gRm

gRm is a native R implementation of the GLLRM-oriented parts of DIGRAM. The public API is a compact statistical modeling workflow:

analysis <- gRm(data, items = c("I1", "I2", "I3"), exogenous = "site")
model <- gllrm(analysis, ld = ~ I1:I2, dif = ~ I3:site)
fit0 <- fit(model)

fit0
summary(fit0)
summary(fit0, which = "parameters")
summary(fit0, which = "thresholds")

item_fit(fit0)
item_fit(fit0, which = "items")

ld <- local_dependence(fit0)
ld
summary(ld)

dif_tests <- dif(fit0)
dif_tests
summary(dif_tests)

global <- global_homogeneity(fit0)
global
summary(global)

score_effects(analysis)

m2(fit0) # experimental
m3(fit0) # experimental

The installed package computes numeric results from data and model objects through native R functions.

Public API

The exported functions are:

• gRm() and read_digram_project() for data setup;
• gllrm() and fit() for model specification and estimation;
• model_graph() for graph extraction from model and fit objects;
• screen() and score_effects() for screening and exogenous score-effect diagnostics;
• item_fit(), local_dependence(), dif(), global_homogeneity(), and ari() for post-fit numeric results;
• experimental m2() and m3() fit diagnostics.

Fitted item parameters and thresholds are reported through summary(fit) and summary(fit, which = "parameters" / "thresholds"). Some diagnostics, such as item_fit() and score_effects(), return their public tables directly. Use summary() for analysis, model, fit, and diagnostic result objects, and use which = only for documented multi-view outputs, such as fitted parameters, item-fit tables, and global-homogeneity sections.

Source Faithfulness

The R implementation is a parallel implementation of selected non-GUI DIGRAM computations in native R. Exported R functions and production computational helpers compute results directly from the original source material.

The current package version covers ordinal item and exogenous variables. Items and exogenous variables are interpreted as ordinal source variables.

Legacy DIGRAM import support covers the simple ordinal category coding subset: category codes in DIGRAM.imv must be contiguous one-based codes matching the values in DIGRAM.csv.

Where a historical DIGRAM value lacks source-backed provenance, R results use NA plus source-status metadata. This includes the global-homogeneity residual cells whose source formula is not available.

Development Notes

Run the package tests with:

R CMD INSTALL .
Rscript -e "testthat::test_local(reporter = 'summary')"

Production R code implements algorithms directly from the original source material.