In specifying a CFA model using the sem package, I got the following warning
message:
In sem.default(ram = ram, S = S, N = N, param.names = pars, var.names vars, :
Could not compute QR decomposition of Hessian.
Optimization probably did not converge.
This is the complete input (including data import):
----------------------------------------------------------------------------
-
DS
read.table("http://www.beltz.de/fileadmin/beltz/downloads/OnlinematerialienP
VU/Statistik_und_Forschungsmethoden/Daten_kap23.dat",
sep="", header = F)
attach(DS)
mycov = cov(DS)
my.model <- specify.model()
eta1 -> V1, NA, 1
eta1 -> V2, lam12, NA
eta1 -> V3, lam13, NA
eta1 -> V4, lam14, NA
eta1 -> V5, lam15, NA
eta1 -> V6, lam16, NA
eta2 -> V4, NA, 1
eta2 -> V5, lam52, NA
eta2 -> V6, lam62, NA
V1 <-> V1, e1, NA
V2 <-> V2, e2, NA
V3 <-> V3, e3, NA
V4 <-> V4, e4, NA
V5 <-> V5, e5, NA
V6 <-> V6, e6, NA
eta1 <-> eta1, var.eta1, NA
eta2 <-> eta2, var.eta2, NA
my.sem <- sem(my.model, mycov, nrow(DS), debug=TRUE)
----------------------------------------------------------------------------
-
The problem should converge easily and does so in Mplus. Also, it converges
if the correlation matrix instead of the variance-covariance matrix is used
and gives the correct standardized coefficients. Any suggestions why it does
not work with the covariance matrix in my example?
[[alternative HTML version deleted]]