MAP: Missing data spike-in in MAP pattern

In missCompare: Intuitive Missing Data Imputation Framework

Description

MAP spikes in missingness using missing-at-assumed (MAP) pattern

Usage

MAP(
  X_hat,
  MD_pattern,
  NA_fraction,
  min_PDM = 10,
  assumed_pattern = c("MAR", "MCAR", "MCAR", "MAR", "MNAR", "MCAR", "MCAR", "MAR",
    "MNAR", "MCAR", "MCAR")
)

Arguments

X_hat	Simulated matrix with no missingness (Simulated_matrix output from the simulate function)
MD_pattern	Missing data pattern in the original dataset (MD_Pattern output from the get_data function)
NA_fraction	Fraction of missingness in the original dataset (Fraction_missingness output from the get_data function)
min_PDM	All patterns with number of observations less than this number will be removed from the missing data generation. This argument is necessary to be carefully set, as the function will fail or generate erroneous missing data patterns with very complicated missing data patterns. The default is 10, but for large datasets this number needs to be set higher to avoid errors. Please select a value based on the min_PDM_thresholds output from the get_data function
assumed_pattern	Vector of missingness types (must be same length as missingness fraction per variable)

Details

This function uses the generated simulated matrix and generates missing datapoints in a missing-at-assumed pattern for each variable using the ampute function, considering the fraction of missingness in the original dataset and the original missingness pattern. In the MAP function, the user needs to define a character vector (of length the same as the fraction the number of columns in the dataset) that specifies which missingness pattern corresponds to the variables. In case the first four columns are assumed missing at random, the next one missing completely at random and the last two column not at random, the input vector will be: c(rep('MAR', 4), 'MCAR', rep('MNAR',2)) The algorithm will spike in missing values according to the specified pattern. Please note that after the missing data spike-in, the function will remove rows with 100% missing data.

Value

MAP_matrix	Matrix with MAP pre-defined missingness pattern
Summary	Summary of MAP_matrix including number of missing values per variable

Examples

cleaned <- clean(clindata_miss, missingness_coding = -9)
metadata <- get_data(cleaned)
simulated <- simulate(rownum = metadata$Rows, colnum = metadata$Columns,
cormat = metadata$Corr_matrix)

MAP(simulated$Simulated_matrix,
    MD_pattern = metadata$MD_Pattern,
    NA_fraction = metadata$Fraction_missingness,
    min_PDM = 10,
    assumed_pattern = c('MAR', 'MCAR', 'MCAR', 'MAR', 'MNAR', 'MCAR',
                        'MAR', 'MCAR', 'MCAR', 'MAR', 'MNAR'))

missCompare documentation built on Dec. 1, 2020, 9:09 a.m.