Regression Imputation - Deterministic & Stochastic

This is the simulation to show the difference in using the technique of regression imputation.

Deterministic: This method assumes that the imputed values appear close to the regression line.

Stochastic: It is a improvement to the previous method by aiming to preserve the variability of data by adding an error(residual) factor to the predicted value.

install.packages("mice")

WARNING: Rtools is required to build R packages but is not currently installed. Please download and install the appropriate version of Rtools before proceeding:

https://cran.rstudio.com/bin/windows/Rtools/
Installing package into 㤼㸱C:/Users/anilp/OneDrive/Documents/R/win-library/3.6㤼㸲
(as 㤼㸱lib㤼㸲 is unspecified)
trying URL 'https://cran.rstudio.com/bin/windows/contrib/3.6/mice_3.6.0.zip'
Content type 'application/zip' length 1855619 bytes (1.8 MB)
downloaded 1.8 MB

package ‘mice’ successfully unpacked and MD5 sums checked

The downloaded binary packages are in
    C:\Users\anilp\AppData\Local\Temp\RtmpuYzw6r\downloaded_packages

# Install and load the R package mice
 
install.packages("mice") # Needs to be done only once

WARNING: Rtools is required to build R packages but is not currently installed. Please download and install the appropriate version of Rtools before proceeding:

https://cran.rstudio.com/bin/windows/Rtools/
Installing package into 㤼㸱C:/Users/anilp/OneDrive/Documents/R/win-library/3.6㤼㸲
(as 㤼㸱lib㤼㸲 is unspecified)
trying URL 'https://cran.rstudio.com/bin/windows/contrib/3.6/mice_3.6.0.zip'
Content type 'application/zip' length 1855619 bytes (1.8 MB)
downloaded 1.8 MB

package ‘mice’ successfully unpacked and MD5 sums checked

The downloaded binary packages are in
    C:\Users\anilp\AppData\Local\Temp\RtmpuYzw6r\downloaded_packages

library("mice") # Load package

Loading required package: lattice

Attaching package: 㤼㸱mice㤼㸲

The following objects are masked from 㤼㸱package:base㤼㸲:

    cbind, rbind

MyData <- read.csv(file="C:/Development/diabetes_test.csv", header=TRUE, sep=";")
data <- data.frame(MyData)

# Deterministic regression imputation

imp <- mice(data, method = "norm.predict", m = 1) # Impute data

 iter imp variable
  1   1  Glucose  BloodPressure  SkinThickness  Insulin  BMI
  2   1  Glucose  BloodPressure  SkinThickness  Insulin  BMI
  3   1  Glucose  BloodPressure  SkinThickness  Insulin  BMI
  4   1  Glucose  BloodPressure  SkinThickness  Insulin  BMI
  5   1  Glucose  BloodPressure  SkinThickness  Insulin  BMI

data_det <- complete(imp) # Store data

# Stochastic regression imputation
 
imp <- mice(data, method = "norm.nob", m = 1) # Impute data

 iter imp variable
  1   1  Glucose  BloodPressure  SkinThickness  Insulin  BMI
  2   1  Glucose  BloodPressure  SkinThickness  Insulin  BMI
  3   1  Glucose  BloodPressure  SkinThickness  Insulin  BMI
  4   1  Glucose  BloodPressure  SkinThickness  Insulin  BMI
  5   1  Glucose  BloodPressure  SkinThickness  Insulin  BMI

data_sto <- complete(imp) # Store data


# Graphical comparison of deterministic and stochastic regression imputation
 
par(mfrow = c(1, 2)) # Both plots in one graphic
 
# Deterministic regression imputation plot
plot(data$Glucose[!is.na(data$Insulin)], data_det$Insulin[!is.na(data$Insulin)],# Plot of observed values
     xlim = c(40, 200), ylim = c(10, 900),
     main = "Deterministic Regression",
     xlab = "Glucose", ylab = "Insulin")
points(data$Glucose[is.na(data$Insulin)], data_det$Insulin[is.na(data$Insulin)],# Plot of missing values
       col = "red")

abline(lm(data$Insulin ~ data$Glucose, data_det), col = "#1b98e0", lwd = 1.5) # Regression slope


# Stochastic regression imputation plot
plot(data$Glucose[!is.na(data$Insulin)], data_sto$Insulin[!is.na(data$Insulin)],# Plot of observed values
     xlim = c(40, 200), ylim = c(10, 900),
     main = "Stochastic Regression",
     xlab = "Glucose", ylab = "Insulin")

points(data$Glucose[is.na(data$Insulin)], data_sto$Insulin[is.na(data$Insulin)],# Plot of missing values
       col = "red")
abline(lm(data$Insulin ~ data$Glucose, data_sto), col = "#1b98e0", lwd = 1.5) # Regression slope