This function writes Mplus input files for conducting latent profile analysis
(LPA) based on six different variance-covariance structures. By default, the
function creates folders in the current working directory for each of the six
sets of analysis, writes Mplus input files for conducting LPA with k = 1
to k = 6 profiles into these folders, and writes the matrix or data frame
specified in x into a Mplus data file in the current working directory.
Optionally, all models can be estimated by setting the argument run.mplus
to TRUE.
mplus.lpa(x, ind = NULL, cluster = NULL,
folder = c("A_Invariant-Theta_Diagonal-Sigma",
"B_Varying-Theta_Diagonal-Sigma",
"C_Invariant-Theta_Invariant-Unrestrictred-Sigma",
"D_Invariant-Theta_Varying-Unrestricted-Sigma",
"E_Varying-Theta_Invariant-Unrestricted-Sigma",
"F_Varying-Theta_Varying-Unrestricted-Sigma"),
file = "Data_LPA.dat", write = c("all", "folder", "data", "input"),
missing = -99, classes = 6, estimator = "MLR", starts = c(500, 100),
stiterations = 50, lrtbootstrap = 1000, lrtstarts = c(0, 0, 100, 50),
processors = 8, output = c("all", "SVALUES", "TECH7", "TECH8",
"TECH11", "TECH14"),
run.mplus = FALSE, Mplus = "Mplus",
replace = c("always", "never", "modifiedDate"), check = TRUE)Returns an object of class misty.object, which is a list with following entries:
callfunction call
typetype of analysis
xmatrix or data frame specified in the argument x
argsspecification of function arguments
resultlist with six entries for each of the variance-covariance structures and the Mplus inputs based on different number of profiles
a matrix or data frame. Note that all variable names must be no longer than 8 character.
a character vector indicating the variables names of the
latent profile indicators in x.
a character string indicating the cluster variable in the
matrix or data frame specified in x representing the
nested grouping structure for computing cluster-robust
standard errors. Note that specifying a cluster variables
does not have any effect on the information criteria, but
for the Vuong-Lo-Mendell-Rubin likelihood ratio test of model fit.
a character vector with six character strings for specifying the names of the six folder representing different variance-covariance structures.
a character string naming the Mplus file with or without
the file extension '.dat', e.g., "Data_LPA.dat" (default)
or "Data_LPA".
a character string or character vector indicating whether
to create the six folders specified in the argument folder
("folder"), to write the matrix or data frame specified
in x into a Mplus data file ("data"), and
write the Mplus input files into the six folders specified
in the argument folder ("input"). By default,
the function creates the folders, writes the Mplus data file,
and writes the Mplus input files into the folders.
a numeric value or character string representing missing
values (NA) in the Mplus data set. This values or
character string will be specified in the Mplus input file
as MISSING IS ALL(missing). By default, -99
is used to represent missing values.
an integer value specifying the maximum number of profiles for the latent profile analysis. By default, a LPA with a maximum of 6 profiles are specified (i.e., k = 1 to k = 6).
a character string for specifying the ESTIMATOR
option in Mplus. By default, the estimator "MLR" is
used.
a vector with two integer values for specifying the
STARTS option in Mplus. The first number represents
the number of random sets of starting values to generate
in the initial stage and the second number represents the
optimizations to use in the final stage. By default, 500
random sets of starting values are generated and 100
optimizations are carried out in the final stage.
an integer value specifying the STITERATIONS option
in Mplus. The numeric value represents the maximum number
of iterations allowed in the initial stage. By default,
50 iterations are requested.
an integer value for specifying the LRTBOOTSTRAP option
in Mplus when reqeuesting a parametric bootstrapped likelihood
ratio test (i.e., output = "TECH14"). The value
represents the number of bootstrap draws to be used in
estimating the p-value of the parametric bootstrapped
likelihood ratio test. By default, 1000 bootstrap draws are
requested.
a vector with four integer values for specifying the
LRTSTARTS option in Mplus when reqeuesting a parametric
bootstrapped likelihood ratio test (i.e., output = "TECH14").
The values specify the number of starting values to use in
the initial stage and the number of optimizations to use
in the final stage for the k - 1 and k profile
model when the data generated by bootstrap draws are analyzed.
By default, 0 random sets of starting values in the initial
stage and 0 optimizations in the final stage are used for
the k - 1 profile model and 100 random sets of starting
values in the initial stage and 50 optimizations in the
final stage are used for the k profile model.
an integer value for specifying the PROCESSORS option
in Mplus. The value specifies the number of processors
and threads to be used for parallel computing to increase
computational speed. By default, 8 processors and threads
are used for parallel computing.
a character string or character vector specifying the TECH
options in the OUTPUT section in Mplus, i.e., codeSVALUES
to request input statements that contain parameter estimates
from the analysis, TECH7 to request sample statistics
for each profile using raw data weighted by the estimated
posterior probabilities for each profile, TECH8 to
request the optimization history in estimating the model,
TECH11 to request the Vuong-Lo-Mendell-Rubin
likelihood ratio test of model fit, and TECH14 to
request a parametric bootstrapped likelihood ratio test.
By default, SVALUES and TECH11 are requested.
Note that TECH11 is only available for the MLR
estimator.
logical: if TRUE, all models in the folders specified
in the argument folder are estimated by using the
run.mplus function in the R package misty.
a character string for specifying the name or path of the Mplus executable to be used for running models. This covers situations where Mplus is not in the system's path, or where one wants to test different versions of the Mplus program. Note that there is no need to specify this argument for most users since it has intelligent defaults.
a character string for specifying three settings:
"always" (default), which runs all models, regardless
of whether an output file for the model exists, "never",
which does not run any model that has an existing output file,
and "modifiedDate", which only runs a model if the
modified date for the input file is more recent than the
output file modified date.
logical: if TRUE, argument specification is checked.
Takuya Yanagida takuya.yanagida@univie.ac.at
Latent profile analysis is a model-based clustering and classification method used to identify qualitatively different classes of observations which are unknown and must be inferred from the data. The within-profile variance-covariance structures represt different assumptions regarding the variance and covariance of the indicator variables both within and between latent profiles. As the best within-profile variance-covariance structure is not known a priori, all of the different structures must be tested to identify the best model (Masyn, 2013). This function specifies six different six different variance-covariance structures labeled A to F (see Table 1 in Patterer et al, 2023):
The within-profile variance is constrained to be profile-invariant and covariances are constrained to be 0 in all profiles (i.e., equal variances across profiles and no covariances among indicator variables).
The within-profile variance profile-varying and covariances are constrained to be 0 in all profiles (i.e., unequal variances across profiles and no covariances among indicator variables).
The within-profile variance is constrained to be profile-invariant and covariances are constrained to be equal in all profiles (i.e., equal variances and covariances across profiles).
The within-profile variance is constrained to be profile-invariant and covariances are profile-varying (i.e., equal variances across profiles and unequal covariances across profiles).
The within-profile variances are profile-varying and covariances are constrained to be equal in all profiles (i.e., unequal variances across profiles and equal covariances across profiles).
The within-class variance and covariances both profile-varying (i.e., unequal variances and covariances across profiles).
Masyn, K. E. (2013). Latent class analysis and finite mixture modeling. In T. D. Little (Ed.), The Oxford handbook of quantitative methods: Statistical analysis (pp. 551–611). Oxford University Press.
Muthen, L. K., & Muthen, B. O. (1998-2017). Mplus User's Guide (8th ed.). Muthen & Muthen.
Patterer, A. S., Yanagida, T., Kühnel, J., & Korunka, C. (2023). Daily receiving and providing of social support at work: Identifying support exchange patterns in hierarchical data. Journal of Work and Organizational Psychology, 32(4), 489-505. https://doi.org/10.1080/1359432X.2023.2177537
result.lpa, run.mplus, read.mplus,
write.mplus
if (FALSE) {
# Load data set "HolzingerSwineford1939" in the lavaan package
data("HolzingerSwineford1939", package = "lavaan")
# LPA with k = 1 to k = 8 profiles
# Input statements that contain parameter estimates
# Lo-Mendell-Rubin LRT and bootstrapped LRT
mplus.lpa(HolzingerSwineford1939, ind = c("x1", "x2", "x3", "x4"),
classes = 8, output = c("SVALUES", "TECH11", "TECH14"))
}
Run the code above in your browser using DataLab