The CALIS Procedure

 

Example 26.2 Estimating Covariances and Means Simultaneously

This example uses the same data set that is used in Example 26.1 and estimates the means and covariances. Use the MSTRUCT model specification as shown in the following statements:

proc calis data=sales meanstr nostand;                    
   mstruct var=q1-q4;
   run;

In the PROC CALIS statement, specify the MEANSTR option to request the mean structure analysis in addition to the default covariance structure analysis. If you are not interested in the standardized solution, specify the NOSTAND option in the PROC CALIS statement to suppress computation of the standardized estimates. Without further model specification (such as the MATRIX statement), PROC CALIS assumes a saturated structural model with all means, variances, and covariances as model parameters.

Output 26.2.1 shows the modeling information. With the MEANSTR option specified in the PROC CALIS statement, the current analysis type is Means and Covariances, instead of the default Covariances in Example 26.1.

Output 26.2.1 Modeling Information of the Saturated Mean and Covariance Structure Model for the Sales Data
Saturated Means and Covariance Structures Using MSTRUCT

The CALIS Procedure
Mean and Covariance Structures: Model and Initial Values

Modeling Information
Data Set WORK.SALES
N Records Read 14
N Records Used 14
N Obs 14
Model Type MSTRUCT
Analysis Means and Covariances

Variables in the Model
q1 q2 q3 q4
Number of Variables = 4

Output 26.2.2 shows the fit summary of the current model. Again, this is a perfect model fit with 0 chi-square value and 0 degrees of freedom.

Output 26.2.2 Fit Summary of the Saturated Mean and Covariance Structure Model for the Sales Data
Fit Summary
Chi-Square 0.0000
Chi-Square DF 0
Pr > Chi-Square .

Output 26.2.3 shows the estimates of the means, together with the standard error estimates and the values. These estimated means are exactly the same as the sample means, which are not shown here.

Output 26.2.3 Mean Estimates for the Sales Data
MSTRUCT _Mean_ Vector
Variable Estimate Standard
Error
t Value
q1 1.36714 0.16132 8.47491
q2 2.07429 0.13146 15.77902
q3 2.20286 0.21596 10.20008
q4 3.65500 0.45281 8.07176

Output 26.2.4 shows the variance and covariance estimates. These estimates are exactly the same as the elements in the sample covariance matrix. In addition, these estimates match the estimates in Output 26.1.5 of Example 26.1, where only the covariance structures are analyzed.

Output 26.2.4 Variance and Covariance Estimates for the Sales Data
MSTRUCT _COV_ Matrix: Estimate/StdErr/t-value
  q1 q2 q3 q4
q1
0.3383
0.1327
2.5495
0.000198
0.0765
0.002587
0.0361
0.1260
0.2865
0.2214
0.2704
0.8186
q2
0.000198
0.0765
0.002587
0.2247
0.0881
2.5495
0.1265
0.1082
1.1693
0.2443
0.2251
1.0853
q3
0.0361
0.1260
0.2865
0.1265
0.1082
1.1693
0.6063
0.2378
2.5495
0.6301
0.3935
1.6012
q4
0.2214
0.2704
0.8186
0.2443
0.2251
1.0853
0.6301
0.3935
1.6012
2.6655
1.0455
2.5495

These estimates are essentially the same as the sample means, variances, and covariances. This kind of analysis is much easier using PROC CORR with the NOMISS option. However, the main purpose of Example 26.1 and Example 26.2 is to introduce the MSTRUCT modeling language and some basic but important options in PROC CALIS. You can apply the MSTRUCT modeling language to more sophisticated situations that are beyond the saturated mean and covariance structure models. Example 26.3 and Example 26.4 fit some patterned covariance models that are nonsaturated. Also, options such as NOSE, NOSTAND, and MEANSTR are useful for all modeling languages in PROC CALIS.