Understanding mental health in the workplace

Our primary intervention is a Work-Related Cognitive Behavioural Therapy (WCBT) program delivered in group workshops over 8 two-hour sessions. It is delivered during the six-week employment task. The intervention uses core techniques from cognitive behavioural therapy to help people deal with feelings of anxiety or depression in work-related situations. Control group participants do placebo workshops of the same length and format on an educational non-psychological topic (environmental sustainability).

Most of our planned causal analysis focuses on effects of WCBT, but we have two ancillary between-person variations in participants' work conditions, which are cross-randomized with WCBT.

Firstly, while all participants receive constructive critical performance feedback, we randomize whether this feedback is 'psychologically enhanced' by adding motivational messaging and contextualisation.

Secondly, we randomize the day of week on which participants receive study payments. This is to estimate the effect of financial worries on performance, following Kaur et al. (2025), against which we can benchmark the effects of WCBT which targets a broader range of worries.

2025-07-03

2026-06-19

Mental health (a first stage outcome); earnings, productivity, and hours worked, both overall and in each task; performance in an interview/skills assessment for a promotion; (incentivized) preferences to do the skills assessment, manage others, and present in front of others; preferences to work under a competitive pay scheme; willingness to negotiate pay and final negotiated outcome; number of times participants speak up in a work meeting.

Our pre-analysis plan (which will be added before outcomes are measured) will describe how we implement corrections for multiple hypothesis testing, and the construction of indices and summary measures for different outcome families.

Mental health is measured using the PHQ-8 (depression) and GAD-7 (anxiety) scales.

Earnings: the total amount earned (overall or in each task) in the period after the intervention has finished. Productivity is measured as earnings per hour. The tasks for which we measure earnings are: data entry, under standard pay and competitive pay; customer service; an adaptation of the manager task from Deming et al. (2024); a team 'tower-building' task; a public presentation task. Some tasks have a fixed time limit so we do not use hours worked or productivity as an outcome for these tasks.

Preferences to do the skills test, manage others or present in front of others are measured on a scale from 0 to 10. Each preference is measured on two occasions: immediately after the intervention has finished, and immediately before doing each task.

Preference to work under a competitive pay scheme is measured by using a design similar to Niederle and Vesterlund (2007) overlaid on our data entry task.

Willingness to negotiate pay: We conduct a collaborative version of the data entry task in which, at the end, the default joint surplus split is unfavorable to some collaborators. We measure these participants' preference to negotiate over the joint surplus, rated from 0 to 10, and also the actual negotiated outcome.

Speed and accuracy in data entry; speed in customer service; the underlying performance ratings (by enumerators/audience members/automated grading) in the customer service and presentation tasks; the number of times participants ask enumerators for help; beliefs about performance; cognitive function (Raven's matrices); measures of self-efficacy (Generalised Self Efficacy Scale), self-esteem (Rosenberg Self-Esteem Scale) and concentration at work (Stanford Presenteeism Scale-6); time, risk and social preferences; positive and negative reciprocity; daily positive and negative affect; daily feelings of financial strain; social anxiety (mini-SPIN); social skills (Reading the Mind in the Eyes test).

Quality of a 'mock job application' as assessed by recruiters; intended job search effort (hours and applications) and reservation wage.

Beliefs about performance are measured by asking participants to predict both before and after feedback their relative performance (how they will rank relative to other participants) and absolute performance (their earnings). We also elicit beliefs about relative performance (rank) in the presentation task, skills assessment and manager task as well as beliefs about absolute performance in the presentation and skills assessment (how they are rated by the audience / interviewer).

Risk preferences are elicited using the method of Eckel and Grossman (2002) with hypothetical options; we also elicit self-reported overall comfort with risk using the question from the Global Preferences Survey (Falk et al. 2023). Time preferences are measured by using a multiple price list method to identify participants' switching points (between a risky vs safe option and between a sooner and smaller vs larger and later amount), in hypothetical choices.

Social preferences are measured using an anonymous one-shot dictator game. Positive and negative reciprocity are measured using the self-report questions from the Global Preferences Survey (Falk et al. 2018).

Daily positive and negative affect is measured using 10 items from the Positive and Negative Affect Scale asked twice each day. Participants' responses on a five-point scaled are averaged.

Daily feelings of financial strain are measured following Kaur et al. (2025) by a question asking participants if they were thinking about any worries or finances while working and if so, what.

Application quality: we use a survey of local HR recruiters who see mock job applications by participants. We measure: the recruiter guess of candidate performance in skills tests (we use both the raw guess and its absolute deviation from the truth as outcomes), and the recruiter's subjective rating of the candidate's employability and personal qualities.

We conduct the trial in multiple phases. In each phase we aim for the trial be completed by a cohort of 60 participants, although if there is attrition then the number will be smaller. Within each cohort, 50% are randomized to the intervention and 50% are randomized to a control condition in which they do placebo workshops. Randomization into WCBT is further stratified on:
- Gender
- Baseline mental health (binary indicator for being above/below the threshold for clinical diagnosis)
- Baseline income (binary indicator for being above/below median)

Participants work for 30 days doing a variety of tasks. After a baseline period in days 1-3, participants do WCBT and placebo workshops over days 4-20, alongside alternating days of data entry and customer service work.

Outcome measurement starts from day 19. From days 19-30, each participant does all of the following tasks:
- Data entry work
- Customer service work
- Delivering a 10-minute presentation
- A 'manager task' based on Deming et al. (2024)
- An interview/skills assessment for promotion to a more challenging version of customer service
- A collaborative version of data entry in which participants supervise other participants' work by correcting their mistakes, and they negotiate over joint earnings at the end
- A tournament version of data entry based on Niederle and Vesterlund (2007)
- A team tower-building game
- A meeting in which we seek participants' ideas for changes to improve productivity in their work

Different participants may do some of these tasks on different days; if so the order in which participants complete each task is independently randomized.

All participants also report preferences to do the presentation, be a manager, and do the interview/skills assessment. A small holdout sample of participants (approx. 5%) will have their preferences implemented and as a result may not do these tasks.

All performance feedback is implemented after day 20.

Not available

Randomization is done by computer.

Randomization is at the individual level.

No

480 people

480 people

240 participants receive WCBT (treatment). Of these:
- 60 receive earlier payment; 60 receive later payment; 60 receive earlier payment and psychologically enhanced feedback; 60 receive later payment and psychologically enhanced feedback.

240 participants receive the placebo. Of these:
- 60 receive earlier payment; 60 receive later payment; 60 receive earlier payment and psychologically enhanced feedback; 60 receive later payment and psychologically enhanced feedback.