MCAR Test

mcar_test_combined shows if a missing data mechanism is MCAR by performing three classical statistic tests: - the dummy variable test dummy_test, Little's MCAR test mcar_test from package naniar and - the normality tests (Hawkin's test and non-parametric test included) - TestMCARNormality from package MissMech. Our null hypothesis H0 is that the missing data mechanism is MCAR. These three tests focus on testing different aspects of MCAR mechanism. The dummy variable test is to verify if the value of one variable is correlated with the missingnness of another variable. Little's MCAR tet focuses on testing the if the mean vector for each missig pattern is part of the total mean vector. And the normality tests focuses on testing the normality and homoscedasticity according to the variance matrix. The conclusion of these three tests as well as their statistics will be returned.

mcar_test_combined(df, col_cat = c(), p_val = 0.1)

Arguments

df

Incomplete dataframe.

col_cat

Categorical columns index.

p_val

Significance level.

Value

test_results The final results from the three tests mentioned in 'Description'. TRUE means that we don't have enough evidence to reject our H0: the missing data mechanism is MCAR. And FALSE means that H0 is rejected, the missing data mechanism is not MCAR. p_values The p-values of our three tests. result_dummy_test The detailed result from dummy variable test. result_little_test The detailed result from Little's MCAR test. result_missmech The detailed result from the normality tests (Hawkin's test and non-parametric test).

Examples

n <- 10000
mu.X <- c(1, 2, 3)
Sigma.X <- matrix(c(9, 3, 2, 3, 4, 0, 2, 0, 1), nrow = 3)
X.complete.cont <- MASS::mvrnorm(n, mu.X, Sigma.X)
rs <- generate_miss(X.complete.cont, 0.5, mechanism = "MAR1")
mcar_test_combined(data.frame(rs$X.incomp))$test_result
#>   Dummy.variable.test Little.s.MCAR.test MissMech.test
#> 1               FALSE              FALSE          TRUE