Skip to contents

Turns a knowledge graph into a confirmatory multidimensional model for mirt::mirt(): Each concept that items measure becomes a factor, each item loads on the concepts it measures, and the graph decides which factors may correlate.

Usage

mv_kg_model(
  kg,
  data = NULL,
  structure = c("correlated", "graph", "orthogonal")
)

Arguments

kg

a knowledge graph from mv_kg_build() or as_mv_kg().

data

optional response data; only its items that are in the graph are used, in the order of the data. Defaults to the items of the graph.

structure

"correlated", "graph" or "orthogonal".

Value

A mirt::mirt.model() object. The mirt syntax is in its "item_syntax" attribute, and print() shows it.

Details

The structure sets the factor correlations:

  • "correlated" (the default): All concepts correlate, the usual confirmatory model.

  • "graph": Only concepts joined directly by a prerequisite or related edge correlate; the others are fixed to be uncorrelated. This is a strict reading of the graph. Concepts on the same learning path usually correlate even when they are not neighbours, so expect anova() to favour "correlated"; the size of the gap shows how much the indirect links matter.

  • "orthogonal": No correlations.

Examples

spec <- mv_kg_model(fractions_kg)
cat(attr(spec, "item_syntax"))
#> meaning = F01, F02, F03, F04
#> equivalent = F05, F06, F07, F08
#> mixed = F09, F10, F11, F12, F23
#> compare = F13, F14, F15, F16
#> common_denominator = F17, F18, F19, F22, F24
#> add_subtract = F20, F21, F22, F23, F24
#> COV = meaning*equivalent, meaning*mixed, meaning*compare, meaning*common_denominator, meaning*add_subtract, equivalent*mixed, equivalent*compare, equivalent*common_denominator, equivalent*add_subtract, mixed*compare, mixed*common_denominator, mixed*add_subtract, compare*common_denominator, compare*add_subtract, common_denominator*add_subtract