipulation in this condition, on a scale from 0 (not at all) to 100 (very much) |
| session_order | Order of stress and control conditions experienced by this participant (1 = control then stress, 2 = stress then control) |
| OSPAN_math_errors_[condition] | Number of errors in the math equations in the Operation Span task |
| OSPAN_score_[condition] | Score in the Operation Span task |
| OSPAN_total_[condition] | Total remembered letters in the Operation Span task |
Questions
Examine how treatments were allocated. Some subjects were stressed on their first day; others on the second day (
session_order). Look at the distribution of treatment assignment by participant gender. Based on what you see, describe the experimental design: what kind of study was this and how were subjects randomized?In an experiment, it’s important to verify that the intervention does what it was supposed to. In this case, the Trier social stress test (TSST) was supposed to stress the participants. This was evaluated with three questions asking the participants about the difficulty, unpleasantness, and stressfulness of the task (the three
TSST_variables). Test whether participants reported higher difficulty, unpleasantness, and stress in the treatment condition than in the control.Salivary cortisol also measures the body’s stress response. Test whether salivary cortisol levels were higher in the treatment condition than in the control.
The outcome measure is whether participants chose the low-effort or high-effort options. Calculate the percentage of time each participant chose the low-effort option in each experimental condition. Test whether the proportions are different between conditions.
We may expect there to be subject-level effects: some subjects generally prefer the higher-effort option more, other prefer it less. This suggests using a random or mixed effects model. Fit a logistic regression predicting the subject’s choice, using a fixed effect for condition and a random effect for subject.
There could also be block and session effects. (Recall that the DST task is done in four successive blocks, so subjects may gradually get better at it, or change their preference over time. They may also have different performance on their second day in the lab.) Extend your model to account for this. Is there evidence of block and session effects?
Any effect might be heterogeneous: different subjects might respond differently to stress. One possible hypothesis is that subjects who enjoy cognitive effort are less likely to switch to the low-effort option, regardless of stress level. The Need for Cognition scale (NFC), available in the demographic data, tries to measure how much the subjects participate in and enjoy cognitive effort. Evaluate whether the effects depend on NFC score.
(This analysis was not conducted in the paper, which did not mention NFC despite it being present in the data.)
Besides effort preference, it may also be interesting to know if stressed subjects perform more poorly on the task (are less accurate). Use a mixed effects logistic regression to answer this question.
It also may be interesting to know whether stressed subjects take longer to decide on the effort and then to do the task, due to stress slowing their cognitive effort. You can use the
effort_choice_RTandTS_RTvariables to measure these times.Evaluate the distribution of these variables. What kind of model would be necessary to answer the question, and would transformations be necessary? Decide on a model and answer the question for each reaction time.
References
Bogdanov, M., Nitschke, J. P., LoParco, S., Bartz, J. A., & Otto, A. R. (2021). Acute Psychosocial Stress Increases Cognitive-Effort Avoidance. Psychological Science, 32(9), 1463-1475. https://doi.org/10.1177/09567976211005465
Data via OSF, at https://doi.org/10.17605/OSF.IO/26W4U
Footnotes
Despite the name, the test does not involve watching Lars von Trier films with your parents, which would also serve as an effective stressor.↩︎