1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
|
#options(error = browser)
library(OpenMx)
library(rpf)
set.seed(9)
numItems <- 20
items <- list()
items[1:numItems] <- list(rpf.grm(outcomes=2)) # equivalent to 2PL
# create random item parameters
correct <- sapply(items, rpf.rparam, version=1)
# simulate data
data.g1 <- rpf.sample(500, items, correct)
data.g2 <- rpf.sample(500, items, correct, mean=-1, cov=matrix(5,1,1))
data.g3 <- rpf.sample(500, items, correct, mean=1, cov=matrix(.5,1,1))
if (0) {
# for flexMIRT, write CSV
write.table(sapply(data.g1, unclass)-1, file="drm-mg2-g1.csv", quote=FALSE, row.names=FALSE, col.names=FALSE)
write.table(sapply(data.g2, unclass)-1, file="drm-mg2-g2.csv", quote=FALSE, row.names=FALSE, col.names=FALSE)
write.table(sapply(data.g3, unclass)-1, file="drm-mg2-g3.csv", quote=FALSE, row.names=FALSE, col.names=FALSE)
}
# This create an IFA model for each group. Item parameters are
# constrained equal across groups.
mkgroup <- function(model.name, data, latent.free) {
ip.mat <- mxMatrix(name="item", nrow=2, ncol=numItems,
values=c(1,0), free=TRUE)
colnames(ip.mat) <- colnames(data)
rownames(ip.mat) <- c('f1', 'b')
for (ix in 1:numItems) {
for (px in 1:2) {
name <- paste(c('p',ix,',',px), collapse='')
ip.mat$labels[px,ix] <- name
}
}
dims <- 1
m.mat <- mxMatrix(name="mean", nrow=1, ncol=dims, values=0)
rownames(m.mat) <- "f1"
cov.mat <- mxMatrix(name="cov", nrow=dims, ncol=dims, values=diag(dims))
dimnames(cov.mat) <- list("f1", "f1")
mean <- "mean"
cov <- "cov"
if (latent.free) {
lm <- paste(model.name, "latent", sep="")
mean <- paste(lm, "expMean", sep=".")
cov <- paste(lm, "expCov", sep=".")
}
m1 <- mxModel(model=model.name, ip.mat, m.mat, cov.mat,
mxData(observed=data, type="raw"),
mxExpectationBA81(ItemSpec=items, mean=mean, cov=cov,
verbose=ifelse(latent.free, 0L, 0L)),
mxFitFunctionML())
m1
}
g1 <- mkgroup("g1", data.g1, FALSE)
g2 <- mkgroup("g2", data.g2, TRUE)
g3 <- mkgroup("g3", data.g3, TRUE)
groups <- paste("g", 1:3, sep="")
# Cannot test derivatives at starting values because Hessian starts very close to singular.
# This create a latent distribution model that can be used to impose
# equality constraints on latent distribution parameters.
mklatent <- function(name) {
mMat <- mxMatrix(nrow=1, ncol=1, free=T, values=0, name="expMean")
rownames(mMat) <- "f1"
covMat <- mxMatrix(type="Symm", nrow=1, ncol=1, free=T, values=1, name="expCov")
dimnames(covMat) <- list("f1", "f1")
m1 <- mxModel(paste(name, "latent", sep=""), mMat, covMat,
mxDataDynamic("cov", expectation=paste(name, "expectation", sep=".")),
mxExpectationNormal(covariance="expCov", means="expMean"),
mxFitFunctionML())
m1
}
latent <- mxModel("latent", mxFitFunctionMultigroup(paste0(groups[-1],"latent")))
g2.latent <- mklatent("g2")
g3.latent <- mklatent("g3")
latent.vargroup <- apply(expand.grid(paste(groups[-1], "latent", sep=""), c('expMean','expCov')),
1, paste, collapse='.')
latent.plan <- NULL # need a plan for latent distribution parameters
if (0) {
# Copy latent distribution parameters from current estimates without transformation.
latent.plan <- mxComputeSequence(list(mxComputeOnce(paste(groups, 'expectation', sep='.')),
mxComputeOnce(paste(groups, 'expectation', sep='.'),
"latentDistribution", "copy"), # c('mean','covariance')
mxComputeOnce('fitfunction', "set-starting-values")),
freeSet=latent.vargroup)
} else {
# Obtain latent distribution parameters via mxExpectationNormal.
# This permits equality constraints (and potentially more complex latent structure).
# latent.plan <- mxComputeGradientDescent(latent.vargroup, fitfunction="latent.fitfunction")
latent.plan <- mxComputeNewtonRaphson(latent.vargroup, fitfunction="latent.fitfunction")
}
grpModel <- mxModel(model="groupModel", g1, g2, g3, g2.latent, g3.latent, latent,
mxFitFunctionMultigroup(groups),
mxComputeSequence(list(
mxComputeEM(paste(groups, 'expectation', sep='.'), 'scores',
mxComputeSequence(list(
mxComputeNewtonRaphson(freeSet=paste(groups,'item',sep=".")),
latent.plan)),
information="mr1991", verbose=0L,
tolerance=1e-10,
infoArgs=list(fitfunction=c("fitfunction", "latent.fitfunction"))),
mxComputeStandardError(),
mxComputeHessianQuality(),
mxComputeOnce(c('fitfunction', paste0('g',2:3,'latent.fitfunction')),
'gradient'),
mxComputeReportDeriv())))
#grpModel <- mxOption(grpModel, "Number of Threads", 1)
grpModel <- mxOption(grpModel, "Checkpoint Units", "iterations")
grpModel <- mxOption(grpModel, "Checkpoint Count", 1)
grpModel <- mxRun(grpModel) #, checkpoint = TRUE
#dm <- grpModel$compute$steps[[1]]$debug$rateMatrix
omxCheckCloseEnough(max(abs(grpModel$output$gradient)), 0, .17)
omxCheckCloseEnough(grpModel$output$fit, 30114.94, .02)
omxCheckEquals(grpModel$output$fitUnits, "-2lnL")
omxCheckCloseEnough(AIC(logLik(grpModel)), 30202.95, .02)
omxCheckCloseEnough(BIC(logLik(grpModel)), 30436.73, .02)
omxCheckCloseEnough(summary(grpModel)$informationCriteria['AIC:','par'], 30202.95, .02)
omxCheckCloseEnough(summary(grpModel)$informationCriteria['BIC:','par'], 30436.73, .02)
omxCheckCloseEnough(grpModel$submodels$g2latent$matrices$expMean$values, -.834, .01)
omxCheckCloseEnough(grpModel$submodels$g2latent$matrices$expCov$values, 3.93, .01)
omxCheckCloseEnough(grpModel$submodels$g3latent$matrices$expMean$values, .933, .01)
omxCheckCloseEnough(grpModel$submodels$g3latent$matrices$expCov$values, .444, .01)
emstat <- grpModel$compute$steps[[1]]$output
if (0) {
# Optimization path has too much variance
omxCheckCloseEnough(emstat$EMcycles, 50, 4)
omxCheckCloseEnough(emstat$totalMstep, 380, 12)
omxCheckCloseEnough(emstat$semProbeCount, 90, 15)
# cat(deparse(round(grpModel$output$standardErrors, 3)))
semse <- c(0.069, 0.077, 0.074, 0.077, 0.094, 0.097, 0.125, 0.111, 0.069, 0.074,
0.132, 0.116, 0.08, 0.081, 0.209, 0.163, 0.102, 0.133, 0.114, 0.107,
0.205, 0.151, 0.068, 0.077, 0.073, 0.138, 0.078, 0.081, 0.088, 0.087,
0.061, 0.068, 0.125, 0.11, 0.084, 0.09, 0.094, 0.094, 0.092, 0.089,
0.11, 0.399, 0.068, 0.055)
omxCheckCloseEnough(c(grpModel$output$standardErrors), semse, .03)
omxCheckCloseEnough(log(grpModel$output$conditionNumber), 5.1, .5)
omxCheckTrue(grpModel$output$infoDefinite)
}
refModels <- mxRefModels(grpModel, run=TRUE)
ind <- refModels[['Independence']]
omxCheckCloseEnough(ind$output$fit, 38588.19, .01)
omxCheckCloseEnough(summary(ind)$informationCriteria['AIC:','par'], 38628.187, .02)
omxCheckCloseEnough(summary(ind)$informationCriteria['BIC:','par'], 38734.451, .02)
i1 <- mxModel(grpModel,
mxComputeSequence(steps=list(
mxComputeOnce('fitfunction', 'information', "meat"),
mxComputeStandardError(),
mxComputeHessianQuality())))
i1 <- mxRun(i1)
#cat(deparse(round(i1$output$standardErrors,3)))
se <- c(0.071, 0.078, 0.076, 0.079, 0.097, 0.099, 0.132, 0.117, 0.075,
0.077, 0.135, 0.121, 0.081, 0.083, 0.215, 0.169, 0.111, 0.141,
0.121, 0.113, 0.213, 0.159, 0.074, 0.082, 0.077, 0.139, 0.084,
0.087, 0.095, 0.09, 0.064, 0.07, 0.135, 0.115, 0.091, 0.095, 0.097,
0.098, 0.096, 0.093, 0.12, 0.512, 0.072, 0.057)
omxCheckCloseEnough(c(i1$output$standardErrors), se, .01)
omxCheckCloseEnough(log(i1$output$conditionNumber), 5.6, .5)
if (0) {
library(mirt)
rdata <- sapply(data, unclass)-1
# for flexMIRT, write CSV
#write.table(rdata, file="ifa-drm-mg.csv", quote=FALSE, row.names=FALSE, col.names=FALSE)
pars <- mirt(rdata, 1, itemtype="2PL", D=1, quadpts=49, pars='values')
pars[pars$name=="a1",'value'] <- 1
pars[pars$name=="a1",'est'] <- FALSE
pars[pars$name=="COV_11",'est'] <- TRUE
fit <- mirt(rdata, 1, itemtype="2PL", D=1, quadpts=49, pars=pars)
# LL -7064.519 * -2 = 14129.04
got <- coef(fit)
print(got$GroupPars)
# COV 4.551
got$GroupPars <- NULL
round(m2$matrices$item$values - simplify2array(got), 2)
}
|
