Advances

2022; 3(3): 140-152

http://www.sciencepublishinggroup.com/j/advances
doi: 10.11648/j.advances.20220303.24

Detecting of Multicollinearity, Autocorrelation and

Heteroscedasticity in Regression Analysis
Abeer Mohamed Abd El Razek Youssef
Faculty of Graduate Studies for Statistical Research, Cairo University, Giza, Egypt

Email address:

To cite this article:

Abeer Mohamed Abd El Razek Youssef. Detecting of Multicollinearity, Autocorrelation and Heteroscedasticity in Regression Analysis.
Advances. Vol. 3, No. 3, 2022, pp. 140-152. doi: 10.11648/j.advances.20220303.24

Received: August 15, 2022; Accepted: September 1, 2022; Published: September 29, 2022

Abstract: When we rely on the general linear regression model to represent the data, we use the ordinary least squares
method to estimate the parameters of this model. This method, when applied, depends on the fulfillment of certain basic
assumptions and conditions so that there is an accuracy in estimating the parameters of the regression model, and in many
practical applications this hypothesis cannot be achieved, which makes the method of least squares ineffective in giving correct
and accurate results, and this leads to falling into econometric problems. The estimated parameters lose the property of
credibility, unbiased and make them not have the lowest possible variance and not expressive of the original theory. Most
econometric models suffer from the problems of autocorrelation, multicollinearity, and heteroscedasticity. This paper presents
a brief on these problems, their causes, how can be detected, tested, and minimized. The OLS method is based on several
assumptions, and if these assumptions are fulfilled, we obtain unbiased, consistent, and efficient estimates (less variance
compared to other methods). We discuss these problems as follows: First: the problem of multicollinearity Second: The
problem of autocorrelation Third: Variation Heteroscedasticity. This article presents inference for many commonly used
estimators - Variance Inflation Factors, Coefficient covariance matrix, Correlogram of Residuals, Normality Test for
Residuals. Serial correlation LM test, Heteroskedasticity Test: Harvey, Actual and Estimated Residuals.
Keywords: Multicollinearity, Autocorrelation, Heteroscedasticity

1. Introduction
2. Material and Methods
There is no doubt that the assumptions of the linear
regression model may or may not be available. If available, 2.1. Autocorrelation and Detection Tests
the ordinary least squares method is valid for use in Causes of Autocorrelation
measuring the economic relations under study. But if it is not 1) Deleting some explanatory variables from the regression
available, the method of ordinary squares does not become model results in the so-called deletion error, which in
the appropriate method for estimating the parameters of turn is reflected in the values of the random term.
economic relations, and these results in the emergence of 2) Misidentification of the mathematical form of the model.
some econometric problems that make this method an For example, if the real relationship of a dependent
inappropriate method, and it is necessary to search in this variable is nonlinear, but the researcher has used a linear
case for other more appropriate standard methods. formula. Hence, without a doubt, the use of the linear
I will present some standard problems encountered in the formula instead of the non-linear one involves a certain
search: type of error and is reflected in the random term [1].
1. The problem of autocorrelation Below are a few examples of some nonlinear formulas.
2. Variation homoskedasticity problem
3. The problem of lack of normal distribution Yi=A+BXi2+Ui
4. The problem of multicollinearity
Yi2=C + D(1/X) Ui
141 Abeer Mohamed Abd El Razek Youssef: Detecting Of Multicollinearity, Autocorrelation and
Heteroscedasticity in Regression Analysis

Yi=F XiM Ui Log Yi=Log F+m Log X+ Log Ui

As: A, B, C, D and F are constants whose value is 3-Data processing. In some cases, the published data may
estimated in the respective model. These formulas indicate be monthly, and the researcher wants data on a quarterly
that there is a nonlinear relationship between Y and the basis, so he collects it and obtains an average of it. Perhaps it
explanatory variable X in the three formulas. However, it is will provide fewer fluctuating data, which involves a kind of
noted that redefining the variable X2 in the form numbered error that will be repeated from one observation to another
(1), as if we put X2=W, converts the original nonlinear because of the approximation process, which leads to the
relationship to a linear relationship: existence of autocorrelation [2].
Y=A+BWi+U i 2.2. Autocorrelation Tests
The use of the mathematical logarithmic transformation 1. Durban Watson Test
transforms the relationship with the number (3) into a linear 2. Durban h test
relationship as well: 3. Breusch-Godfrey serial correlation LM test
2.2.1. Steps of Residual Series

Source: Prepared by the researcher based on the statistical program EViews 10th Edition
Figure 1. Steps of Residual Series.

2.2.2. Using the Durban Watson Statistic

Source: Prepared by the researcher based on the statistical program EViews 10th Edition
Figure 2. Durban Watson.
 Advances 2022; 3(3): 140-152 142

2.3. Breusch-Godfrey Serial Correlation LM Test f-statistic 2.52 is smaller than the tabular one, which
means accepting the null hypothesis and rejecting the
To perform the Breusch-Godfrey test we have two alternative hypothesis, i.e., rejecting the existence of
possibilities [3]: autocorrelation.
As for the classical f test, we note that the calculated

Figure 3. Steps of Breusch-Godfrey test.

Source: Prepared by the researcher based on the statistical program EViews 10th Edition
Figure 4. Steps serial correlation LM test.
143 Abeer Mohamed Abd El Razek Youssef: Detecting Of Multicollinearity, Autocorrelation and
Heteroscedasticity in Regression Analysis

But if we take the second case, the calculated Obs*R In the practical aspect of conducting this test, we follow the
squared statistic of 5.41 is greater than the tabular statistic following steps [6]:
which has a chi-square distribution, rejecting the null The first stage: calculating the error term t in the
hypothesis and accepting the alternative hypothesis that is, regression model.
the existence of autocorrelation [4]. The second stage: calculating et2.
The third stage: autoregression of residuals through p
2.4. Inconsistency of Variance (Heterogeneity of Variance) periods of deceleration. Significant delays are preserved.
ARCH test: Test the variance hypothesis, using an Fourth stage: calculating the LaGrange multiplier statistic,
autoregressive conditional variance test. We judge the results, where LM=N*R2, with N sample size and R2 representing the
whether the possibility of accepting the null hypothesis coefficient of determination [7].
which states that the variance of the random error term in the
estimated model is constant, or the rejection of the null 3. Steps of Heteroscedasticity Test
hypothesis and the acceptance of the alternative hypothesis
“absence of the homo of the variance.” [5]. The result decides to accept the null hypothesis with the
We conduct the ARCH Test to test the relationship constant of the variance and the rejection of the inconstant of
between the square of residuals as a dependent variable and the variance for the error term series. This is the confirmation
the square of the slowed residuals for one period to test the of the validity of the second hypothesis of the least squares
null hypothesis saying constant of variance. This test is based method, which states constant of variance [8].
on either the classic Fisher test or the LaGrange multiple tests.

Source: Prepared by the researcher based on the statistical program EViews 10th Edition
Figure 5. Steps of heteroscedasticity test.
 Advances 2022; 3(3): 140-152 144

3.1. Steps of Heteroscedasticity Test

Source: Prepared by the researcher based on the statistical program EViews 10th Edition
Figure 6. Calculation method of heteroscedasticity LM Test.

3.2. Test to Verify The Normalization of Residuals of the Regression Equation (Jarque-Bera Test)

For checking the normal distribution of residuals of the regression equation we use the Jarque-Bera test.
The null hypothesis that the residuals of the regression equation are normally distributed based on the statistic of this choice
can be rejected or accepted [9].
We reject the null hypothesis if the JB statistic value is greater than the tabular value of the chi-square distribution [10].

Source: Prepared by the researcher based on the statistical program EViews 10th Edition
Figure 7. Steps of Normal Distribution Test.

4. Practical Application for Diagnostic Tests: Empirical Results

First hypothesis: There is a negative significant relationship with statistical significance between emissions of methane,
nitrous and GDP.
Study variables: The variables used in estimating the model can be defined as follows:
145 Abeer Mohamed Abd El Razek Youssef: Detecting Of Multicollinearity, Autocorrelation and
Heteroscedasticity in Regression Analysis

Table 1. Definition of Study variables.

variable name Definition measruing unit Variable type

LGDP GDP (current US$) dependent variable
LCO2_EMISSIONS CO2 emissions (kt) independent variable
LMETHANE_EMISSIONS Agricultural methane emissions (Thousand metric tons of CO2 equivalent) independent variable
LNITROUS_OXIDE_EMISSIONS Agricultural nitrous oxide emissions (Thousand metric tons of CO2 equivalent) independent variable

Study population and sample:

GDP was chosen as an indicator of production in Egypt 4.1. Application for Multiple Regression
and as a response variable (dependent); While were methane Overall Significance: Through the model, we find that the
and nitrous included as an independent and explanatory value of the F-statistic (0.000) is less than (0.05) indicating
variable, the study covers Egypt country with during the the overall Significance of the model, which is significant at
period 1990 to 2022, thus the number of observations used in the level of significance of 5%, meaning that the model is
the total sample is 33. The study was applied to Egypt [3]. totally significant.

Table 2. Estimation of Multiple Regression.

Variable Coefficient Std. Error t-Statistic Prob.

C 2.590248 0.662695 3.908660 0.0005
LCO2_EMISSIONS 0.338262 0.076663 4.412349 0.0001
LNITROUS_OXIDE_EMISSIONS -0.569029 0.142645 -3.989127 0.0004
LMANUFACTURING 0.920630 0.032070 28.70723 0.0000
LMETHANE_EMISSIONS 0.257899 0.123110 2.094861 0.0454
R-squared 0.997566 Mean dependent var 25.61279
Adjusted R-squared 0.997218 S. D. dependent var 0.753666
S. E. of regression 0.039749 Akaike info criterion -3.473725
Sum squared resid 0.044240 Schwarz criterion -3.246982
Log likelihood 62.31647 Hannan-Quinn criter. -3.397433
F-statistic 2869.009 Durbin-Watson stat 0.832508
Prob (F-statistic) 0.000000

Dependent Variable: LGDP

Method: Least Squares
Date: 07/26/22 Time: 04:26
Sample: 1 33
Included observations: 33
Source: Prepared by the researcher based on the statistical program EViews 10th Edition

As shown in the table significance of parameters this means

that the model coefficients with statistical significance are Table 3. Variance Inflation Factors.
represented the gross domestic production and emissions, Coefficient Uncentered Centered
Variable
where we find the probability of the fixed coefficient for each Variance VIF VIF
of them (0.0001), (0.0004), (0.0000), (0.0454) respectively, C 0.439164 9172.407 NA
LCO2_EMISSIONS 0.005877 17576.45 17.57123
because they are less than the level of significance (0.05) [11].
LMANUFACTURING 0.001028 12181.52 11.37272
It is clear from the Estimation of regression that the model LMETHANE_EMISSIONS 0.015156 29277.58 4.378833
suffers from problems to detect them, we follow the following LNITROUS_OXIDE_EMISSI
0.020348 38518.83 9.943755
tests: ONS

4.2. Variance Inflation Factors Date: 07/26/22 Time: 09:02

Sample: 1 33
The variance inflation factor (VIF) is a measure of the Included observations: 33
Source: Prepared by the researcher based on the statistical program EViews
amount of multiple linearity in a set of multiple regression 10th Edition
variables. Mathematically, the VIF of a regression model
variable is equal to the ratio of the total model variance to the 4.3. Coefficient Covariance Matrix
model variance that includes only that single independent
variable [12]. The variance-covariance matrix forms the keystone artifact
VIF values appear in the Centered VIF column (green of regression models. The variance-covariance matrix of the
values), VIF values show the variables that may be the cause regression model’s errors is used to determine whether the
of multicollinearity problem and whose value is higher than model’s error terms are homoscedastic (constant variance)
10 (yellow values). Based on the highest value of VIF for an and uncorrelated. The variance-covariance matrix of the
independent variable, it is the main cause of the fitted regression model’s coefficients is used to derive the
multicollinearity [13]. standard errors and confidence intervals of the fitted model’s
 Advances 2022; 3(3): 140-152 146

coefficient estimates. Both matrices are used in forming the prediction intervals of the model’s forecasts [14].

Table 4. Coefficient Covariance Matrix.

LMETHANE_E LNITROUS_OXIDE_E
C LCO2_EMISSIONS LMANUFACTURING
MISSIONS MISSIONS
C 0.439164 0.020861 -0.004496 -0.039445 -0.021249
LCO2_EMISSIONS 0.020861 0.005877 -0.002078 0.001139 -0.005530
LMANUFACTURING -0.004496 -0.002078 0.001028 -5.54E-05 0.000568
LMETHANE_EMISSIONS -0.039445 0.001139 -5.54E-05 0.015156 -0.012460
LNITROUS_OXIDE_EMISSIONS -0.021249 -0.005530 0.000568 -0.012460 0.020348

Source: Prepared by the researcher based on the statistical program EViews 10th Edition.

The variance-covariance matrix is a square matrix i.e., it has

the same number of rows and columns. The elements of the 4.4. Correlation Matrix
matrix that lie along its main diagonal i.e., the one that goes A correlation matrix is a table showing correlation
from top-left to bottom-right contain the variances while all coefficients between sets of variables. Each random variable
other elements contain the co-variances. Thus, the variance- (Xi) in the table is correlated with each of the other values in
covariance matrix of the fitted coefficients of a regression model the table (Xj). This allows you to see which pairs have the
contains the variances of the fitted model’s coefficient estimates highest correlation [15].
and the pairwise covariances between coefficient estimates.

Table 5. Correlation matrix.

LMETHANE_EMI LNITROUS_OXIDE
LGDP LCO2_EMISSIONS LMANUFACTURING
SSIONS _EMISSIONS
LGDP 1.000000 0.958840 0.997754 0.674685 0.817970
LCO2_EMISSIONS 0.958840 1.000000 0.953773 0.731382 0.889963
LMANUFACTURING 0.997754 0.953773 1.000000 0.675877 0.826621
LMETHANE_EMISSIONS 0.674685 0.731382 0.675877 1.000000 0.872754
LNITROUS_OXIDE_EMISSIONS 0.817970 0.889963 0.826621 0.872754 1.000000

Source: Prepared by the researcher based on the statistical program EViews 10th Edition

Table 5 shows at intersection of each cell, three values are

given: Correlation, t-statistic, Prob. Table 6. Correlogram of Residuals.
The order from top to bottom (green values) and thus the Partial
Autocorrelation AC PAC Q-Stat Prob
degree of correlation of the independent variables X's with Correlation
the dependent variable Y. . |*** | . |*** | 1 0.451 0.451 7.3469 0.007
. |*. | . |. | 2 0.152 -0.065 8.2029 0.017
If <5% (Correlation) value P, we conclude that there is a
. |. | . |. | 3 0.015 -0.036 8.2113 0.042
significant association between the variable. . *|. | . *|. | 4 -0.152 -0.171 9.1285 0.058
The function F and the rest of the independent variables, .**|. | . *|. | 5 -0.250 -0.139 11.703 0.039
which means a successful selection of the variables. .**|. | . *|. | 6 -0.316 -0.177 15.984 0.014
There is an indication of a significant linear correlation .**|. | . *|. | 7 -0.313 -0.134 20.324 0.005
.**|. | . *|. | 8 -0.295 -0.172 24.350 0.002
between the values of the independent variables – before the
. *|. | . |. | 9 -0.149 -0.021 25.425 0.003
application of the regression model (red values), which leads . |. | . |. | 10 -0.007 -0.025 25.428 0.005
to the emergence of the problem of the linear correlation . |*. | . |*. | 11 0.158 0.083 26.744 0.005
between Multicollinearity variables. . |*. | . |. | 12 0.184 -0.054 28.598 0.005
It is not preferable to have a significant association in the . |** | . |. | 13 0.219 0.047 31.371 0.003
. |*. | . |. | 14 0.172 -0.052 33.178 0.003
regression models between X's.
. |*. | . |*. | 15 0.162 0.082 34.870 0.003
. |. | . *|. | 16 0.015 -0.125 34.886 0.004
4.5. Correlogram of Residuals
Date: 07/26/22 Time: 09:12
According to Table 6 the graphic representation of the Sample: 1 33
functions of correlation and partial correlation, the series is Included observations: 33
not characterized by any format, initially (and graphically
only) it can be said that the series is stable, where all 4.6. Normality Test for Residuals
parameters of the functions of correlation and partial
Normality tests are used to determine if a data set is well-
correlation are within the confidence range, except for the
modeled by a normal distribution and measures a goodness
first parameter of each statistical function [16].
of fit of a normal model to the data [17].
147 Abeer Mohamed Abd El Razek Youssef: Detecting Of Multicollinearity, Autocorrelation and
Heteroscedasticity in Regression Analysis

Source: Prepared by the researcher based on the statistical program EViews 10th Edition

Figure 8. Normality Test for Residuals.

From Figure 8, data of residuals are given. Normality of mean value (0.037 <3.000), data were considered unnormally
the above data was assessed. Result showed that data were distributed.
not normally distributed as skewness (0.06772) and kurtosis
(3.308) individually were within ±1. Jarque-Bera test (P = 2. 4.7. Autocorrelation Test
654) were statistically significant, that is, data were An identifiable relationship (positive or negative) exists
considered unnormal distributed. between the values of the error in one period and the values
Although both methods indicated that data were not of the error in another period.
normally distributed. As SD of residuals was less than half

Table 7. Serial correlation LM test.

Variable Coefficient Std. Error t-Statistic Prob.

C -0.079918 0.585771 -0.136432 0.8925
LCO2_EMISSIONS 0.011606 0.070225 0.165269 0.8700
LMANUFACTURING -0.013838 0.029776 -0.464722 0.6460
LMETHANE_EMISSIONS -0.076020 0.111119 -0.684131 0.4999
LNITROUS_OXIDE_EMISSIONS 0.105422 0.130050 0.810629 0.4249
RESID (-1) 0.592287 0.204035 2.902872 0.0074
RESID (-2) 0.076091 0.245170 0.310360 0.7588
R-squared 0.278753 Mean dependent var -3.45E-15
Adjusted R-squared 0.112312 S. D. dependent var 0.037182
S. E. of regression 0.035032 Akaike info criterion -3.679287
Sum squared resid 0.031908 Schwarz criterion -3.361846
Log likelihood 67.70824 Hannan-Quinn criter. -3.572478
F-statistic 1.674782 Durbin-Watson stat 1.705451
Prob (F-statistic) 0.167141

F-statistic: 5.024346; Prob. F (2,26): 0.0143

Obs*R-squared: 9.198859; Prob. Chi-Square (2): 0.0101
Test Equation:
Dependent Variable: RESID
Method: Least Squares
Date: 07/26/22 Time: 09:17
Sample: 1 33
Included observations: 33
Presample missing value lagged residuals set to zero.
Source: Prepared by the researcher based on the statistical program EViews 10th Edition.

As shown in table Serial correlation LM test since Prob. between X 's and residuals.
Chi-Square (2) less than 5% is (0.0101) We reject H0 that
there is autocorrelation. Obs*R2 =9.198859 =NR2 4.8. Heteroskedasticity and Serial Correlation
H0: ρ = 0 In statistics, heteroskedasticity (or heteroscedasticity)
H1: ρ ≠ 0 happens when the standard deviations of a predicted variable,
P value (Coefficient) > 5% It means there is no correlation monitored over different values of an independent variable or
 Advances 2022; 3(3): 140-152 148

as related to prior time periods, are non-constant. With the residual errors is that they will tend to fan out over time,
heteroskedasticity, the tell-tale sign upon visual inspection of as depicted in the image below.

Table 8. Types of Heteroskedasticity.

Unconditional Unconditional heteroskedasticity occurs when the heteroskedasticity is uncorrelated with the values of the independent variables.
Heteroskedasticity Although this is a violation of the homoscedasticity assumption, it does not present major problems to statistical inference.
Conditional Conditional heteroskedasticity occurs when the error variance is related/conditional on the values of the independent variables. It poses
Heteroskedasticity significant problems for statistical inference. Fortunately, many statistical software packages can diagnose and correct this error.

Table 9. Heteroskedasticity Test: Harvey.

Variable Coefficient Std. Error t-Statistic Prob.

C -27.44860 30.62924 -0.896157 0.3778
LCO2_EMISSIONS -10.95344 3.543287 -3.091322 0.0045
LMANUFACTURING 5.328096 1.482235 3.594638 0.0012
LMETHANE_EMISSIONS -1.402768 5.690064 -0.246529 0.8071
LNITROUS_OXIDE_EMISSIONS 3.894966 6.592940 0.590778 0.5594
R-squared 0.316628 Mean dependent var -8.043523
Adjusted R-squared 0.219004 S. D. dependent var 2.078869
S. E. of regression 1.837179 Akaike info criterion 4.193067
Sum squared resid 94.50640 Schwarz criterion 4.419811
Log likelihood -64.18561 Hannan-Quinn criter. 4.269360
F-statistic 3.243329 Durbin-Watson stat 2.457017
Prob (F-statistic) 0.026316

F-statistic: 3.243329; Prob. F (4,28): 0.0263

Obs*R-squared: 10.44874; Prob. Chi-Square (4): 0.0335
Scaled explained SS: 8.873285; Prob. Chi-Square (4): 0.0643
Test Equation:
Dependent Variable: LRESID2
Method: Least Squares
Date: 07/26/22 Time: 09:21
Sample: 1 33
Included observations: 33
Source: Prepared by the researcher based on the statistical program EViews 10th Edition.

Figure 9. Heteroskedasticity and vs Homoskedasticity.

As shown in table Prob. Chi-Square (4) less than 5% is (0.0335) We reject H0 that there is Heteroskedasticity. Obs*R2 =
10.44874 = NR2
H0: σ12 = σ22 …σ n2 = 0 The constant homogeneity of variances.
H1: σ i 2 ≠ 0 at least heterogeneity of variances

5. Draw the Estimated Equation: Substituted Coefficients

LGDP = 2.59024769764 + 0.338262093151*LCO2_EMISSIONS + 0.920630344411*LMANUFACTURING +
0.257898931733*LMETHANE_EMISSIONS - 0.56902863672*LNITROUS_OXIDE_EMISSIONS
149 Abeer Mohamed Abd El Razek Youssef: Detecting Of Multicollinearity, Autocorrelation and
Heteroscedasticity in Regression Analysis

Source: Prepared by the researcher based on the statistical program EViews 10th Edition
Figure 10. Simple regression.

As appears in figure 10 diagram between independent when the natural logarithm is taken for it. This method is
variable and corresponding to the measurement of the LGDP used when the variable values are large, and it is intended to
(horizontal axis) and the LMETHANE_EMISSIONS and simplify them to reduce dispersion and variance between
LNITROUS_OXIDE_EMISSIONS (vertical axis). shows other variables. The log is characterized by not changing the
increasing positive relation among variables. shape of the distribution but changing the shape of the scale.
The figure 11 shows the variables over time. It is clear
6. Trends of the Variables over Time from the figure that the dependent and independent variables
are in a continuous direction through time .and that there is
To avoid the problem of different scales for each variable, an increasing direct relationship through time between the
the log is chosen. The values of the variable are not affected dependent and independent variable.

Source: Prepared by the researcher based on the statistical program EViews 10th Edition
Figure 11. logarithm of dependent and independent variables.
 Advances 2022; 3(3): 140-152 150

6.1. Actual and Estimated Residuals

Figure 12. Actual and Estimated Residuals. Figure 13. LGDP Residuals.

Source: Prepared by the researcher based on the statistical program EViews Source: Prepared by the researcher based on the statistical program EViews
10th Edition. 10th Edition

The curve in red represents the "true" actual values of the The figure 13 shows the residual time series of the
time series of the dependent variable in the figure 12. logarithm variable of GDP, which means that the residual
The curve in green represents the estimated values of the series is unstable and often there is no co-integration.
dependent variable according to the estimated equation.
The blue curve represents the residuals of the regression 7. Conclusion
equation "random error term perturbation"
From the Figure 12, we notice that there are "extreme" From the above we conclude the following:
abnormal values in the random error term, and I think that the 1. The model suffers from the problem of heterogeneity of
estimated model suffers from a disruption of variance. variance, and this leads to that the predictions in the
There is no essential difference between the actual values, variable Y depending on the estimators Βˆ 's (the
which are the values before including the explanatory coefficients of the independent variables) from the
variables, and the estimated values, which are the values after original data will have large variances, and this means
including the explanatory variables. that the prediction will be inefficient and the reason for
From the figure, we notice an almost similarity. As for the this is that the variance The predictions will include the
curve in blue, it indicates the behavior of residuals, which U variance as well as the parameters variance.
can theoretically be divided into three sections. 2. The model suffers from a problem of autocorrelation,
Where it is noticed that there is relative stability of the which means that Cov (uj, ui) ≠ 0, and therefore the
behavior of the residuals in the middle region of the chain, standard errors σ 2 are rather large, which means that
which in turn explained the absence of the imbalance the accuracy in the model is low and therefore the
between the actual values and the estimated values. confidence intervals and the model’s significance will
As for the sides of the series parties, we notice a be unacceptable and unreliable in and inefficient.
fluctuation, which in turn affects the quality of the overall 3. The model suffers from the problem of linear
model. It is believed that the problem should be treated with interference. This means that the estimators’ values are
the effect of outliers, as you can detect them through the very large and biased, as well as the variances of these
mahaleb’s test. estimators and the covariances are very large, so the
properties of estimators are not BLUE.
6.2. Logarithm of LGDP Residuals

To avoid the problem of different scales for each variable, Acknowledgements

the log is chosen. The values of the variable are not affected
I thank my father and mother for bearing hardships and
when the natural logarithm is taken for it. This method is
stress on me during the exams, and for their service in
used when the variable values are large, and it is intended to
difficult times, and for providing me the effort and time to
simplify them to reduce dispersion and variance between
allow me the opportunity to work hard and diligently day and
other variables. The log is characterized by not changing
night. I hope that God accepts their work and makes them
the shape of the distribution but changing the shape of the
worship the righteous and places them in the Paradise of
scale.
Paradise, in the company of the prophets and friends.
I thank Dr. Hossam Elden M. Abdelkader, Associate
151 Abeer Mohamed Abd El Razek Youssef: Detecting Of Multicollinearity, Autocorrelation and
Heteroscedasticity in Regression Analysis

professor Economic Dep., Faculty of Administrative Sciences, teaching us advanced econometrics lectures in the preparatory
Ain Shams University, Egypt .I thank him for teaching the year for the doctorate program, He is a distinguished young
monetary policy course, benefiting from his knowledge, and doctor, genius and a role model for the youth.

Appendix of Study
Table 10. Data of the Study Variables.

Country year GDP methane emissions nitrous oxide emissions CO2 emissions GDP growth (annual%) Manufacturing FDI
Egypt 1990 4.3E+10 11270 9450 87750 2.900791 7.3E+09 1.058425
Egypt 1991 3.74E+10 11980 10130 89370 3.973172 5.99E+09 2.420133
Egypt 1992 4.19E+10 12490 10100 90900 4.642459 6.54E+09 0.994028
Egypt 1993 4.66E+10 12760 10890 92660 4.988731 7.33E+09 0.940415
Egypt 1994 5.19E+10 13000 10030 87900 5.492355 8.31E+09 1.135376
Egypt 1995 6.02E+10 13140 11730 93720 5.575497 9.83E+09 1.268437
Egypt 1996 6.76E+10 13400 12120 98940 6.053439 1.12E+10 1.174393
Egypt 1997 7.84E+10 14100 11760 106060 6.370004 1.28E+10 1.236997
Egypt 1998 8.48E+10 13210 12290 110980 3.535252 1.44E+10 0.527385
Egypt 1999 9.07E+10 14640 12400 116540 2.390204 1.63E+10 0.759753
Egypt 2000 9.98E+10 15100 13170 114610 3.193455 1.8E+10 0.295684
Egypt 2001 9.67E+10 14760 13450 126700 4.092072 1.71E+10 1.590836
Egypt 2002 8.51E+10 15940 14140 129440 4.471744 1.53E+10 5.999509
Egypt 2003 8.03E+10 16040 14540 133020 6.843838 1.39E+10 9.348567
Egypt 2004 7.88E+10 16360 15680 144500 7.087827 1.36E+10 8.876336
Egypt 2005 8.96E+10 16350 15500 162220 7.156284 1.5E+10 5.831413
Egypt 2006 1.07E+11 16940 15180 170750 4.6736 1.72E+10 3.548351
Egypt 2007 1.3E+11 17630 14670 183400 5.147235 2E+10 2.916017
Egypt 2008 1.63E+11 17820 14910 189940 1.764572 2.53E+10 0.204543
Egypt 2009 1.89E+11 15810 14800 197660 2.2262 2.99E+10 1.002341
Egypt 2010 2.19E+11 14880 14400 200310 2.185466 3.53E+10 1.453434
Egypt 2011 2.36E+11 16110 14990 205770 2.915912 3.72E+10 1.50925
Egypt 2012 2.79E+11 16730 14560 215000 4.372019 4.51E+10 2.102581
Egypt 2013 2.88E+11 16090 14630 213860 4.346643 4.79E+10 2.438563
Egypt 2014 3.06E+11 15990 14740 219120 4.181221 5.13E+10 3.142826
Egypt 2015 3.29E+11 15380 15320 226280 5.314121 5.5E+10 3.260263
Egypt 2016 3.32E+11 15570 15670 231230 5.557684 5.6E+10 2.972837
Egypt 2017 2.36E+11 14770 15460 242230 3.569669 3.88E+10 1.602124
Egypt 2018 2.5E+11 13180 15070 247910 3.326742 4.04E+10 1.602124
Egypt 2019 3.03E+11 16800 15650 249370 3.326742 4.82E+10 1.602124
Egypt 2020 3.65E+11 16800 15650 249370 3.326742 5.88E+10 1.602124
Egypt 2021 4.04E+11 16800 15650 249370 3.326742 5.88E+10 1.602124
Egypt 2022 4.04E+11 16800 15650 249370 3.326742 5.88E+10 1.602124

Source: Data collected by researcher from world bank.

Theory and Methods, 4 (3), 277-292.

[7] Anderson, R. L. (1954). The problem of autocorrelation in
References regression analysis. Journal of the American Statistical
Association, 49 (265), 113-129.
[1] Alin, A. (2010). Multicollinearity. Wiley interdisciplinary
reviews: computational statistics, 2 (3), 370-374. [8] Kadiyala, K. R. (1968). A transformation used to circumvent
the problem of autocorrelation. Econometrica: Journal of the
[2] Mansfield, E. R., & Helms, B. P. (1982). Detecting Econometric Society, 93-96.
multicollinearity. The American Statistician, 36 (3a), 158-160.
[9] Griffith, D. A., Fischer, M. M., & LeSage, J. (2017). The
[3] Daoud, J. I. (2017, December). Multicollinearity and spatial autocorrelation problem in spatial interaction
regression analysis. In Journal of Physics: Conference Series modelling: a comparison of two common solutions. Letters in
(Vol. 949, No. 1, p. 012009). IOP Publishing. Spatial and Resource Sciences, 10 (1), 75-86.
[4] Farrar, D. E., & Glauber, R. R. (1967). Multicollinearity in [10] Stimson, R. J., Mitchell, W., Rohde, D., & Shyy, P. (2011).
regression analysis: the problem revisited. The Review of Using functional economic regions to model endogenous
Economic and Statistics, 92-107. regional performance in Australia: Implications for addressing
the spatial autocorrelation problem. Regional Science Policy
[5] Kim, J. H. (2019). Multicollinearity and misleading statistical & Practice, 3 (3), 131-144.
results. Korean journal of anesthesiology, 72 (6), 558-569.
[11] Shrestha, N. (2020). Detecting multicollinearity in regression
[6] Gunst, R. F., & Webster, J. T. (1975). Regression analysis and analysis. American Journal of Applied Mathematics and
problems of multicollinearity. Communications in Statistics- Statistics, 8 (2), 39-42.
 Advances 2022; 3(3): 140-152 152

[12] Obite, C. P., Olewuezi, N. P., Ugwuanyim, G. U., & [15] Aslam, M., & Ahmad, S. (2020). The modified Liu-ridge-type
Bartholomew, D. C. (2020). Multicollinearity effect in estimator: a new class of biased estimators to address
regression analysis: A feed forward artificial neural network multicollinearity. Communications in Statistics-Simulation
approach. Asian journal of probability and statistics, 6 (1), 22- and Computation, 1-20.
33.
[16] Negret, P. J., Marco, M. D., Sonter, L. J., Rhodes, J.,
[13] Zhang, T., Zhou, X. P., & Liu, X. F. (2020). Reliability Possingham, H. P., & Maron, M. (2020). Effects of spatial
analysis of slopes using the improved stochastic response autocorrelation and sampling design on estimates of protected
surface methods with multicollinearity. Engineering Geology, area effectiveness. Conservation Biology, 34 (6), 1452-1462.
271, 105617.
[17] Stojkoski, V., Sandev, T., Kocarev, L., & Pal, A. (2022).
[14] Vörösmarty, G., & Dobos, I. (2020, October). Green Autocorrelation functions and ergodicity in diffusion with
purchasing frameworks considering firm size: a stochastic resetting. Journal of Physics A: Mathematical and
multicollinearity analysis using variance inflation factor. In Theoretical, 55 (10), 104003.
Supply Chain Forum: An International Journal (Vol. 21, No. 4,
pp. 290-301). Taylor & Francis.