Need to check (e.g., cov) and pretty-up: The LISREL software may be obtained at www.ssicentral.com. Other software packages include Eqs (www.mvsoft.com), and Amos (available through SPSS). Useful documentation is provided by Barbara Byrne (1998) for Lisrel (and other books on Eqs and Amos). We focus on Lisrel because it is the most comprehensive of the packages, and the matrix notation in the literature and SEM texts derive from Lisrel. Need to check (e.g., cov) and pretty-up: Confirmatory Factor Analysis Syntax: Confirmatory Factor Analysis example. da ni=15 no=100 ma=cm la q1 q2 q3 c1 c2 c3 v1 v2 v3 cs1 cs2 cs3 r1 r2 r3 cm sy 2.813 2.264 1.973

1.961 1.670 1.528 etc. ... 1.530 mo nx=15 nk=5 lx=fu,fr td=di,fr ph=st,fr pa lx 10000 Need to check (e.g., cov) and pretty-up: mo nx=15 nk=5 lx=fu,fr td=di,fr ph=st,fr pa lx 10000 10000 10000 01000

01000 01000 00100 00100 00100 00010 00010 00010 00001 00001 00001 pd out me=ml rs mi Need to check (e.g., cov) and pretty-up: Commentary: first line is a title, arbitrary but ends with a period. da stands for dataline, with number of

input variables (ni=15), number of observations or sample size (no=100) and that the matrix to be analyzed is a covariance matrix (ma=cm). la says that the variable labels follow on the next line. cm sy says that a covariance matrix follows, in symmetric format (lower triangle of the matrix). mo = model, nx = number of x variables, nk = number of ksi (constructs). lx is the lambda x (factor loadings matrix) which begins as a fu,fr full matrix (15x5) with all elements free to be estimated (this status is modified shortly). td is theta delta and all its diagonal elements are free to be estimated. ph is phi, the factor intercorrelation matrix, it is st standardized to be a correlation matrix. pa lx tells Lisrel that the pattern for the lx matrix follows; 0s mean the loading is not to be estimated and is set by default to zero; 1s mean the loading is free to be estimated (to any value). pd draws a path diagram for the user (very cool). out requests the appropriate outputs. me=ml says the estimation method is maximum likelihood (the default). rs produces residuals and mi produces modification indices. Need to check (e.g., cov) and pretty-up: Path Analysis Syntax Path Analysis.

da ni=5 no=100 ma=cm la qual price value cs intent cm sy 1.00 -0.03 1.00 0.58 -0.53 1.00 0.62 -0.22 0.39 1.00 0.39 -0.43

0.19 0.59 1.00 se value cs intent qual price mo ny=3 ne=3 nx=2 nk=2 ly=id,fi te=ze,fi lx=id,fi td=ze,fi ph=st be=fu,fr ga=fu,fr pa be 000 100 010 pa ga Need to check (e.g., cov) and pretty-up: pa ga 11

10 01 pd out me=ml rs mi Need to check (e.g., cov) and pretty-up: Commentary: the new lines include se which can serve two functions: first, it can be used to select a subset of the variables in the la list, and second, it can be used to rearrange the variables, to put the endogenous variables first, which is a rule required by Lisrel. In the model statement, the factor loadings matrices, lx and ly, are fixed to be identity matrices (=id,fi), and their corresponding error matrices, td and te, are fixed to be zero matrices (=ze,fi). The patterns for beta and gamma follow beta is ne x ne, and gamma is nk x ne. Need to check (e.g., cov) and pretty-up: Full SEM Syntax big example woohoo.

da ni=15 no=100 ma=cm la q1 q2 q3 c1 c2 c3 v1 v2 v3 cs1 cs2 cs3 r1 r2 r3 cm sy 2.813 2.264 1.973 1.961 1.670 1.528 etc. ... 1.530 se v1 v2 v3 cs1 cs2 cs3 r1 r2 r3 q1 q2 q3 c1 c2 c3 mo ny=9 ne=3 nx=6 nk=2 ly=fu,fr te=di,fr lx=fu,fr td=di,fr be=fu,fr ga=fu,fr pa ly 000 100

Need to check (e.g., cov) and pretty-up: mo ny=9 ne=3 nx=6 nk=2 ly=fu,fr te=di,fr lx=fu,fr td=di,fr be=fu,fr ga=fu,fr pa ly 000 100 100 000 010 010 000 001 001 pa lx 00

10 10 00 01 01 pa be 000 100 010 Need to check (e.g., cov) and pretty-up: pa be 000 100 010 pa ga

11 10 01 fi te(1) te(4) te(7) va 0.0 te(1) te(4) te(7) va 1.0 ly(1,1) ly(4,2) ly(7,3) pd out me=ml rs mi Need to check (e.g., cov) and pretty-up: Commentary: the new lines include fi which says fix (do not estimate) the parameters that follow. The lines va 0.0 and va 1.0 fix the theta epsilons to 0.0, and three lambda ys to 1.0. Need to check (e.g., cov) and pretty-up: Commentary: first line is a title, arbitrary but ends with a period. da stands for dataline, with number of input variables (ni=15), number of observations or sample size (no=100) and that the matrix to be

analyzed is a covariance matrix (ma=cm). la says that the variable labels follow on the next line. cm sy says that a covariance matrix follows, in symmetric format (lower triangle of the matrix). mo = model, nx = number of x variables, nk = number of ksi (constructs). lx is the lambda x (factor loadings matrix) which begins as a fu,fr full matrix (15x5) with all elements free to be estimated (this status is modified shortly). td is theta delta and all its diagonal elements are free to be estimated. ph is phi, the factor intercorrelation matrix, it is st standardized to be a correlation matrix. pa lx tells Lisrel that the pattern for the lx matrix follows; 0s mean the loading is not to be estimated and is set by default to zero; 1s mean the loading is free to be estimated (to any value). pd draws a path diagram for the user (very cool). out requests the appropriate outputs. me=ml says the estimation method is maximum likelihood (the default). rs produces residuals and mi produces modification indices.