When a direct cause-and-effect relationship is obscured by hidden factors that affect both your treatment and outcome, instrumental variables help isolate the impact. Use this method when you have a secondary variable—the instrument—that influences your treatment but has no direct connection to the outcome itself.
Understanding 2SLS Estimation
The instrumental variables method utilizes a two-stage least squares (2SLS) approach. In the first stage, the platform estimates the treatment variable using your instruments and exogenous covariates. This captures the part of the treatment variation that is independent of unobserved factors.
In the second stage, the method uses these estimated treatment values to calculate the effect on the outcome. This process yields the Local Average Treatment Effect (LATE), providing a clearer view of the causal relationship than standard linear regression could offer in the presence of endogeneity.
Automatic Diagnostic Checks
Lattice performs automated diagnostics to ensure your results remain credible. Every analysis includes a first-stage F-test to confirm the strength of your instruments. If your model uses more instruments than endogenous variables, the platform also automatically runs a Sargan J test.
The Sargan test checks the validity of your over-identifying restrictions. If this test rejects the null hypothesis, it suggests that one or more of your instruments may directly impact the outcome, violating the exclusion restriction. In such cases, the platform alerts you to review your instrument selection.
Interpreting Model Robustness
Beyond coefficients, the platform provides HC0 robust standard errors by default to account for heteroskedasticity. This ensures your p-values and confidence intervals are calculated using a reliable foundation, reducing the risk of over-optimistic conclusions.
Because causal analysis relies heavily on the quality of the data and the validity of assumptions, Lattice flags concerns if diagnostics like the F-test or Sargan test suggest potential issues. Always check these flags to determine if your results require further sensitivity analysis or a different set of instruments.
1 · Intent → method
An LLM picks causal_iv from a fixed catalog.
2 · Method → numbers
Deterministic Python engine runs the math. Same input → same output.
3 · Numbers → plain language
A second LLM translates the result into your domain’s vocabulary.
What is an instrument?
An instrument is a variable that influences your treatment choice but doesn't directly affect the outcome. It acts as a proxy to filter out the 'noisy' part of the treatment variable, allowing you to estimate a cleaner causal effect.
How does Lattice detect weak instruments?
Lattice automatically runs a first-stage F-test for each endogenous variable. If the F-statistic falls below 10, it flags a concern because a weak instrument leads to significant bias in 2SLS estimates, meaning the results shouldn't be interpreted as a reliable causal effect.
Tool input schema
Schema for causal_iv not exported yet (run pnpm export:registry).