---
title: "Replicating the Mplus DIFFTEST Procedure"
subtitle: "An R Function to Reproduce Nested Model Comparisons"
author:
- name: Ben A. Kite
affiliation: CRMDA, University of Kansas
email: bakite@ku.edu
- name: Paul E. Johnson
affiliation: CRMDA, University of Kansas
email: pauljohn@ku.edu
- name: Chong Xing
affiliation: CRMDA, University of Kansas
email: cxing@ku.edu
advertise: >
**CRMDA Guide #44**. Please visit
[http://crmda.ku.edu/guides](http://crmda.ku.edu/guides) for updates.
keywords: >
Structural Equation Modeling (SEM), Chi-Square Difference test,
Categorical Data. Weighted Least Squares (WLSMV), Mplus DIFFTEST
abstract: >
In the Mplus statistical software suite
([Muthén & Muthén, 1998-2017](https://www.statmodel.com/download/usersguide/MplusUserGuideVer_8.pdf)),
the $T_3$ method
([Asparouhov & Muthén, 2010](https://www.statmodel.com/download/WLSMV_new_chi21.pdf))
is used to conduct nested model comparisons ("difftest") with data that is not
normally distributed. The Mplus $T_3$ method is widely used for comparison
of models fitted with Likert-type ordinal variables, and yet it
is not widely understood. This technical report describes the
anatomy of the Mplus $T_3$ DIFFTEST procedure.
We offer an R (http://r-project.org) function
`crmda_difftest` which will import Mplus output files and reconstitute
matrices needed to calculate $T_3$. We provide an exhaustive
collection of Mplus examples, along with their output, to demonstrate
the fact that the $T_3$ estimates from Mplus are reproduced exactly.
checked_by: "Paul Johnson"
date: "`r format(Sys.time(), '%Y %B %d')`"
output:
crmda::crmda_html_document:
toc: true
toc_depth: 3
highlight: haddock
number_sections: true
css: theme/kutils.css
template: theme/guide-boilerplate.html
logoleft: theme/jayhawk.png
logoright: theme/CRMDAlogo-vert.png
---
```{r setup, include=FALSE}
##This Invisible Chunk is required in all CRMDA documents
outdir <- paste0("tmpout")
if (!file.exists(outdir)) dir.create(outdir, recursive = TRUE)
knitr::opts_chunk$set(echo=TRUE, comment=NA, fig.path=paste0(outdir, "/p-"))
options(width = 75)
```
```{r themecopy, include = FALSE}
library(kutils)
library(crmda)
## If theme directory does not have required images or TeX files
## we need to retrieve them and put them in "theme" directory.
logos <- c(logoleft = "jayhawk.png",
logoright = "CRMDAlogo-vert.png")
files <- c("kutils.css",
"guide-boilerplate.html")
getFiles(logos, pkg = "crmda")
getFiles(files, pkg = "crmda")
```
```{r, echo=T, include=FALSE}
pdf.options(onefile=FALSE, family="Times", paper="special", height=4,
width=6, pointsize=10)
## Reading in `crmda_difftest` and table making function
source("difftest_functions.R")
library(xtable)
library(MplusAutomation)
```
# Motivation
The Mplus[`DIFFTEST`](http://www.statmodel.com/download/usersguide/Chapter13.pdf)
is a popular procedure (a chi-square difference based test) for
comparing nested structural equation models. It is available for
models estimated by the "MLMV", "WLSMV", and "ULSMV" procedures in
Mplus. A model including a larger number of parameters is estimated,
and then a second model is estimated with some parameters that are
restricted (for example, to be equal to each other, or equal to
0). Then the two models are compared using a statistic, referred to as
$T_3$, that is supposed to be distributed as a $\chi^2$ random
variable. When the test comparing models indicates that a difference
is not statistically significant, or the equal-fit hypothesis is not
rejected, researchers generally prefer the simpler model, the one which
estimates fewer parameters.
In Mplus, each input file can be used to estimate only one model. To
communicate the results of estimation from the first to the second
model, the Mplus
[`SAVEDATA`](http://www.statmodel.com/download/usersguide/Chapter18.pdf)
command creates a text file that is a single column of numbers that must be
sorted into various vectors and matrices. This file, commonly referred
to as a `DIFFTEST` file, is a snapshot summary of the internal state
of the model. This file is then put to use by a `DIFFTEST` statement
in the input file for the second model.
There are two elements in the process that seem to be rather like a
"black box". First, the format of the `DIFFTEST` file itself is not
formally documented. It is simply a column of numbers. It is not
apparent to the user how these values can be used to reconstruct the
matrices needed to calculate the difference of models test. Second,
the details required to calculate $T_3$ are elusive. There are a number
of elements in the calculation, especially for multi-group models,
that are not publicly documented.
As we illustrate in this report, it is possible to unbox both of these
elements. We parse the `DIFFTEST` files from two models and from them it
is possible to recalculate the value of $T_3$ reported by Mplus. To
the best of our knowledge, our R function, dubbed the
`crmda_difftest`, provides values of the $T_3$ statistic that matches
the Mplus `DIFFTEST` results perfectly (Mplus reports values to the
third decimal place).
The following sections first describe the two-step procedure in Mplus
for conducting $T_3$ `DIFFTEST`. Second, calculation steps in
`crmda_difftest` for reproducing $T_3$ statistics are outlined - from
importing a `DIFFTEST` file, to separating the values into a collection
of matrices, and to using these matrices for $T_3$ calculation. The last
section reports a comprehensive series of models which demonstrate that the
`crmda_difftest` procedures reproduce/match the values reported by Mplus.
Applications of `crmda_difftest` to 36 model specifications are
included to demonstrate the exact match with
Mplus results. These examples employ data from the Human Behavior in
School-Aged Children (HBSC) project
([Iannotti, 2005-2006](https://www.icpsr.umich.edu/icpsrweb/ICPSR/studies/28241?q=HBSC)).
The Mplus input (`inp`), output (`out`), and difftest (`dif`) files
for all 36 models are included in
[the folder with this report](http://crmda.dept.ku.edu/guides/44.difftest).
The user should find a number of files.
1. `44.difftest.Rmd`. The R markdown file with which the HTML is
prepared.
2. `difftest_functions.R`. Includes the main function `crmda_difftest` and
several auxiliary functions.
3. A plethora of Mplus input files named `m*.inp`, along with their
associated output and snapshot `DIFFTEST` files.
# The $T_3$ Method
The $T_3$ (second order) chi-square difference correction has been
implemented in Mplus since Version 6
([Asparouhov & Muthén, 2010](https://www.statmodel.com/download/WLSMV_new_chi21.pdf)). It was proposed as a replacement of the procedure suggested
by Satorra ([1999](https://econ-papers.upf.edu/papers/395.pdf),
[2000](https://link.springer.com/chapter/10.1007/978-1-4615-4603-0_17)),
which we refer to as $T_2$
([Asparouhov & Muthén, 2006](https://www.statmodel.com/download/webnotes/webnote10.pdf)).
One criticism of $T_2$ is that the degrees of freedom for the test is
a floating point number (which is typically rounded to the nearest
integer). The $T_2$ method is still available in Mplus when
`Satterthwaite = ON` is specified under the `ANALYSIS` command in Mplus
model syntax. Instead of involving a $df$ approximation, the degrees
of freedom for a $T_3$ statistic is an integer, the difference in the
number of parameters in two models.
The calculation of the $T_3$ statistic is somewhat complicated. One
must first calculate a summary of the difference between the two models.
The difference between models can be referred to as $T.chisq$.
This value is not in fact distributed as a $\chi^2$ statistic,
an issue which has been widely discussed in the structural equation
modeling (SEM) literature.
To remedy that fact, Asparouhov and Muthén
([2010](https://www.statmodel.com/download/WLSMV_new_chi21.pdf))
propose a linear transformation of $T.chisq$. To perform the
transformation, a 'scale factor', referred to as $a$, and a 'shift
parameter', $b$, are needed. The $T_3$ statistic/transformation is computed with
the following equation:
\[
T_3 = T.chisq \times a + b. \tag{1}
\]
Asparouhov and Muthén suggest that this shifting and scaling procedure
produces a test statistic that is more desirable than the previously
proposed methods (e.g., $T_2$), given the statistical property of
$T_3$ is more similar to a $\chi^2$ statistic than the other
methods. The importance of this claim is that researchers who rely on
other methods may mistakenly detect differences between models.
## The two step estimation procedure
When estimating structural equation models in Mplus, the estimators
known as "MLMV", "WLSMV", and "ULSMV" include a feature to create a
model summary snapshot file (the so-called DIFFTEST file). The syntax
of the
[SAVEDATA](http://www.statmodel.com/download/usersguide/Chapter18.pdf)
command is:
```
SAVEDATA:
DIFFTEST = m0.dif;
```
This saves a model snapshot that we have chosen to name `m0.dif` (for
clarity). A second model is then estimated, and the input file (say,
`m1.inp`) includes an `ANALYSIS` stanza that requests a comparison
against `m0.dif` by a
[`DIFFTEST`](http://www.statmodel.com/download/usersguide/Chapter13.pdf)
statement:
```
ANALYSIS:
ESTIMATOR = WLSMV;
DIFFTEST = m0.dif
```
In this nested model comparison, following Asparouhov and Muthén (
[2006](https://www.statmodel.com/download/webnotes/webnote10.pdf))
terminology, the first model is called the parent (or the $H_1$) model.
The `dif` file for that model (`m0.dif`) is necessary to calculate
the `DIFFTEST` which compares the parent against the second, nested
model ($H_0$). In order to calculate the $T_3$ estimate with R, we
need to save the `dif` snapshot files for both $H_1$ and $H_0$ models.
# Calculating $T_3$
## Step 1 - Extracting model information from an Mplus `DIFFTEST` File
An Mplus `DIFFTEST` file contains a single column vector. It combines,
without punctuation or labels, the values of 4 scalars, $T$, $g$, $s$
and $p$, along with consecutive sets of values that must be separated
and rearranged into 3 matrices, $\mathbf{\Delta}_i$, $\mathbf{P}_i$,
and $\mathbf{V}_i$.
1) $T$ is the minimized value of the fit function; this matches the
value reported for the last iteration in the
[TECH5](http://www.statmodel.com/download/usersguide/Chapter18.pdf) output.
2) The number of groups ($g$).
3) The number of sample statistics ($s$).
4) The number of model parameters ($p$).
5) $\mathbf{\Delta}_i$, an $s \cdot g \times p$ matrix to be
reconstructed from the next $s \cdot g \cdot p$ elements in the `dif`
file (there is one stanza for each group to be constructed).
6) $\mathbf{P}_i$, a symmetric $p \times p$ matrix, reconstructed from
the next $p(p + 1)/2$ elements in the `dif` file.
7) $\mathbf{V}_i$, a symmetric $p \times p$ matrix, reconstructed from
the final $p(p + 1)/2$ elements in the `dif` file.
Because the `DIFFTEST` file provides the one long column of numbers,
some trial and error was necessary to separate the values and place
them into $\mathbf{\Delta}_i$, $\mathbf{P}_i$ and $\mathbf{V}_i$. The
individual values are placed into the matrices row wise. The matrix
$\mathbf{\Delta}_i$ has separate stanzas for each of $g$
groups. Because $\mathbf{P}$ and $\mathbf{V}$ are symmetric matrices,
the `dif` file provides only $p(p+1)/2$ distinct values for each
one. The matrix is reconstructed by filling in a lower triangle
row-wise (e.g., insert $\mathbf{P}[1,1]$, then $\mathbf{P}[2,1]$,
$\mathbf{P}[2,2]$, $\mathbf{P}[3,1]$ and so forth). The fact that
these are symmetric is relied upon to complete their reconstruction.
## Step 2 - Calculating $T.chisq$
*T.chisq* is computed by first taking the difference between the $T$
values (the minimized fit function),
\[
T_\Delta = T_{nested} - T_{parent}, \tag{2}
\]
and then rescaling $T_\Delta$ by multiplying $N*2$, where *N* is the
total sample size:
\[
T.chisq = T_\Delta*N*2. \tag{3}
\]
The value of $T.chisq$ cannot be reconstructed from the output of the
two fitted models. In particular, *T.chisq* is not based on the model
$\chi^2$ values that are reported in Mplus output. When a model is
estimated with "WLSMV", the chi-square value reported is a mean and
variance adjusted version of $T$. In the $T_3$ test, *T.chisq*
incorporates unscaled values of $T_{nested}$ and $T_{parent}$.
## Step 3 - Calculating the $\mathbf{H}$ Matrix
The calculation of $T_3$ will require the construction of several
separate matrices that will be used to derive the scaling and shifting
parameters. Here we are concerned with $\mathbf{H}$.
This matrix is mentioned in
Asparouhov and Muthén ([2006](https://www.statmodel.com/download/webnotes/webnote10.pdf)),
where the details of its calculation are left as an exercise for the
reader.
For that reason, we do not know how this is calculated in Mplus, but
the following will prove to be correct:
\[
H = (\Delta'_1\Delta_1)^{-1}\Delta_1'\Delta_0. \tag{4}
\]
This method is suggested by Yves Rosseel in the source code of the R
package `lavaan` ([Rosseel, 2017](http://lavaan.ugent.be/)). Note
that to obtain the two matrices, $\mathbf{\Delta}_0$ and
$\mathbf{\Delta}_1$, it is necessary parse both of the model `dif`
files.
## Step 4 - Calculating The $\mathbf{M}$ Matrix
$\mathbf{M}$ is an ingredient in the equations for the scaling factor
and shift parameter.
As shown by
Asparouhov and Muthen ([2006](https://www.statmodel.com/download/webnotes/webnote10.pdf)),
$\mathbf{M}$ can be calculated as:
\[
M = (P_1 - P_1(H(H'P_1H)^{-1}H'P_1)V_1 \tag{5}
\]
using the matrices constructed previously.
## Step 5 - Calculating the Scaling Factor and the Shift Parameter
The scaling factor $a$ is computed with the following formula:
\[
a = \sqrt{\frac{D}{Tr(M*M)}}. \tag{6}
\]
Here $D$ is the degrees of freedom, calculated as the difference
between the numbers of parameters estimated in the two
models.
The shift parameter $b$ is computed as
\[
b = D - \sqrt{\frac{D*Tr(M)^2}{Tr(M*M)}}. \tag{7}
\]
In these formulas, $Tr$ is the trace operator, the
sum of the elements along the main diagonal.
## Step 6 - Compute the Scaled $T_3$ Statistic
Once $a$ and $b$ have been computed, they are used in Equation 1 along
with $T.chisq$ to compute the $T_3$ chi-square difference
statistic. Asparouhov and Muthén
([2010](https://www.statmodel.com/download/WLSMV_new_chi21.pdf)) suggest
that the $T_3$ statistic follows a chi-square distribution with $D$
degrees of freedom.
## Using the `crmda_difftest` Function
The `crmda_difftest` function takes the file names of the Mplus output
files for two nested models as input. For example:
```{r, eval = FALSE}
crmda_difftest("parent.out", "nested.out")
```
We use functions provided by the MplusAutomation package
([Hallquist and Wiley, 2017](https://CRAN.R-project.org/package=MplusAutomation))
for R to parse the Mplus output files.
The function detects the location of the companion `dif` files,
`parent.dif` and `nested.dif`. In our examples, these are available in
the same folder as the output files. The `dif` files are parsed by an
auxiliary function named `diffReader`. If there are multiple groups,
the user must also specify the total sample size as a third argument
to `crmda_difftest`. The function returns the $T_3$ value calculated
by the aforementioned formulas, along with the estimate of $T_3$
retrieved from the Mplus output file.
# Matching Mplus `DIFFTEST` Calculations
The `crmda_difftest` function imports the `dif` files and reproduces
estimates of the $\chi^2$ test that match the $T_3$ estimates reported
in Mplus
([Asparouhov and Muthén, 2010](https://www.statmodel.com/download/WLSMV_new_chi21.pdf)).
This section shows the results obtained from the `crmda_difftest`
function match the values reported by the Mplus DIFFTEST procedure on
a variety of models.
## The demonstration data
A subset of the HBSC data
([Health Behavior in School-Aged Children; Iannotti, 2005-2006](http://www.uib.no/en/hbscdata/94259/data-documentation-20052006))
was used to obtain the demonstration results. The HBSC data contains
responses from students in grades 6, 7, 8, 9, and 10. The subset data
being used here contains responses to ordinal measures (five-point
Likert-type) on self-reported depression level, physical well-being,
bullying experiences (got bullied by others and bullied others), and
alcohol use. A total of 25 nested model comparison scenarios were
devised. The Mplus estimator "WLSMV" was used for all of the model
estimations.
## Example models
The following models were fitted to the HBSC data.
```{r modeltable, include = FALSE, echo = FALSE, cache = TRUE}
mplusdir <- "Mplus"
outs <- list.files(mplusdir, pattern = "\\.out$", full.names = TRUE)
models <- lapply(outs, function(x) readModels(x)$summaries)
sums <- lapply(models, function(x) gsub(";", "", x[c("Title")]))
sums2 <- lapply(sums, function(x) t(cbind(unlist(strsplit(x, " - ")))))
titles <- do.call("rbind", sums2)
titles <- cbind(titles, basename(outs))
colnames(titles) <- c("Model Number", "Model Description", "Output File")
modelTable <- print(xtable(titles), type = "html",
include.rownames = FALSE, file = "tmpout/modeltable.html")
```
```{r printmods, results = "asis", echo = FALSE, dependson='modeltable'}
cat(modelTable)
```
## Nested model comparisons
In the following table, we summarize the $T_3$ values calculated with
`crmda_difftest`, based on the `dif` files from Mplus, and compare
against the values of the $T_3$ estimates reported in the Mplus output
files. These results show that across a variety of model comparison
scenarios the `crmda_difftest` function reproduces the test statistic
provided by Mplus exactly to three decimal places.
```{r, include = FALSE}
## Creating a Data Frame Containing References for Mplus Out Files
## Labels for 25 Nested Model Comparisons
mplusOutRef <- data.frame("comparison" = c("Model.00vs01",
"Model.02vs03",
"Model.04vs05",
"Model.06vs07",
"Model.06vs08",
"Model.09vs10",
"Model.09vs11",
"Model.10vs12",
"Model.11vs13",
"Model.14vs15",
"Model.17vs18",
"Model.18vs19",
"Model.19vs20",
"Model.21vs22",
"Model.22vs23",
"Model.23vs24",
"Model.25vs26",
"Model.26vs27",
"Model.27vs28",
"Model.29vs30",
"Model.30vs31",
"Model.31vs32",
"Model.33vs34",
"Model.34vs35",
"Model.35vs36"), stringsAsFactors = FALSE)
## Labels for the Parent Models' Mplus Output Files
mplusOutRef[ , "parent"] <- c("m00-2_indicators.out",
"m02-3_indicators.out",
"m04-5_6_indicators.out",
"m06-3_factors.out",
"m06-3_factors.out",
"m09-structural.out",
"m09-structural.out",
"m10-structural.out",
"m11-structural.out",
"m14-indirect.out",
"m17-1_factor-2_groups-config.out",
"m18-1_factor-2_groups-weak.out",
"m19-1_factor-2_groups-strong.out",
"m21-1_factor-3_groups-config.out",
"m22-1_factor-3_groups-weak.out",
"m23-1_factor-3_groups-strong.out",
"m25-1_factor-4_groups-config.out",
"m26-1_factor-4_groups-weak.out",
"m27-1_factor-4_groups-strong.out",
"m29-1_factor-5_groups-config.out",
"m30-1_factor-5_groups-weak.out",
"m31-1_factor-5_groups-strong.out",
"m33-5_factors-5_groups-config.out",
"m34-5_factors-5_groups-weak.out",
"m35-5_factors-5_groups-strong.out")
## Model Specifications for the Parent Models
mplusOutRef[ , "parentSpec"] <- c("2 Factors_2 Indicators_Factor Correlation Free",
"2 Factors_3 Indicators_Factor Correlation Free",
"2 Factors_5/6 Indicators_Factor Correlation Free",
"3 Factors_5/6/9 Indicators_Factor Correlations Free",
"3 Factors_5/6/9 Indicators_Factor Correlations Free",
"3 Factors_2 IVs 1 DV_IVs Correlation Free_Regressive Paths Free",
"3 Factors_2 IVs 1 DV_IVs Correlation Free_Regressive Paths Free",
"3 Factors_2 IVs 1 DV_IVs Correlation Fixed 0_Regressive Paths Free",
"3 Factors_2 IVs 1 DV_IVs Correlation Fixed 0_Regressive Paths Equated",
"3 Factors_1 IV 1 DV 1 ME_Regressive Paths Free",
"1 Factor_2 Groups_config",
"1 Factor_2 Groups_weak",
"1 Factor_2 Groups_strong",
"1 Factor_3 Groups_config",
"1 Factor_3 Groups_weak",
"1 Factor_3 Groups_strong",
"1 Factor_4 Groups_config",
"1 Factor_4 Groups_weak",
"1 Factor_4 Groups_strong",
"1 Factor-5 Groups_config",
"1 Factor-5 Groups_weak",
"1 Factor-5 Groups_strong",
"5 Factors-5 Groups_config",
"5 Factors-5 Groups_weak",
"5 Factors-5 Groups_strong")
## Labels for the Child Models' Mplus Output Files
mplusOutRef[ , "child"] <- c("m01-2_indicators.out",
"m03-3_indicators.out",
"m05-5_6_indicators.out",
"m07-3_factors.out",
"m08-3_factors.out",
"m10-structural.out",
"m11-structural.out",
"m12-structural.out",
"m13-structural.out",
"m15-indirect.out",
"m18-1_factor-2_groups-weak.out",
"m19-1_factor-2_groups-strong.out",
"m20-1_factor-2_groups-strict.out",
"m22-1_factor-3_groups-weak.out",
"m23-1_factor-3_groups-strong.out",
"m24-1_factor-3_groups-strict.out",
"m26-1_factor-4_groups-weak.out",
"m27-1_factor-4_groups-strong.out",
"m28-1_factor-4_groups-strict.out",
"m30-1_factor-5_groups-weak.out",
"m31-1_factor-5_groups-strong.out",
"m32-1_factor-5_groups-strict.out",
"m34-5_factors-5_groups-weak.out",
"m35-5_factors-5_groups-strong.out",
"m36-5_factors-5_groups-strict.out")
## Model Specifications for the Child Models
mplusOutRef[ , "ChildSpec"] <- c("2 Factors_2 Indicators_Factor Correlation Fixed 0",
"2 Factors_3 Indicators_Factor Correlation Fixed 0",
"2 Factors_5/6 Indicators_Factor Correlation Fixed 0",
"3 Factors_5/6/9 Indicators_ 1 Factor Correlation Fixed 0",
"3 Factors_5/6/9 Indicators_2 Factor Correlations Equated",
"3 Factors_2 IVs 1 DV_IVs Correlation Fixed 0_Regressive Paths Free",
"3 Factors_2 IVs 1 DV_IVs Correlation Free_Regressive Paths Equated",
"3 Factors_2 IVs 1 DV_IVs Correlation Fixed 0_Regressive Paths Equated",
"3 Factors_2 IVs 1 DV_IVs Correlation and _Regressive Paths Equated",
"3 Factors_1 IV 1 DV 1 ME_Direct Path Fixed Zero",
"1 Factor_2 Groups_weak",
"1 Factor_2 Groups_strong",
"1 Factor_2 Groups_strict",
"1 Factor_3 Groups_weak",
"1 Factor_3 Groups_strong",
"1 Factor_3 Groups_strict",
"1 Factor_4 Groups_weak",
"1 Factor_4 Groups_strong",
"1 Factor_4 Groups_strict",
"1 Factor-5 Groups_weak",
"1 Factor-5 Groups_strong",
"1 Factor-5 Groups_strict",
"5 Factors-5 Groups_weak",
"5 Factors-5 Groups_strong",
"5 Factors-5 Groups_strict")
## Number of Groups for each Nested Model Comparison
mplusOutRef[ , "groupSize"] <- c(1, 1, 1, 1, 1,
1, 1, 1, 1, 1,
2, 2, 2, 3, 3,
3, 4, 4, 4, 5,
5, 5, 5, 5, 5)
## Total Sample Size for each Nested Model Comparison
## Has to Be Supplied to crmda_difftest When the Number of Groups Greater than 1
mplusOutRef[ , "totalSampleSize"] <- c(9135, 9136, 9137, 9139, 9139,
9139, 9139, 9139, 9139, 9139,
4221, 4221, 4221, 6019, 6019,
6019, 7485, 7485, 7485, 9087,
9087, 9087, 9151, 9151, 9151)
## Inspecting the Mplus Out Reference
mplusOutRef
str(mplusOutRef)
res <- vector("list", length = 25)
for (i in seq_along(rownames(mplusOutRef))) {
x <- mplusOutRef[i, ]
if(x["groupSize"] != 1) {
res[[i]] <- crmda_difftest(file.path(mplusdir, x["parent"]),
file.path(mplusdir, x["child"]),
samplesize = x["totalSampleSize"])
} else {
res[[i]] <- crmda_difftest(file.path(mplusdir, x["parent"]),
file.path(mplusdir, x["child"]))
}
res[[i]][["comparison"]] <- mplusOutRef$comparison[[i]]}
names(res) <- mplusOutRef$comparison
```
```{r comparisons, results = "asis", echo = FALSE, include = FALSE, cache=TRUE, eval = TRUE}
tests <- lapply(res, function(x){
c(x[["comparison"]],
unlist(x[["R computations"]]),
x[["Mplus computations"]][[1]])
})
tests <- do.call("rbind", tests)
tests <- as.data.frame(tests, stringsAsFactors = FALSE)
colnames(tests) <- c("Comparison", "crmda_difftest_T3", "df", "p", "Mplus_DIFFTEST_T3")
tests <- tests[ , c("Comparison", "crmda_difftest_T3", "Mplus_DIFFTEST_T3", "df", "p")]
rownames(tests) <- NULL
tests <- data.frame(tests, stringsAsFactors = FALSE)
tests[ , "Comparison"] <- gsub("\\.", " ", tests[,"Comparison"])
tests[ , "Comparison"] <- gsub("vs", " vs. ", tests[,"Comparison"])
tests[ , "p"] <- as.numeric(tests[,"p"])
tests[ , "p"] <- sprintf("%.3f", tests[,"p"])
tests[ , "crmda_difftest_T3"] <-
sprintf("%.3f", as.numeric(tests[ , "crmda_difftest_T3"]))
tests[ , "df"] <- sprintf("%.3f", as.numeric(tests[ , "df"]))
comparisonTable <- print(xtable(tests), type = "html", include.rownames = FALSE,
file = "tmpout/comparisonTable.html")
```
```{r printcomp, results = "asis", echo = FALSE, eval=TRUE, dependson='comparisons'}
cat(comparisonTable)
```
# Summary
We have outlined the procedures necessary to import and parse the
Mplus `DIFFTEST` files and to calculate a difference of fit test known
as $T_3$. While some elements in the calculations are based on our
best guess about calculations in Mplus, we believe that the
combination of elements is correct. The difference in our estimates
and the values reported by Mplus is smaller than $10^{-4}$.
The larger project from which this report is drawn will compare the
hypothesis tests based on $T_2$, $T_3$, and others. With the
completion of this phase, we are one step closer to the final
goal. The next phase in this project is to create R functions that
provide replacements for $\mathbf{\Delta}_i$, $\mathbf{P}_i$,
$\mathbf{V}_i$.
## R Session Info
```{r sessionInfo, echo = FALSE}
sessionInfo()
```
```{r warnings, echo = FALSE}
if(!is.null(warnings())){
print("Warnings:")
warnings()
}
```
Available under
[Created Commons license 3.0 ](http://creativecommons.org/licenses/by/3.0/)