Calculate latent SSM parameters from a fitted lavaan measurement model

The low-level adapter behind ssm_sem(): take an already fitted lavaan model of a fixed-angle circumplex measurement structure (as generated by ssm_sem_syntax(), possibly user-modified – e.g., a partial-invariance respecification) and compute latent SSM parameter estimates with in-package confidence intervals. Compatibility with the expected parameter structure is checked structurally (the named loading, factor-covariance, and measure-covariance parameters must be present), not by provenance.

Usage

ssm_sem_parameters(
  fit,
  scales,
  angles = octants(),
  measures = NULL,
  ci_method = c("mvn", "boot"),
  boots = 2000,
  interval = 0.95,
  contrast = FALSE,
  parallel = "no",
  ncpus = 1
)

Arguments

fit: Required. A fitted lavaan object whose model preserves the ssm_sem_syntax() parameter structure (factors g, cx, cy; the measures covarying with them). For ci_method = "mvn", fit the model with robust (sandwich) standard errors (lavaan's se = "robust.huber.white", ssm_sem()'s default) so the propagated covariance stays valid when the fixed-angle model is an approximation; see the se argument of ssm_sem().
scales: Required. A character vector with the scale (indicator) names, in the same order as angles.
angles: Optional. A numeric vector of the scales' theoretical angles in degrees (default = octants()). Must be the angles the model was generated with.
measures: Optional for multi-group fits, required otherwise. A character vector of the measure names; NULL on a multi-group fit selects the latent MEAN path (the fit must carry the mean structure: scale intercepts and latent means).
ci_method, boots, interval, contrast, parallel, ncpus: See ssm_sem(). Note that for a multi-group fit the CONTRAST DIRECTION (and the group labels in the output) follows the fit's own group order – lavaan's default is order of appearance in the data unless group.label was supplied at fitting time – so read the direction from the output's Group column, not from factor-level conventions.

Value

A circumplex_ssm_sem object; see ssm_sem().

Details

Important: multi-group fits are supported here as the partial-invariance escape hatch, and this path bypasses the invariance gating that ssm_sem() applies. Where ssm_sem() fits a configural-metric-scalar ladder and refuses a latent group contrast when the required rung is rejected, ssm_sem_parameters() computes the contrast from whatever multi-group fit you supply without testing invariance at all. You own the comparability claim: the groups are compared on this instrument's latent metric only to the extent the model you fitted makes them comparable.

Reproducibility

This function consumes R's random number stream for both ci_method settings ("mvn" through the package's own draws; "boot" through a seed handed to lavaan's bootstrap). Call set.seed() immediately before ssm_sem() for reproducible confidence intervals.

Examples

# \donttest{
data("jz2017")
scales <- c("PA", "BC", "DE", "FG", "HI", "JK", "LM", "NO")
syn <- ssm_sem_syntax(scales = scales, angles = octants(), measures = "NARPD")
# Robust (sandwich) SEs so the mvn engine propagates a
# misspecification-consistent covariance (ssm_sem()'s default)
fit <- lavaan::cfa(syn, data = jz2017, se = "robust.huber.white")
set.seed(12345)
ssm_sem_parameters(fit, scales = scales, measures = "NARPD", boots = 500)
#> 
#> # Latent (SEM-based) SSM
#> 
#> Measurement model:	 scaled fixed-angle circumplex
#> Global fit (N = 1166): chisq(17) = 356.099, p < 0.001 
#> 			CFI = 0.93, RMSEA = 0.131, SRMR = 0.072
#> 
#> # Profile [NARPD]:
#> 
#>                Estimate   Lower CI   Upper CI
#> Elevation         0.249      0.210      0.295
#> X-Value          -0.009     -0.054      0.033
#> Y-Value           0.231      0.189      0.273
#> Amplitude         0.232      0.192      0.274
#> Displacement     92.132     82.515    104.461
#> Model Fit         0.975                      
#> 
# }