quid (version 0.0.1)

constraintBF: Function to compute Bayes factors for ordinal constraints

Description

This function uses Bayesian mixed models to estimate individual effect sizes and to test theoretical order constraints.

Usage

constraintBF(
  formula,
  data,
  whichRandom,
  ID,
  whichConstraint,
  rscaleEffects,
  iterationsPosterior = 10000,
  iterationsPrior = iterationsPosterior * 10,
  burnin = 1000,
  ...
)

Arguments

formula

a formula containing the full model.

data

a data.frame containing the data with all variables defined in the formula.

whichRandom

a character vector specifying which factors are random.

ID

a character vector of length one specifying which variable holds the subject ID.

whichConstraint

a named character vector specifying the constraints placed on certain factors; see Details.

rscaleEffects

a named vector of prior settings for individual factors. Values are scales, names are factor names; see Details.

iterationsPosterior

the number of iterations to sample from the posterior of the full model.

iterationsPrior

the number of iterations to sample from the prior of the full model.

burnin

the number of initial iterations to discard from posterior sampling.

...

further arguments to be passed to generalTestBF.

Value

An object of class BFBayesFactorConstraint-class.

Details

This function provides a way of testing whether theoretical constraints on certain effects hold for all subjects. The backend is provided by the generalTestBF function from the BayesFactor-package. The input formula is the full model to be tested. It usually contains an interaction term between the subject ID and the effect for which constraints are tested (e.g. ID:condition). The ID variable is to be specified in ID and is usually a random factor to be specified in whichRandom.

Order constraints on effects should be specified in whichConstraint, as a named character vector. Each constraint in the vector can take 2 levels of the effect. They are of the form: "effect name" = "condition A" < "condition B". In order to impute more than 2 levels, the same effect name has to be entered with different conditions as the value. For instance, for testing whether conditions A < B < C, the input should be: "effect name" = "condition A" < "condition B", "effect name" = "condition B" < "condition C". At this point, constraints can only be tested for the same effect.

Priors have to be specified for all factors in whichConstraint, for ID, and for the interaction between the two. A Detailed description of the models, priors and methods is given in the documentation of anovaBF and more extensively in Rouder et al. (2012).

References

Rouder, J. N., Morey, R. D., Speckman, P. L., Province, J. M., (2012) Default Bayes Factors for ANOVA Designs. Journal of Mathematical Psychology. 56. p. 356-374.

Examples

# NOT RUN {
data(stroop)

resStroop <- constraintBF(rtS ~ ID*cond,
                          data = stroop,
                          whichRandom = "ID",
                          ID = "ID",
                          whichConstraint = c(cond = "2 > 1"),
                          rscaleEffects = c("ID" = 1, "cond" = 1/6, "ID:cond" = 1/10))
# }
# NOT RUN {
# }