MAAR_mechanism_noindep: MAAR missing data mechanism
In bojinov/diagMAAR: Diagnostic tests for missing always at random

Description Usage Arguments Details Value See Also Examples

MAAR_mechanism takes a matrix and deletes some of the entries using a missing always at random mechanism that allows the missingness in one variable to depend on a partially observed variable.

1 2	MAAR_mechanism_noindep(samples, miss.coef, miss.nvar, miss.var = NULL, prob.coef, seed = NULL)

`samples`	A numeric matrix with no missing values.
`miss.nvar`	An integer specifying the number of variables that will have missing values. miss.nvar < ncol(samples).
`miss.var`	An integer vector of length equal to miss.nvar specifying which variables will have missing values. If left blank a random sample will be taken from the columns of samples will be taken.
`prob.coef`	A numeric matrix that represents the regression coefficients that will be used to generate the missing data pattern. The nrow(miss.coef) == miss.var and ncol(miss.coef) == ncol(samples).
`seed`	A numeric value used to set a seed, leave blank for NULL.
`miss.coef:`	A numeric value or a vector of length equal to miss.nvar. The missingness coefficient(s) determines the proportion of missing values for each variable.
`self.dep`	A logical values indicating if the probability of variable j missing depends on the values that variable j took.

MAAR_mechanism is part of the data generating functions used when conducting simulation studies involving missing data.The probability that a variable is missing is only a function of the observed variables. For example, assume we have 3 variables, Y_1 and Y_2 can have missing values and Y_3 is always fully observed. Let R_1 and R_2 be the response indicators for Y_1 and Y_2 respectively Then the probability that the i-th variable is missing is

p(R_{i,1}=1|Y_{i,\cdot}) = \text{logit}^{-1}(α_1+β_1Y_{i,3}).

p(R_{i,2}=1|Y_{i,\cdot},R_{i,1}) = \text{logit}^{-1}(α_2+ γ_1R_{i,1}Y_{i,1}+β_2Y_{i,3}).

The vector β is specified using the prob.coef parameter, and α is selected to ensure that the proportion of missing values in each variable is equal to the miss.coef parameter.

samples.obs A data matrix of dimensions equal to samples with some missing values.

Other data generating functions: MAAR_mechanism, MANAR_mechanism, logistic_prob, sample_mvn

# Generate 100 iid samples from a MVN with correlation equal to 0.3
samples.mvn <- sample_mvn(5, 0.3, 100)
# Take the Gaussian data and and delete some values from the fourth row.
obs.mvn <- MAAR_mechanism_noindep(samples = samples.mvn, miss.coef = 0.2, 
                                  miss.nvar = 1, miss.var = 1, 
                                  prob.coef = matrix(c( - 1, 0.5, 0.7, 
                                                       - 0.2), 1, 4))