Skip to contents

Simulate CFA model fit across sample sizes

Source: R/GLM_SEM.R
simulate_cfa_fit.Rd

Runs a confirmatory factor analysis (CFA) repeatedly across a range of sample sizes and returns fit indices for each. Useful for power analysis and sample size planning in SEM.

Two workflows are supported:

  • Coefficient-based: supply model_sim with fixed loadings - data are generated from those population parameters at each sample size.

  • Correlation-based: supply df - data are bootstrapped from the observed correlation structure of df at each sample size.

Iterations run in parallel via future.apply. Results are written to a PDF (scatter plots of fit indices vs. sample size) and an Excel file.

Usage

simulate_cfa_fit(
  model_sim = NULL,
  model = NULL,
  df = NULL,
  minnobs = 50,
  maxnobs = 1000,
  stepping = 10,
  file = NULL,
  w = 10,
  h = 10
)

Arguments

model_sim

A lavaan model string with fixed coefficients (e.g. "F =~ 1*x1 + 0.8*x2"). Used to generate population data. Mutually exclusive with df; supply one or the other.

model

A lavaan model string with free coefficients specifying the CFA structure to be estimated at each sample size (e.g. "F =~ x1 + x2").

df

A data frame of observed data. When supplied, each iteration simulates data by sampling from the observed correlation structure. Mutually exclusive with model_sim.

minnobs

Integer. Smallest sample size to evaluate. Default 50.

maxnobs

Integer. Largest sample size to evaluate. Default 1000.

stepping

Integer. Increment between sample sizes. Default 10.

file

Character. Base name (without extension) for the output PDF and Excel files. If NULL no files are written.

w

Numeric. Width of the output PDF in inches. Default 10.

h

Numeric. Height of the output PDF in inches. Default 10.

Value

A list of two elements:

  • [[1]] - a data frame with one row per sample size and one column per lavaan fit index (CFI, RMSEA, SRMR, etc.).

  • [[2]] - a named list of ggplot scatter plots, one per fit index, showing how the index changes with sample size.

Examples

model_sim <- 'LATENT =~ 1*X1 + 0.5*X2 + 1.5*X3 + 1.5*X4 + X5'
model     <- 'LATENT =~ X1 + X2 + X3 + X4 + X5'

# Coefficient-based: generate data from known population parameters
result<-simulate_cfa_fit(model_sim=model_sim, model=model,
                         minnobs=50, maxnobs=1000, stepping=100, file="report")











































plot_multiplot(plotlist=result[[2]], cols=4)

# Correlation-based: resample from observed data
df <- lavaan::simulateData(model=model_sim, model.type="cfa",
                           return.type="data.frame", sample.nobs=1000)
result<-simulate_cfa_fit(model=model, df=df,
                         minnobs=50, maxnobs=10000, stepping=100, file="report")











































plot_multiplot(plotlist=result[[2]], cols=4)