The CALIS Procedure
-
Overview
-
Getting Started
-
SyntaxClasses of Statements in PROC CALISSingle-Group Analysis SyntaxMultiple-Group Multiple-Model Analysis SyntaxPROC CALIS StatementBOUNDS StatementBY StatementCOSAN StatementCOV StatementDETERM StatementEFFPART StatementFACTOR StatementFITINDEX StatementFREQ StatementGROUP StatementLINCON StatementLINEQS StatementLISMOD StatementLMTESTS StatementMATRIX StatementMEAN StatementMODEL StatementMSTRUCT StatementNLINCON StatementNLOPTIONS StatementOUTFILES StatementPARAMETERS StatementPARTIAL StatementPATH StatementPCOV StatementPVAR StatementRAM StatementREFMODEL StatementRENAMEPARM StatementSAS Programming StatementsSIMTESTS StatementSTD StatementSTRUCTEQ StatementTESTFUNC StatementVAR StatementVARIANCE StatementVARNAMES StatementWEIGHT Statement
-
DetailsInput Data SetsOutput Data SetsDefault Analysis Type and Default ParameterizationThe COSAN ModelThe FACTOR ModelThe LINEQS ModelThe LISMOD Model and SubmodelsThe MSTRUCT ModelThe PATH ModelThe RAM ModelNaming Variables and ParametersSetting Constraints on ParametersAutomatic Variable SelectionEstimation CriteriaRelationships among Estimation CriteriaGradient, Hessian, Information Matrix, and Approximate Standard ErrorsCounting the Degrees of FreedomAssessment of FitCase-Level Residuals, Outliers, Leverage Observations, and Residual DiagnosticsTotal, Direct, and Indirect EffectsStandardized SolutionsModification IndicesMissing Values and the Analysis of Missing PatternsMeasures of Multivariate KurtosisInitial EstimatesUse of Optimization TechniquesComputational ProblemsDisplayed OutputODS Table NamesODS Graphics
-
ExamplesEstimating Covariances and CorrelationsEstimating Covariances and Means SimultaneouslyTesting Uncorrelatedness of VariablesTesting Covariance PatternsTesting Some Standard Covariance Pattern HypothesesLinear Regression ModelMultivariate Regression ModelsMeasurement Error ModelsTesting Specific Measurement Error ModelsMeasurement Error Models with Multiple PredictorsMeasurement Error Models Specified As Linear EquationsConfirmatory Factor ModelsConfirmatory Factor Models: Some VariationsResidual Diagnostics and Robust EstimationThe Full Information Maximum Likelihood MethodComparing the ML and FIML EstimationPath Analysis: Stability of AlienationSimultaneous Equations with Mean Structures and Reciprocal PathsFitting Direct Covariance StructuresConfirmatory Factor Analysis: Cognitive AbilitiesTesting Equality of Two Covariance Matrices Using a Multiple-Group AnalysisTesting Equality of Covariance and Mean Matrices between Independent GroupsIllustrating Various General Modeling LanguagesTesting Competing Path Models for the Career Aspiration DataFitting a Latent Growth Curve ModelHigher-Order and Hierarchical Factor ModelsLinear Relations among Factor LoadingsMultiple-Group Model for Purchasing BehaviorFitting the RAM and EQS Models by the COSAN Modeling LanguageSecond-Order Confirmatory Factor AnalysisLinear Relations among Factor Loadings: COSAN Model SpecificationOrdinal Relations among Factor LoadingsLongitudinal Factor Analysis
- References
Missing Values and the Analysis of Missing Patterns
If the DATA= data set contains raw data (rather than a covariance or correlation matrix), in general observations with missing values for any variables in the analysis are omitted from the computations. The only exception is with METHOD=FIML. Incomplete observations with at least one nonmissing variables in the analysis are also used for the estimation.
If a covariance or correlation matrix is read, missing values are allowed as long as every pair of variables has at least one nonmissing value. Unlike the raw data input, METHOD=FIML does not allow missing values in the covariance or correlation matrix.
When you use METHOD=FIML, PROC CALIS provide several analyses on the missing patterns of the raw input data sets. First, PROC CALIS shows the coverage results for the means and covariances. The coverage results refer to the proportions of data present for computing the means and the covariances. Because distinct missing patterns in the data sets are possible, the coverage proportions for the individual means and covariances could vary. Average coverage proportions of the means and covariances give you an overall idea about the missingness (or the lack of). In order to help locate the problematic means and covariances that have the low coverage, PROC CALIS shows the rank orders of the smallest coverages of mean and covariance elements. The number of smallest coverages shown for the means is equal to half of the total number of variables. The number of smallest coverages shown for the covariances is equal to half of the total number of the distinct elements in the lower triangular of the covariance matrix. However, in both cases at most 10 smallest coverages would be shown.
Second, PROC CALIS ranks the most frequent missing patterns in the data set (the nonmissing pattern is excluded in the ranking). Because the number of missing patterns could be quite large, PROC CALIS displays only a limited number of most frequent missing patterns in the output. You can use the MAXMISSPAT= and the TMISSPAT= options to control the number of missing patterns to display. See these options for details.
Third, PROC CALIS shows the means of the most frequent missing patterns, along with the means for the nonmissing pattern for comparison.
See Example 27.15 for an illustration of the use of the full information maximum likelihood method and the analysis of missing patterns.