R Graphical Manual

Browse All

Last data update: 2014.03.03

R: Conduct a Correlation/Covariance Structure Analysis with WLS

wls	R Documentation

Conduct a Correlation/Covariance Structure Analysis with WLS

Description

It fits a correlation or covariance structure with weighted least squares (WLS) estimation method where the inverse of the asymptotic covariance matrix is used as the weight matrix. tssem2 conducts the second stage analysis of the two-stage strutural equation modeling (TSSEM). tssem2 is a wrapper of wls.

Usage

wls(Cov, asyCov, n, Amatrix=NULL, Smatrix=NULL, Fmatrix=NULL,
    diag.constraints=FALSE, cor.analysis=TRUE, intervals.type=c("z","LB"),
    mx.algebras=NULL, model.name=NULL, suppressWarnings=TRUE,
    silent=TRUE, run=TRUE, ...)
tssem2(tssem1.obj, Amatrix=NULL, Smatrix=NULL, Fmatrix=NULL,
    diag.constraints=FALSE, intervals.type=c("z", "LB"), mx.algebras=NULL,
    model.name=NULL, suppressWarnings=TRUE, silent=TRUE, run=TRUE, ...)

Arguments

`tssem1.obj`	An object of either class `tssem1FEM`, class `tssem1FEM.cluster` or class `tssem1REM` returned from `tssem1()`
`Cov`	A p x p sample correlation/covariance matrix where p is the number of variables.
`asyCov`	A p* x p* asymptotic sampling covariance matrix of either `vechs` `(Cov)` or `vech` `(Cov)` where p = p(p-1)/2* for correlation matrix and p = p(p+1)/2* for covariance matrix.
`n`	Sample size.
`Amatrix`	An asymmetric matrix in the RAM approach with `MxMatrix-class`. If it is `NULL`, a matrix of zero will be created. If it is a matrix, it will be converted into `MxMatrix-class` by the `as.mxMatrix` function.
`Smatrix`	A symmetric matrix in the RAM approach with `MxMatrix-class`. If it is a matrix, it will be converted into `MxMatrix-class` by the `as.mxMatrix` function.
`Fmatrix`	A filter matrix in the RAM approach with `MxMatrix-class`. If it is `NULL` (the default), an identity matrix with the same dimensions of `Cov` will be created. If it is a matrix, it will be converted into `MxMatrix-class` by the `as.mxMatrix` function. It is not required when there is no latent variable.
`diag.constraints`	Logical. This argument is ignored when `cor.analysis=FALSE`. If `diag.constraints=TRUE`, the diagonals of the model implied matrix will be constrained at 1 by nonlinear constraints. The drawback is that standard error will not be generated. Parametric bootstrap is used to estimate the standard error by drawing samples from N(vech(Cov), asyCov) for covariance analysis and N(vechs(Cov), asyCov) for correlation analysis while asyCov is treated as fixed. This process is computational intensive. A better approach is to request likelihood-based confidence intervals (CIs) by specifying `intervals.type="LB"`. If `diag.constraints=FALSE` and `cor.analysis=TRUE`, the diagonals are automatically constrained as ones by treating the error variances as computed values rather than as parameters. Since the error variances are not parameters, they are not reported.
`cor.analysis`	Logical. Analysis of correlation or covariance structure. If `cor.analysis=TRUE`, `vechs` is used to vectorize `S`; otherwise, `vech` is used to vectorize `S`.
`intervals.type`	Either `z` (default if missing) or `LB`. If it is `z`, it calculates the 95% Wald CIs based on the z statistic. If it is `LB`, it calculates the 95% likelihood-based CIs on the parameter estimates. Please note that the z values and their associated p values are based on the z statistic. They are not related to the likelihood-based CIs.
`mx.algebras`	A list of `mxMatrix` or `mxAlgebra` objects on the `Amatrix`, `Smatrix` and `Fmatrx`. It can be used to define new functions of parameters and their LBCIs. For example, if the regression coefficients to calculate an indirect effect are stored in A[1,2] and A[1,3], we may define `list(ind=mxAlgebra(Amatrix[1,2]*Amatrix[1,3], name="ind"))` See the examples in `Becker92` and `Hunter83`. It should be noted that Fmatrix, Amatrix, Smatrix, Iden (a p x p identity matrix), sampleS (sample correlation or covariance matrix), impliedS1, impliedS (model implied correlation or covariance matrix), vecS, invAcov, obj, One, select and constraint and Ematrix (computed error variances when `diag.constraints=FALSE`) have been defined internally. You should not create new matrices using these names.
`model.name`	A string for the model name in `mxModel`. If it is missing, the default is "TSSEM2 (or WLS) Analysis of Correlation Structure" for `cor.analysis=TRUE` and "TSSEM2 (or WLS) Analysis of Covariance Structure" for `cor.analysis=FALSE`.
`suppressWarnings`	Logical. If `TRUE`, warnings are suppressed. Argument to be passed to `mxRun`.
`silent`	Logical. Argument to be passed to `mxRun`
`run`	Logical. If `FALSE`, only return the mx model without running the analysis.
`...`	Futher arguments to be passed to `mxRun`.

Value

An object of class wls with a list of

`call`	The matched call
`Cov`	Input data of either a covariance or correlation matrix
`asyCov`	Asymptotic covariance matrix of the input data
`noObservedStat`	Number of observed statistics
`n`	Sample size
`cor.analysis`	logical
`noConstraints`	Number of constraints imposed on S
`indepModelChisq`	Chi-square statistic of the independent model returned by `.indepwlsChisq`
`indepModelDf`	Degrees of freedom of the independent model returned by `.indepwlsChisq`
`mx.fit`	A fitted object returned from `mxRun`

Note

If the input is a list of tssem1.obj, it returns a list of results for each cluster.

Author(s)

Mike W.-L. Cheung <mikewlcheung@nus.edu.sg>

References

Bentler, P.M., & Savalei, V. (2010). Analysis of correlation structures: current status and open problems. In Kolenikov, S., Thombs, L., & Steinley, D. (Eds.). Recent Methodological Developments in Social Science Statistics (pp. 1-36). Hoboken, NJ: Wiley.

Cheung, M. W.-L. (2010). Fixed-effects meta-analyses as multiple-group structural equation models. Structural Equation Modeling, 17, 481-509.

Cheung, M. W.-L. (2014). Fixed- and random-effects meta-analytic structural equation modeling: Examples and analyses in R. Behavior Research Methods, 46, 29-40.

Cheung, M. W.-L., & Chan, W. (2005). Meta-analytic structural equation modeling: A two-stage approach. Psychological Methods, 10, 40-64.

Cheung, M. W.-L., & Chan, W. (2009). A two-stage approach to synthesizing covariance matrices in meta-analytic structural equation modeling. Structural Equation Modeling, 16, 28-53.

Joreskog, K. G., Sorbom, D., Du Toit, S., & Du Toit, M. (1999). LISREL 8: New Statistical Features. Chicago: Scientific Software International.

McArdle, J. J., & MacDonald, R. P. (1984). Some algebraic properties of the Reticular Action Model for moment structures. British Journal of Mathematical and Statistical Psychology, 37, 234-251.

Examples

#### Analysis of correlation structure
R1 <- matrix(c(1.00, 0.22, 0.24, 0.18,
               0.22, 1.00, 0.30, 0.22,
               0.24, 0.30, 1.00, 0.24,
               0.18, 0.22, 0.24, 1.00), ncol=4, nrow=4)
n <- 1000
acovR1 <- asyCov(R1, n)

## One-factor CFA model
(A1 <- cbind(matrix(0, nrow=5, ncol=4),
             matrix(c("0.2*a1","0.2*a2","0.2*a3","0.2*a4",0),
             ncol=1)))

(S1 <- Diag(c("0.2*e1","0.2*e2","0.2*e3","0.2*e4",1)))

## The first 4 variables are observed while the last one is latent.
(F1 <- create.Fmatrix(c(1,1,1,1,0), name="F1"))
wls.fit1 <- wls(Cov=R1, asyCov=acovR1, n=n, Fmatrix=F1, Smatrix=S1, Amatrix=A1,
                 cor.analysis=TRUE, intervals="LB")
summary(wls.fit1)


#### Multiple regression analysis
## Variables in R2: y, x1, x2
R2 <- matrix(c(1.00, 0.22, 0.24, 
               0.22, 1.00, 0.30, 
               0.24, 0.30, 1.00, 
               0.18, 0.22, 0.24), ncol=3, nrow=3)
acovR2 <- asyCov(R2, n)

## A2: Regression coefficents
#    y x1 x2
# y  F T  T 
# x1 F F  F 
# x2 F F  F 
(A2 <- mxMatrix("Full", ncol=3, nrow=3, byrow=TRUE,
               free=c(FALSE, rep(TRUE, 2), rep(FALSE, 6)), name="A2"))

## S2: Covariance matrix of free parameters
#    y x1 x2
# y  T F  F 
# x1 F F  F 
# x2 F T  F
(S2 <- mxMatrix("Symm", ncol=3, nrow=3, values=c(0.2,0,0,1,0.2,1),
               free=c(TRUE,FALSE,FALSE,FALSE,TRUE,FALSE), name="S2"))

## F may be ignored as there is no latent variable.
wls.fit2 <- wls(Cov=R2, asyCov=acovR2, n=n, Amatrix=A2, Smatrix=S2,
                cor.analysis=TRUE, intervals="LB")
summary(wls.fit2)


#### Analysis of covariance structure
R3 <- matrix(c(1.50, 0.22, 0.24, 0.18,
               0.22, 1.60, 0.30, 0.22,
               0.24, 0.30, 1.80, 0.24,
               0.18, 0.22, 0.24, 1.30), ncol=4, nrow=4)
n <- 1000
acovS3 <- asyCov(R3, n, cor.analysis=FALSE)

(A3 <- cbind(matrix(0, nrow=5, ncol=4),
             matrix(c("0.2*a1","0.2*a2","0.2*a3","0.2*a4",0),ncol=1)))

(S3 <- Diag(c("0.2*e1","0.2*e2","0.2*e3","0.2*e4",1)))

F3 <- c(TRUE,TRUE,TRUE,TRUE,FALSE)
(F3 <- create.Fmatrix(F3, name="F3", as.mxMatrix=FALSE))

wls.fit3 <- wls(Cov=R3, asyCov=acovS3, n=n, Amatrix=A3, Smatrix=S3,
                Fmatrix=F3, cor.analysis=FALSE)
summary(wls.fit3)