Test62: Finding case deletions required to change fit metric using approximate method

Description

Remove a fixed percentage of samples (determined by the approximate influence) at a time and refit the model to observe the change in the fit metric of interest. Use method 2 - Approximate Method.

Usage

Test62(
  df,
  model,
  threshold,
  fit,
  max_final,
  N,
  equalCons = 0L,
  measureTest = "cfi",
  fitThreshold = 0.9,
  highGood = TRUE,
  ...
)

Value

A list of class TestResult62 containing:

methodname: The name of the method used.
testindex: The index of the test performed.
original_fit_value: The original value of the fit measurement.
final_fit_value: The fit value after dropping the influential points using the approximate method.
num_drops: The number of data points dropped.
threshold_crossed: Logical indicating whether the threshold was crossed.
final_drops: The indices of the most influential data points dropped using the approximate method.
measureTest: The name of the fit measurement used.
fitThreshold: The threshold of the fit measurement used.
highGood: Logical indicating if a higher value is better for the fit measurement.
appx_threshold_tally: The tally value required to cross the threshold using the approximate method.
model_appx_threshold_final: The final fit value after dropping points sufficient to cross the threshold using the approximate method.
max_final: The maximum number of influential data points to consider.
N: The total number of data points.
equalCons: Logical indicating whether equality constraints exist in the model.

Arguments

df: A data frame containing the dataset.
model: A specified SEM model.
threshold: The threshold for percentage of data dropped.
fit: The SEM object.
max_final: The maximum number of influential data points to consider.
N: The total number of data points.
equalCons: Logical; whether equality constraints exist in the model. The approximate method can only be run if no equality constraints are present (default is 0).
measureTest: The fit measurement name. Can be "cfi", "chisq", "tli", "rmsea".
fitThreshold: The threshold of the fit measurement to be a "good" model. For example, for CFI (measureTest = "cfi"), this threshold can be 0.9.
highGood: A boolean argument stating if the fit measurement is higher the better. For CFI, this argument is TRUE.
...: Other arguments.

Examples

Run this code

# \donttest{
library(lavaan)
library(dplyr)
library(semfindr)
library(R.utils)


# Import data
df <- PoliticalDemocracy

# Build Model
model <- '
  # measurement model
  ind60 =~ x1 + x2 + x3
  dem60 =~ y1 + y2 + y3 + y4
  dem65 =~ y5 + y6 + y7 + y8
  # regressions
  dem60 ~ ind60
  dem65 ~ ind60 + dem60
  # residual correlations
  y1 ~~ y5
  y2 ~~ y4 + y6
  y3 ~~ y7
  y4 ~~ y8
  y6 ~~ y8
'

threshold <- 10

# Fit SEM model
fit <- lavaan::sem(model, data = df)
summary(fit)

# Compute max number of points to be dropped
max_final <- ceiling(threshold * nrow(df) / 100)
N <- nrow(df)

Test62_result <- Test62(df, model, threshold, fit, max_final, N, equalCons = 0L,
 measureTest = "cfi", fitThreshold = 0.9, highGood = TRUE)
summary(Test62_result)
# }

Run the code above in your browser using DataLab