Pivot-Style Statistical Tables for Large-Scale Assessment Data

Turns assessment and survey sample data (demographics, scores, sampling weights, ability estimates with individual item response theory standard errors, replicate weights and plausible values) into fully customizable pivot-style statistical tables in which every cell carries a design-appropriate standard error. Provides weighted means, proportions above cut scores, proficiency-level percentages and quantiles; sampling variance via linearization, Woodruff (1952) intervals for quantiles, or balanced repeated replication and jackknife replicate weights including Fay's method; measurement variance via delta-method propagation of individual standard errors or Rubin (1987) combination across plausible values. Imports data from 'CSV', 'Excel', 'SPSS', 'SAS', 'Stata' and item response theory software person files ('Winsteps', 'ConQuest'); a configuration-file interface serves non-programmers and automation; results export to formatted 'Excel', 'JSON' and standalone 'HTML' reports with rule-based plain-language interpretation.

LISTC

Pivot-style statistical tables for large-scale assessment data — every cell with the right standard error.

中文文档 → · Quick start for non-programmers · Advanced guide · Design document

LISTC turns assessment and survey microdata — demographics, scores, sampling weights, IRT ability estimates with individual standard errors, replicate weights, plausible values — into Excel-pivot-style tables where you freely choose the row variables, column variables and cell statistics, and every cell carries a design-appropriate standard error.

No existing R package combines both halves: survey packages (EdSurvey, intsvy, Rrepest, BIFIEsurvey) compute correct statistics but offer no free-form table layout; pivot packages (pivottabler) lay out tables but know nothing about weights, replicate designs or measurement error. LISTC does both, at scale: 5 million respondents, full table set, ~10 seconds on a laptop.

Installation

# install.packages("LISTC")                      # once on CRAN
# install.packages("remotes")
remotes::install_github("weiandata/LISTC")       # development version

Quick example

library(LISTC)

# 1. Declare what each column means
x <- lst_data(students,
  id       = student_id,
  group    = c(region, gender),
  weight   = w_final,
  theta    = c(math = th_math),      # IRT ability estimate
  theta_se = c(math = se_math)       # its individual standard error
)

# 2. Lay out the pivot table: rows, columns, cell statistics
lv <- c(below = -Inf, basic = -0.5, good = 0.5, excellent = 1.2)
tab <- lst_table(x,
  rows = region, cols = gender,
  values = list(
    mean      = st_mean(math),
    excellent = st_prop_above(math, cutoff = 1.2, method = "prob"),
    levels    = st_level_prop(math, breaks = lv, method = "prob"),
    n         = st_count()
  ),
  margins = TRUE
)

# 3. Use the results
tab                          # wide layout: "0.52 (0.03)" per cell
as_long(tab)                 # tidy data with se_sampling / se_measurement
lst_to_excel(tab, "results.xlsx")
lst_to_html(tab, "report.html")
lst_interpret(tab)           # rule-based plain-language conclusions

One engine, three audiences

Variance engines

Total variance = sampling + measurement, always reported as separate components (se_sampling, se_measurement, se_total).

All formulas are validated against Monte Carlo simulations in the test suite (96.3% coverage; core engine files > 95%). A notable result baked into the design: for EAP estimates with posterior SDs, probabilistic classification is already calibrated — while WLE/ML estimates need the "latent" empirical-Bayes correction. See design doc §4.1/§6.

Data in, results out

In: csv/tsv (data.table::fread), Excel, SPSS/SAS/Stata with value labels (haven), Winsteps PFILE, ConQuest person files. Out: styled Excel workbooks (Chinese-friendly fonts and widths), tidy JSON with computation metadata, standalone HTML reports — each with rule-based plain-language interpretation that guards non-experts against misreading standard errors.

Performance

data.table backend, column pruning on import, per-item chunking. Measured on Apple Silicon (see scripts/benchmark.R):

Documentation

• Vignette: vignette("LISTC-intro")
• Non-programmer guides (EN/中文): docs/guides/
• Design document with all methods and Monte Carlo evidence: docs/design-v1.md
• For LLM agents: system.file("llms.txt", package = "LISTC")

License

GPL (>= 2). Copyright (c) 2026 WEIAN DATA TECH (Beijing) Co., Ltd. See inst/COPYRIGHTS.