Replaces deprecated ctmaOptimizeInit, which was limited to initial fitting
(i.e., applies ctmaInit) of a primary study reFits times to capitalize on chance for obtaining
a hard-to-find optimal fit.
Now, optimizing a CoTiMA model generated with ctmaFit can also be done.
Using ctmaOptimizeFit could be helpful if a model yields out-of-range estimates, which could happen if the fitting
algorithm unfortunately used random start values that resulted in a locally but not globally optimal fit. Essentially, using
ctmaOptimizeFit is like gambling, hoping that at least one set of starting values (the number it tries is specified in the reFits argument)
enables finding the global optimal fit. On unix-like machines (e.g. MacOS), this could be done in parallel mode if coresToUse > 1.
ctmaOptimizeFit(
activateRPB = FALSE,
activeDirectory = NULL,
checkSingleStudyResults = FALSE,
coresToUse = c(2),
CoTiMAStanctArgs = NULL,
ctmaFitFit = NULL,
ctmaInitFit = NULL,
customPar = FALSE,
finishsamples = NULL,
indVarying = FALSE,
lambda = NULL,
manifestMeans = 0,
manifestVars = NULL,
n.latent = NULL,
posLL = TRUE,
primaryStudies = NULL,
problemStudy = NULL,
randomPar = FALSE,
randomScaleTime = c(1, 1),
reFits = NULL,
scaleMod = NULL,
scaleTime = NULL,
T0means = 0,
transfMod = NULL
)returns a list with bestFit (= the best fit achieved), all_minus2ll (= all -2ll values for all fitted models), and summary, which is printed if the summary function is applied to the returned object, and which shows the summary information of the ctsem model with the best fit.
set to TRUE to receive push messages with 'CoTiMA' notifications on your phone
activeDirectory
displays estimates from single study 'ctsem' models and waits for user input to continue.
if neg., the value is subtracted from available cores, else value = cores to use
parameters that can be set to improve model fitting of the ctStanFit Function
a object fitted with ctmaFit
the ctmaInitFit object that was used to create the ctmaFitFit object with ctmaFit
logical. If set TRUE (default) leverages the first pass using priors and ensure that the drift diagonal cannot easily go too negative (helps since ctsem > 3.4)
number of samples to draw (either from hessian based covariance or posterior distribution) for final results computation (default = 1000).
control for unobserved heterogeneity by having randomly (inter-individually) varying manifest means
R-type matrix with pattern of fixed (=1) or free (any string) loadings.
Default 0 (assuming standardized variables). Can be assigned labels to estimate them freely.
define the error variances of the manifests within a single time point using R-type lower triangular matrix with nrow=n.manifest & ncol=n.manifest. Useful to check estimates before they are saved.
number of latent variables of the model (hast to be specified)!
logical. Allows (default = TRUE) of positive loglik (neg -2ll) values
list of primary study information created with ctmaPrep or ctmaFitToPrep
number (position in list) where the problem study in primaryStudies is found
logical. Overrides arguments used fo customPar and randomly selects customPar either TRUE or FALSE
lower and upper limit of uniform distribution from which timeScale argument for ctmaInit is uniformly shuffled (integer)
how many reFits should be done
scale moderator variables - TRUE (default) recommended for continuous and categorical moderators, to separate withing and betwen efeccts
scale time (interval) - sometimes desirable to improve fitting
Default 0 (assuming standardized variables). Can be assigned labels to estimate them freely.
more general option to change moderator values. A vector as long as number of moderators analyzed (e.g., c("mean(x)", "x - median(x)"))
if (FALSE) {
optimFit313 <- ctmaOptimizeFit(primaryStudies=CoTiMAstudyList_3,
activeDirectory="/Users/tmp/", # adapt!
problemStudy=which(CoTiMAstudyList_3$studyNumbers == 313),
reFits=10,
n.latent=2)
summary(optimFit313)
}
Run the code above in your browser using DataLab