data.ecpe.RdECPE dataset from the Templin and Hoffman (2013) tutorial of specifying cognitive diagnostic models in Mplus.
data(data.ecpe)The format of the data is a list containing the dichotomous item
response data data (2922 persons at 28 items)
and the Q-matrix q.matrix (28 items and 3 skills):
List of 2 $ data :'data.frame': ..$ id : int [1:2922] 1 2 3 4 5 6 7 8 9 10 ... ..$ E1 : int [1:2922] 1 1 1 1 1 1 1 0 1 1 ... ..$ E2 : int [1:2922] 1 1 1 1 1 1 1 1 1 1 ... ..$ E3 : int [1:2922] 1 1 1 1 1 1 1 1 1 1 ... ..$ E4 : int [1:2922] 0 1 1 1 1 1 1 1 1 1 ... [...] ..$ E27: int [1:2922] 1 1 1 1 1 1 1 0 1 1 ... ..$ E28: int [1:2922] 1 1 1 1 1 1 1 1 1 1 ... $ q.matrix:'data.frame': ..$ skill1: int [1:28] 1 0 1 0 0 0 1 0 0 1 ... ..$ skill2: int [1:28] 1 1 0 0 0 0 0 1 0 0 ... ..$ skill3: int [1:28] 0 0 1 1 1 1 1 0 1 0 ...
The skills are
skill1: Morphosyntactic rules
skill2: Cohesive rules
skill3: Lexical rules.
The dataset has been used in Templin and Hoffman (2013), and Templin and Bradshaw (2014).
The dataset was downloaded from http://psych.unl.edu/jtemplin/teaching/dcm/dcm12ncme/.
Templin, J., & Bradshaw, L. (2014). Hierarchical diagnostic classification models: A family of models for estimating and testing attribute hierarchies. Psychometrika, 79, 317-339.
Templin, J., & Hoffman, L. (2013). Obtaining diagnostic classification model estimates using Mplus. Educational Measurement: Issues and Practice, 32, 37-50.
GDINA::ecpe
if (FALSE) {
data(data.ecpe, package="CDM")
dat <- data.ecpe$data[,-1]
Q <- data.ecpe$q.matrix
#*** Model 1: LCDM model
mod1 <- CDM::gdina( dat, q.matrix=Q, link="logit")
summary(mod1)
#*** Model 2: DINA model
mod2 <- CDM::gdina( dat, q.matrix=Q, rule="DINA")
summary(mod2)
# Model comparison using likelihood ratio test
anova(mod1,mod2)
## Model loglike Deviance Npars AIC BIC Chisq df p
## 2 Model 2 -42841.61 85683.23 63 85809.23 86185.97 206.0359 18 0
## 1 Model 1 -42738.60 85477.19 81 85639.19 86123.57 NA NA NA
#*** Model 3: Hierarchical LCDM (HLCDM) | Templin and Bradshaw (2014)
# Testing a linear hierarchy
hier <- "skill3 > skill2 > skill1"
skill.names <- colnames(Q)
# define skill space with hierarchy
skillspace <- CDM::skillspace.hierarchy( hier, skill.names=skill.names )
skillspace$skillspace.reduced
## skill1 skill2 skill3
## A000 0 0 0
## A001 0 0 1
## A011 0 1 1
## A111 1 1 1
zeroprob.skillclasses <- skillspace$zeroprob.skillclasses
# define user-defined parameters in LCDM: hierarchical LCDM (HLCDM)
Mj.user <- mod1$Mj
# select items with require two attributes
items <- which( rowSums(Q) > 1 )
# modify design matrix for item parameters
for (ii in items){
m1 <- Mj.user[[ii]]
Mj.user[[ii]][[1]] <- (m1[[1]])[,-2]
Mj.user[[ii]][[2]] <- (m1[[2]])[-2]
}
# estimate model
# note that avoid.zeroprobs is set to TRUE to avoid algorithmic instabilities
mod3 <- CDM::gdina( dat, q.matrix=Q, link="logit",
zeroprob.skillclasses=zeroprob.skillclasses, Mj=Mj.user,
avoid.zeroprobs=TRUE )
summary(mod3)
#*****************************************
#** estimate further models
#*** Model 4: RRUM model
mod4 <- CDM::gdina( dat, q.matrix=Q, rule="RRUM")
summary(mod4)
# compare some models
IRT.compareModels(mod1, mod2, mod3, mod4 )
#*** Model 5a: GDINA model with identity link
mod5a <- CDM::gdina( dat, q.matrix=Q, link="identity")
summary(mod5a)
#*** Model 5b: GDINA model with logit link
mod5b <- CDM::gdina( dat, q.matrix=Q, link="logit")
summary(mod5b)
#*** Model 5c: GDINA model with log link
mod5c <- CDM::gdina( dat, q.matrix=Q, link="log")
summary(mod5c)
# compare models
IRT.compareModels(mod5a, mod5b, mod5c)
}