# Control Function for multordRS Control function for multordRS, a model for multivariate ordinal responses including response styles ```r 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. ## Returns Returns list of control parameters used in `multordRS`. ## Examples ```r 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") ``` ## References Schauberger, Gunther and Tutz, Gerhard (2021): Multivariate Ordinal Random Effects Models Including Subject and Group Specific Response Style Effects, **Statistical Modelling**, tools:::Rd_expr_doi("10.1177/1471082X20978034") ## See Also `multordRS` `MultOrdRS-package` `plot.MultOrdRS` ## Author(s) Gunther Schauberger gunther.schauberger@tum.de [https://orcid.org/0000-0002-0392-1580](https://orcid.org/0000-0002-0392-1580)

ctrl.multordRS function