MCMCmixfactanal {MCMCpack} | R Documentation |
This function generates a sample from the posterior distribution of a mixed data (both continuous and ordinal) factor analysis model. Normal priors are assumed on the factor loadings and factor scores, improper uniform priors are assumed on the cutpoints, and inverse gamma priors are assumed for the error variances (uniquenesses). The user supplies data and parameters for the prior distributions, and a sample from the posterior distribution is returned as an mcmc object, which can be subsequently analyzed with functions provided in the coda package.
MCMCmixfactanal(x, factors, lambda.constraints=list(), data=parent.frame(), burnin = 1000, mcmc = 20000, thin=1, tune=NA, verbose = 0, seed = NA, lambda.start = NA, psi.start=NA, l0=0, L0=0, a0=0.001, b0=0.001, store.lambda=TRUE, store.scores=FALSE, std.mean=TRUE, std.var=TRUE, ... )
x |
A one-sided formula containing the
manifest variables. Ordinal (including dichotomous) variables must
be coded as ordered factors. Each level of these ordered factors must
be present in the data passed to the function. NOTE: data input is different in
MCMCmixfactanal than in either MCMCfactanal or
MCMCordfactanal . |
factors |
The number of factors to be fitted. |
lambda.constraints |
List of lists specifying possible equality
or simple inequality constraints on the factor loadings. A typical
entry in the list has one of three forms: varname=list(d,c) which
will constrain the dth loading for the variable named varname to
be equal to c, varname=list(d,"+") which will constrain the dth
loading for the variable named varname to be positive, and
varname=list(d, "-") which will constrain the dth loading for the
variable named varname to be negative. If x is a matrix without
column names defaults names of ``V1", ``V2", ... , etc will be
used. Note that, unlike MCMCfactanal , the
Lambda matrix used here has factors +1
columns. The first column of Lambda corresponds to
negative item difficulty parameters for ordinal manifest variables
and mean parameters for continuous manifest variables and should
generally not be constrained directly by the user.
|
data |
A data frame. |
burnin |
The number of burn-in iterations for the sampler. |
mcmc |
The number of iterations for the sampler. |
thin |
The thinning interval used in the simulation. The number of iterations must be divisible by this value. |
tune |
The tuning parameter for the Metropolis-Hastings
sampling. Can be either a scalar or a k-vector (where
k is the number of manifest variables). tune must be
strictly positive. |
verbose |
A switch which determines whether or not the progress of
the sampler is printed to the screen. If verbose is great
than 0 the iteration number and
the Metropolis-Hastings acceptance rate are printed to the screen
every verbose th iteration. |
seed |
The seed for the random number generator. If NA, the Mersenne
Twister generator is used with default seed 12345; if an integer is
passed it is used to seed the Mersenne twister. The user can also
pass a list of length two to use the L'Ecuyer random number generator,
which is suitable for parallel computation. The first element of the
list is the L'Ecuyer seed, which is a vector of length six or NA (if NA
a default seed of rep(12345,6) is used). The second element of
list is a positive substream number. See the MCMCpack
specification for more details. |
lambda.start |
Starting values for the factor loading matrix
Lambda. If lambda.start is set to a scalar the starting value for
all unconstrained loadings will be set to that scalar. If
lambda.start is a matrix of the same dimensions as Lambda then the
lambda.start matrix is used as the starting values (except
for equality-constrained elements). If lambda.start is set to
NA (the default) then starting values for unconstrained
elements in the first column of Lambda are based on the observed
response pattern, the remaining unconstrained elements of Lambda are
set to 0, and starting values for inequality constrained elements
are set to either 1.0 or -1.0 depending on the nature of the
constraints. |
psi.start |
Starting values for the error variance (uniqueness)
matrix. If psi.start is set to a scalar then the starting
value for all diagonal elements of Psi that represent error
variances for continuous variables are set to this value. If
psi.start is a k-vector (where k is the
number of manifest variables) then the staring value of Psi
has psi.start on the main diagonal with the exception that
entries corresponding to error variances for ordinal variables are
set to 1.. If psi.start is set to NA (the default) the
starting values of all the continuous variable uniquenesses are set
to 0.5. Error variances for ordinal response variables are always
constrained (regardless of the value of psi.start to have an
error variance of 1 in order to achieve identification. |
l0 |
The means of the independent Normal prior on the factor
loadings. Can be either a scalar or a matrix with the same
dimensions as Lambda . |
L0 |
The precisions (inverse variances) of the independent Normal
prior on the factor loadings. Can be either a scalar or a matrix with
the same dimensions as Lambda . |
a0 |
Controls the shape of the inverse Gamma prior on the
uniqueness. The actual shape parameter is set to a0/2 . Can be
either a scalar or a k-vector. |
b0 |
Controls the scale of the inverse Gamma prior on the
uniquenesses. The actual scale parameter is set to b0/2 . Can
be either a scalar or a k-vector. |
store.lambda |
A switch that determines whether or not to store the factor loadings for posterior analysis. By default, the factor loadings are all stored. |
store.scores |
A switch that determines whether or not to store the factor scores for posterior analysis. NOTE: This takes an enormous amount of memory, so should only be used if the chain is thinned heavily, or for applications with a small number of observations. By default, the factor scores are not stored. |
std.mean |
If TRUE (the default) the continuous manifest
variables are rescaled to have zero mean. |
std.var |
If TRUE (the default) the continuous manifest
variables are rescaled to have unit variance. |
... |
further arguments to be passed |
The model takes the following form:
Let 1=1,...,n index observations and j=1,...,K index response variables within an observation. An observed variable x_ij can be either ordinal with a total of C_j categories or continuous. The distribution of X is governed by a N by K matrix of latent variables Xstar and a series of cutpoints gamma. Xstar is assumed to be generated according to:
xstar_i = Lambda phi_i + epsilon_i
epsilon_i ~ N(0, Psi)
where xstar_i is the k-vector of latent variables specific to observation i, Lambda is the k by d matrix of factor loadings, and phi_i is the d-vector of latent factor scores. It is assumed that the first element of phi_i is equal to 1 for all i.
If the jth variable is ordinal, the probability that it takes the value c in observation i is:
pi_ijc = pnorm(gamma_jc - Lambda'_j phi_i) - pnorm(gamma_j(c-1) - Lambda'_j phi_i)
If the jth variable is continuous, it is assumed that xstar_{ij} = x_{ij} for all i.
The implementation used here assumes independent conjugate priors for each element of Lambda and each phi_i. More specifically we assume:
Lambda_ij ~ N(l0_ij, L0_ij^-1), i=1,...,k, j=1,...,d
phi_i(2:d) ~ N(0, I), i=1,...,n
MCMCmixfactanal
simulates from the posterior distribution using
a Metropolis-Hastings within Gibbs sampling algorithm. The algorithm
employed is based on work by Cowles (1996). Note that
the first element of phi_i is a 1. As a result, the
first column of Lambda can be interpretated as negative
item difficulty parameters. Further, the first
element gamma_1 is normalized to zero, and thus not
returned in the mcmc object.
The simulation proper is done in compiled C++ code to maximize
efficiency. Please consult the coda documentation for a comprehensive
list of functions that can be used to analyze the posterior sample.
As is the case with all measurement models, make sure that you have plenty of free memory, especially when storing the scores.
An mcmc object that contains the posterior sample. This object can be summarized by functions provided by the coda package.
Kevin M. Quinn. 2004. ``Bayesian Factor Analysis for Mixed Ordinal and Continuous Responses.'' Political Analysis. 12: 338-353.
M. K. Cowles. 1996. ``Accelerating Monte Carlo Markov Chain Convergence for Cumulative-link Generalized Linear Models." Statistics and Computing. 6: 101-110.
Valen E. Johnson and James H. Albert. 1999. ``Ordinal Data Modeling." Springer: New York.
Andrew D. Martin, Kevin M. Quinn, and Daniel Pemstein. 2004. Scythe Statistical Library 1.0. http://scythe.wustl.edu.
Martyn Plummer, Nicky Best, Kate Cowles, and Karen Vines. 2002. Output Analysis and Diagnostics for MCMC (CODA). http://www-fis.iarc.fr/coda/.
plot.mcmc
, summary.mcmc
,
factanal
, MCMCfactanal
,
MCMCordfactanal
,
MCMCirt1d
, MCMCirtKd
## Not run: data(PErisk) post <- MCMCmixfactanal(~courts+barb2+prsexp2+prscorr2+gdpw2, factors=1, data=PErisk, lambda.constraints = list(courts=list(2,"-")), burnin=5000, mcmc=1000000, thin=50, verbose=500, L0=.25, store.lambda=TRUE, store.scores=TRUE, tune=1.2) plot(post) summary(post) library(MASS) data(Cars93) attach(Cars93) new.cars <- data.frame(Price, MPG.city, MPG.highway, Cylinders, EngineSize, Horsepower, RPM, Length, Wheelbase, Width, Weight, Origin) rownames(new.cars) <- paste(Manufacturer, Model) detach(Cars93) # drop obs 57 (Mazda RX 7) b/c it has a rotary engine new.cars <- new.cars[-57,] # drop 3 cylinder cars new.cars <- new.cars[new.cars$Cylinders!=3,] # drop 5 cylinder cars new.cars <- new.cars[new.cars$Cylinders!=5,] new.cars$log.Price <- log(new.cars$Price) new.cars$log.MPG.city <- log(new.cars$MPG.city) new.cars$log.MPG.highway <- log(new.cars$MPG.highway) new.cars$log.EngineSize <- log(new.cars$EngineSize) new.cars$log.Horsepower <- log(new.cars$Horsepower) new.cars$Cylinders <- ordered(new.cars$Cylinders) new.cars$Origin <- ordered(new.cars$Origin) post <- MCMCmixfactanal(~log.Price+log.MPG.city+ log.MPG.highway+Cylinders+log.EngineSize+ log.Horsepower+RPM+Length+ Wheelbase+Width+Weight+Origin, data=new.cars, lambda.constraints=list(log.Horsepower=list(2,"+"), log.Horsepower=c(3,0), weight=list(3,"+")), factors=2, burnin=5000, mcmc=500000, thin=100, verbose=500, L0=.25, tune=3.0) plot(post) summary(post) ## End(Not run)