Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Estimate causal effects from existing data. Use when fitting or interpreting DiD, ITS, synthetic control, regression discontinuity, or other treatment-effect analyses, including robustness checks and counterfactual plots. For choosing a study design before analysis, use designing-experiments instead.
.claude/skills/foryourhealth111-pixel-performing-causal-analysis/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 17% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -50% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -53% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -76% | 0% |
Executes causal analysis on existing data. This skill owns model setup, treatment-effect estimation, counterfactual comparison, robustness checks, and interpretation of fitted causal results.
It does not own the earlier question of which experiment or quasi-experiment should be designed before analysis begins.
summary(), print_coefficients(), and plot().experiment.summary(): Prints model summary and main results.experiment.plot(): Visualizes observed vs. counterfactual.experiment.print_coefficients(): Shows model coefficients.Detailed usage for specific methods:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 9,232 | 9,433 | +2% | 1 | 1 | 0% | 1,714 | 2,010 | +17% | 0 | 0 | — |
case-02 | fail→pass | 17,490 | 8,368 | -52% | 1 | 1 | 0% | 3,562 | 1,798 | -50% | 0 | 0 | — |
case-03 | fail→pass | 18,395 | 21,286 | +16% | 1 | 1 | 0% | 3,626 | 4,525 | +25% | 0 | 0 | — |
case-04 | pass→fail | 17,432 | 3,413 | -80% | 1 | 1 | 0% | 3,013 | 762 | -75% | 0 | 0 | — |
case-05 | fail→pass | 13,919 | 5,590 | -60% | 1 | 1 | 0% | 2,626 | 1,223 | -53% | 0 | 0 | — |
case-06 | pass→pass | 17,941 | 3,744 | -79% | 1 | 1 | 0% | 2,997 | 813 | -73% | 0 | 0 | — |
case-07 | pass→pass | 12,719 | 5,142 | -60% | 1 | 1 | 0% | 2,226 | 1,176 | -47% | 0 | 0 | — |
case-08 | fail→pass | 12,534 | 1,567 | -87% | 1 | 1 | 0% | 2,099 | 512 | -76% | 0 | 0 | — |
case-09 | fail→pass | 12,540 | 1,800 | -86% | 1 | 1 | 0% | 2,191 | 528 | -76% | 0 | 0 | — |
case-10 | pass→pass | 8,720 | 1,613 | -82% | 1 | 1 | 0% | 1,507 | 498 | -67% | 0 | 0 | — |
case-11 | pass→pass | 11,324 | 2,864 | -75% | 1 | 1 | 0% | 2,054 | 675 | -67% | 0 | 0 | — |
case-12 | fail→pass | 9,358 | 2,026 | -78% | 1 | 1 | 0% | 1,546 | 595 | -62% | 0 | 0 | — |
case-13 | pass→pass | 7,320 | 1,902 | -74% | 1 | 1 | 0% | 1,176 | 540 | -54% | 0 | 0 | — |
case-14 | pass→pass | 8,094 | 1,967 | -76% | 1 | 1 | 0% | 1,416 | 543 | -62% | 0 | 0 | — |
case-15 | pass→pass | 14,514 | 3,295 | -77% | 1 | 1 | 0% | 2,599 | 756 | -71% | 0 | 0 | — |
case-16 | fail→pass | 10,917 | 2,372 | -78% | 1 | 1 | 0% | 1,627 | 612 | -62% | 0 | 0 | — |
case-17 | fail→pass | 15,542 | 2,330 | -85% | 1 | 1 | 0% | 2,698 | 646 | -76% | 0 | 0 | — |
case-18 | fail→pass | 5,839 | 2,620 | -55% | 1 | 1 | 0% | 1,035 | 754 | -27% | 0 | 0 | — |
case-19 | fail→pass | 16,346 | 6,033 | -63% | 1 | 1 | 0% | 2,785 | 1,211 | -57% | 0 | 0 | — |
case-20 | pass→pass | 17,407 | 6,659 | -62% | 1 | 1 | 0% | 2,892 | 1,323 | -54% | 0 | 0 | — |
case-21 | pass→pass | 12,260 | 1,605 | -87% | 1 | 1 | 0% | 2,060 | 457 | -78% | 0 | 0 | — |
case-22 | pass→pass | 21,019 | 11,227 | -47% | 1 | 1 | 0% | 3,755 | 2,273 | -39% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +45 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.