imputation: Impute the missing data with the method selected under the condition.

Description

This function provides eight methods for imputation with categorical varaibles as conditions.

Usage

imputation(origdata, method, vartype = NULL, missingpct = NULL, condition = NULL,  knn = 5, mi.n = 3, mi.seed = 1234567, row_var = NULL)

Arguments

origdata

A data frame whose missing values need to be imputed. This data frame should be selected from the missing data GUI.

method

The imputation method selected from the missing data GUI. Must be one of 'Below 10 'MI:areg','MI:norm','MI:mice','MI:mi'. If method='MI:mice', then the methods of the variables containing NA's must be attached with argument method. If not, then default methods are used.

vartype

A vector of the classes of origdata. The length is the same as the number of columns of origdata. The value should be from "integer", "numeric", "logical", "character", "factor", and "ordered".

missingpct

A vector of the percentage of missings of the variables in origdata. The length is the same as the number of columns of origdata. The values should be between 0 and 1.

condition

A vector of categorical variables. The dataset will be partitioned based on those variables, and then the imputation is implemented in each group. There are no missing values in those variables. If it is null, then there is no division. The imputation is based on the whole dataset.

knn

number of the neighbors.

mi.n

number of the imputation sets for multiple imputation

mi.seed

random number seed for multiple imputation

row_var

A column name (character) that defines the ID of rows.

Value

The imputed data frame with the last column being the row number from the original dataset. During the procedure of the function, rows may be exchanged, thus a column of row number could keep track of the original row number and then help to find the shadow matrix.

Details

The imputation methods: This list displays all the imputation methods. Users can make one selection. (1) 'Below 10 NA's of one variable will be replaced by the value which equals to the minimum of the variable minus 10 For categorical variables, NA's are treated as a new category. Under this status the selected conditioning variables are ignored. If the data are already imputed, then this item will show the imputed result. (2) 'Simple' will create three tabs: Median, Mean, and Random Value. 'Median' means NA's will be replaced by the median of this variable (omit NA's). 'Mean' means NA's will be replaced by the mean of the variable (omit NA's). The median does not apply to the nominal variable, neither does the mean to the categorical variable. In these cases the mode (omit NA's) is provided. 'Random Value' means NA's will be replaced by any values of this variable (omit NA's) which are randomly selected. (3) 'Neighbor' contains two methods: 'Average Neighbor' and 'Random Neighbor'. 'Average Neighbor' will replace the NA's by the mean of the nearest neighbors. 'Random Neighbor' substitutes the missing for a random sample of the k nearest neighbors. The number of neighbors is default to 5, and can be changed by argument knn. The Neighbor methods require at lease one case to be complete, at least two variables to be selected, and no factor/character variables. The ordered factors are treated as integers. The method will return the overall mean or a global random sample value if the observation only contains NA's. (4) 'MI:areg' uses function aregImpute from package Hmisc. It requires at lease one case to be complete, and at least two variables to be selected. (5) 'MI:norm' uses function imp.norm from package norm. It requires all selected variables to be numeric(at least integer), and at least two variables to be selected. Sometimes it cannot converge, then the programme will leave NA's without imputation. (6) 'MI:mice' uses the mice package. The methods of the variables containing NA's must be attached with argument method. If not, then default methods are used. (7) 'MI:mi' employes the mi package.