ctmaFit: ctmaFit

In CoTiMA: Continuous Time Meta-Analysis ('CoTiMA')

ctmaFit

Description

Fits a ctsem model with invariant drift effects across primary studies, possible multiple moderators (but all of them of the the same type, either "cont" or "cat"), and possible cluster (e.g., countries where primary studies were conducted).

Usage

ctmaFit(
  ctmaInitFit = NULL,
  primaryStudyList = NULL,
  cluster = NULL,
  activeDirectory = NULL,
  activateRPB = FALSE,
  digits = 4,
  drift = NULL,
  invariantDrift = NULL,
  moderatedDrift = NULL,
  equalDrift = NULL,
  mod.number = NULL,
  mod.type = "cont",
  mod.names = NULL,
  indVarying = FALSE,
  coresToUse = c(1),
  scaleTI = TRUE,
  scaleMod = NULL,
  transfMod = NULL,
  scaleClus = NULL,
  scaleTime = NULL,
  optimize = TRUE,
  nopriors = TRUE,
  finishsamples = NULL,
  iter = NULL,
  chains = NULL,
  verbose = NULL,
  allInvModel = FALSE,
  customPar = FALSE,
  inits = NULL,
  modsToCompare = NULL,
  catsToCompare = NULL,
  driftsToCompare = NULL,
  useSampleFraction = NULL,
  T0means = 0,
  manifestMeans = 0,
  CoTiMAStanctArgs = NULL,
  lambda = NULL,
  manifestVars = NULL
)

Arguments

ctmaInitFit	object to which all single ctsem fits of primary studies has been assigned to (i.e., what has been returned by ctmaInit)
primaryStudyList	could be a list of primary studies compiled with ctmaPrep that defines the subset of studies in ctmaInitFit that should actually be used
cluster	vector with cluster variables (e.g., countries). Has to be set up carfully. Will be included in ctmaPrep in later 'CoTiMA' versions.
activeDirectory	defines another active directory than the one used in ctmaInitFit
activateRPB	set to TRUE to receive push messages with 'CoTiMA' notifications on your phone
digits	Number of digits used for rounding (in outputs)
drift	labels for drift effects. Have to be either of the type 'V1toV2' or '0' for effects to be excluded.
invariantDrift	drift labels for drift effects that are set invariant across primary studies (default = all drift effects).
moderatedDrift	labels for drift effects that are moderated (default = all drift effects)
equalDrift	Not enabled
mod.number	which in the vector of moderator values shall be used (e.g., 2 for a single moderator or 1:3 for 3 moderators simultaneously)
mod.type	'cont' or 'cat' (mixing them in a single model not yet possible)
mod.names	vector of names for moderators used in output
indVarying	allows continuous time intercepts to vary at the individual level (random effects model, accounts for unobserved heterogeneity)
coresToUse	if negative, the value is subtracted from available cores, else value = cores to use
scaleTI	scale TI predictors - not recommended until version 0.5.3.1. Does not change aggregated results anyways, just interpretation of effects for dimmies representing primary studies.
scaleMod	scale moderator variables - TRUE (default) recommended for continuous and categorical moderators, to separate withing and betwen efeccts
transfMod	more general option to change moderator values. A vector as long as number of moderators analyzed (e.g., c("mean(x)", "x - median(x)"))
scaleClus	scale vector of cluster indicators - TRUE (default) yields avg. drift estimates, FALSE yields drift estimates of last cluster
scaleTime	scale time (interval) - sometimes desirable to improve fitting
optimize	if set to FALSE, Stan’s Hamiltonian Monte Carlo sampler is used (default = TRUE = maximum a posteriori / importance sampling) .
nopriors	if TRUE, any priors are disabled – sometimes desirable for optimization
finishsamples	number of samples to draw (either from hessian based covariance or posterior distribution) for final results computation (default = 1000).
iter	number of iterations (defaul = 1000). Sometimes larger values could be required fom Bayesian estimation
chains	number of chains to sample, during HMC or post-optimization importance sampling.
verbose	integer from 0 to 2. Higher values print more information during model fit – for debugging
allInvModel	estimates a model with all parameters invariant (DRIFT, DIFFUSION, T0VAR) if set TRUE (defautl = FALSE)
customPar	logical. If set TRUE leverages the first pass using priors and ensure that the drift diagonal cannot easily go too negative (helps since ctsem > 3.4)
inits	vector of start values
modsToCompare	when performing contrasts for categorical moderators, the moderator numbers (position in mod.number) that is used
catsToCompare	when performing contrasts for categorical moderators, the categories (values, not positions) for which effects are set equal
driftsToCompare	when performing contrasts for categorical moderators, the (subset of) drift effects analyzed
useSampleFraction	to speed up debugging. Provided as fraction (e.g., 1/10).
T0means	Default 0 (assuming standardized variables). Can be assigned labels to estimate them freely.
manifestMeans	Default 0 (assuming standardized variables). Can be assigned labels to estimate them freely.
CoTiMAStanctArgs	parameters that can be set to improve model fitting of the ctStanFit Function
lambda	R-type matrix with pattern of fixed (=1) or free (any string) loadings.
manifestVars	define the error variances of the manifests with a single time point using R-type lower triangular matrix with nrow=n.manifest & ncol=n.manifest.

Value

ctmaFit returns a list containing somearguments supplied, the fitted model, different elements summarizing the main results, model type, and the type of plot that could be performed with the returned object. The arguments in the returned object are activeDirectory, coresToUse, moderator names (mod.names), and moderator type (mod.type). Further arguments, which are just copied from the init-fit object supplied, are, n.latent, studyList, parameterNames, and statisticsList. The fitted model is found in studyFitList, which is a large list with many elements (e.g., the ctsem model specified by CoTiMA, the rstan model created by ctsem, the fitted rstan model etc.). Further results returned are n.studies = 1 (required for proper plotting), data (created pseudo raw data), and a list with modelResults (i.e., DRIFT=model_Drift_Coef, DIFFUSION=model_Diffusion_Coef, T0VAR=model_T0var_Coef, CINT=model_Cint_Coef, MOD=modTI_Coeff, and CLUS=clusTI_Coeff). Possible invariance constraints are included in invariantDrift. The number of moderators simultaneously analyzed are included in ' n.moderators. The most important new results are returned as the list element "summary", which is printed if the summary function is applied to the returned object. The summary list element comprises "estimates" (the aggregated effects), possible randomEffects (not yet fully working), the minus2ll value and its n.parameters, the opt.lag sensu Dormann & Griffin (2015) and the max.effects that occur at the opt.lag, clus.effects and mod.effects, and possible warning messages (message). Plot type is plot.type=c("drift") and model.type="stanct" ("omx" was deprecated).

Examples

## Not run: 
# Example 1. Fit a CoTiMA to all primary studies previously fitted one by one
# with the fits assigned to CoTiMAInitFit_6
CoTiMAFullFit_6 <- ctmaFit(ctmaInitFit=CoTiMAInitFit_6)
summary(CoTiMAFullFit_6)

## End(Not run)

## Not run: 
# Example 2. Fit a CoTiMA with only 2 cross effects invariant (not the auto
# effects) to all primary studies previously fitted one by one with the fits
# assigned to CoTiMAInitFit_6
CoTiMAInitFit_6$activeDirectory <- "/Users/tmp/" # adapt!
CoTiMAFullInv23Fit_6 <- ctmaFit(ctmaInitFit=CoTiMAInitFit_6,
                        invariantDrift=c("V1toV2", "V2toV1"))
summary(CoTiMAFullInv23Fit_6)

## End(Not run)

## Not run: 
# Example 3. Fit a moderated CoTiMA
CoTiMAInitFit_6$activeDirectory <- "/Users/tmp/" # adapt!
CoTiMAMod1onFullFit_6 <- ctmaFit(ctmaInitFit=CoTiMAInitFit_6,
                                 mod.number=1, mod.type="cont",
                                 mod.names=c("Control"))
summary(CoTiMAMod1onFullFit_6)

## End(Not run)

CoTiMA documentation built on Nov. 10, 2022, 5:16 p.m.