Uncompensated wage elasticity of labor supply
Labor and tax subpanel · pooled centers and 90 percent intervals from 15 independent runs per model · shaded region marks the review range [0.05, 0.3].
What the models were asked
Percent change in annual hours worked from a 1 percent change in the net-of-tax wage, allowing both substitution and income effects and excluding participation responses.
- In standard notation:
- εM = ∂ ln h / ∂ ln wnet (uncompensated: substitution + income effects) (shorthand for display; the models received only the prose definition above)
- Population:
- Prime-age workers in the United States who are already working
- Interpretation:
- Intensive-margin uncompensated annual-hours elasticity
The exact prompt, verbatim
Answer from your current memory and background knowledge only.
Do not use tools, files, the web, code, or external resources.
Do not try to reconstruct a literature review or search for a consensus estimate.
Report the belief you currently endorse.
Quantity of interest:
- Name: Uncompensated wage elasticity of labor supply
- Definition: Percent change in annual hours worked from a 1 percent change in the net-of-tax wage, allowing both substitution and income effects and excluding participation responses.
- Target interpretation: Intensive-margin uncompensated annual-hours elasticity
- Population/context: Prime-age workers in the United States who are already working
- Units: elasticity
Task:
1. Use exactly the target interpretation above. In `interpretation`, restate it briefly.
2. Give your subjective quantiles p05, p25, p50, p75, and p95 for this quantity.
3. Set `point_estimate` equal to `p50`.
4. Make the quantiles weakly increasing and numerically coherent.
5. In `citations`, list up to 3 source anchors from memory that influenced your belief. These are recall anchors only. If none come to mind confidently, return `[]`.
6. Keep `reasoning_summary` brief and substantive.
Return valid JSON only with exactly this shape:
{
"interpretation": "...",
"point_estimate": <number>,
"quantiles": {
"p05": <number>,
"p25": <number>,
"p50": <number>,
"p75": <number>,
"p95": <number>
},
"citations": ["..."],
"reasoning_summary": "..."
}Read from the archived request logs; all 29 models received this identical text. How the JSON response is enforced varies by provider — see the Methods harness table and the Process page.
Dot: pooled center (mean of run point estimates). Bar: pooled 90 percent mixture interval. Faint underlay: each run's elicited p05–p95. Models sorted by center; color = provider family. Filters change which models render; the axis stays fixed to the full panel.
Review-range sources: CBO 2012; Blundell and MaCurdy 1999. These are hand-coded literature anchors, not benchmark truths.
Alternative estimators (REML and Bayesian hierarchical)
| Model | Pooled 90% | REML predictive 90% | Bayes predictive 90% |
|---|---|---|---|
| gemini-3.1-pro-preview | [-0.15, 0.30] | [-0.13, 0.21] | [-0.06, 0.14] |
| gemini-3.1-flash-lite-preview | [-0.10, 0.30] | [-0.13, 0.24] | [-0.05, 0.16] |
| gpt-5.5 | [-0.10, 0.30] | [-0.11, 0.22] | [-0.04, 0.14] |
| claude-sonnet-5 | [-0.05, 0.29] | [-0.09, 0.20] | [-0.03, 0.14] |
| gpt-5.6-sol | [-0.14, 0.30] | [-0.12, 0.24] | [-0.05, 0.16] |
| gemini-3.6-flash | [-0.10, 0.35] | [-0.13, 0.25] | [-0.05, 0.17] |
| gpt-5.6-luna | [-0.30, 0.47] | [-0.22, 0.36] | [-0.10, 0.23] |
| gemini-3.5-flash | [-0.15, 0.35] | [-0.10, 0.29] | [-0.03, 0.21] |
| qwen-3.7-max | [-0.10, 0.47] | [-0.12, 0.27] | [-0.04, 0.19] |
| gpt-5.4-mini | [-0.10, 0.58] | [-0.13, 0.31] | [-0.04, 0.21] |
| claude-fable-5 | [-0.05, 0.35] | [-0.08, 0.28] | [-0.01, 0.20] |
| gpt-5.4 | [-0.10, 0.40] | [-0.13, 0.33] | [-0.03, 0.23] |
| claude-opus-4.7 | [-0.05, 0.40] | [-0.10, 0.31] | [-0.01, 0.22] |
| claude-sonnet-4.6 | [-0.10, 0.40] | [-0.10, 0.31] | [-0.02, 0.22] |
| gemini-3-flash-preview | [-0.15, 0.50] | [-0.18, 0.39] | [-0.06, 0.26] |
| claude-opus-4.8 | [-0.10, 0.40] | [-0.12, 0.33] | [-0.02, 0.23] |
| gpt-5.6-terra | [-0.14, 0.53] | [-0.15, 0.38] | [-0.04, 0.26] |
| minimax-m3 | [-0.12, 0.49] | [-0.09, 0.32] | [-0.01, 0.22] |
| grok-4.1-fast | [-0.19, 0.78] | [-0.23, 0.48] | [-0.09, 0.31] |
| kimi-k3 | [-0.17, 0.50] | [-0.13, 0.38] | [-0.03, 0.26] |
| grok-4.3 | [-0.17, 0.57] | [-0.14, 0.40] | [-0.03, 0.28] |
| kimi-k2.6 | [-0.05, 0.57] | [-0.13, 0.38] | [-0.02, 0.27] |
| deepseek-v4-pro | [-0.16, 0.75] | [-0.14, 0.39] | [-0.03, 0.27] |
| claude-opus-5 | [-0.05, 0.49] | [-0.10, 0.38] | [0.00, 0.27] |
| claude-haiku-4.5 | [0.01, 0.44] | [-0.04, 0.33] | [0.03, 0.24] |
| gpt-5.4-nano | [-0.39, 0.85] | [-0.19, 0.48] | [-0.12, 0.41] |
| grok-4.20 | [-0.18, 0.66] | [-0.14, 0.47] | [-0.01, 0.33] |
| glm-5.2 | [-0.08, 0.59] | [-0.09, 0.42] | [0.01, 0.31] |
| grok-4.5 | [0.00, 0.57] | [-0.06, 0.40] | [0.03, 0.30] |
The paper's headline object is the pooled mixture interval; the alternatives are robustness estimators (paper, Appendix A2).
Run-level responses
Each model answered 15 independent times. Pick a model to read every run: its elicited 90 percent interval, point estimate, and stated reasoning.
435 successful runs · elicited April and July 2026 · v4 prompts · 15 runs per model-quantity cell. Code · Raw responses · Paper (PDF)
Result directories: gemini-3.1-pro-preview-elasticities-batch15, gemini-3.1-flash-lite-preview-elasticities-batch15, gpt-5.5-elasticities-batch15, and 26 more.