1 Pendahuluan

Dokumen ini memuat analisis perbandingan metode Ordinary Least Squares (OLS) dan Robust MM-Estimator pada model regresi linear. Data Gallstone digunakan sebagai data sekunder untuk menentukan karakteristik variabel dan parameter acuan simulasi. Selanjutnya dilakukan simulasi Monte Carlo dengan beberapa skenario persentase pencilan.

2 Persiapan

rm(list = ls())
set.seed(123)

library(readr)
library(robustbase)
library(ggplot2)

3 Import Data

Gallstone_dataset_uci <- read_csv2(
  "C:/SEMESTER 4/Gallstone dataset-uci.csv"
)

# Melihat nama seluruh variabel
names(Gallstone_dataset_uci)
##  [1] "Gallstone Status"                              
##  [2] "Age"                                           
##  [3] "Gender"                                        
##  [4] "Comorbidity"                                   
##  [5] "Coronary Artery Disease (CAD)"                 
##  [6] "Hypothyroidism"                                
##  [7] "Hyperlipidemia"                                
##  [8] "Diabetes Mellitus (DM)"                        
##  [9] "Height"                                        
## [10] "Weight"                                        
## [11] "Body Mass Index (BMI)"                         
## [12] "Total Body Water (TBW)"                        
## [13] "Extracellular Water (ECW)"                     
## [14] "Intracellular Water (ICW)"                     
## [15] "Extracellular Fluid/Total Body Water (ECF/TBW)"
## [16] "Total Body Fat Ratio (TBFR) (%)"               
## [17] "Lean Mass (LM) (%)"                            
## [18] "Body Protein Content (Protein) (%)"            
## [19] "Visceral Fat Rating (VFR)"                     
## [20] "Bone Mass (BM)"                                
## [21] "Muscle Mass (MM)"                              
## [22] "Obesity (%)"                                   
## [23] "Total Fat Content (TFC)"                       
## [24] "Visceral Fat Area (VFA)"                       
## [25] "Visceral Muscle Area (VMA) (Kg)"               
## [26] "Hepatic Fat Accumulation (HFA)"                
## [27] "Glucose"                                       
## [28] "Total Cholesterol (TC)"                        
## [29] "Low Density Lipoprotein (LDL)"                 
## [30] "High Density Lipoprotein (HDL)"                
## [31] "Triglyceride"                                  
## [32] "Aspartat Aminotransferaz (AST)"                
## [33] "Alanin Aminotransferaz (ALT)"                  
## [34] "Alkaline Phosphatase (ALP)"                    
## [35] "Creatinine"                                    
## [36] "Glomerular Filtration Rate (GFR)"              
## [37] "C-Reactive Protein (CRP)"                      
## [38] "Hemoglobin (HGB)"                              
## [39] "Vitamin D"
# Melihat struktur dataset
str(Gallstone_dataset_uci)
## spc_tbl_ [319 × 39] (S3: spec_tbl_df/tbl_df/tbl/data.frame)
##  $ Gallstone Status                              : num [1:319] 0 0 0 0 0 0 0 0 0 0 ...
##  $ Age                                           : num [1:319] 50 47 61 41 42 96 37 41 38 38 ...
##  $ Gender                                        : num [1:319] 0 0 0 0 0 0 0 0 0 0 ...
##  $ Comorbidity                                   : num [1:319] 0 1 0 0 0 0 0 0 0 0 ...
##  $ Coronary Artery Disease (CAD)                 : num [1:319] 0 0 0 0 0 0 0 0 0 0 ...
##  $ Hypothyroidism                                : num [1:319] 0 0 0 0 0 0 0 0 0 0 ...
##  $ Hyperlipidemia                                : num [1:319] 0 0 0 0 0 0 0 0 0 0 ...
##  $ Diabetes Mellitus (DM)                        : num [1:319] 0 0 0 0 0 0 0 0 0 0 ...
##  $ Height                                        : num [1:319] 185 176 171 168 178 155 185 176 186 171 ...
##  $ Weight                                        : num [1:319] 92.8 94.5 91.1 67.7 89.6 49 67.1 114 93.7 68.6 ...
##  $ Body Mass Index (BMI)                         : num [1:319] 27.1 30.5 31.2 24 28.3 20.4 19.6 36.8 27.1 23.5 ...
##  $ Total Body Water (TBW)                        : num [1:319] 52.9 43.1 47.2 41.4 51.4 34 41.8 57.6 51.9 39.5 ...
##  $ Extracellular Water (ECW)                     : num [1:319] 21.2 19.5 20.1 17 20 15.7 17.1 23.5 20.9 16.6 ...
##  $ Intracellular Water (ICW)                     : num [1:319] 31.7 23.6 27.1 24.4 31.4 18.3 24.7 34.1 31 22.9 ...
##  $ Extracellular Fluid/Total Body Water (ECF/TBW): num [1:319] 40 45 43 41 39 46 41 41 40 42 ...
##  $ Total Body Fat Ratio (TBFR) (%)               : num [1:319] 19.2 32.8 27.3 15.8 20 6.3 11.3 31.8 21.5 19.2 ...
##  $ Lean Mass (LM) (%)                            : num [1:319] 80.8 67.2 72.7 84.2 80 ...
##  $ Body Protein Content (Protein) (%)            : num [1:319] 18.9 16.7 16.4 16.9 16.8 ...
##  $ Visceral Fat Rating (VFR)                     : num [1:319] 9 15 15 6 8 12 3 16 8 6 ...
##  $ Bone Mass (BM)                                : num [1:319] 3.7 3.2 3.3 2.9 3.5 2.4 3 3.8 3.6 2.8 ...
##  $ Muscle Mass (MM)                              : num [1:319] 71.4 60.3 62.9 54.1 68.2 43.5 56.5 73.9 70 52.6 ...
##  $ Obesity (%)                                   : num [1:319] 23.4 38.8 41.7 9 28.6 7.4 10.9 67.4 23.1 6.7 ...
##  $ Total Fat Content (TFC)                       : num [1:319] 17.8 31 24.9 10.7 17.9 3.1 7.6 36.3 20.1 13.2 ...
##  $ Visceral Fat Area (VFA)                       : num [1:319] 10.6 18.4 16.2 6.5 10.4 0.9 4.6 23.7 12.7 8.2 ...
##  $ Visceral Muscle Area (VMA) (Kg)               : num [1:319] 39.7 32.7 34 29.2 37.4 25.9 31.4 37.5 37.8 28.8 ...
##  $ Hepatic Fat Accumulation (HFA)                : num [1:319] 0 0 0 1 2 0 0 3 2 0 ...
##  $ Glucose                                       : num [1:319] 102 94 103 69 109 74 94 125 93 93 ...
##  $ Total Cholesterol (TC)                        : num [1:319] 250 172 179 173 205 133 166 246 290 239 ...
##  $ Low Density Lipoprotein (LDL)                 : num [1:319] 175 108 124 73 154 ...
##  $ High Density Lipoprotein (HDL)                : num [1:319] 40 43 43 59 30 30 40 34 33 43 ...
##  $ Triglyceride                                  : num [1:319] 134 103 69 53 326 65 60 464 679 129 ...
##  $ Aspartat Aminotransferaz (AST)                : num [1:319] 20 14 18 20 27 13 15 26 68 19 ...
##  $ Alanin Aminotransferaz (ALT)                  : num [1:319] 22 13 14 12 54 13 14 28 102 34 ...
##  $ Alkaline Phosphatase (ALP)                    : num [1:319] 87 46 66 34 71 60 72 69 70 75 ...
##  $ Creatinine                                    : num [1:319] 0.82 0.87 1.25 1.02 0.82 1.46 0.77 1.3 0.91 0.91 ...
##  $ Glomerular Filtration Rate (GFR)              : num [1:319] 112.5 107.1 65.5 94.1 112.5 ...
##  $ C-Reactive Protein (CRP)                      : num [1:319] 0 0 0 0 0 0 0 0.11 1.57 0 ...
##  $ Hemoglobin (HGB)                              : num [1:319] 16 14.4 16.2 15.4 16.8 11 13.8 16.5 16.5 16.6 ...
##  $ Vitamin D                                     : num [1:319] 33 25 30.2 35.4 40.6 45.8 20 24.5 22.7 15.6 ...
##  - attr(*, "spec")=
##   .. cols(
##   ..   `Gallstone Status` = col_double(),
##   ..   Age = col_double(),
##   ..   Gender = col_double(),
##   ..   Comorbidity = col_double(),
##   ..   `Coronary Artery Disease (CAD)` = col_double(),
##   ..   Hypothyroidism = col_double(),
##   ..   Hyperlipidemia = col_double(),
##   ..   `Diabetes Mellitus (DM)` = col_double(),
##   ..   Height = col_double(),
##   ..   Weight = col_double(),
##   ..   `Body Mass Index (BMI)` = col_double(),
##   ..   `Total Body Water (TBW)` = col_double(),
##   ..   `Extracellular Water (ECW)` = col_double(),
##   ..   `Intracellular Water (ICW)` = col_double(),
##   ..   `Extracellular Fluid/Total Body Water (ECF/TBW)` = col_double(),
##   ..   `Total Body Fat Ratio (TBFR) (%)` = col_double(),
##   ..   `Lean Mass (LM) (%)` = col_double(),
##   ..   `Body Protein Content (Protein) (%)` = col_double(),
##   ..   `Visceral Fat Rating (VFR)` = col_double(),
##   ..   `Bone Mass (BM)` = col_double(),
##   ..   `Muscle Mass (MM)` = col_double(),
##   ..   `Obesity (%)` = col_double(),
##   ..   `Total Fat Content (TFC)` = col_double(),
##   ..   `Visceral Fat Area (VFA)` = col_double(),
##   ..   `Visceral Muscle Area (VMA) (Kg)` = col_double(),
##   ..   `Hepatic Fat Accumulation (HFA)` = col_double(),
##   ..   Glucose = col_double(),
##   ..   `Total Cholesterol (TC)` = col_double(),
##   ..   `Low Density Lipoprotein (LDL)` = col_double(),
##   ..   `High Density Lipoprotein (HDL)` = col_double(),
##   ..   Triglyceride = col_double(),
##   ..   `Aspartat Aminotransferaz (AST)` = col_double(),
##   ..   `Alanin Aminotransferaz (ALT)` = col_double(),
##   ..   `Alkaline Phosphatase (ALP)` = col_double(),
##   ..   Creatinine = col_double(),
##   ..   `Glomerular Filtration Rate (GFR)` = col_double(),
##   ..   `C-Reactive Protein (CRP)` = col_double(),
##   ..   `Hemoglobin (HGB)` = col_double(),
##   ..   `Vitamin D` = col_double()
##   .. )
##  - attr(*, "problems")=<externalptr>
# Melihat 10 baris pertama
head(Gallstone_dataset_uci, 10)

4 Memilih Variabel Penelitian

Variabel respons yang digunakan adalah Triglyceride, sedangkan variabel prediktornya adalah Age, Body Mass Index (BMI), Glucose, dan Total Cholesterol.

data_set <- data.frame(
  Y = Gallstone_dataset_uci$Triglyceride,
  X1 = Gallstone_dataset_uci$Age,
  X2 = Gallstone_dataset_uci$`Body Mass Index (BMI)`,
  X3 = Gallstone_dataset_uci$Glucose,
  X4 = Gallstone_dataset_uci$`Total Cholesterol (TC)`
)

head(data_set, 10)
# Melihat struktur data
str(data_set)
## 'data.frame':    319 obs. of  5 variables:
##  $ Y : num  134 103 69 53 326 65 60 464 679 129 ...
##  $ X1: num  50 47 61 41 42 96 37 41 38 38 ...
##  $ X2: num  27.1 30.5 31.2 24 28.3 20.4 19.6 36.8 27.1 23.5 ...
##  $ X3: num  102 94 103 69 109 74 94 125 93 93 ...
##  $ X4: num  250 172 179 173 205 133 166 246 290 239 ...
# Melihat jumlah observasi dan variabel
dim(data_set)
## [1] 319   5

5 Analisis Deskriptif

# Rata-rata
colMeans(data_set)
##         Y        X1        X2        X3        X4 
## 144.50216  48.06897  28.87712 108.68871 203.49530
# Standar deviasi
sapply(data_set, sd)
##         Y        X1        X2        X3        X4 
## 97.904493 12.114558  5.313707 44.848714 45.758501
# Nilai minimum
sapply(data_set, min)
##     Y    X1    X2    X3    X4 
##  1.39 20.00 17.40 69.00 60.00
# Nilai maksimum
sapply(data_set, max)
##     Y    X1    X2    X3    X4 
## 838.0  96.0  49.7 575.0 360.0
# Median
sapply(data_set, median)
##     Y    X1    X2    X3    X4 
## 119.0  49.0  28.3  98.0 198.0
# Korelasi antarvariabel
cor(data_set)
##             Y         X1          X2         X3          X4
## Y  1.00000000 0.05804325 0.173438727 0.34037768 0.326705964
## X1 0.05804325 1.00000000 0.235835013 0.14850426 0.159004883
## X2 0.17343873 0.23583501 1.000000000 0.12536774 0.004509984
## X3 0.34037768 0.14850426 0.125367744 1.00000000 0.083041839
## X4 0.32670596 0.15900488 0.004509984 0.08304184 1.000000000

6 Menentukan Parameter Simulasi

Model regresi linear awal diestimasi menggunakan data penelitian. Koefisien hasil estimasi digunakan sebagai parameter acuan dalam proses simulasi.

model_awal <- lm(
  Y ~ X1 + X2 + X3 + X4,
  data = data_set
)

summary(model_awal)
## 
## Call:
## lm(formula = Y ~ X1 + X2 + X3 + X4, data = data_set)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -282.51  -46.49  -22.97   29.68  528.71 
## 
## Coefficients:
##              Estimate Std. Error t value Pr(>|t|)    
## (Intercept) -116.0800    36.2438  -3.203  0.00150 ** 
## X1            -0.5870     0.4205  -1.396  0.16370    
## X2             2.7776     0.9449   2.939  0.00353 ** 
## X3             0.6687     0.1102   6.071 3.67e-09 ***
## X4             0.6678     0.1077   6.199 1.78e-09 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 86.56 on 314 degrees of freedom
## Multiple R-squared:  0.2282, Adjusted R-squared:  0.2183 
## F-statistic: 23.21 on 4 and 314 DF,  p-value: < 2.2e-16
# Parameter acuan simulasi
beta0 <- coef(model_awal)[1]
beta1 <- coef(model_awal)[2]
beta2 <- coef(model_awal)[3]
beta3 <- coef(model_awal)[4]
beta4 <- coef(model_awal)[5]

# Standar deviasi error
sigma_e <- sd(residuals(model_awal))

# Menyusun parameter
parameter <- c(
  beta0 = beta0,
  beta1 = beta1,
  beta2 = beta2,
  beta3 = beta3,
  beta4 = beta4
)

parameter
## beta0.(Intercept)          beta1.X1          beta2.X2          beta3.X3 
##      -116.0800356        -0.5870008         2.7776473         0.6687474 
##          beta4.X4 
##         0.6678434
sigma_e
## [1] 86.0123

7 Simulasi Data

Variabel prediktor dibangkitkan dari distribusi normal dengan rata-rata dan standar deviasi masing-masing variabel pada data penelitian. Variabel respons dibentuk berdasarkan parameter acuan dan komponen error acak.

nindividu <- nrow(data_set)

# Membuat variabel bebas X
X1 <- rnorm(
  nindividu,
  mean = mean(data_set$X1),
  sd = sd(data_set$X1)
)

X2 <- rnorm(
  nindividu,
  mean = mean(data_set$X2),
  sd = sd(data_set$X2)
)

X3 <- rnorm(
  nindividu,
  mean = mean(data_set$X3),
  sd = sd(data_set$X3)
)

X4 <- rnorm(
  nindividu,
  mean = mean(data_set$X4),
  sd = sd(data_set$X4)
)

# Membuat error
epsilon <- rnorm(nindividu, 0, sigma_e)

# Variabel respons Y
Y <- beta0 +
  beta1 * X1 +
  beta2 * X2 +
  beta3 * X3 +
  beta4 * X4 +
  epsilon

data_set_simulasi <- data.frame(
  Y = Y,
  X1 = X1,
  X2 = X2,
  X3 = X3,
  X4 = X4
)

head(data_set_simulasi, 10)

8 Pemodelan OLS dan Robust MM-Estimator

# Model OLS
model_ols <- lm(
  Y ~ X1 + X2 + X3 + X4,
  data = data_set_simulasi
)

summary(model_ols)
## 
## Call:
## lm(formula = Y ~ X1 + X2 + X3 + X4, data = data_set_simulasi)
## 
## Residuals:
##      Min       1Q   Median       3Q      Max 
## -224.665  -50.856    2.709   56.172  267.348 
## 
## Coefficients:
##             Estimate Std. Error t value Pr(>|t|)    
## (Intercept) -43.7244    41.9774  -1.042   0.2984    
## X1           -0.8715     0.3973  -2.194   0.0290 *  
## X2            1.8769     0.8726   2.151   0.0323 *  
## X3            0.7915     0.1008   7.853 6.49e-14 ***
## X4            0.4460     0.1006   4.433 1.29e-05 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 81.64 on 314 degrees of freedom
## Multiple R-squared:  0.228,  Adjusted R-squared:  0.2182 
## F-statistic: 23.18 on 4 and 314 DF,  p-value: < 2.2e-16
# Model Robust MM-Estimator
model_mm <- lmrob(
  Y ~ X1 + X2 + X3 + X4,
  data = data_set_simulasi,
  method = "MM"
)

summary(model_mm)
## 
## Call:
## lmrob(formula = Y ~ X1 + X2 + X3 + X4, data = data_set_simulasi, method = "MM")
##  \--> method = "MM"
## Residuals:
##      Min       1Q   Median       3Q      Max 
## -226.454  -51.670    2.055   54.776  268.998 
## 
## Coefficients:
##             Estimate Std. Error t value Pr(>|t|)    
## (Intercept) -44.0576    45.8851  -0.960   0.3377    
## X1           -0.8644     0.4043  -2.138   0.0333 *  
## X2            1.7378     0.8795   1.976   0.0490 *  
## X3            0.8074     0.1015   7.956 3.26e-14 ***
## X4            0.4604     0.1039   4.432 1.29e-05 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Robust residual standard error: 80.55 
## Multiple R-squared:  0.2331, Adjusted R-squared:  0.2233 
## Convergence in 10 IRWLS iterations
## 
## Robustness weights: 
##  24 weights are ~= 1. The remaining 295 ones are summarized as
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##  0.2419  0.8701  0.9541  0.9072  0.9845  0.9989 
## Algorithmic parameters: 
##        tuning.chi                bb        tuning.psi        refine.tol 
##         1.548e+00         5.000e-01         4.685e+00         1.000e-07 
##           rel.tol         scale.tol         solve.tol          zero.tol 
##         1.000e-07         1.000e-10         1.000e-07         1.000e-10 
##       eps.outlier             eps.x warn.limit.reject warn.limit.meanrw 
##         3.135e-04         6.352e-10         5.000e-01         5.000e-01 
##      nResample         max.it       best.r.s       k.fast.s          k.max 
##            500             50              2              1            200 
##    maxit.scale      trace.lev            mts     compute.rd fast.s.large.n 
##            200              0           1000              0           2000 
##                   psi           subsampling                   cov 
##            "bisquare"         "nonsingular"         ".vcov.avar1" 
## compute.outlier.stats 
##                  "SM" 
## seed : int(0)
# Parameter OLS
parameter_ols <- coef(model_ols)
parameter_ols
## (Intercept)          X1          X2          X3          X4 
## -43.7243947  -0.8714623   1.8768792   0.7914869   0.4460443
# Parameter MM-Estimator
parameter_mm <- coef(model_mm)
parameter_mm
## (Intercept)          X1          X2          X3          X4 
## -44.0575996  -0.8644209   1.7377901   0.8073651   0.4604435

9 Parameter dan Perbandingan Hasil Estimasi

parameter <- c(
  beta0 = beta0,
  beta1 = beta1,
  beta2 = beta2,
  beta3 = beta3,
  beta4 = beta4
)

# Parameter simulasi OLS
parameter_simulasi_ols <- coef(model_ols)

# Parameter simulasi MM-Estimator
parameter_simulasi_mm <- coef(model_mm)

# Hasil simulasi
hasil_simulasi <- data.frame(
  parameter = names(parameter),
  Asli = as.numeric(parameter),
  OLS = as.numeric(parameter_simulasi_ols),
  MM_Estimator = as.numeric(parameter_simulasi_mm)
)

hasil_simulasi
# Error estimasi
error_ols <- hasil_simulasi$Asli - hasil_simulasi$OLS
error_mm <- hasil_simulasi$Asli - hasil_simulasi$MM_Estimator

error_ols
## [1] -72.3556408   0.2844615   0.9007681  -0.1227396   0.2217990
error_mm
## [1] -72.0224359   0.2774200   1.0398572  -0.1386178   0.2073999

10 Validasi Model

Grafik berikut memperlihatkan hubungan variabel respons dengan Age dan BMI pada data simulasi, beserta garis regresi linear.

ggplot(data_set_simulasi, aes(x = X1, y = Y)) +
  geom_point(alpha = 0.5) +
  geom_smooth(method = "lm", se = FALSE, color = "black") +
  theme_minimal() +
  labs(x = "Age", y = "Triglyceride")

ggplot(data_set_simulasi, aes(x = X2, y = Y)) +
  geom_point(alpha = 0.5) +
  geom_smooth(method = "lm", se = FALSE, color = "black") +
  theme_minimal() +
  labs(x = "BMI", y = "Triglyceride")

11 Membentuk Skenario Pencilan

Pencilan ditambahkan pada variabel respons (pencilan vertikal) sebesar delapan kali standar deviasi error. Bagian ini membentuk contoh data dengan pencilan sebesar 5%.

persentase <- c(0, 0.05, 0.10, 0.15)

# Membuat salinan data simulasi
data_pencilan <- data_set_simulasi

# Menentukan jumlah pencilan untuk contoh 5%
jumlah_pencilan <- round(nindividu * 0.05)

# Memilih observasi yang menjadi pencilan
indeks_pencilan <- sample(
  1:nindividu,
  jumlah_pencilan
)

# Menambahkan pencilan pada variabel Y
data_pencilan$Y[indeks_pencilan] <-
  data_pencilan$Y[indeks_pencilan] + 8 * sigma_e

head(data_pencilan, 10)

12 Pemodelan Data dengan Pencilan

# Model OLS
model_ols_pencilan <- lm(
  Y ~ X1 + X2 + X3 + X4,
  data = data_pencilan
)

summary(model_ols_pencilan)
## 
## Call:
## lm(formula = Y ~ X1 + X2 + X3 + X4, data = data_pencilan)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -285.66  -86.16  -24.65   33.89  759.96 
## 
## Coefficients:
##               Estimate Std. Error t value Pr(>|t|)    
## (Intercept) -95.610496  90.411058  -1.058 0.291092    
## X1            0.002541   0.855689   0.003 0.997633    
## X2            0.603888   1.879469   0.321 0.748191    
## X3            0.818473   0.217078   3.770 0.000195 ***
## X4            0.826043   0.216714   3.812 0.000166 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 175.8 on 314 degrees of freedom
## Multiple R-squared:  0.08826,    Adjusted R-squared:  0.07665 
## F-statistic: 7.599 on 4 and 314 DF,  p-value: 7.439e-06
# Model Robust MM-Estimator
model_mm_pencilan <- lmrob(
  Y ~ X1 + X2 + X3 + X4,
  data = data_pencilan,
  method = "MM"
)

summary(model_mm_pencilan)
## 
## Call:
## lmrob(formula = Y ~ X1 + X2 + X3 + X4, data = data_pencilan, method = "MM")
##  \--> method = "MM"
## Residuals:
##      Min       1Q   Median       3Q      Max 
## -222.464  -48.773    8.329   62.888  834.060 
## 
## Coefficients:
##             Estimate Std. Error t value Pr(>|t|)    
## (Intercept) -33.8992    46.1344  -0.735 0.463015    
## X1           -0.8106     0.4159  -1.949 0.052193 .  
## X2            1.8084     0.8806   2.054 0.040828 *  
## X3            0.8031     0.1009   7.958 3.21e-14 ***
## X4            0.3827     0.1053   3.635 0.000324 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Robust residual standard error: 86.79 
## Multiple R-squared:  0.2116, Adjusted R-squared:  0.2016 
## Convergence in 11 IRWLS iterations
## 
## Robustness weights: 
##  16 observations c(34,51,61,90,113,116,124,125,139,162,166,176,182,190,293,308)
##   are outliers with |weight| = 0 ( < 0.00031); 
##  25 weights are ~= 1. The remaining 278 ones are summarized as
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##  0.3438  0.8872  0.9582  0.9182  0.9872  0.9989 
## Algorithmic parameters: 
##        tuning.chi                bb        tuning.psi        refine.tol 
##         1.548e+00         5.000e-01         4.685e+00         1.000e-07 
##           rel.tol         scale.tol         solve.tol          zero.tol 
##         1.000e-07         1.000e-10         1.000e-07         1.000e-10 
##       eps.outlier             eps.x warn.limit.reject warn.limit.meanrw 
##         3.135e-04         6.352e-10         5.000e-01         5.000e-01 
##      nResample         max.it       best.r.s       k.fast.s          k.max 
##            500             50              2              1            200 
##    maxit.scale      trace.lev            mts     compute.rd fast.s.large.n 
##            200              0           1000              0           2000 
##                   psi           subsampling                   cov 
##            "bisquare"         "nonsingular"         ".vcov.avar1" 
## compute.outlier.stats 
##                  "SM" 
## seed : int(0)

13 Simulasi Monte Carlo

Simulasi dilakukan sebanyak 500 replikasi untuk setiap skenario pencilan, yaitu 0%, 5%, 10%, dan 15%. Pada setiap replikasi, model OLS dan MM-Estimator diestimasi pada data yang dibangkitkan.

# Jumlah replikasi
nreplikasi <- 500

# Skenario pencilan
skenario <- c(0, 0.05, 0.10, 0.15)

# Menyiapkan hasil simulasi
hasil_replikasi <- data.frame()

# Melakukan simulasi
for (s in skenario) {

  for (i in 1:nreplikasi) {

    # Membuat variabel bebas X
    X1 <- rnorm(
      nindividu,
      mean = mean(data_set$X1),
      sd = sd(data_set$X1)
    )

    X2 <- rnorm(
      nindividu,
      mean = mean(data_set$X2),
      sd = sd(data_set$X2)
    )

    X3 <- rnorm(
      nindividu,
      mean = mean(data_set$X3),
      sd = sd(data_set$X3)
    )

    X4 <- rnorm(
      nindividu,
      mean = mean(data_set$X4),
      sd = sd(data_set$X4)
    )

    # Membuat error
    epsilon <- rnorm(nindividu, 0, sigma_e)

    # Variabel respons Y
    Y <- beta0 +
      beta1 * X1 +
      beta2 * X2 +
      beta3 * X3 +
      beta4 * X4 +
      epsilon

    data_set_simulasi <- data.frame(
      Y = Y,
      X1 = X1,
      X2 = X2,
      X3 = X3,
      X4 = X4
    )

    # Menentukan jumlah pencilan
    jumlah_pencilan <- round(nindividu * s)

    # Menambahkan pencilan
    if (jumlah_pencilan > 0) {

      indeks_pencilan <- sample(
        1:nindividu,
        jumlah_pencilan
      )

      data_set_simulasi$Y[indeks_pencilan] <-
        data_set_simulasi$Y[indeks_pencilan] +
        8 * sigma_e
    }

    # Model OLS
    model_ols <- lm(
      Y ~ X1 + X2 + X3 + X4,
      data = data_set_simulasi
    )

    # Model Robust MM-Estimator
    model_mm <- lmrob(
      Y ~ X1 + X2 + X3 + X4,
      data = data_set_simulasi,
      method = "MM"
    )

    # Parameter acuan
    parameter <- c(
      beta0 = beta0,
      beta1 = beta1,
      beta2 = beta2,
      beta3 = beta3,
      beta4 = beta4
    )

    # Parameter estimasi OLS dan MM
    parameter_ols <- coef(model_ols)
    parameter_mm <- coef(model_mm)

    # Hasil simulasi OLS
    hasil_ols <- data.frame(
      replikasi = i,
      skenario = s * 100,
      metode = "OLS",
      parameter = names(parameter),
      Asli = as.numeric(parameter),
      Estimasi = as.numeric(parameter_ols)
    )

    # Hasil simulasi MM-Estimator
    hasil_mm <- data.frame(
      replikasi = i,
      skenario = s * 100,
      metode = "MM-Estimator",
      parameter = names(parameter),
      Asli = as.numeric(parameter),
      Estimasi = as.numeric(parameter_mm)
    )

    # Menggabungkan hasil simulasi
    hasil_replikasi <- rbind(
      hasil_replikasi,
      hasil_ols,
      hasil_mm
    )
  }
}

# Melihat hasil simulasi
head(hasil_replikasi, 10)
# Melihat dimensi hasil simulasi
dim(hasil_replikasi)
## [1] 20000     6

14 Menghitung Bias, MSE, dan RMSE

Bias dihitung sebagai rata-rata selisih estimasi dengan parameter acuan. Mean Squared Error (MSE) merupakan rata-rata kuadrat selisih, sedangkan Root Mean Squared Error (RMSE) adalah akar kuadrat dari MSE.

# Menghitung error
hasil_replikasi$error <-
  hasil_replikasi$Estimasi -
  hasil_replikasi$Asli

# Menghitung kuadrat error
hasil_replikasi$error2 <-
  hasil_replikasi$error^2

# Menghitung Bias
hasil_bias <- aggregate(
  error ~ skenario + metode + parameter,
  data = hasil_replikasi,
  FUN = mean
)

names(hasil_bias)[4] <- "Bias"
hasil_bias
# Menghitung MSE
hasil_mse <- aggregate(
  error2 ~ skenario + metode + parameter,
  data = hasil_replikasi,
  FUN = mean
)

names(hasil_mse)[4] <- "MSE"
hasil_mse
# Menghitung RMSE
hasil_mse$RMSE <- sqrt(hasil_mse$MSE)

# Menggabungkan hasil evaluasi
hasil_evaluasi <- merge(
  hasil_bias,
  hasil_mse,
  by = c("skenario", "metode", "parameter")
)

hasil_evaluasi

15 Visualisasi Hasil Simulasi

Grafik berikut membandingkan Bias, MSE, dan RMSE OLS dengan MM-Estimator pada setiap skenario persentase pencilan.

ggplot(hasil_evaluasi, aes(
  x = factor(skenario),
  y = Bias,
  fill = metode
)) +
  geom_col(position = "dodge") +
  facet_wrap(~parameter, scales = "free_y") +
  theme_minimal() +
  labs(
    x = "Persentase Pencilan (%)",
    y = "Bias",
    title = "Perbandingan Bias OLS dan MM-Estimator"
  )

ggplot(hasil_evaluasi, aes(
  x = factor(skenario),
  y = MSE,
  fill = metode
)) +
  geom_col(position = "dodge") +
  facet_wrap(~parameter, scales = "free_y") +
  theme_minimal() +
  labs(
    x = "Persentase Pencilan (%)",
    y = "MSE",
    title = "Perbandingan MSE OLS dan MM-Estimator"
  )

ggplot(hasil_evaluasi, aes(
  x = factor(skenario),
  y = RMSE,
  fill = metode
)) +
  geom_col(position = "dodge") +
  facet_wrap(~parameter, scales = "free_y") +
  theme_minimal() +
  labs(
    x = "Persentase Pencilan (%)",
    y = "RMSE",
    title = "Perbandingan RMSE OLS dan MM-Estimator"
  )

16 Catatan

Pastikan lokasi berkas CSV pada bagian Import Data disesuaikan dengan lokasi penyimpanan di komputer yang digunakan. Parameter hasil regresi awal pada dokumen ini berperan sebagai parameter acuan simulasi.