adjust the overall model fit table

Author: juliuspfadt

R/plssem.R

@@ -309,29 +309,11 @@ checkCSemModel <- function(model, availableVars) {
 .plsSemFitTab <- function(modelContainer, dataset, options, ready) {
   # create model fit table
   if (!is.null(modelContainer[["fittab"]])) return()
-
-  fittab <- createJaspTable(title = gettext("Model Fit"))
-  fittab$dependOn(optionsFromObject = modelContainer,
-                  options = c("group", "bollenStineBootstrapSamples", "significanceLevel", "saturatedStructuralModel"))
-  fittab$position <- 0
-
-  if (options[["group"]] != "")
-    fittab$addColumnInfo(name = "group",  title = gettext("Group"),              type = "string", combine = TRUE)
-
-  fittab$addColumnInfo(name = "measure",  title = gettext("Distance Measure"),                type = "string" )
-  fittab$addColumnInfo(name = "statistic",      title = gettext("Test Statistic"),                type = "number" )
-  fittab$addColumnInfo(name = "critValue",      title = gettext("Critical Value"),                type = "number" )
-
-  modelContainer[["fittab"]] <- fittab
-
   if (!ready) return()
 
   # fill model fit table
   plsSemResults <- .plsSemComputeResults(modelContainer, dataset, options)
-
-  if (modelContainer$getError()) return()
-
-  # we need this for a lot of other tables
+  # we need this for a lot of other tables so we do this once here:
   results <- plsSemResults[[1]]
   msc <- .withWarnings(.computeMSC(results, dataset, options))
   #create jasp state and store msc for additional output tables
@@ -340,33 +322,54 @@ checkCSemModel <- function(model, availableVars) {
   modSelCriteria$dependOn(optionsFromObject = modelContainer)
   modSelCriteria$object <- msc
 
+  if (modelContainer$getError() || !options[["overallModelFit"]]) return()
+
+  fittab <- createJaspTable(title = gettext("Model Fit"))
+  fittab$dependOn(optionsFromObject = modelContainer,
+                  options = c("group", "omfBootstrapSamples", "omfSignificanceLevel", "saturatedStructuralModel"))
+  fittab$position <- 0
+
+  if (options[["group"]] != "")
+    fittab$addColumnInfo(name = "group",  title = gettext("Group"),              type = "string", combine = TRUE)
+
+  fittab$addColumnInfo(name = "measure",    title = gettext("Distance Measure"),                type = "string" )
+  fittab$addColumnInfo(name = "statistic",  title = gettext("Test Statistic"),                type = "number" )
+  fittab$addColumnInfo(name = "critValue",  title = gettextf("Critical Value (%s%% Quantile)", (1 - options[["omfSignificanceLevel"]]) * 100), type = "number" )
+
   vv <- cSEM::verify(plsSemResults[[1]])
   if (any(unlist(vv))) {
     fittab$setError(gettext("At least one result is inadmissible."))
     return()
   }
 
   omf <- .withWarnings(cSEM::testOMF(.object = plsSemResults[[1]],
-                                     .alpha = options[["significanceLevel"]],
-                                     .R = options[["bollenStineBootstrapSamples"]],
+                                     .alpha = options[["omfSignificanceLevel"]],
+                                     .R = options[["omfBootstrapSamples"]],
                                      .saturated = options[["saturatedStructuralModel"]],
                                      .seed = if (options[["setSeed"]]) options[["seed"]]))
   fit <- omf$value
 
   if (options[["group"]] == "") {
-    stat      <- fit$Test_statistic
+    stat <- fit$Test_statistic
     fittab[["measure"]] <- names(stat)
     fittab[["statistic"]] <- stat
     fittab[["critValue"]] <- fit$Critical_value
 
   } else {
-    stats     <- sapply(fit, function(x) x$Test_statistic)
+    stats <- sapply(fit, function(x) x$Test_statistic)
     fittab[["group"]]     <- rep(names(fit), each = nrow(stats))
     fittab[["measure"]] <- rep(rownames(stats), ncol(stats))
     fittab[["statistic"]] <- c(stats)
     fittab[["critValue"]] <- c(sapply(fit, function(x) x$Critical_value))
   }
 
+  if (options[["saturatedStructuralModel"]])
+    fittab$addFootnote(gettext("Denotes the fit of the model including a saturated structural model."))
+
+  modelContainer[["fittab"]] <- fittab
+
+  return()
+
 }
 
 # compute model selection criteria/ fit measures

inst/help/PLSSEM.md

@@ -4,23 +4,33 @@ This document explains how to perform Partial Least Squares Structural Equation
 
 ## 1. Model Setup
 ---
-In the **Model** section, you can specify the structural equation model using lavaan syntax. The following operators can be used: =~ to specify a latent variable 
-measured by a set of indicators; <~ to specify a composite/emergent variable that is made up of a set of indicators; and ~ to specify the structural model, i.e., the relationships 
-between the latent variables and the composites. In specifying the relationships between the latent variables and the composites, 
-they must not be isolated in the structural model. In addition, no indicator may be connected to more than one latent variable/composite. 
-Furthermore, the structural model may not include any observed variables, i.e., to include observed variables in the structural model they must be specified as
-a single-indicator construct. Finally, a grouping variable can be selected. In this case, the model is estimated separately for each group. 
-
-- **Grouping Variable**: You can select the grouping variable for multi-group analysis. The grouping variable is optional and can be left empty if not required.
+In the **Model** section, you can specify the structural equation model using lavaan syntax. 
+The following operators can be used: =~ to specify a latent variable 
+measured by a set of indicators; <~ to specify a composite/emergent variable that is made up of a set of indicators; 
+and ~ to specify the structural model, i.e., the relationships 
+between the latent variables and the composites. In specifying 
+the relationships between the latent variables and the composites, 
+they must not be isolated in the structural model. In addition, 
+no indicator may be connected to more than one latent variable/composite. 
+Furthermore, the structural model may not include any observed variables, i.e.,
+to include observed variables in the structural model they must be specified as
+a single-indicator construct. Finally, a grouping variable can be selected. 
+In this case, the model is estimated separately for each group. 
+
+- **Grouping Variable**: You can select the grouping variable for multi-group analysis. 
+The grouping variable is optional and can be left empty if not required.
 
 ## 2. Estimation Options
 ---
 In the **Estimation** section, the following options are available:
 
-- **Consistent Partial Least Squares**: Enables the option to use consistent partial least squares (PLSc), which, in contrast to traditional PLS, produces consistent estimates for latent variable models (=~). In this case, Mode A weights are transformed to obtain consistent factor loadings. 
-In addition, the correlations between latent variables and other variables of the structural model are corrected for attenuation before they are used to estimate the parameters of the structural model.  
+- **Consistent partial least squares**: Enables the option to use consistent partial least squares (PLSc), which, 
+in contrast to traditional PLS, produces consistent estimates for latent variable models (=~). 
+In this case, Mode A weights are transformed to obtain consistent factor loading estimates. 
+In addition, the correlations between latent variables and other variables of the structural model are corrected 
+for attenuation before they are used to estimate the parameters of the structural model.  
 
-- **Inner Weighting Scheme**: Choose from the following options to calculate inner weights used in the PLS algorithm:
+- **Inner weighting scheme**: Choose from the following options to calculate inner weights used in the PLS algorithm:
   - Path weighting scheme
   - Centroid weighting scheme
   - Factorial weighting scheme
@@ -36,13 +46,11 @@ In case of centroid and factorial inner weighting schemes, the structural model
 
 - **Error calculation method**: Choose from the following options to calculate the standard errors and confidence intervals of the parameter estimates:
   - None
-  - Robust: Use one of the following resample techniques:
-    - Bootstrap
-    - Jackknife
+  - Bootstrap
 
 **Samples**: The number of bootstrap runs can be specified.  
 
-**Repeatabilty**: A seed can be set, to make the analysis reproducible.  
+**Repeatability**: A seed can be set, to make the analysis reproducible.  
 
 
 ## 3. Output Options
@@ -59,6 +67,20 @@ The **Output** section includes options to customize the output you want to gene
 - **Observed construct correlations**: Enable output of the construct correlation matrix. 
 - **Implied construct correlations**: Enable output of the model-implied construct correlation matrix. 
 
+- **Overall model fit** Enables the option to assess the overall fit of the model. 
+In particular, bootstrap in combination with various distance measures, i.e., the geodesic distance (dG), 
+the standardized root mean square residual (SRMR), the squared Euclidean distance (dL), 
+and the distance of the ML fit function (dML) is used to assess 
+the discrepancy between the sample correlation matrix and the model-implied counterpart. 
+In the literature, this approach is also known as Bollen-Stine bootstrap. If the test statistic value
+is below the critical value the discrepancy is so small that equality between the model-implied 
+and the sample correlation matrix can be assumed, that is, the model is a good representation of the structure
+in the population.
+  - Bootstrap runs: Specify the number of bootstrap runs
+  - significance level: Specify the significance level for the bootstrap test. 
+  The significance level is used to determine the critical value. 
+  The critical value is determined as 1-alpha quantile of the bootstrap distance measures.
+  - Saturated structural model: Enables the option to assess the overall fit of a model with a saturated structural model.
 
 You can also add **construct scores** to the dataset for further analysis.
 

inst/qml/PLSSEM.qml

@@ -143,13 +143,7 @@ Form
 				RadioButton { value: "drop"; 	label: qsTr("Drop")						}
 			}
 		}
-		Group 
-		{
-			title: qsTr("Overall Model Fit")
-			IntegerField { name: "bollenStineBootstrapSamples"; label: qsTr("Bollen-Stine bootstrap samples"); fieldWidth: 60; defaultValue: 499; min: 100 }
-			CIField { text: qsTr("Significance level"); name: "significanceLevel"; defaultValue: 5 }
-			CheckBox { name: "saturatedStructuralModel"; label: qsTr("Saturated structural model") }
-		}
+
 		SetSeed {}
 
 	}
@@ -174,6 +168,14 @@ Form
 			CheckBox { name: "impliedConstructCorrelation"; 	label: qsTr("Implied construct correlations")	}
 		}
 
+		Group 
+		{
+			CheckBox { name: "overallModelFit"; label: qsTr("Overall model fit") ; id: omf}
+			IntegerField { visible:omf.checked; name: "omfBootstrapSamples"; label: qsTr("Bootstrap samples"); fieldWidth: 60; defaultValue: 499; min: 100 }
+			CIField { visible: omf.checked; text: qsTr("Significance level"); name: "omfSignificanceLevel"; defaultValue: 5 }
+			CheckBox { visible: omf.checked; name: "saturatedStructuralModel"; label: qsTr("Saturated structural model") }
+		}
+
 		CheckBox
 		{
 			name: "addConstructScores"