SEPA: Subprofile Extraction via Pattern Analysis

SEPA implements the Subprofile Extraction via Pattern Analysis method for ipsatized multivariate data. Given a matrix of domain scores (e.g., WJ-IV broad ability scores), SEPA removes the level-elevation component, fits a rank-K row-isometric SVD biplot, and quantifies each person’s alignment with each domain via direction cosines, with bootstrap confidence intervals.

Installation

# Install from CRAN:
install.packages("SEPA")

Quick start

library(SEPA)

# Run the full pipeline on the bundled synthetic dataset
res <- run_sepa(
  data         = fake_wj,
  K            = 4,
  target_ids   = c(724, 944, 1080, 2117),
  B_dom        = 2000,
  B_cos        = 2000,
  seed         = 20251003
)

print(res)
# SEPA result
#   n = 5127  p = 7  K = 4
#   Domains: LT, ST, CP, AP, VP, CK, FL
#   PA significant dims: 1, 2, 3, 4
#   Marker domains: LT, ST, VP, CK

# Exemplar cosine table for Plane 1
head(res$cosine_tables[["12"]], 14)

# Biplot
draw_sepa_biplot(
  svd_fit      = list(U = res$ref_fit$U, d = res$ref_fit$d_all, V = res$ref_fit$V),
  id_vec       = res$pid,
  domain_names = res$domains,
  ids_highlight = c(724, 944, 1080, 2117)
)

Key functions

Function	Description
`run_sepa()`	Full pipeline wrapper
`sepa_stats_all()`	Plane-fit rho + direction cosines
`parallel_analysis_ipsatized()`	Permutation-based dimensionality selection
`boot_cis_all()`	BCa confidence intervals
`percentile_ci_mat()`	Percentile confidence intervals
`draw_sepa_biplot()`	Row-isometric SVD biplot
`simulate_sepa_fake_wj()`	Synthetic WJ-IV data generator

Data

fake_wj is a 5,127-case synthetic dataset that approximates the WJ-IV marginal distributions. The original norming data are proprietary; this object provides a fully reproducible substitute.

Citation

If you use SEPA in your research, please cite: