estimateGremlinsModel {RGremlinsConjoint}

Estimate Gremlin's Model - Hierarchical MNL

Description

The function estimates the model described in "Gremlin's in the Data: Identifying the Information Content of Research Subjects" (Howell et al. (2021) <doi:10.1177/0022243720965930>) using a hierarchical multinomial logit model

Usage

estimateGremlinsModel(
  data,
  design,
  Priors = NULL,
  R = NULL,
  keepEvery = 1,
  verbose = TRUE,
  num_lambda_segments = 2,
  constraints = NULL,
  startingValues = NULL,
  previous_iterations = 0,
  Atchade_slope_tuning = 0.1,
  Atchade_lambda_tuning = 10
)

Arguments

data

A matrix containing the raw data. The first column the respondent identifier, followed by the design number, the remaining columns indicate the choices for the tasks that coincide to the design file.

design

A matrix representing the coded (dummy of effects) design file. The design file should be formatted as a matrix with number of versions X number of tasks X number of alternatives rows and number of parameters + 3 columns. The first column contains the version number, the second columns contains the task number, the third column contains the alternative, and the remaining columns contain the coded design. A generic Sawtooth Software design file can be converted to this format using the code_sawtooth_design function.

Priors

A data structure that contains the priors for to the model. Can be null indicating the use of default priors or must contain a full prior specification.

R

The number of repetitions in the chain

keepEvery

saves every keepEvery-th draw for output

verbose

Print intermediate results to the screen (default = TRUE)

num_lambda_segments

(Default = 2) The number of segments for the scale factor

constraints

(Optional) a vector of length n-param specifying the constraints to impose on the parameters or NULL. a 1 indicates the parameter is constrained to be positive a -1 constrains to be negative, and a 0 indicates no constraint.

startingValues

(Optional) starting values to use for the MCMC algorithm. This is a list of containing: slope = a nRespondent by nParamter matrix of slopes for the respondent slopeBar = a nParameter vector of the slopeBar parameter slopeCov = a nParameter by nParameter matrix containing the variance covariance matrix for the slopeBar parameter lambda = a nSegment vector containing the starting values for the lambda parameters. The first element in the vector should be 1. segMembership = a nRespondent vector containing the segment membership for each respondent. phi_lambda = a nParameter vector containing the base probabilities that an individual belongs to each segment. Should sum to 1.

previous_iterations

The number of previous iterations run. This parameter is used for the Atchade adaptive MCMC step size algorithm. This is used since the Atchade update does not happen for less than 1000 iterations. (Default = 0)

Atchade_slope_tuning

The tuning factor to use for Atchade step for the slopes parameter. Larger values decrease the acceptance rate. (Default = 0.01)

Atchade_lambda_tuning

The tuning factor to use for the Atchade step for the lambda parameter. Larger values decrease the acceptance rate. (Default = 10)

Value

A data structure containing the draws from the complete MCMC chain

See Also

code_sawtooth_design

Examples

truck_design_file <- system.file("extdata", "simTruckDesign.csv", package = "RGremlinsConjoint")
truck_data_file <- system.file("extdata", "simTruckData.csv", package = "RGremlinsConjoint")

truckDesign <- read.csv(truck_design_file)
truckData <- read.csv(truck_data_file)
outputSimData_burn <- estimateGremlinsModel(truckData,
                                            truckDesign,
                                            R = 10,
                                            keepEvery = 1,
                                            num_lambda_segments = 2)