1. Overview
The Raw Draws export contains the unaggregated simulation draws produced during model estimation. Each draw is one sample from the model's posterior distribution — the distribution of plausible values given the data — so the export gives you the full set of samples that summary tables aggregate. This is what you need to compute uncertainty and run advanced scenario outcomes outside the platform.
Use this export when you want to:
- Quantify uncertainty around utilities and derived metrics.
- Run custom Monte Carlo simulations outside the platform.
- Audit model convergence and draw stability.
For the estimator and sampler that produce these draws, see Utility and simulated share methodology. For summarized respondent-level utilities rather than raw draws, see the Utility Scores Export Format. For the question types this export applies to, see Choice-Based Conjoint and Running MaxDiff.
2. File Structure & Layout
Each row represents one draw for one respondent and one modeled parameter.
A single respondent and level combination therefore appears across many rows, one per draw index.
Example (first 5 rows):
| respondent_id | attribute | level | parameter_key | draw_index | draw_value | model_run_id |
|---|---|---|---|---|---|---|
| r_1001 | Brand | Alpha | utility:Brand:Alpha | 1 | 0.79 | run_2026_04 |
| r_1001 | Brand | Alpha | utility:Brand:Alpha | 2 | 0.88 | run_2026_04 |
| r_1001 | Brand | Alpha | utility:Brand:Alpha | 3 | 0.83 | run_2026_04 |
| r_1001 | Price | $14.99 | utility:Price:14.99 | 1 | -0.61 | run_2026_04 |
| r_1001 | Price | $14.99 | utility:Price:14.99 | 2 | -0.70 | run_2026_04 |
3. Key Columns
- respondent_id - Unique participant identifier.
- attribute - Attribute linked to the sampled parameter (when applicable).
- level - Attribute level linked to the sampled parameter (when applicable).
- parameter_key - Canonical parameter identifier used in model internals.
- draw_index - Sequential draw number for that parameter and respondent.
- draw_value - Numeric value sampled for that draw.
- model_run_id - Identifier for the estimation run.
- chain_id - Optional chain identifier for multi-chain samplers.
4. Data Representation
Draw-level granularity
No averaging is applied in this export. Every retained draw is included to preserve full distribution shape.
Repeated keys
The same parameter_key repeats across many draw_index values by design. This is required for percentile, interval, and stability calculations.
Respondent-level sampling
Where respondent heterogeneity is modeled, draws are stored per respondent rather than at aggregate-only level.
5. Missing & Special Values
- Warm-up or burn-in draws are typically excluded unless explicitly configured.
- Rows with invalid samples may be absent if removed by model diagnostics.
- Always filter to one model_run_id before computing summary statistics.
6. Best Practices
- Summarize with median and credible intervals instead of only means.
- Validate effective sample size and convergence metrics before downstream simulation.
- Preserve
draw_indexandchain_idin diagnostics to detect sampler issues.
7. When to Use Raw Draws Export
- For advanced Bayesian diagnostics.
- For custom risk and uncertainty simulations.
- When model transparency requires full draw-level traceability.