Function that searches for the global maximum of the log-likelihood of different models and selects the optimal number of states.
lmestSearch(responsesFormula = NULL, latentFormula = NULL,
data, index, k,
version = c("categorical", "continuous"),
weights = NULL, nrep = 2, tol1 = 10^-5,
tol2 = 10^-10, out_se = FALSE, seed = NULL, ...)a symbolic description of the model to fit. A detailed description is given in the <U+2018>Details<U+2019> section of lmest
a symbolic description of the model to fit. A detailed description is given in the <U+2018>Details<U+2019> section of lmest
a data.frame in long format
a character vector with two elements, the first indicating the name of the unit identifier, and the second the time occasions
a vector of integer values for the number of latent states
an optional vector of frequencies of the available response configurations
type of responses for the LM model: "categorical" and "continuous"
number of repetitions of each random initialization
tolerance level for checking convergence of the algorithm in the random initializations
tolerance level for checking convergence of the algorithm in the last deterministic initialization
to compute the information matrix and standard errors (FALSE is the default option)
an integer value with the random number generator
Returns an object of class 'LMsearch' with the following components:
Output of every LM model estimated for each number of latent states given in input
Values the Akaike Information Criterion for each number of latent states given in input
Values of the Bayesian Information Criterion for each number of latent states given in input
Values of log-likelihood for each number of latent states given in input.
The function combines deterministic and random initializations strategy to reach the global maximum of the model log-likelihood.
It uses one deterministic initialization (start=0) and a number of random initializations (start=1) proportional to the number of latent states. The tolerance level is set equal to 10^-5. Starting from the best solution obtained in this way, a final run is performed (start=2) with a default tolerance level equal to 10^-10.
Missing responses are allowed according to the model to be estimated.
Bartolucci F., Pandolfi S., Pennoni F. (2017) LMest: An R Package for Latent Markov Models for Longitudinal Categorical Data, Journal of Statistical Software, 81(4), 1-38.
Bartolucci, F., Farcomeni, A. and Pennoni, F. (2013) Latent Markov Models for Longitudinal Data, Chapman and Hall/CRC press.
# NOT RUN {
### Example with data on drug use in wide format
data("data_drug")
long <- data_drug[,-6]
# add labels referred to the identifier
long <- data.frame(id = 1:nrow(long),long)
# reshape data from the wide to the long format
long <- reshape(long,direction = "long",
idvar = "id",
varying = list(2:ncol(long)))
out <- lmestSearch(data = long,
index = c("id","time"),
version = "categorical",
k = 1:3,
weights = data_drug[,6],
modBasic = 1,
seed = 123)
out
summary(out$out.single[[3]])
# }
# NOT RUN {
### Example with data on self rated health
# LM model with covariates in the measurement model
data("data_SRHS_long")
SRHS <- data_SRHS_long[1:1000,]
# Categories rescaled to vary from 1 (<U+201C>poor<U+201D>) to 5 (<U+201C>excellent<U+201D>)
SRHS$srhs <- 5 - SRHS$srhs
out1 <- lmestSearch(data = SRHS,
index = c("id","t"),
version = "categorical",
responsesFormula = srhs ~ -1 +
I(gender - 1) +
I( 0 + (race == 2) + (race == 3)) +
I(0 + (education == 4)) +
I(0 + (education == 5)) + I(age - 50) +
I((age-50)^2/100),
k = 1:2,
out_se = TRUE,
seed = 123)
summary(out1)
summary(out1$out.single[[2]])
# }
Run the code above in your browser using DataLab