Chapter Contents |
Previous |
Next |
The CALIS Procedure |
There are four sets of statements available in the CALIS procedure to specify a model. Since a LISREL analysis can be performed easily by using a RAM, COSAN, or LINEQS statement, there is no specific LISREL input form available in the CALIS procedure.
For COSAN-style input, you can specify the following statements:
For linear equations input, you can specify the following statements:
For RAM-style input, you can specify the following statements:
For (confirmatory) factor analysis input, you can specify the following statements:
The model can also be obtained from an INRAM= data set, which is usually a version of an OUTRAM= data set produced by a previous PROC CALIS analysis (and possibly modified).
If no INRAM= data set is specified, you must use one of the four statements that defines the input form of the analysis model: COSAN, RAM, LINEQS, or FACTOR.
The following statements define the structural model of the alienation example as a COSAN model:
Cosan J(9, Ide) * A(9, Gen, Imi) * P(9, Sym); Matrix A [ ,7] = 1. .833 5 * 0. Beta (.5) , [ ,8] = 2 * 0. 1. .833 , [ ,9] = 4 * 0. 1. Lamb Gam1-Gam2 (.5 2 * -.5); Matrix P [1,1] = The1-The2 The1-The4 (6 * 3.) , [7,7] = Psi1-Psi2 Phi (2 * 4. 6.) , [3,1] = The5 (.2) , [4,2] = The5 (.2) ;
The matrix model specified in the COSAN statement is the RAM model
The MATRIX statement for matrix A specifies the values in columns 7, 8, and 9, which correspond to the three latent variables F1, F2, and F3, in accordance with the RAM model. The other columns of A are assumed to be zero. The initial values for the parameter elements in A are chosen as in the path diagram to be
In accordance with matrix P of the RAM model and the path model, the nine diagonal elements of matrix P are parameters with initial values
See the section "COSAN Model Statement" for more information about the COSAN statement.
The following statements define the structural model of the alienation example as a LINEQS model:
Lineqs V1 = F1 + E1, V2 = .833 F1 + E2, V3 = F2 + E3, V4 = .833 F2 + E4, V5 = F3 + E5, V6 = Lamb (.5) F3 + E6, F1 = Gam1(-.5) F3 + D1, F2 = Beta (.5) F1 + Gam2(-.5) F3 + D2; Std E1-E6 = The1-The2 The1-The4 (6 * 3.), D1-D2 = Psi1-Psi2 (2 * 4.), F3 = Phi (6.) ; Cov E1 E3 = The5 (.2), E4 E2 = The5 (.2);
The LINEQS statement shows the equations in the section "LINEQS Model", except that in this case the coefficients to be estimated can be followed (optionally) by the initial value to use in the optimization process. If you do not specify initial values for the parameters in a LINEQS statement, PROC CALIS tries to assign these values automatically. The endogenous variables used on the left side can be manifest variables (with names that must be defined by the input data set) or latent variables (which must have names starting with F). The variables used on the right side can be manifest variables, latent variables (with names that must start with an F), or error variables (which must have names starting with an E or D). Commas separate the equations. The coefficients to be estimated are indicated by names. If no name is used, the coefficient is constant, either equal to a specified number or, if no number is used, equal to 1. The VAR statement in Bentler's notation is replaced here by the STD statement, because the VAR statement in PROC CALIS defines the subset of manifest variables in the data set to be analyzed. The variable names used in the STD or COV statement must be exogenous (that is, they should not occur on the left side of any equation). The STD and COV statements define the diagonal and off-diagonal elements in the matrix. The parameter specifications in the STD and COV statements are separated by commas. Using k variable names on the left of an equal sign in a COV statement means that the parameter list on the right side refers to all k(k-1)/2 distinct variable pairs in the matrix. Identical coefficient names indicate parameters constrained to be equal. You can also use prefix names to specify those parameters for which you do not need a precise name in any parameter constraint.
See the section "LINEQS Model Statement" for more information about the precise syntax rules for a LINEQS statement.
The following statement defines the structural model of the alienation example as a RAM model:
Ram 1 1 7 1. , 1 2 7 .833 , 1 3 8 1. , 1 4 8 .833 , 1 5 9 1. , 1 6 9 .5 Lamb , 1 7 9 -.5 Gam1 , 1 8 7 .5 Beta , 1 8 9 -.5 Gam2 , 2 1 1 3. The1 , 2 2 2 3. The2 , 2 3 3 3. The1 , 2 4 4 3. The2 , 2 5 5 3. The3 , 2 6 6 3. The4 , 2 1 3 .2 The5 , 2 2 4 .2 The5 , 2 7 7 4. Psi1 , 2 8 8 4. Psi2 , 2 9 9 6. Phi ;
You must assign numbers to the nodes in the path diagram. In the path diagram of Figure 19.1, the boxes corresponding to the six manifest variables V1, ... , V6 are assigned the number of the variable in the covariance matrix (1, ... ,6); the circles corresponding to the three latent variables F1, F2, and F3 are given the numbers 7, 8, and 9. The path diagram contains 20 paths between the nine nodes; nine of the paths are one-headed arrows and eleven are two-headed arrows.
The RAM statement contains a list of items separated by commas. Each item corresponds to an arrow in the path diagram. The first entry in each item is the number of arrow heads (matrix number), the second entry shows where the arrow points to (row number), the third entry shows where the arrow comes from (column number), the fourth entry gives the (initial) value of the coefficient, and the fifth entry assigns a name if the path represents a parameter rather than a constant. If you specify the fifth entry as a parameter name, then the fourth list entry can be omitted, since PROC CALIS tries to assign an initial value to this parameter.
See the section "RAM Model Statement" for more information about the RAM statement.
For a first-order confirmatory factor analysis, you can use MATRIX statements to define elements in the matrices F, P, and U of the more general model
To perform a component analysis, specify the COMPONENT option to constrain the matrix U to a zero matrix; that is, the model is replaced by
Note that the rank of FF' is equal to the number m of components in F, and if m is smaller than the number of variables in the moment matrix C, the matrix of predicted model values is singular and maximum likelihood estimates for F cannot be computed. You should compute ULS estimates in this case.
The HEYWOOD option constrains the diagonal elements of U to be nonnegative; that is, the model is replaced by
If the factor loadings are unconstrained, they can be orthogonally rotated by one of the following methods:
The most common approach to factor analysis consists of two steps:
For default (exploratory) factor analysis, PROC CALIS computes initial estimates. If you use a MATRIX statement together with a FACTOR model specification, initial values are generally computed by McDonald's (McDonald and Hartmann 1992) method or are set by the START= option. See the section "FACTOR Model Statement" and Example 19.3 for more information about the FACTOR statement.
Although much effort has been made to implement reliable and numerically stable optimization methods, no practical algorithm exists that can always find the global optimum of a nonlinear function, especially when there are nonlinear constraints.
Chapter Contents |
Previous |
Next |
Top |
Copyright © 1999 by SAS Institute Inc., Cary, NC, USA. All rights reserved.