tangent/sem

Structural equation modeling for JavaScript (ESM). Browser-first, runs in Node.js and Deno. The first maintained SEM implementation in the JavaScript ecosystem — part of the tangent suite, built entirely on its validated numeric leaves: lina (linear algebra), opt (L-BFGS), proba (test distributions).

• Confirmatory factor analysis, path analysis, and full SEM (latent regressions, correlated residuals)
• lavaan-style model syntax: =~, ~, ~~, 1*x to fix, NA*x to free
• Maximum-likelihood estimation (RAM parameterization), expected- information standard errors, χ²/CFI/TLI/RMSEA/SRMR/AIC/BIC
• lavaan identification defaults: first-indicator markers, automatic residual variances and exogenous-factor covariances

Install

npm install @tangent.to/sem     # npm
deno add jsr:@tangent/sem       # Deno / JSR

Usage

import { sem } from '@tangent.to/sem';

const fit = sem(`
  visual  =~ x1 + x2 + x3
  textual =~ x4 + x5 + x6
  speed   =~ x7 + x8 + x9
`, { data });          // rows as objects; or { cov, n, names }

fit.fit.chisq;         // 85.306 on Holzinger-Swineford, matching lavaan
fit.fit.cfi;           // 0.931
fit.estimates;         // [{lhs, op, rhs, est, se, z, pvalue}, ...]
console.log(fit.summary());

Full SEM with latent regressions:

const pd = sem(`
  ind60 =~ x1 + x2 + x3
  dem60 =~ y1 + y2 + y3 + y4
  dem65 =~ y5 + y6 + y7 + y8
  dem60 ~ ind60
  dem65 ~ ind60 + dem60
  y1 ~~ y5
`, { data });

Validation

tests_compare-to-semopy/ fits the two canonical benchmarks with both tangent/sem and semopy (which reproduces R's lavaan) on identical data:

• Holzinger-Swineford CFA: χ² = 85.306 (lavaan's published value, exactly), estimates and expected-information standard errors matching semopy to ~1e-3 relative; CFI/TLI/RMSEA/SRMR/AIC/BIC match lavaan's published output.
• Bollen political democracy full SEM: χ² = 38.125 (lavaan's published value); our optimum is marginally tighter than semopy's and matches lavaan's published parameter table more closely.

npm run test:semopy    # requires uv and Node

Scope

Covariance-structure ML for complete numeric data: CFA, path analysis, full SEM. Not yet: mean structures/intercepts, multiple groups, missing data (FIML), ordinal indicators (WLSMV), robust corrections. Those are roadmap items, prioritized by demand.

License

GPL-3.0 (application layer of the tangent suite; the numeric leaves it builds on are MIT).