1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
|
---
title: "Model based Imputation Methods"
author: Gregor de Cillia
output: rmarkdown::html_vignette
vignette: >
%\VignetteIndexEntry{Model based Imputation Methods}
%\VignetteEngine{knitr::rmarkdown}
%\VignetteEncoding{UTF-8}
---
```{r, include = FALSE}
knitr::opts_chunk$set(
collapse = TRUE,
comment = "#>",
fig.width = 6,
fig.align = "center"
)
```
This vignette showcases the functions `regressionImp()` and `rangerImpute()`,
which can both be used to generate imputations for several variables in a
dataset using a formula interface.
## Data
For data, a subset of `sleep` is used. The columns have been selected
deliberately to include some interactions between the missing values.
```{r setup, message = FALSE}
library(VIM)
library(magrittr)
dataset <- sleep[, c("Dream", "NonD", "BodyWgt", "Span")]
dataset$BodyWgt <- log(dataset$BodyWgt)
dataset$Span <- log(dataset$Span)
aggr(dataset)
str(dataset)
```
## Imputation
In order to invoke the imputation methods, a formula is used to specify which
variables are to be estimated and which variables should be used as regressors.
We will start by imputing `NonD` based in `BodyWgt` and `Span`.
```{r}
imp_regression <- regressionImp(NonD ~ BodyWgt + Span, dataset)
imp_ranger <- rangerImpute(NonD ~ BodyWgt + Span, dataset)
aggr(imp_regression, delimiter = "_imp")
```
We can see that for `regrssionImp()` there are still missings in `NonD` for all observations where
`Span` is unobserved. This is because the regression model could not be applied
to those observations. The same is true for the values imputed via
`rangerImpute()`.
## Diagnosing the results
As we can see in the next two plots, the correlation structure of `NonD` and
`BodyWgt` is preserved by both imputation methods. In the case of
`regressionImp()` all imputed values almost follow a straight line. This
suggests that the variable `Span` had little to no effect on the model.
```{r, fig.height=5}
imp_regression[, c("NonD", "BodyWgt", "NonD_imp")] %>%
marginplot(delimiter = "_imp")
```
For `rangerImpute()` on the other hand, `Span` played an important role in the
generation of the imputed values.
```{r, fig.height=5}
imp_ranger[, c("NonD", "BodyWgt", "NonD_imp")] %>%
marginplot(delimiter = "_imp")
imp_ranger[, c("NonD", "Span", "NonD_imp")] %>%
marginplot(delimiter = "_imp")
```
## Imputing multiple variables
To impute several variables at once, the formula in `rangerImpute()` and
`regressionImp()` can be specified with more than one column name in the
left hand side.
```{r}
imp_regression <- regressionImp(Dream + NonD ~ BodyWgt + Span, dataset)
imp_ranger <- rangerImpute(Dream + NonD ~ BodyWgt + Span, dataset)
aggr(imp_regression, delimiter = "_imp")
```
Again, there are missings left for both `Dream` and `NonD`.
## Performance of method
In order to validate the performance of `regressionImp()` the `iris` dataset is used. Firstly, some values are randomly set to `NA`.
```{r}
library(reactable)
data(iris)
df <- iris
colnames(df) <- c("S.Length","S.Width","P.Length","P.Width","Species")
# randomly produce some missing values in the data
set.seed(1)
nbr_missing <- 50
y <- data.frame(row=sample(nrow(iris),size = nbr_missing,replace = T),
col=sample(ncol(iris)-1,size = nbr_missing,replace = T))
y<-y[!duplicated(y),]
df[as.matrix(y)]<-NA
aggr(df)
sapply(df, function(x)sum(is.na(x)))
```
We can see that there are missings in all variables and some observations reveal missing values on several points. In the next step we perform a multiple variable imputation and `Species` serves as a regressor.
```{r}
imp_regression <- regressionImp(S.Length + S.Width + P.Length + P.Width ~ Species, df)
aggr(imp_regression, delimiter = "imp")
```
The plot indicates that all missing values have been imputed by the `regressionImp()` algorithm. The following table displays the rounded first five results of the imputation for all variables.
```{r echo=F,warning=F}
results <- cbind("TRUE1" = as.numeric(iris[as.matrix(y[which(y$col==1),])]),
"IMPUTED1" = round(as.numeric(imp_regression[as.matrix(y[which(y$col==1),])]),2),
"TRUE2" = as.numeric(iris[as.matrix(y[which(y$col==2),])]),
"IMPUTED2" = round(as.numeric(imp_regression[as.matrix(y[which(y$col==2),])]),2),
"TRUE3" = as.numeric(iris[as.matrix(y[which(y$col==3),])]),
"IMPUTED3" = round(as.numeric(imp_regression[as.matrix(y[which(y$col==3),])]),2),
"TRUE4" = as.numeric(iris[as.matrix(y[which(y$col==4),])]),
"IMPUTED4" = round(as.numeric(imp_regression[as.matrix(y[which(y$col==4),])]),2))[1:5,]
reactable(results, columns = list(
TRUE1 = colDef(name = "True"),
IMPUTED1 = colDef(name = "Imputed"),
TRUE2 = colDef(name = "True"),
IMPUTED2 = colDef(name = "Imputed"),
TRUE3 = colDef(name = "True"),
IMPUTED3 = colDef(name = "Imputed"),
TRUE4 = colDef(name = "True"),
IMPUTED4 = colDef(name = "Imputed")
),
columnGroups = list(
colGroup(name = "S.Length", columns = c("TRUE1", "IMPUTED1")),
colGroup(name = "S.Width", columns = c("TRUE2", "IMPUTED2")),
colGroup(name = "P.Length", columns = c("TRUE3", "IMPUTED3")),
colGroup(name = "P.Width", columns = c("TRUE4", "IMPUTED4"))
),
striped = TRUE,
highlight = TRUE,
bordered = TRUE
)
```
|
