R/Performance.Dyad.Model.16.lag.R

In ginettelafit/PowerLAPIM: Power Analysis Longitudinal APIM with Quadratic Effects

Performance.Dyad.Model.16.lag = function(N.dyad,T.obs,  
c,rho.Y,a,a.2,p,p.2,d,d.a,d.a2,d.p,d.p2,
sigma.eps.F,sigma.eps.M,rho.eps.FM,
sigma.nu,
mu.XF,sigma.XF,mu.XM,sigma.XM,rho.X,
prob.D,is.center.X,alpha,is.REML){ 

# Generate data  
  
data = Sim.Dyad.Model.16.lag(N.dyad,T.obs,  
c,rho.Y,a,a.2,p,p.2,d,d.a,d.a2,d.p,d.p2,
sigma.eps.F,sigma.eps.M,rho.eps.FM,
sigma.nu,
mu.XF,sigma.XF,mu.XM,sigma.XM,rho.X,
prob.D,is.center.X)

# Fit multilevel model

if(is.center.X==TRUE){
# Person-centered the predictors
data <- data %>% 
group_by(subject.ID,dyad.ID) %>% 
mutate(X.Actor = X.Actor - mean(X.Actor),
       X.Partner = X.Partner - mean(X.Partner))
}

# Compute quadratic variables
data$X.Actor.2 = I(data$X.Actor^2)
data$X.Partner.2 = I(data$X.Partner^2)

# Compute interactions
data$X.Actor.D = data$X.Actor*data$D
data$X.Partner.D = data$X.Partner*data$D
data$X.Actor.D.2 = data$X.Actor.2*data$D
data$X.Partner.D.2 = data$X.Partner.2*data$D

# Model estimation

if (is.REML==TRUE){
fit.Model.16 = lme(fixed = Y ~ 1 + Y.lag + X.Actor + X.Actor.2 + X.Partner 
                  + X.Partner.2 + D +                         
                  X.Actor.D + X.Actor.D.2 + X.Partner.D + X.Partner.D.2,
                  random = list(dyad.ID = pdCompSymm(~ Gender -1)),
                  correlation = corCompSymm(form = ~1|dyad.ID/Obs),
                  weights = varIdent(form = ~1|Gender),
                  data = data, na.action=na.omit, 
                  method = 'REML',
                  control=list(msVerbose=TRUE, maxIter=500, msMaxIter=500)) 
}

if (is.REML==FALSE){
fit.Model.16 = lme(fixed = Y ~ 1 + Y.lag + X.Actor + X.Actor.2 + X.Partner 
                  + X.Partner.2 + D +                         
                  X.Actor.D + X.Actor.D.2 + X.Partner.D + X.Partner.D.2,
                  random = list(dyad.ID = pdCompSymm(~ Gender -1)),
                  correlation = corCompSymm(form = ~1|dyad.ID/Obs),
                  weights = varIdent(form = ~1|Gender),
                  data = data, na.action=na.omit, 
                  method = 'ML',
                  control=list(msVerbose=TRUE, maxIter=500, msMaxIter=500))
}

# Performance measures 

# Fixed Effects
# Estimated values
coef = coef(summary(fit.Model.16))[,1]

# Bias
bias = coef(summary(fit.Model.16))[,1] - c(c,rho.Y,a,a.2,p,p.2,d,d.a,d.a2,d.p,d.p2)

# Power
power = coef(summary(fit.Model.16))[,5]<alpha

# Coverage rate
CI = intervals(fit.Model.16, level = 1-alpha, which = "fixed")$fixed
CI.width = CI[,3] - CI[,1]
CR = CI[,1] < c(c,rho.Y,a,a.2,p,p.2,d,d.a,d.a2,d.p,d.p2) & CI[,3] > c(c,rho.Y,a,a.2,p,p.2,d,d.a,d.a2,d.p,d.p2)

summary.fixed.effect = list(coef=coef,bias=bias,power=power,CI.width=CI.width,CR=CR)

# Variance components
Sigma.hat = VarCorr(fit.Model.16)
Sigma.weight.hat = 1/unique(varWeights(fit.Model.16$modelStruct$varStruct))
sigma.eps.F.hat = as.numeric(Sigma.hat['Residual',2])*Sigma.weight.hat[1]
sigma.eps.M.hat = as.numeric(Sigma.hat['Residual',2])*Sigma.weight.hat[2]
rho.eps.FM.hat = as.numeric(coef(fit.Model.16$modelStruct$corStruct,unconstrained=FALSE)) 
sigma.nu.hat = as.numeric(Sigma.hat['GenderF',2])

sigma.eps.F.bias = sigma.eps.F.hat - sigma.eps.F
sigma.eps.M.bias = sigma.eps.M.hat - sigma.eps.M
rho.eps.FM.bias = rho.eps.FM.hat - rho.eps.FM 
sigma.nu.bias = sigma.nu.hat - sigma.nu

summary.var.hat = c(sigma.eps.F.hat=sigma.eps.F.hat,sigma.eps.M.hat=sigma.eps.M.hat,
rho.eps.FM.hat=rho.eps.FM.hat,sigma.nu.hat=sigma.nu.hat)

summary.var.bias = c(sigma.eps.F.bias=sigma.eps.F.bias,
sigma.eps.M.bias=sigma.eps.M.bias,rho.eps.FM.bias=rho.eps.FM.bias,
sigma.nu.bias=sigma.nu.bias)

return(list(summary.fixed.effect=summary.fixed.effect,summary.var.hat=summary.var.hat,summary.var.bias=summary.var.bias))
}