Previous Page | Next Page

The CALIS Procedure

Example 25.4 Linear Relations among Factor Loadings

In this example, a confirmatory factor-analysis model with linear constraints on loadings is specified by the COSAN statement. SAS programming statements are used to set the constraints. In the context of the current example, the differences between fitting covariance structures and correlation structures will also be discussed.

The correlation matrix of six variables from Kinzer and Kinzer (N=326) is used by Guttman (1957) as an example that yields an approximate simplex. McDonald (1980) uses this data set as an example of factor analysis where he supposes that the loadings of the second factor are linear functions of the loadings on the first factor. Let be the factor loading matrix containing the two factors and six variables so that

     

and

     

The correlation structures are represented by

     

where represents the diagonal matrix of unique variances for the variables.

With parameters and being unconstrained, McDonald (1980) has fitted an underidentified model with seven degrees of freedom. Browne (1982) imposes the following identification condition:

     

In this example, Browne’s identification condition is imposed. The following statements specify the confirmatory factor model:


   Data Kinzer(TYPE=CORR);
   Title "Data Matrix of Kinzer & Kinzer, see GUTTMAN (1957)";
      _TYPE_ = 'CORR'; INPUT _NAME_ $ var1-var6;
      Datalines;
   var1  1.00   .     .     .     .     .
   var2   .51  1.00   .     .     .     .
   var3   .46   .51  1.00   .     .     .
   var4   .46   .47   .54  1.00   .     .
   var5   .40   .39   .49   .57  1.00   .
   var6   .33   .39   .47   .45   .56  1.00
     ;
   proc calis data=Kinzer method=max nobs=326 nose;
      cosan B(2,Gen) * I(2,Ide) + U(6,Ide) * PSI(6,DIA);                                   
      matrix B                                                     
             [ ,1]= b11 b21 b31 b41 b51 b61,                                        
             [ ,2]= b12 b22 b32 b42 b52 b62;
      matrix Psi                                       
             [1,1]= psi1-psi6;                                      
      parameters alpha = .5;                                   
      /* SAS Programming Statements */
      b12  = alpha - b11;
      b22  = alpha - b21;
      b32  = alpha - b31;
      b42  = alpha - b41;
      b52  = alpha - b51;
      b62  = alpha - b61;
      vnames B  Fact1 Fact2,
             I  Fact1 Fact2,
             U  var1-var6,
             PSI var1-var6;                             
   run;
   

In the COSAN statement, you specify the model matrices for the correlation structures. Matrix is the factor loading matrix. Matrix is a fixed identity matrix, representing the factor variances and covariances. In the next term, matrix is a identity matrix, appearing there only because it makes the specification conform to the format of the generalized COSAN model. The PSI matrix is the matrix for variances and covariances for errors. The matrix specification in the COSAN statement realizes the following matrix model:

     

which is the intended correlation structures.

In the MATRIX statements, parameters are defined. The loading parameters are all specified with prefix "b" and are located in the matrix. Error variances are all specified with prefix "psi" and are located in the PSI matrix. In this model, all manifest variables have nonzero loadings on the two factors. These loading parameters are specified after the equal signs and are named with the prefix "b".


An additional parameter alpha is specified in the PARAMETERS statement with an initial value of 0.5. Then, six SAS programming statements are used to define the loadings on the second factor as functions of the loadings on the first factor.

Finally, the VNAMES statement is used to name the columns of the model matrices.

In the specification, there are 12 loadings in matrix and six unique variances in matrix PSI. Adding the parameter alpha in the list, there are 19 parameters in total. However, the loading parameters are not all independent of each other. As defined in the SAS programming statements, six loadings are dependent. This reduces the number of free parameters to 13. Hence the degrees of freedom for the model is .

In Output 25.4.1, a fit summary table is shown. The chi-square test statistic of model fit is 10.337 with =8 ( = 0.242). This indicates a good fit.

Output 25.4.1 Fit of the Correlation Structures
Data Matrix of Kinzer & Kinzer, see GUTTMAN (1957)
Fitting Correlations As Covariance Structures Directly

The CALIS Procedure
Covariance Structure Analysis: Maximum Likelihood Estimation

Fit Function 0.0318
Goodness of Fit Index (GFI) 0.9897
GFI Adjusted for Degrees of Freedom (AGFI) 0.9730
Root Mean Square Residual (RMR) 0.0409
Standardized Root Mean Square Residual (SRMR) 0.0409
Parsimonious GFI (Mulaik, 1989) 0.5278
Chi-Square 10.3374
Chi-Square DF 8
Pr > Chi-Square 0.2421
Independence Model Chi-Square 682.87
Independence Model Chi-Square DF 15
RMSEA Estimate 0.0300
RMSEA 90% Lower Confidence Limit .
RMSEA 90% Upper Confidence Limit 0.0756
ECVI Estimate 0.1136
ECVI 90% Lower Confidence Limit .
ECVI 90% Upper Confidence Limit 0.1525
Probability of Close Fit 0.7137
Bentler's Comparative Fit Index 0.9965
Normal Theory Reweighted LS Chi-Square 10.1441
Akaike's Information Criterion -5.6626
Bozdogan's (1987) CAIC -43.9578
Schwarz's Bayesian Criterion -35.9578
McDonald's (1989) Centrality 0.9964
Bentler & Bonett's (1980) Non-normed Index 0.9934
Bentler & Bonett's (1980) NFI 0.9849
James, Mulaik, & Brett (1982) Parsimonious NFI 0.5253
Z-Test of Wilson & Hilferty (1931) 0.7019
Bollen (1986) Normed Index Rho1 0.9716
Bollen (1988) Non-normed Index Delta2 0.9965
Hoelter's (1983) Critical N 489

The estimated factor loading matrix and error covariance matrix are presented in Output 25.4.2. The estimates for alpha and other dependent parameters are presented in Output 25.4.3.

Output 25.4.2 Estimates
Data Matrix of Kinzer & Kinzer, see GUTTMAN (1957)
Fitting Correlations As Covariance Structures Directly

The CALIS Procedure
Covariance Structure Analysis: Maximum Likelihood Estimation

Estimated Parameter Matrix B[6:2]
General Matrix
  Fact1 Fact2
var1
0.3609
[b11]
0.6174
var2
0.3212
[b21]
0.6571
var3
0.4859
[b31]
0.4923
var4
0.5745
[b41]
0.4038
var5
0.7985
[b51]
0.1797
var6
0.6736
[b61]
0.3046

Data Matrix of Kinzer & Kinzer, see GUTTMAN (1957)
Fitting Correlations As Covariance Structures Directly

The CALIS Procedure
Covariance Structure Analysis: Maximum Likelihood Estimation

Estimated Parameter Matrix PSI[6:6]
Diagonal Matrix
  var1 var2 var3 var4 var5 var6
var1
0.5304
[psi1]
0
 
0
 
0
 
0
 
0
 
var2
0
 
0.4499
[psi2]
0
 
0
 
0
 
0
 
var3
0
 
0
 
0.4876
[psi3]
0
 
0
 
0
 
var4
0
 
0
 
0
 
0.4728
[psi4]
0
 
0
 
var5
0
 
0
 
0
 
0
 
0.3112
[psi5]
0
 
var6
0
 
0
 
0
 
0
 
0
 
0.5382
[psi6]


Estimated Parameter Matrix U[6:6]
Identity Matrix
Constant Model Matrix

Output 25.4.3 Additional Parameter
Additional PARMS and
Dependent Parameters
The Number of Dependent
Parameters is 6
Parameter Estimate
alpha 0.97825
b12 0.61736
b22 0.65709
b32 0.49234
b42 0.40378
b52 0.17973
b62 0.30462


All these estimates are essentially the same as reported in Browne (1982). Notice that there are no standard error estimates in the output, as requested by the NOSE option in the PROC CALIS statement. Standard error estimates are not of interest in the current example.

Fitting the Correct Correlation Structures

In specifying the previous factor model by using COSAN, covariance structures rather than the correlation structures have been incorrectly assumed. That is, when fitting the correlation structures, the diagonal elements of must always be fixed ones. But these fixed ones have not been handled properly in the previous code, resulting in a specification of covariance structures rather than correlation structures. To fix the problem, additional model constraints on the diagonal elements of the correlation structures represented by are required. For example, it can be done by constraining the error variances

     

Other constraints could also serve the same purpose, but these are the most convenient and therefore are used subsequently. Without setting up these constraints, the previous PROC CALIS run is analyzing covariance structures. As pointed out in Browne (1982), fitting such covariance structures directly is not entirely appropriate for analyzing correlations. For the current data example, according to Browne (1982), the estimates obtained from fitting the wrong covariance structure model are still consistent (as if they were estimating the population parameters in the correlation structures). However, the chi-square test statistic as reported in the preceding section is not correct.

In addition, due to the fact that discrepancy functions used in PROC CALIS are derived for covariance matrices rather than correlation matrices, PROC CALIS is essentially set up for analyzing covariance structures (with or without mean structures), but not correlation structures. Hence, the statistical theory behind PROC CALIS applies to covariance structure analysis, but might not generalize to correlation structure analysis. Despite that, can PROC CALIS somehow fit the correct correlation structures to the current data without compromising the statistical theory? The answer is yes.

First, recall that the correlation structures are represented by:

     

As before, in the matrix, there are six linear constraints on the factor loadings. In addition, the diagonal elements of must be ones. To analyze the correlation structures by using PROC CALIS, a covariance structure model with such correlation structures embedded is now specified. That is, the covariance structure to be fitted by PROC CALIS is:

     

where is a diagonal matrix containing the population standard deviations for the manifest variables. This form of covariance structures implies the specification of the following COSAN model:


   proc calis data=Kinzer method=max nobs=326 nose;
      cosan D(6,DIA) * B(2,Gen) * I(2,Ide) + D(6,DIA) * PSI(6,DIA);                                   
      matrix B                                                     
             [ ,1]= b11 b21 b31 b41 b51 b61,                                        
             [ ,2]= b12 b22 b32 b42 b52 b62;
      matrix Psi                                       
             [1,1]= psi1-psi6;                                      
      matrix D                                       
             [1,1]= d1-d6 (6 * 1.);                                      
      parameters alpha (1.);
      /* SAS Programming Statements */
      /* 12 Constraints on Correlation structures */
      b12  = alpha - b11;
      b22  = alpha - b21;
      b32  = alpha - b31;
      b42  = alpha - b41;
      b52  = alpha - b51;
      b62  = alpha - b61;
      psi1 = 1. - b11 * b11 - b12 * b12;
      psi2 = 1. - b21 * b21 - b22 * b22;
      psi3 = 1. - b31 * b31 - b32 * b32;
      psi4 = 1. - b41 * b41 - b42 * b42;
      psi5 = 1. - b51 * b51 - b52 * b52;
      psi6 = 1. - b61 * b61 - b62 * b62;
      vnames D  var1-var6,
             B  Fact1 Fact2,
             I  Fact1 Fact2,
             Psi var1-var6;                             
   run;
   

In the COSAN statement, the covariance structures are described in the form of the generalized COSAN model. The most significant change is the inclusion of matrix for the population standard deviations. Accompanying this change is the addition of a MATRIX statement for defining the error variance parameters in matrix . Also, in this new specification, Psi1-Psi6 are defined as functions of the loading parameters:

     

This ensures that the correlation structures will have fixed ones for the diagonal elements.

Other constraints remain the same way as defined in the previous run. For example, the loadings on the second factor are functions of the loadings on the first factor:

     

The fit summary is presented in Output 25.4.4. The chi-square test statistic is 14.63 with =8 ( = 0.067). This shows that the previous chi-square test based on fitting a wrong covariance structure model is indeed questionable.

Output 25.4.4 Model Fit of the Correlation Structures
Data Matrix of Kinzer & Kinzer, see GUTTMAN (1957)
Fitting Correlations By Using Correct Covariance Structures

The CALIS Procedure
Covariance Structure Analysis: Maximum Likelihood Estimation

Fit Function 0.0450
Goodness of Fit Index (GFI) 0.9849
GFI Adjusted for Degrees of Freedom (AGFI) 0.9604
Root Mean Square Residual (RMR) 0.0378
Standardized Root Mean Square Residual (SRMR) 0.0378
Parsimonious GFI (Mulaik, 1989) 0.5253
Chi-Square 14.6269
Chi-Square DF 8
Pr > Chi-Square 0.0668
Independence Model Chi-Square 682.87
Independence Model Chi-Square DF 15
RMSEA Estimate 0.0505
RMSEA 90% Lower Confidence Limit .
RMSEA 90% Upper Confidence Limit 0.0908
ECVI Estimate 0.1268
ECVI 90% Lower Confidence Limit .
ECVI 90% Upper Confidence Limit 0.1729
Probability of Close Fit 0.4382
Bentler's Comparative Fit Index 0.9901
Normal Theory Reweighted LS Chi-Square 14.9199
Akaike's Information Criterion -1.3731
Bozdogan's (1987) CAIC -39.6683
Schwarz's Bayesian Criterion -31.6683
McDonald's (1989) Centrality 0.9899
Bentler & Bonett's (1980) Non-normed Index 0.9814
Bentler & Bonett's (1980) NFI 0.9786
James, Mulaik, & Brett (1982) Parsimonious NFI 0.5219
Z-Test of Wilson & Hilferty (1931) 1.5034
Bollen (1986) Normed Index Rho1 0.9598
Bollen (1988) Non-normed Index Delta2 0.9902
Hoelter's (1983) Critical N 346

Estimates of the loadings, population standard deviations, error variances, and the alpha parameter are presented in Output 25.4.5 and Output 25.4.6.

Output 25.4.5 Estimates of Loadings and Standard Deviations
Estimated Parameter Matrix B[6:2]
General Matrix
  Fact1 Fact2
var1
0.3422
[b11]
0.6318
var2
0.3210
[b21]
0.6531
var3
0.4918
[b31]
0.4822
var4
0.5755
[b41]
0.3985
var5
0.7769
[b51]
0.1971
var6
0.6666
[b61]
0.3074

Estimated Parameter Matrix D[6:6]
Diagonal Matrix
  var1 var2 var3 var4 var5 var6
var1
1.0077
[d1]
0
 
0
 
0
 
0
 
0
 
var2
0
 
0.9971
[d2]
0
 
0
 
0
 
0
 
var3
0
 
0
 
0.9908
[d3]
0
 
0
 
0
 
var4
0
 
0
 
0
 
0.9909
[d4]
0
 
0
 
var5
0
 
0
 
0
 
0
 
0.9964
[d5]
0
 
var6
0
 
0
 
0
 
0
 
0
 
1.0169
[d6]

Output 25.4.6 Estimates of Error Variances and Alpha
Estimated Parameter Matrix PSI[6:6]
Diagonal Matrix
  var1 var2 var3 var4 var5 var6
var1
0.4837
0
 
0
 
0
 
0
 
0
 
var2
0
 
0.4705
0
 
0
 
0
 
0
 
var3
0
 
0
 
0.5256
0
 
0
 
0
 
var4
0
 
0
 
0
 
0.5100
0
 
0
 
var5
0
 
0
 
0
 
0
 
0.3576
0
 
var6
0
 
0
 
0
 
0
 
0
 
0.4612

Additional PARMS and
Dependent Parameters
The Number of Dependent
Parameters is 12
Parameter Estimate
alpha 0.97400
b12 0.63184
b22 0.65305
b32 0.48222
b42 0.39849
b52 0.19715
b62 0.30741
psi1 0.48370
psi2 0.47051
psi3 0.52561
psi4 0.50999
psi5 0.35763
psi6 0.46115

Except for the population standard deviation parameter d’s, all other parameters estimated in the current model can be compared with those from the previous fitting of an incorrect covariance structure model. Although estimates in the current model do not differ very much from those in the previous analysis, it is assuring that they are obtained from fitting a correctly specified covariance structure model with the intended correlation structures embedded.

Previous Page | Next Page | Top of Page