brycewang-stanford/Auto-Empirical-Research-Skills / skills/11-James-Traina-compound-science/skills/identification-proofs
identification-proofs
VerifiedThis skill covers formal identification arguments and proofs in structural and reduced-form econometrics. Use when the user needs to prove or formalize that a parameter is identified — including writing identification propositions, stating regularity conditions, deriving rank conditions, or showing observational equivalence fails. Triggers on "identification proof", "identification argument", "identify the parameter", "show identification", "identification condition", "exclusion restriction proof", "rank condition", "order condition", "identification strategy formal", "nonparametric identification", "parametric identification", "local identification", "global identification", "observational equivalence", "identification at infinity", "completeness condition", "regularity conditions", "Rothenberg", "proof of identification", "identification result", "identified parameter", "point identified", "set identified", "partial identification".
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 9.7 KB |
| references/derivation-tools.md | markdown | 9.4 KB |
| references/proof-template.md | markdown | 3.1 KB |
| references/regularity-and-partial-id.md | markdown | 4.2 KB |
references/regularity-and-partial-id.md
View on GitHub# Regularity Conditions and Partial Identification
Detailed checklists for regularity conditions and methods for partial identification when point identification fails. Referenced from the main SKILL.md.
---
## Regularity Conditions Checklist
For every identification argument, verify each condition before claiming the result.
- [ ] **Support condition**: Does the instrument $Z$ have sufficient support? For binary $Z$: $\Pr(Z=1) \in (0,1)$. For continuous $Z$ in nonparametric IV: full support on $\mathcal{Z}$.
- [ ] **Rank condition**: Is the Jacobian $G(\theta_0) = \partial\mathbb{E}[m]/\partial\theta'|_{\theta_0}$ full column rank? Check this numerically at your preliminary estimates — a near-singular Jacobian signals weak identification.
- [ ] **Order condition**: Number of moment conditions $L \geq$ number of parameters $K$.
- [ ] **Compactness**: Is the parameter space $\Theta$ compact? Required for most consistency theorems and for applying extreme value theorems in the proof.
- [ ] **Continuity**: Are the moment functions $\theta \mapsto \mathbb{E}[m(X;\theta)]$ continuous (or differentiable where needed) in $\theta$?
- [ ] **Integrability**: Are all expectations $\mathbb{E}[m(X;\theta)]$ finite? Check that $m(X;\theta)$ is dominated by an integrable function uniformly in $\theta \in \Theta$.
- [ ] **Unique zero (global)**: For global identification, is $\theta_0$ the unique solution to $\mathbb{E}[m(X;\theta)] = 0$ over all of $\Theta$?
- [ ] **Monotonicity (for IV LATE)**: Is $D(Z=1) \geq D(Z=0)$ a.s.? This rules out defiers. Often defended by design (one-sided non-compliance) or institutional argument.
- [ ] **No anticipation (for DiD)**: $Y_{it}(g) = Y_{it}(\infty)$ for $t < g$ (potential outcomes before treatment are unaffected by future treatment status).
- [ ] **Overlap / common support (for ATE, ATT)**: $0 < \Pr(D=1|X) < 1$ over the support of $X$.
**What to do when a condition fails:**
- Rank condition fails: the model is not identified — revisit assumptions or add instruments
- Support condition fails (e.g., instrument has limited support): partial identification may be possible; see below
- Global uniqueness hard to verify: report local identification, note the caveat, use multiple starting values to search for alternative solutions
---
## Partial Identification
When point identification fails, characterize what the data *do* identify: the identified set $\Theta^* = \{\theta \in \Theta : P_\theta = P_{\theta_0}\}$.
### Manski (1990) Sharp Bounds
The canonical example is the average treatment effect under sample selection or missing data. With binary outcome $Y \in \{0,1\}$ and binary treatment $D$:
$$E[Y(1)] \in \left[\frac{E[YD]}{E[D]}, \quad \frac{E[YD]}{E[D]} + \frac{E[1-D]}{1}\right]$$
(the lower bound uses $Y(1)=0$ for non-treated; the upper bound uses $Y(1)=1$). These *sharp* bounds — the tightest possible given the observable distribution — define the identified set without assuming missing-at-random.
Tightening sharp bounds requires additional assumptions: monotone treatment response, monotone treatment selection, or an instrument. Each additional assumption narrows $\Theta^*$. The partial identification approach is explicit about the trade-off between assumptions and set width.
### Intersection Bounds
When multiple identifying assumptions each imply an upper or lower bound on $\theta_0$, the intersection of these bounds is sharper:
$$\theta_0 \in \bigcap_{k} \left[L_k, U_k\right].$$
Chernozhukov, Lee, and Rosen (2013) provide inference methods for intersection bounds.
### Interval Regression
With interval-censored data — $Y \in [Y_L, Y_U]$ — the regression coefficient is set-identified. Stoye (2010) characterizes the identified set and provides valid confidence regions.
### Connection to Sensitivity Analysis
Partial identification and sensitivity analysis are related: Oster (2019) bounds on treatment effects under proportional selection on observables are equivalent to characterizing the identified set under a restriction on the degree of selection. The `empirical-playbook` skill (`sensitivity-analysis.md`) covers Oster bounds and related sensitivity exercises. Both approaches answer the same question: "How much can the unidentified component vary, and what does that imply for the parameter?"
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |