Estimate weather’s effect on mental health
Quick Overview
This question evaluates causal inference and applied statistical modeling skills—specifically defining outcomes and treatments, addressing confounding and selection biases, choosing identification strategies, and conducting robustness checks with observational data.
Scenario
You are studying whether weather (e.g., temperature, precipitation, sunlight, air pressure) affects mental health outcomes (e.g., depression score, anxiety index, crisis hotline calls, therapy app usage).
You have observational data at either the person-day level or region-day level.
Task
Describe how you would estimate the causal effect of weather on mental health while addressing confounding.
Your answer should include:
-
Define the outcome and treatment
- What is the outcome variable (continuous score vs binary event)?
- What is the “treatment” (e.g., +5°C temperature shock, rainy day indicator, hours of sunlight)?
- Consider possible lagged effects (weather today impacts mental health over the next few days).
-
Confounders & biases
- List plausible confounders (seasonality, holidays, location, socioeconomic status, pollution, day-of-week, long-term trends).
- Discuss selection bias (who appears in the dataset) and measurement error.
-
Identification strategy
- Propose an approach to identify a causal effect (e.g., fixed effects, difference-in-differences/event study, instrumental variables, propensity scores), and state assumptions.
-
Model specification and checks
- Provide a concrete regression / model form.
- How would you test key assumptions (parallel trends, no unobserved time-varying confounders, SUTVA)?
-
Interpretation
- How would you interpret the effect size?
- What sensitivity analyses would you run to assess robustness?
Output
Provide a structured plan and at least one model equation (LaTeX is OK).