sfareg-methods.Rd: a method table up front, and correct what vcov and nobs do (issue #3)

[davidhbernstein]

This continues the issue #3 readability pass with sfareg-methods.Rd (coef, vcov, logLik, nobs, estfun, bread). Three statements on the page were stale; each was checked against R/sfareg_methods.R and R/sandwich_methods.R.

Corrected facts

the page said	the code does
vcov() "is solve(object$opt$hessian)"	.sfa_solve_equilibrated() scales the Hessian to unit diagonal (when that diagonal is positive) before inverting. .sfa_map_vcov() then maps the inverse onto the reported scale. The diagonal fallback warns only when inversion or mapping fails. A fit with no Hessian falls back to the diagonal silently, or, with no SEs either, returns NA with a warning (checked on an optHessian = FALSE fit).
nobs() reads object$data, or else re-evaluates the call's data	it takes the stored count (object$nobs) first, then object$data, then a per-observation vector, and only then re-evaluates data. Re-evaluated data must pass the A47 data check. Checked: an sfm fit with one NA row reports 299 of 300.
"sfareg" is returned by psfm, sfm, zsfm, ttsfm	also by lcsfm, copsfm, ivsfm, selsfm and esfm (class c("esfm", "sfareg")). npsfm returns "npsfareg", so it is not listed.

Layout

• \description now has a method / returns table.
• \details opens with "What each method returns", which moved from the bottom of the page.
• The rest is under titled subsections: four covariances, BHHH, sandwich, which fits supply scores, the scale the scores live on, two easily-confused quantities, and estimators with no log-likelihood.

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

🤖 Generated with Claude Code

https://claude.ai/code/session_014oJL17ewiahA8pWbemnqNp

---

Generated by Claude Code