SimSem monte carlo simulation for sample size of a new tested sem model

[Gb]

Hello everyone, I am having some troubles in defing the correct code for calculating the sample size for a new model that I built. I attached the hypothesized model. I have a few questions: Is it the code that I am using the correct one for my aim? 2. Should I include the residual variances and covariances? If yes, how can I set them since I don't have any references for it? I used the studies that I cited to support my hypotheses to set both factor loadings and coefficients, but I don't have any info about the residual covariances and variances. Should I set them according to standard value? Thank you very much for your help!

Here above the syntax that I was thinking to use:

#I set the  factor loadings according to literature

popModel <- "
#measurament model
sti =~ 0.49*s1 + 0.72*s2 + 0.88*s3
su =~ 0.6*su1 + 0.6*su2 + 0.6*su3
pc =~ 0.85*p1 + 0.87*p2 + 0.79*p3
ic =~ 0.65*i1 + 0.63*i2 + 0.83*i3
att =~ 0.84*a1 + 0.86*a2 + 0.86*a3
sn =~ 0.87*sn1 + 0.91*sn2 + 0.89*sn3
pbc =~ 0.83*pb1 + 0.92*pb2 + 0.80*pb3 + 0.85*pb4
int =~ 0.90*in1 + 0.90*in2 + 0.91*in3
pb =~ 1*pb1

#structural model


att ~ -0.67*sti + 0.19*su + 0.25*pc + 0.25*ic
sn ~ 0.20*su
pbc ~ 0.19*su
int ~ 0.30*att + 0.28*sn + 0.37*pbc + 0.18*pb


"
analysis.mod  <- '
# define latent variables
sti =~ s1 + s2 + s3
su =~ su1 + su2 + su3
pc =~ p1 + p2 + p3
ic =~ i1 + i2 + i3
att =~ a1 + a2 + a3
sn =~ sn1 + sn2 + sn3
pbc =~ pb1 + pb2 + pb3 + pb4
int =~ in1 + in2 + in3
pb =~ pb1
                 

                 
# specify structural model

att ~ b8*sti + b1*su + b10*pc + b11*ic
sn ~ b3*su
pbc ~ b5*su
int ~ b2*att + b4*sn + b6*pbc + b7*pb + b12*su
                 
# indirect effects
su.att.int := b1*b2
su.sn.int := b3*b4
su.pbc.int := b5*b6
                 
# total effects
                 
att.int := b12 + b1*b2
sn.int := b12 +b3*b4
pbc.int := b12 + b5*b6'

# A) run simulation (varying sample sizes)
med.n <- simsem::sim(n=rep(seq(100,500,20),100), # varying sample sizes (each 100x)
                     model=analysis.mod,
                     generate=pop.mod,
                     lavaanfun="sem")


summary(med.n)
med.n <- sim(5, n=100, model = analysis.mod, generate =  pop.mod, lavaanfun = "sem", seed=1234)


?seq
# B) The varying parameter value that provides the power just over the specified power value
# (the adjacent value of varying parameter provides lower power than the specified power value).
powTable2 <- getPower(med.n)
findPower(powTable2, "N", 0.85)

# su.att.int --> N1
# su.sn.int  --> N2
# su.pbc.int  --> N3



# C) power curves
plotPower(med.n, powerParam = "su.att.int")
abline(h = 0.85, lwd = 2, lty = 2)

plotPower(med.n, powerParam = "su.sn.int ")
abline(h = 0.85, lwd = 2, lty = 2)

plotPower(med.n, powerParam = "su.pbc.int")
abline(h = 0.85, lwd = 2, lty = 2)




# D) 2 simulations with n=264 (10.000 repetitions) with two different seeds
med.n.neu1 <- simsem::sim(nRep=10000, model=analysis.mod, n=264, generate=pop.mod, lavaanfun="sem", seed=9999)
med.n.neu2 <- simsem::sim(nRep=10000, model=analysis.mod, n=264, generate=pop.mod, lavaanfun="sem", seed=99)

round(summaryParam(med.n.neu1, alpha=0.05, detail=TRUE), digits=3)
round(summaryParam(med.n.neu2, alpha=0.05), digits=3)

# in both cases: power for indirect effects >= .849

[Terrence Jorgensen]

You could use the simstandard package, or you could use the built-in simsem functionality (e.g., see RPS= argument in ?model help page).  You can find examples in the attached slides.

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

http://www.uva.nl/profile/t.d.jorgensen