ctrl.multordRS: Control Function for multordRS

In MultOrdRS: Model Multivariate Ordinal Responses Including Response Styles

Description

Control function for multordRS, a model for multivariate ordinal responses including response styles

Usage

ctrl.multordRS(
  RS = TRUE,
  thresholds.acat = c("full", "shift", "minimal"),
  XforRS = TRUE,
  opt.method = c("L-BFGS-B", "nlminb"),
  Q = 10,
  cores = 5,
  lambda = 0.01
)

Arguments

RS	Logical value indicating whether response style should be modelled.
thresholds.acat	Type of parametrization used for thresholds: thresholds = "full" implies separate estimates of threshold values for each response variable; thresholds = "shift" implies equal threshold parameter across all response variables modified by shift parameters for each response variable; thresholds = "minimal" implies equal threshold parameter across all response variables. This option only applies for adjacent categories models (model = "acat" and is not implemented for cumulative models.)
XforRS	Logical value indicating whether also covariate effects on the response style should be considered. Only relevant if RS = TRUE.
opt.method	Specifies optimization algorithm used by optim, either L-BFGS-B or nlminb.
Q	Number of nodes to be used (per dimension) in Gauss-Hermite-Quadrature. If RS = TRUE, Gauss-Hermite-Quadrature is two-dimensional.
cores	Number of cores to be used in parallelized computation.
lambda	Tuning parameter for internal ridge penalty. It is supposed to be set to a small value to stabilize estimates.

Value

Returns list of control parameters used in multordRS.

Author(s)

Gunther Schauberger
[e-mail]
https://orcid.org/0000-0002-0392-1580

References

Schauberger, Gunther and Tutz, Gerhard (2021): Multivariate Ordinal Random Effects Models Including Subject and Group Specific Response Style Effects, Statistical Modelling, https://journals.sagepub.com/doi/10.1177/1471082X20978034

See Also

multordRS MultOrdRS-package plot.MultOrdRS

Examples

data(tenseness)

## create a small subset of the data to speed up calculations
set.seed(1860)
tenseness <- tenseness[sample(1:nrow(tenseness), 300),]

## scale all metric variables to get comparable parameter estimates
tenseness$Age <- scale(tenseness$Age)
tenseness$Income <- scale(tenseness$Income)

## two formulas, one without and one with explanatory variables (gender and age)
f.tense0 <- as.formula(paste("cbind(",paste(names(tenseness)[1:4],collapse=","),") ~ 1"))
f.tense1 <- as.formula(paste("cbind(",paste(names(tenseness)[1:4],collapse=","),") ~ Gender + Age"))



####
## Adjacent Categories Models
####

## Multivariate adjacent categories model, without response style, 
## without explanatory variables
m.tense0 <- multordRS(f.tense0, data = tenseness, control = ctrl.multordRS(RS = FALSE))
m.tense0

## Multivariate adjacent categories model, with response style as a random effect, 
## without explanatory variables
m.tense1 <- multordRS(f.tense0, data = tenseness)
m.tense1

## Multivariate adjacent categories model, with response style as a random effect, 
## without explanatory variables for response style BUT for location
m.tense2 <- multordRS(f.tense1, data = tenseness, control = ctrl.multordRS(XforRS = FALSE))
m.tense2

## Multivariate adjacent categories model, with response style as a random effect, with 
## explanatory variables for location AND response style
m.tense3 <- multordRS(f.tense1, data = tenseness)
m.tense3

plot(m.tense3)



####
## Cumulative Models
####

## Multivariate cumulative model, without response style, without explanatory variables
m.tense0.cumul <- multordRS(f.tense0, data = tenseness, 
  control = ctrl.multordRS(RS = FALSE), model = "cumulative")
m.tense0.cumul

## Multivariate cumulative model, with response style as a random effect, 
## without explanatory variables
m.tense1.cumul <- multordRS(f.tense0, data = tenseness, model = "cumulative")
m.tense1.cumul

## Multivariate cumulative model, with response style as a random effect, 
## without explanatory variables for response style BUT for location
m.tense2.cumul <- multordRS(f.tense1, data = tenseness, 
  control = ctrl.multordRS(XforRS = FALSE), model = "cumulative")
m.tense2.cumul

## Multivariate cumulative model, with response style as a random effect, with 
## explanatory variables for location AND response style
m.tense3.cumul <- multordRS(f.tense1, data = tenseness, model = "cumulative")
m.tense3.cumul

plot(m.tense3.cumul)

################################################################
## Examples from Schauberger and Tutz (2020) on 
## Data from the German Longitudinal Election Study (GLES) 2017
#################################################################

####
## Source: German Longitudinal Election Study 2017 
## Rossteutscher et al. 2017, https://doi.org/10.4232/1.12927
####

## load GLES data
data(GLES17)

## scale data
GLES17[,7:11] <- scale(GLES17[,7:11])

## define formula
f.GLES <- as.formula(cbind(RefugeeCrisis, ClimateChange, Terrorism, 
                       Globalization, Turkey, NuclearEnergy) ~ 
                       Age + Gender + Unemployment + EastWest + Abitur)

## fit adjacent categories model without and with response style parameters
m.GLES0 <- multordRS(f.GLES, data = GLES17, control =  ctrl.multordRS(RS = FALSE, cores = 6))
m.GLES <- multordRS(f.GLES, data = GLES17, control =  ctrl.multordRS(cores = 6))

m.GLES0
m.GLES

plot(m.GLES, main = "Adjacent categories model")


## fit cumulative model without and with response style parameters (takes pretty long!!!)
m.GLES20 <- multordRS(f.GLES, data = GLES17,  model="cumul", 
control = ctrl.multordRS(opt.method = "nlminb", cores = 6, RS = FALSE))

m.GLES2 <- multordRS(f.GLES, data = GLES17,  model="cumul", 
control = ctrl.multordRS(opt.method = "nlminb", cores = 6))

m.GLES20
m.GLES2

plot(m.GLES2, main = "Cumulative model")

MultOrdRS documentation built on March 30, 2021, 1:07 a.m.