ivsfm.Rd: an estimator table up front, and say what C2SLS returns (issue #3)

[davidhbernstein]

This continues the issue #3 readability pass with ivsfm.Rd. I held it until #46 merged, since #46 also edits this page's uhet entry.

Corrected facts

These were measured on the page's own example (n = 600) on main.

the page said	what the fit returns
parameters are "the frontier coefficients, sigma_u, sigma_v, …, one rho_ per endogenous variable"	true for "IVLIML" and "IVCF". "C2SLS" reports sigma_u and sigma_v first (rownames(out): sigma_u sigma_v (Intercept) x1 x2), all with NA standard errors
rho, vcov_rho, sigma_c listed as components of every fit	"C2SLS" fits carry none of them
Pi, Sigma_xi: "reduced-form coefficient matrix / error covariance"	estimated jointly only for "IVLIML"; "IVCF" and "C2SLS" return the least-squares values (Pi_ols, Sxx_ols in R/ivsfm.R)

The example's claim that ignoring endogeneity biases x2 toward zero holds. The truth is −0.6; IVLIML gives −0.609, and the NHN sfm() fit gives −0.466.

Layout

• \description now has a table of the three estimators with what each is for. It records that "C2SLS" takes no uhet, per ivsfm: refuse uhet for C2SLS instead of silently ignoring it #46.
• \details is under titled subsections: formula syntax, the model, the estimators, inefficiency, what is assumed, and standard errors for ρ.
• The 1.2.0 note on ρ's standard errors moved from \value to the last subsection.

No technical content was removed. \usage is byte-identical, and tools::checkRd() is clean. The rendered text was checked with Rdconv -t txt.

🤖 Generated with Claude Code

https://claude.ai/code/session_014oJL17ewiahA8pWbemnqNp

---

Generated by Claude Code