parSim_example_psychonetrics_2

library("parSim")

parSim(
  ### SIMULATION CONDITIONS
  # Vary sample size:
  sampleSize = c(250, 500, 1000),
  
  # Vary missingness:
  missing = c(0, 0.1, 0.25),
  
  reps = 100, # 5 repititions
  write = TRUE, # Writing results to a file
  name = "missingdata_sims_2", # Name of the file
  nCores = 8, # I use 8 cores here
  expression = {
    ### SIMULATION CODE ###
    # Packages:
    library("psychonetrics")
    library("dplyr")
    library("bootnet")
    library("mice")
    
    # Function needed:
    source("compareNetworks.R")
    
    # Generate true network:
    trueNet <- genGGM(10, nei = 2, propPositive = 0.8, p = 0.25)
    
    # Generate data:
    generator <- ggmGenerator()
    Data <- generator(sampleSize, trueNet)
    
    # Add missings:
    for (i in 1:ncol(Data)){
      Data[runif(sampleSize) < missing,i] <- NA
    }
    
    ## Method 1: psychonetrics with FIML
    # Estimate model (FDR prune):
    mod <- ggm(Data, estimator = "FIML") %>% 
      runmodel %>% 
      prune(adjust = "fdr", alpha = 0.05, recursive = FALSE)
    
    # Extract network:
    estNet <- getmatrix(mod, "omega")
    
    # Compare to true network:
    results_psychonetrics <- compareNetworks(trueNet, estNet)
    
    # Add method:
    results_psychonetrics$method <- "psychonetrics_FIML"
    
    ## Method 2: ggmModSelect with mice imputation
    # Impute data using mice:
    miceres <- mice(Data, m = 1)
    imputedData <- complete(miceres)
    
    # Run ggmModSelect:
    net <- estimateNetwork(imputedData, default = "ggmModSelect")
    estNet <- net$graph
    
    # Compare to true network:
    results_ggmModSelect <- compareNetworks(trueNet, estNet)
    
    # Add method:
    results_ggmModSelect$method <- "ggmModSelect_mice"
    
    # Combine both in a data frame:
    results <- bind_rows(results_psychonetrics,results_ggmModSelect)
    
    # Return:
    results
  }
)