Data Panel Modelling with Fixed Effect Model (FEM) Approach to
Analyze the Influencing Factors of DHF in Pasuruan Regency
Achmad Faridz Jauhari
1
, Yoyon K. Suprapto
1
and Achmad Mauludiyanto
1
1
Department of Electrical Engineering Institut Teknologi Sepuluh Nopember Surabaya, Indonesia
Keywords:
Panel Data, Fixed Effect Model, Regression Analysis, Ordinary Least Square, Dengue Hemorrhagic Fever.
Abstract:
Dengue Hemorrhagic Fever (DHF) is one of the endemic diseases caused by the bites of Aedes mosquitoes
which are infected with the dengue virus. This disease can cause death. The DHF mortality rate in Pasu-
ruan Regency is high (above 1% per year) in the last four years. Therefore, this study aimed to find a model
that can explain the influencing factors of DHF incidence. The variables used were the number of DHF pa-
tients(Y), waste volume(X
1
), rainy days(X
2
), health facilities(X
3
), temperature(X
4
), number of high-educated
population(X
5
), population density(X
6
), and rainfall(X
7
). The data used were ranging from 2015 to 2018 and
obtained from several agencies in Pasuruan Regency. In this study, the method used was the Panel Data Re-
gression with Fixed Effect Model approach. The results of the model showed R
2
: 0.804 meaning that the
seven variables were able to explain the effect on the incidence of DHF by 80.4% while the remaining 19.6%
was influenced by other unknown variables. Of the seven predictor variables, there are six variables that have
a significant effect consist of Waste Volume, Health Facilities, Temperature, Number of High-Educated Popu-
lation, Population Density, and Rainfall. Henceforth, future DHF prevention and control policies can be more
emphasized on these factors.
1 INTRODUCTION
Dengue hemorrhagic fever (DHF) is a disease caused
by the dengue virus and is transmitted by Aedes ae-
gypti and Aedes albopictus mosquitos. Both types of
mosquitoes are found in all corners of Indonesia, ex-
cept in places with altitudes above 1000 masl (meters
above sea level) (Arsin, 2013). The symptoms pa-
tients is a high fever for 2-7 days (38-400°C). At the
acute level, this disease can cause death. Besides that,
DHF can appear throughout the year and can affect all
age groups (Kemenkes, 2017).
Indonesia is the country which has the highest
DHF cases in Southeast Asia (Kemenkes, 2010). In
the last five years, the highest number of dengue cases
was occurred in 2016 reaching up to 204,171 cases
with 1,598 deaths. The number of dengue cases in
2016 increased by 57.5% compared to the number
of dengue cases in 2015 which was 129,650 cases.
The number of DHF deaths in 2016 also worsens by
49.2% compared to the number of deaths in 2015
which was only 1,071 people. It was also reported
that the Incident Rate (IR) in 2015 increased from
50.75 to 78.85 per 100,000 population. However, the
Case Fatality Rate (CFR) has decreased from 0.83%
in 2015 to 0.78% in 2016. DHF has spread in 34
provinces and 463 districts/cities in Indonesia (Ke-
menkes, 2017).
Pasuruan Regency is one of the regions in East
Java where its DHF mortality rate is above 1% in this
past four years. Since 2015, DHF is determined as an
Extraordinary Situation and as a result, the efforts to
control DHF were carried out intensively. Based on
the data released by the Public Health Office of Pa-
suruan Regency, the number of DHF cases in 2015
was 686 cases with 28 deaths whereas, in 2016, the
number of DHF cases increased by 11% to 764 cases
with a total death of 27 people. In 2017, the number
of DHF cases decreased by 59% to 317 cases with
13 deaths and continued to decline by 38% with four
deaths in 2018. The Incident Rate (IR) of DHF in
2015 was 43.31 per a 100,000 population while the
Case Fatality Rate (CFR) was 4.1%. In 2016, the
IR increased to 48.23 per a 100,000 population while
the CFR dropped to 3.5%. On the other hand, in
2017, the IR decreased to 19.75 per 100,000 popula-
tion while the CFR increased to 4.1%. In 2018, the IR
was 11.90 per 100,000 population while the CFR was
2.1%. Based on the above data, the mortality/Case
Fatality Rate of DHF in Pasuruan for the past four
224
Jauhari, A., Suprapto, Y. and Mauludiyanto, A.
Data Panel Modelling with Fixed Effect Model (FEM) Approach to Analyze the Influencing Factors of DHF in Pasuruan Regency.
DOI: 10.5220/0009881702240232
In Proceedings of the 2nd International Conference on Applied Science, Engineering and Social Sciences (ICASESS 2019), pages 224-232
ISBN: 978-989-758-452-7
Copyright
c
2020 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
years is included in the high category (CFR ¿ 1%)
(Kemenkes, 2017).
The spread of DHF in Indonesia is influenced by
many factors such as climate, community behaviour,
environment, as well as demographic and socioeco-
nomic conditions (Arsin, 2013). The climate factors
include temperature, rainfall, rainy days, humidity,
and wind speed; behavioural factor is basically a per-
son’s response towards stimuli that is related to illness
and disease including knowledge, actions, and beliefs
associated with DHF, availability of health resources,
adequate and affordability of health facilities, com-
munity support, as well as the concern from govern-
ment and health workers; environmental factors are
the number of places or containers of DHF vector pro-
liferation; demographic factors consist of the popula-
tion density and population mobility; socio-economic
factors are the level of education, employment, and
the number of family members. Of the factors men-
tioned above, knowing which of the factors that have
a significant effect on DHF is very important so that
future efforts to prevent and control DHF can be done
more effectively and efficiently.
Modelling is a method that can be used to de-
termine the factors that have a significant effect on
DHF in Pasuruan Regency. This study aimed to find
a model that is able to explain the influencing fac-
tors of DHF in 21 sub-districts of Pasuruan within the
period of 2015 to 2018. Different regional character-
istics in each sub-district are thought to influence the
incidence of DHF. By that, this study used Panel Data
structure. There has never been a research which ex-
amined the factors of DHF in Pasuruan Regency so
that the results of this modelling are expected to be
used as early warnings or basis for the formulation of
strategic policies to prevent and eradicate DHF in the
future.
Several studies related to the analysis of DHF have
been done before. Research conducted by (Ariani,
2018) used the Negative Binomial regression model
to produce variables that significantly influence DHF,
namely population density, number of health workers
and rainfall. Subsequent research by (Rustiani, 2017)
using multiple linear regression resulted in R
2
values
of 67 percent, significant variables included popula-
tion density, rainfall and larva free index. Research
conducted by (Rasmanto, 2016) using Linear Regres-
sion resulted in R
2
values of 24.1 percent, variables
that had a significant effect consisted of air temper-
ature and relative humidity. The study by (Hasirun,
2016) used a Spatial regression model with spatial
error resulting in R
2
values of 43.34 percent, sig-
nificant variables covering rainfall, health facilities,
the percentage of PHBS houses, and the percentage
of healthy houses. Subsequent research by (Martha,
2015) used panel data regression with a fixed effect
model approach resulted in R
2
values of 72,76 per-
cent, significant variables included population den-
sity, population mobility, the average age of DHF pa-
tients, and the number of DHF patients in the previous
time period. Based on these studies we use several
variables that have been used previously namely pop-
ulation density, rainfall, rainy days, temperature, and
the number of health facilities. Then we add two new
variables namely the waste volume and the Number
of High-Educated Population.
2 LITERATURE REVIEW
2.1 Panel Data
Panel Data is a collection of data consisting of cross-
section data and time series data. Time series data
usually includes one individual observed in a certain
period of time while cross-section data is obtained
from several individuals observed in a certain period
of time. Therefore, it can be said that the Panel Data
presents larger and more informative data and gener-
ates a greater degree of freedom. The general form of
Panel Data regression models can be seen in Equation
1 (Hsiao, 2003; I. et al., 2017).
γ
it
= α
it
+ β
0
X
it
+ ε
it
(1)
2.2 Estimation of Panel Data
Regression Model
In making parameter estimation, the model depends
on the assumptions about intercepts and slope coef-
ficients. Using Panel Data allows different intercepts
and slope coefficients to occur in each individual and
each time period. In this concept, there are three ap-
proaches that can be used, namely CEM (Common
Effect Models), FEM (Fixed Effect Model) and REM
(Random Effect Models) (Widarjono, 2013; I. et al.,
2017)
Data Panel Modelling with Fixed Effect Model (FEM) Approach to Analyze the Influencing Factors of DHF in Pasuruan Regency
225
2.2.1 CEM (Common Effect Models)
The approach with CEM model assumes that inter-
cepts and slope coefficients have the same value for
all individuals at all time periods. In other words, this
model ignores individual dimensions and time. The
equation of CEM model is illustrated in this follow-
ing Equation 2.
γ
it
= α + β
0
X
it
+ ε
it
(2)
2.2.2 FEM (Fixed Effect Model)
This approach believes that differences in character-
istics between individuals are represented in the in-
tercepts. Thus, intercepts for each individual will be
different but the slope coefficients remain constant at
all time periods. Equation 3 below presents the equa-
tion of FEM model.
γ
it
= α
i
+ β
0
X
it
+ ε
it
(3)
2.2.3 REM (Random Effect Models)
In REM, the differences in individual characteristics
are accommodated by error terms. Error terms may
correlate between individuals and between times. For
more details, the equation of REM model is formu-
lated in Equation 4.
γ
it
= α + β
0
X
it
+W
it
(4)
α is the mean of intercept from error terms cross
section and time series. W
it
: µ
i
+ ε
it
where µ
i
is the
random error cross section deviation, which explains
the differences between individuals and each other.
2.3 Selection of Panel Data Regression
Model
In choosing the model that will be used to manage
Panel Data, several tests need to be done, namely:
2.3.1 Chow Test
Chow Test is used to choose a better model between
CEM and FEM (Greene, 2000). The hypothesis in the
Chow Test can be written as follows.
H
0
: α
1
= α
2
= ·· · = α
n
= 0 (CEM)
H
1
: minimumo f oneα
i
6= 0 (FEM)
Next, the equation for Chow Test Statistics is:
F
count
(RRSS URSS)/(N 1)
URSS/(nT n K)
(5)
Description: RRSS: restricted residual sum of
square CEM URSS: unrestricted residual sum of
square FEM n: number of individuals(cross section)
T: time periods (Time Series) If the value of F
c
ount
F
t
able = F
(n1,nT nK)
or p value 0.05, then hy-
pothesis H
0
is rejected. This means that FEM is a
better model than CEM.
2.3.2 Hausman Test
Hausman test is used to choose a better model be-
tween FEM and REM (Baltagi, 2010; Sutikno et al.,
2017). The hypothesis in the Hausman Test is as fol-
lows.
H
0
: correlation(µ
i
,X
it
) = 0(REM)
H
1
: correlation(µ
i
,C
it
) 6= 0(FEM)
This following equation 6 displays the Hausman
Test Statistics.
W =(
ˆ
β
FEM
ˆ
β
REM
)
0
[var(
ˆ
β
FEM
ˆ
β
REM
)]
1
(
ˆ
β
FEM
ˆ
β
REM
)
(6)
If the value of W χ
2
table
= χ
2
(a,K)
or p value
0.05, then hypothesis H
0
rejected, meaning that FEM
is a better model than REM. But if the value of W <
χ
2
table
= χ
2
(a,K)
or p value > 0.05, then hypothesis
H
0
, is accepted. This points out that the best model
between FEM and REM is REM.
2.3.3 Lagrange Multiplier Test
LM Test is used to choose a better model between
CEM and REM. LM Test does not need to be done
if FEM is found to be the best in the Chow Test and
Hausman Test. This test is only done when the best
model found in the Chow Test is FEM whereas, in the
Hausman Test, the best model is REM. These are the
hypothesis in the LM test:
H
0
: σ
2
ε
= 0(CEM).
H
1
: σ
2
ε
6= 0(REM).
In details, the statistics of LM Test can be under-
stood through Equation 7 (Widarjono, 2013) below.
LM =
nT
2(T 1)
(
Σ
n
i=1
(T ˆe
i
)
2
Σ
n
i=1
Σ
T
t=1
ˆe
2
it
1)
2
(7)
Description:
n:number of individuals
T:time period
ˆe:error on CEM
ICASESS 2019 - International Conference on Applied Science, Engineering and Social Science
226
Hypothesis H
0
is rejected if the value of LM
χ
2
table
= χ
2
(a,K)
or p value 0.05. This indicates that
REM is a better model than CEM. Otherwise, if the
value of LM < χ
2
table
= χ
2
(a,K))
or p value > 0.05,
then the hypothesis H
0
is accepted which means that
the correct model between the two models is CEM.
2.3.4 Classical Assumptions Test
To get the estimator of the parameter model in the
CEM and FEM approach, Ordinary Least Square
(OLS) method is used. The classical assumption is
done to get the best OLS estimator known as BLUE
(Best Linear Unbiased Estimator). Classical assump-
tion tests include 1. Multicollinearity Test, 2. Nor-
mality Test, 3. Heteroscedasticity Test, and 4. Au-
tocorrelation Test. From all the four tests, only test
3 and 4 that must be met to obtain the best estimator
(BLUE) (Gujarati, 2004).
2.3.5 Multicollinearity Test
One of the assumptions used in OLS is the absence of
a strong relationship between predictor variables. The
presence of multicollinearity makes the model to have
a large variant (Widarjono, 2013)). However, multi-
collinearity can still produce BLUE estimators. One
way to detect multicollinearity is to use the sample of
correlation coefficient (Gujarati, 2004; Sutikno et al.,
2017). The sample of correlation coefficient is shown
in the next Equation 8.
r
12
=
(8)
If the correlation coefficient is > 0.85, then it
is suspected that multicollinearity occurred in the
model. Conversely, if the correlation coefficient is
0.85, it is assumed that the model did not contain
multicollinearity (Widarjono, 2013).
2.3.6 Normality Test
This assumption test is optional, but it is better if in-
cluded in an effort to get a good OLS estimator. The
hypotheses in this test are:
H
0
: E(ε
it
) = 0(error, distributed normally)
H
1
: E(ε
it
) = 0(error, not distributed normally)
The Normality Test Statistics adopted the Jarque-Bera
formula as we can see in this following Equation 9
(Gujarati, 2004)).
JB = NT (
S
2
k
6
+
(K
r
3)
2
24
)
(9)
Description,
S
k
: slope(Skewness)
K
r
: height(kurtosis)
If the value of JB < χ
2
tabel
= χ
2
(a,2)
or p value >
0.05, then hypothesis H
0
is accepted. This clarifies
that the error is normally distributed.
2.3.7 Heteroscedasticity Test
The assumption that the error variant has a constant
(homoscedastic) nature is very important to produce
the best OLS estimator. The Heteroscedasticity Test
hypothesis is:
H
0
error does not contain heteroscedasticity
H
1
error contain heteroscedasticity
Glejser Method is used for the Heteroscedastic-
ity Test Statistics, which is by regressing the abso-
lute value of the error with its independent variable
(Widarjono, 2013). If β
1
is not significant in t-test
or the Prob value is> 0.05, then hypothesis H
0
is ac-
cepted. This explains that the error did not contain
heteroscedasticity or was constant.
2.3.8 Autocorrelation Test
This assumption test is used to see if there is no serial
correlation on the error. This test is crucial to get a
BLUE OLS estimator. One way to detect autocorre-
lation is to use a Durbin Watson test method. These
are the hypothesis for the autocorrelation test in this
study:
H
0
: Cov(ε
it
,ε
i,t1
) = 0 (no autocorrelation)
H
1
: Cov(ε
it
,ε
i,t1
) 6= 0 (there is autocorrelation)
The statistics of the Durbin Watson autocorrela-
tion test can be seen in equation 10 (Gujarati and P,
2010).
DW =
Σ
N
i=1
Σ
2
t=2
(ε
it
ε
i,t1
)
2
Σ
N
i=1
Σ
2
t=2
ε
2
it
(10)
Data Panel Modelling with Fixed Effect Model (FEM) Approach to Analyze the Influencing Factors of DHF in Pasuruan Regency
227
Figure 1 below presents the critical value in the
autocorrelation test:
Figure 1: Autocorrelation critical value of durbin watson
justified.
2.3.9 Significance Test of Regression Parameters
The parameter significance test is used to find out the
level of influence that the predictor variable has on
the response variable. There are two tests carried out,
namely simultaneous test and partial test.
2.3.10 Simultaneous Test
A simultaneous test is applied to determine the ef-
fect of predictor variables on the response variable
together. The simultaneous test hypothesis is as fol-
lows.
H
0
: β
1
= β
2
= ··· = β
k
= 0 (all predictor variables
have no effect on the response variable)
H
1
:minimum of one β
k
6= 0 (there is at least one
predictor variable that affects the response vari-
able)
F
count
=
R
2
/(n + K 1)
(1 R
2
)/(nT n K)
(11)
R2 is the coefficient of determination. If the value
of F
count
> F
table
= F
(a,n+K1,nT nK)
or p value
0.05, then hypothesis H
0
, is rejected, meaning that
in the model there is at least one predictor variable
which has a significant effect on the response vari-
able.
2.3.11 Partial Test
A partial test is used to determine the effect of each
predictor variable on the response variable. The par-
tial test hypothesis is:
H
0
: β
k
= 0 (k-predictor variable does not signifi-
cantly influence the response variable)
H
1
: β
k
6= 0 (k-predictor variable significantly in-
fluences the response variable)
The statistics of the partial test is formulated in
Equation 12.
t
count
=
β
k
se(β
k
)
(12)
If the value of t
count
> t
table
= t
(a/2,nT nK)
or
p value 0.05, then hypothesis H
0
is rejected. This
describes that k-predictor variable has a significant ef-
fect on the response variable.
3 LITERATURE REVIEW
In this research, the data used are secondary data ob-
tained from several agencies in Pasuruan Regency in-
cluding Department of Health, Department of Popu-
lation and Civil Registration, Department of Environ-
ment, Department of Irrigation, and Meteorological,
Climatological, and Geophysical Agency. The scope
of the research is limited to 21 sub-districts in Pasu-
ruan Regency which are at altitudes below 1000 masl
in a span of four years (2015-2018). The steps of anal-
ysis can be seen in Figure 2.
Figure 2: Steps of analysis.
Based on Figure 2, The steps of analysis in this
study can be described as follows:
1. Conducting a descriptive analysis of the variables
used in the study.
2. Performing multicollinearity test on all predictor
variables (X) using the sample correlation coeffi-
cient.
3. Selecting the Panel Data regression model by con-
ducting the Chow Test, Hausman Test, and La-
grange Multiplier Test
ICASESS 2019 - International Conference on Applied Science, Engineering and Social Science
228
4. Estimating the Panel Data regression model with
the approach chosen in step 3.
5. Carrying classic assumption tests which include
normality test, Heteroscedasticity test, and auto-
correlation test.
6. Testing the significance of the regression parame-
ters through Simultaneous Test (f-test) and Partial
Test (t-test).
7. Making conclusions and suggestions.
The variables which are suspected of having an
effect on DHF are presented in Figure 3.
Figure 3: Variables assumed to influence dhf.
4 RESULTS AND DISCUSSION
4.1 Descriptive Analysis
The first step in this research is to do a descriptive
analysis that is useful to know the characteristics of
each research variable. Descriptive analysis used in-
cludes average, maximum, and minimum values. The
results of the descriptive analysis are shown in Figure
4.
Figure 4: Descriptive statistics of research variables.
Based on Figure 4, the average number of DHF
patients (Y) in 21 sub-districts of Pasuruan in this
past four years (2015-2018) is 23 people. Lumbang
and Puspo sub-districts have the least number of pa-
tients (none) while Bangil has the highest number of
patients (115 people). The average waste (X
1
) that
is not transported is 54.26 where Kraton becomes the
sub-district that has the most non-transported waste
reaching up to 95.12 dam
3
. In contrast, Bangil has
the least non-transported waste which is only 11.04
dam
3
. The range of maximum and minimum values
of the waste volume that is not transported is very
high. This indicates that the waste services are still
focused on certain sub-districts or in other words, not
comprehensive in all sub-districts. The average num-
ber of rainy days (X
2
) is 110 where Purwosari has the
most number of rainy days reaching up to 161 days
of rain in a year. Meanwhile, the sub-district with
the least number of rainy days is Gondangwetan and
Winongan, 71 days a year. On the other hand, the av-
erage number of health facilities (X
3
) is 53. The high-
est number of health facilities is owned by Pandaan
sub-district (135 units) while the least health facili-
ties are owned by Puspo (13 units). The average tem-
perature (X
6
) is 26.32°C. It is reported that the high-
est temperature occurs in Beji reaching up to 27.23°C
while the lowest temperature is in Purwodadi reach-
ing up to 22.96°C. The average number of educated
population (X
5
) is 253 where the area with the highest
number of educated people is Bangil (712 people per
a 10,000 population) while the lowest number of edu-
cated people is Lekok (75 people per a 10,000 popu-
lation). The average population density (X
6
) is 1,631.
The most densely populated area is Pohjentrek with a
population density of 2,954 people/km
2
whereas the
lowest population density can be found in Lumbang,
only 325 people/km
2
. The range between the max-
imum and minimum population density is very big.
This means that the population distribution is uneven
or concentrated in certain sub-districts. Last but not
least, the average value of Rainfall (X
7
) is 1,640.41
where the area with the highest rainfall intensity is
Purwosari reaching up to 3,309 mm in a year. Mean-
while, the area with the lowest rainfall intensity is
Winongan, only 707 mm in a year.
4.2 Multicollinearity Test
Multicollinearity test is applied to see the correla-
tion of each predictor variable used in the regression
model. The existence of perfect multicollinearity can
cause many predictor variables do not have a signifi-
cant effect although the coefficient of determination
is high. To detect multicollinearity, the correlation
coefficient method is used (r). The results of multi-
collinearity test are presented in Figure 5.
Figure 5: Correlation coefficient.
From Table 4, the value of r in each column is
< 0.85 so that all predictor variables are free from
multicollinearity problems.
Data Panel Modelling with Fixed Effect Model (FEM) Approach to Analyze the Influencing Factors of DHF in Pasuruan Regency
229
4.3 Selection of Panel Data Regression
Model
To choose a model that fits the research data, re-
searchers can conduct the Chow Test, Hausman Test,
and Lagrange Multiplier Test. Chow Test is used to
determine whether or not the differences between the
characteristics of the districts are seen from the inter-
cept on each regression model. From the Chow Test,
it is obtained that F
count
= 5.618 > F
0.05(20;56)
= 1.76
dan p value = 0.0000 0.05 with a level of reli-
ability by 95%. This means that hypothesis H0 re-
jected, there are differences in the regional charac-
teristics presented in the DHF Panel Data regression
equation model of Pasuruan Regency so that the ap-
propriate model must be FEM.
Moreover, the Hausman Test is performed to test
whether or not there is a correlation between the er-
ror component in the model and its predictor vari-
able. The Hausman Test results showed that W
count
=
79.223 > χ
2
(005;7)
= 14.07 and p value = 0.0000
0.05 with a level of reliability by 95%. It can be said
that the hypothesis H0 is rejected. This points out
that there is no correlation between the error com-
ponent predictor variables on Pasuruan DHF Panel
Data regression equation. Therefore, the correspond-
ing model is FEM.
Because of the two tests above refer FEM as the
appropriate model, it is not necessary to do the La-
grange Multiplier Test.
4.4 Model Estimation with FEM
Approach
Here is the estimation of Panel Data regression model
formed by using FEM approach.
ˆ
γ
it
=
ˆ
α + 1.785X
1
it
+ 0.186X
2
it
+ 1.470X
3
it
+ 32.267X
4
it
+ 3.186X
5
it
0.177X
6
it
0.014X
7
it
(13)
ˆ
α
i
is the intercept that has different values in each
region. This value will distinguish the prediction of
the number of DHF patients in between sub-districts.
Because each region has different characteristics, the
characteristic differences are represented by an inter-
cept variable in the fixed effect Panel Data regression
model. The intercept value for each district is shown
in Figure 6.
Equation 13 shows that the increase in variable Y
is influenced by the increase in variable X
1
, X
2
, X
3
,
X
4
, and X
5
. Meanwhile, the increase in variable X
6
and X
7
will affect the decrease in variable Y.
Figure 6: Estimated intercept value of each region.
4.5 Classical Assumptions Test
After the estimation of the Panel Data regression
model is obtained, the next step is to carry out the
classical assumption tests which include the Normal-
ity Test, Heteroscedasticity Test, and Autocorrelation
Test.
The Normality Test is done by using the Jarque-
Bera formula. From this test, it is obtained JB =
0.372 < χ
2
(005;2)
= 5.991 and p value = 0.83 > 0.05
with a level of reliability by 95%. By that, hypothesis
H
0
is accepted meaning that the error is normally dis-
tributed. Then, a Heteroscedasticity Test is performed
to see whether the error variant has a constant nature
or not. Glejser Test is used for the Heteroscedasticity
Test and the values obtained can be seen in Figure 7
below.
Figure 7: Simultaneous test result.
Based on Figure 7, the p-value for each predictor
variable is > 0.05 making the hypothesis H
0
accepted.
This explains that the error in the model does not con-
tain heteroscedasticity or is constant.
The last classical assumption test is the Autocor-
relation Test that is done by using the Durbin Wat-
son test method. It is obtained that DW = 2.056.
ICASESS 2019 - International Conference on Applied Science, Engineering and Social Science
230
The value is in the area of Hypothesis H
0
which is
d
u
= 1.8291 DW = 2.101 4d
u
= 2.1709, mean-
ing that the error in the model is free from the prob-
lem of autocorrelation.
4.6 Significance of Parameters Test
After the model has passed all classical assumption
tests, the next step is to do a simultaneous test and par-
tial test to determine the effect of predictor variables
on the response variable both simultaneously and par-
tially. The simultaneous test results are presented in
Figure 8.
Figure 8: Simultaneous test result.
As shown in Figure 8, the value of F
count
=
8.528 > F
(0.05;27;56)
= 1.686 and p value = 0.000 <
0.05 and R
2
= 0.804 with a level of reliability by
95%. This clarifies that hypothesis H0 is rejected,
the predictor variable simultaneously affects the re-
sponse variable. The value of R
2
= 0.804 shows that
the seven predictor variables can affect the number of
DHF patients in Pasuruan by 80.4% while the remain-
ing 19.6% are influenced by other variables outside
the model.
After that, a partial test is carried out. The results
of the partial test are shown in Figure 9.
Figure 9: Partial test results.
As presented in Figure 9 with a reliability level
of 95%, there are six variables that have a significant
effect on the response variable (p value < 0.05),
namely variable X
1
, X
3
, X
4
, X
5
, X
6
, and X
7
while the
variable X
2
does not have a significant influence on
the response variable (p value > 0.05).
5 CONCLUSION AND
SUGGESTION
Based on the results of the analysis, it can be con-
cluded that the Panel Data regression model with
Fixed Effect Model approach can explain the effect of
predictor variables on the response variable. There-
fore, the model equation is:
ˆ
γ
it
=
ˆ
α + 1.785X
1
it
+ 0.186X
2
it
+ 1.470X
3
it
+ 32.267X
4
it
+ 3.186X
5
it
0.177X
6
it
0.014X
7
it
(14)
From the Equation 14 above, the value of R
2
=
0.804 shows that the percentage of all seven predic-
tor variables able to affect the number of DHF pa-
tients in Pasuruan by 80.4% while the other 19.6% are
influenced by other variables outside this study. Of
the seven predictor variables, there are six variables
that significantly influence the increase in the num-
ber of DHF patients such as waste volume, number of
health facilities, temperature, high-educated popula-
tion, population density, and rainfall.
Last but not least, the Department of Health of Pa-
suruan Regency is suggested to plan the programs and
activities to control and eradicate future DHF by fo-
cusing on the variables that have a significant effect
on DHF patients.
In the future the model can be developed into an
application that can be utilized by government or in-
terested organizations.
REFERENCES
Ariani, P. M. (2018). Analisis Faktor yang Berpengaruh
Terhadap Pencegahan Penyakit DBD di Provinsi Jawa
Tengah Menggunakan Regresi Binomial Negatif. Tu-
gas Akhir. Universitas Islam Indonesia, Yogyakarta.
Arsin, A. A. (2013). Epidemiologi demam berdarah dengue
(DBD. Masagena Press, di indonesia.Makassar.
Baltagi, B. H. (2010). Econometric Analysis of Panel Data.
John Wiley & Sons, Ltd, Sussex, 4th edition. west edi-
tion.
Greene, W. (2000). Econometric Analysis. Prentice Hall,
Inc, 4th edition.
Gujarati, D. (2004). Basic Econometrics. The McGraw-Hill
Companies, New York, 4th edition.
Gujarati, D. and P, D. C. (2010). Dasar-dasar Ekonometrika
Edisi 5. Salemba Empat, Jakarta.
Hasirun (2016). Model Spasial Faktor Risiko Kejadian De-
mam Berdarah Dengue di Provinsi Jawa Timur Tahun
2014. Tesis. Universitas Airlangga, Surabaya.
Hsiao, C. (2003). Analysis of Panel Data. Cambridge Uni-
versity Press, New York.
I., A. W., Ratnasari, V., and Wibowo, W. (2017). Analisis
faktor yang berpengaruh terhadap tingkat penganggu-
ran terbuka di provinsi jawa timur menggunakan re-
gresi data panel. Jurnal Sains dan Seni ITS, 6(1).
Kemenkes, R. (2010). Jendela Epidemiologi. Pusat Data
dan Surveilans Epidemiologi Kementerian Kesehatan
Republik Indonesia, Jakarta.
Data Panel Modelling with Fixed Effect Model (FEM) Approach to Analyze the Influencing Factors of DHF in Pasuruan Regency
231
Kemenkes, R. (2017). Profil Kesehatan Indonesia Tahun
2017. Setjen Kementerian Kesehatan Republik In-
donesia, Jakarta.
Martha, Z. (2015). Pemodelan Regresi Data Panel Pada
Kasus Jumlah Penderita Demam Berdarah Dengue
(DBD) di Kota Bogor.Tesis. Institut Pertanian Bogor,
Bogor.
Rasmanto, M. (2016). Model prediksi kejadian demam
berdarah dengue (dbd) berdasarkan unsur iklim di
kota kendari tahun 2000-2005. Jurnal Ilmiah Maha-
siswa Kesehatan Masyarakat, 1.
Rustiani, N. N. (2017). Pemodelan penyebaran penyakit
demam berdarah dengue (dbd) dengan pendekatan re-
gresi linear berganda. In Prosiding Seminar Nasional
Sainstek 2017. Fakultas MIPA Universitas Udayana.
Sutikno, B., F.k, A., and D, O. (2017). Penerapan regresi
panel data komponen satu arah untuk menentukan
faktor-faktor yang mempengaruhi indeks pembangu-
nan manusia.jurnal matematika integratif unpad.
Widarjono, A. (2013). Ekonometrika Teori dan Aplikasi
Untuk Ekonomi dan Bisnis. Ekonosia FE UII, Yo-
gyakarta.
ICASESS 2019 - International Conference on Applied Science, Engineering and Social Science
232