Skip to contents

Writes the model with mv_kg_model() and fits it with mirt::mirt(). The result is an ordinary mirt model, so tidy(), glance(), anova(), mirt::M2() and the other mirtverse and mirt functions work on it; the graph is kept in its "mv_kg" attribute.

Usage

mv_kg_irt(
  kg,
  data,
  structure = c("correlated", "graph", "orthogonal"),
  itemtype = "2PL",
  method = NULL,
  ...
)

Arguments

kg

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

data

response data. Columns that are not items of the graph, such as an ID column, are dropped.

structure

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

itemtype

item type passed to mirt::mirt(), such as "2PL" or "graded".

method

estimation method; defaults to "EM" with up to three concepts and "QMCEM" with more.

...

further arguments passed to mirt::mirt().

Value

A fitted mirt model (class SingleGroupClass) with the graph in attribute "mv_kg" and the structure in attribute "mv_kg_structure".

Details

With more than three concepts the default estimation method is quasi-Monte Carlo EM ("QMCEM"), which is much faster than standard EM in many dimensions; set method to change it.

Examples

# \donttest{
# about two minutes each: six concepts, 1,200 students
published <- mv_kg_irt(fractions_kg, fractions)
glance(published)
#> # A tibble: 1 × 20
#>    nobs nitems nfact itemtype  npar  logLik    AIC    BIC  SABIC converged
#>   <int>  <int> <int> <chr>    <int>   <dbl>  <dbl>  <dbl>  <dbl> <lgl>    
#> 1  1200     24     6 2PL         66 -16742. 33617. 33953. 33743. TRUE     
#> # ℹ 10 more variables: method <chr>, fit_stat <chr>, M2 <dbl>, df <int>,
#> #   p.value <dbl>, RMSEA <dbl>, SRMSR <dbl>, CFI <dbl>, TLI <dbl>,
#> #   reliability <dbl>
# the strict reading of the graph: Only neighbours correlate
strict <- mv_kg_irt(fractions_kg, fractions, structure = "graph")
#> Warning: Observed log-likelihood may still be increasing.
#>   Final log-likelihood values were: 
#>   -17122.247, -17122.244, -17122.241
anova(strict, published)
#>                AIC    SABIC       HQ      BIC    logLik      X2 df p
#> strict    34356.48 34463.64 34463.85 34641.52 -17122.24             
#> published 33616.61 33742.92 33743.16 33952.56 -16742.31 759.865 10 0
# the true graph (lab_truth) fits better than the published one
true <- mv_kg_irt(lab_truth("fractions")$graph, fractions)
#> Warning: Log-likelihood was decreasing near the ML solution. EM method may be unstable
anova(published, true)
#>                AIC    SABIC       HQ      BIC    logLik     X2 df   p
#> published 33616.61 33742.92 33743.16 33952.56 -16742.31              
#> true      33565.68 33691.99 33692.23 33901.63 -16716.84 50.927  0 NaN
# }
# a quick two-concept model
kg2 <- mv_kg_build(c("meaning", "equivalent"),
                   items = data.frame(item = sprintf("F%02d", 1:8),
                                      concept = rep(c("meaning", "equivalent"), each = 4)))
mod <- mv_kg_irt(kg2, fractions)
tidy(mod)
#> # A tibble: 24 × 5
#>    item  itemtype term  estimate std.error
#>    <chr> <chr>    <chr>    <dbl>     <dbl>
#>  1 F01   2PL      a1        1.53        NA
#>  2 F01   2PL      a2        0           NA
#>  3 F01   2PL      d         1.75        NA
#>  4 F02   2PL      a1        1.32        NA
#>  5 F02   2PL      a2        0           NA
#>  6 F02   2PL      d         1.32        NA
#>  7 F03   2PL      a1        1.79        NA
#>  8 F03   2PL      a2        0           NA
#>  9 F03   2PL      d         2.16        NA
#> 10 F04   2PL      a1        1.42        NA
#> # ℹ 14 more rows