FitMaxDiff: 'FitMaxDiff'
In erikerhardt/flipMaxDiff: MaxDiff Experimental Design and Analysis

Description Usage Arguments

View source: R/max-diff.R

Fits a latent class rank-ordered logit model with ties to a max-diff experiment.

FitMaxDiff(design, version = NULL, best, worst, alternative.names,
  n.classes = 1, subset = NULL, weights = NULL, characteristics = NULL,
  seed = 123, initial.parameters = NULL, trace = 0,
  sub.model.outputs = FALSE, lc = TRUE, output = "Probabilities",
  tasks.left.out = 0, is.mixture.of.normals = FALSE,
  algorithm = "Default", normal.covariance = "Full",
  pool.variance = FALSE, lc.tolerance = 1e-04, n.draws = 100,
  is.tricked = FALSE, hb.iterations = 100, hb.chains = 1,
  hb.max.tree.depth = 10, hb.adapt.delta = 0.8)

`design`	A `data.frame`, where the first variable is called 'Version', the second is called 'Task' or 'Question', and the remaining variables contain the alternatives shown in each task.
`version`	A vector of integers showing the version of the design shown to each respondent.
`best`	A data frame of factors or a matrix of integers showing the choices made by each respondent on each of the questions. One column for each task. The integers need to correspond to the `design` vector of integers showing the version of the design shown to each respondent. Coerced to a matrix if a `data.frame`.
`worst`	As with 'best', except denoting worst..
`alternative.names`	A `character` vector of the alternative names. Where `best` and `worst` are factors or characters, these names must match them.
`n.classes`	The number of latent classes.
`subset`	An optional vector specifying a subset of observations to be used in the fitting process.
`weights`	An optional vector of sampling or frequency weights.
`characteristics`	Data frame of characteristics on which to run varying coefficients by latent class boosting.
`seed`	Seed for initial random class assignments.
`initial.parameters`	Specify initial parameters intead of starting at random.
`trace`	Non-negative integer indicating the detail of outputs provided when fitting models: 0 indicates no outputs, and 6 is the most detailed outputs.
`sub.model.outputs`	If TRUE, prints diagnostics on interim models.
`lc`	Whether to run latent class step at the end if characteristics are supplied.
`output`	Output type. Can be "Probabilities" or "Classes".
`tasks.left.out`	Number of questions to leave out for cross-validation.
`is.mixture.of.normals`	Whether to model with mixture of normals instead of LCA.
`algorithm`	If "HB", Hierarchical Bayes with a MVN prior is used and the other parameters are ignored.
`normal.covariance`	The form of the covariance matrix for mixture of normals. Can be 'Full, 'Spherical', 'Diagonal'.
`pool.variance`	Whether to pool parameter covariances between classes in mixture of normals.
`lc.tolerance`	The tolerance used for defining convergence in latent class analysis.
`n.draws`	The number of draws when fitting mixture of normals.
`is.tricked`	Whether to use tricked logit instead of rank-ordered logit with ties.
`hb.iterations`	The number of iterations in Hierarchical Bayes.
`hb.chains`	The number of chains in Hierarchical Bayes.
`hb.max.tree.depth`	http://mc-stan.org/misc/warnings.html#maximum-treedepth-exceeded
`hb.adapt.delta`	http://mc-stan.org/misc/warnings.html#divergent-transitions-after-warmup