Fit a structural equation model from lavaan syntax.

The workhorse function of fastsemR. Passes lavaan model syntax and a data matrix to the fastsem compiled engine and returns a named list of fit statistics and parameter estimates.

Usage

fastsem_fit(syntax, data, col_names = NULL, control = list())

Arguments

syntax: Character string of lavaan model syntax.
data: Data frame or numeric matrix. Non-numeric columns are coerced to NA. Row order is preserved.
col_names: Character vector of column names. Defaults to colnames(data). Must include every observed variable named in syntax.

Value

A named list with the following elements:

estimator: Character. "ML", "FIML", or "DWLS".
seType: Character. SE method, e.g. "OIM" or "cluster-robust".
nobs: Integer. Number of observations (or effective N for weighted models).
nFreeParams: Integer. Number of free parameters.
df: Integer. Model degrees of freedom.
chi2: Numeric. Chi-square test statistic.
pvalue: Numeric. Chi-square p-value.
neg2logL: Numeric. \(-2 \ln L\) value.
aic: Numeric. Akaike information criterion.
bic: Numeric. Bayesian information criterion.
paramNames: Character vector. Structural names of all free parameters (e.g. "f->y1", "Var(y1)", "y1~1").
estimates: Numeric vector. Parameter estimates in the same order as paramNames. Variance parameters are on the original (not log) scale.
se: Numeric vector. Standard errors.
z: Numeric vector. z-statistics (estimates / se).
stdEst: Numeric vector. StdAll standardised estimates.
stdSE: Numeric vector. Delta-method SEs for stdEst.
groupEstimates: List of numeric vectors (multi-group models). groupEstimates[[g]] contains the per-group estimates for group g in the shared parameter space.
groupSEs: List of numeric vectors (multi-group models).

Estimators

The estimator is chosen automatically based on the data and model:

ML — complete data, no definition variables, no ordinal indicators.
FIML — missing data (NA) present, or definition variables used.
DWLS / WLSMV — ordinal indicators declared with ordered: or threshold syntax (y | t1 + t2).

Extended lavaan syntax

Beyond standard lavaan, fastsem supports:

Parameter labels — b1*var: assign the label b1 to the path.
Equality constraints — two params with the same label are equated, or b1 == b2 on a standalone line.
Bounds — b1 > 0.5 / b1 < 2.0.
Starting values — start(0.8)*var.
Derived parameters — indirect := a * b (delta-method SE).
Definition variables — data.colname * var on the RHS of =~.
Multi-group — group: colname directive plus c(v1, v2)*var per-group annotations.
Weighted observations — weight: colname.
Cluster-robust SEs — cluster: colname.

Examples

if (FALSE) { # \dontrun{
# Single-factor CFA on simulated data
set.seed(1)
n  <- 200
f  <- rnorm(n)
df <- data.frame(y1 = 0.8*f + rnorm(n, sd=0.6),
                 y2 = 0.9*f + rnorm(n, sd=0.5),
                 y3 = 0.7*f + rnorm(n, sd=0.7))

syntax <- "
  f =~ 1*y1 + y2 + y3
  f ~~ f
  y1 ~~ y1;  y2 ~~ y2;  y3 ~~ y3
"
res <- fastsem_fit(syntax, df)
print_fastsem(res)

# Mediation with derived indirect effect
df2 <- as.data.frame(scale(mtcars[, c("mpg","hp","wt")]))
syntax2 <- "
  wt  ~ a*hp
  mpg ~ b*wt + c_prime*hp
  hp ~~ hp;  wt ~~ wt;  mpg ~~ mpg
  hp ~ 1;  wt ~ 1;  mpg ~ 1
  ab := a * b
"
res2 <- fastsem_fit(syntax2, df2)
cat("Indirect hp->wt->mpg:", res2$estimates[res2$paramNames == "ab"],
    "+/-", res2$se[res2$paramNames == "ab"], "\n")
} # }