Content uploaded by Anik Djuraidah
Author content
All content in this area was uploaded by Anik Djuraidah on Feb 06, 2023
Content may be subject to copyright.
AIP Conference Proceedings 2662, 020010 (2022); https://doi.org/10.1063/5.0111046 2662, 020010
© 2022 AIP Publishing LLC.
The modeling of poverty in Java Island 2012
until 2018 using dynamic spatial panel
Cite as: AIP Conference Proceedings 2662, 020010 (2022); https://doi.org/10.1063/5.0111046
Published Online: 22 December 2022
Asti Nuswantari, Anik Djuraidah and Utami Dyah Syafitri
The Modeling of Poverty in Java Island 2012 until 2018
using Dynamic Spatial Panel
Asti Nuswantari1, a), Anik Djuraidah1, b), and Utami Dyah Syafitri1, c)
1Department of Statistics, IPB University, Bogor, 16680, Indonesia
a) astinuswantari@gmail.com
b) Corresponding author: anikdjuraidah@apps.ipb.ac.id
c) utamids@apps.ipb.ac.id
Abstract. Three provinces in Java had poverty percentages above the national average at 10.19% in 2020. One of the
methods to identify the factors that affect poverty is the dynamic spatial panel model. This research aims to determine the
best dynamic spatial panel model based on several spatial weighting matrices and identify the variables that affect the
percentage of poverty in Java Island. The weighted matrices compared in this research were queen contiguity, two nearest
neighbors, inverse distance, and exponential distance. The data used in this research were the percentage of poverty as a
response variable, meanwhile the gross domestic product, the percentage of education completed by elementary school,
the literacy rate, expenditure per capita, percentage of the productive age population, expected of length school, the average
of length school, and percentage of health centers as predictor variables observed in 2012-2018 on Java Island. The methods
used were data exploration, dependency test, and dynamic spatial panel analysis. The spatial autocorrelation test results
show a spatial dependency on the dependent variable and predictors, so the Durbin spatial model was used. The best model
was a dynamic spatial Durbin panel with a fixed effect and two nearest neighbor weighting matrices. Predictors that
significantly affect poverty were the literacy rate, percentage of the productive age population, and the percentage of health
centers. The marginal effect shows that an increase of gross domestic product, the percentage of education completed by
elementary school, the literacy rate, and expenditure per capita could reduce the percentage of the city in Java.
INTRODUCTION
Poverty is a problem in almost all countries in the world. Therefore, world governments agree that ending poverty
is sustainable development’s main goal. This goal is considered important because it becomes the foundation behind
various other development goals such as energy, tourism, education, etc. According to the [1], Java Island is the island
with the most considerable economic contribution in Indonesia, namely 59.03% in 2019. Even so, the poverty rate is
also relatively high. This is indicated by the percentage of poverty in three provinces on the island of Java, namely
Central Java, East Java, and Yogyakarta Province, which is still above the national average percentage of 10.19% in
2020. Therefore, research is needed to identify the factors that influence the percentage of poverty in Java.
Modeling the percentage of poverty requires suitable methods and variables in order to provide the correct
predictions. Previous research which concluded that the Human Development Index (HDI) variable affects poverty in
Java [11]. Research about poverty in Java in 2009-2016, and the result is that the gross regional domestic product
variable has a significant effect on poverty [5].
International Conference on Statistics and Data Science 2021
AIP Conf. Proc. 2662, 020010-1–020010-13; https://doi.org/10.1063/5.0111046
Published by AIP Publishing. 978-0-7354-4249-8/$30.00
020010-1
Panel data analysis is an analysis that includes cross section and time series data. This analysis can be dynamic in
order to capture the existing time trends. That means the response variable is not only influenced by the explanatory
variables, but also the response variable in the previous period. The dynamic panel model must satisfy the assumption
that the individual units are independent of each other. However, sometimes there are relationships between individual
units, so that this assumption is violated. According to Tobler's first law, the closer the distance between two regions,
the stronger the relationship [14]. That results in a relationship between geographically adjacent areas which can
indicate spatial dependencies. Therefore, the panel data model is assumed to be more accurate in modeling poverty if
it adds dynamic and spatial elements so that a dynamic spatial panel model appears. This study aims to determine the
best dynamic spatial panel model based on several spatial weighting matrices and identify the variables that affect the
percentage of poverty in Java Island.
METHODOLOGY
Data
This study uses secondary data obtained from the results of BPS publications from 2012 to 2018. The unit of
observation in this study is 118 districts/cities on the island of Java. The variables used are shown in Table 1 as follows:
TABLE 1. Variables Used
Code
Variables
Unit
References
MISKIN
Percentage of poverty
Percent
PDRB
Gross domestic product
Billion rupiah
[5, 11]
PSD
Percentage of education completed by elementary
school
Percent
[11]
AMH
The literacy rate
Percent
[4]
PKP
Percentage of expenditure per capita
Percent
[11, 13]
PKU
Percentage of the productive age population
Percent
[8]
HLS
Expected of length school
Year
[8]
RLS
Average of length school
Year
[8]
PKM
Percentage of health centers
Percent
[11]
Data Analysis Procedure
The stages carried out in this research are as follows:
1. Exploring data to see the general characteristics of the data.
2. Choose a panel data model with the following steps:
a. Choose the model used between the common effect model or the fixed-effect model using the Chow test [2].
H0: Common effect model
H1: Fixed effect model
Test statistics: ܨൌሺܴܴܵܵെܷܴܵܵሻ ሺݏെͳ
Τሻ
ሺܷܴܵܵሻ ሺݏݐെݏെሻ
Τሺͳሻ
020010-2
where RRSS (Restricted Residual Sums of Square) is the residual sum of square resulting from the common
effect model estimation and URSS (Unrestricted Residuals Sums of Square) is the residual sum of square
resulting from the fixed-effect model estimation,ܰ is the number of individual units, ܶ is the number of time
series units and is the number of explanatory variables. The test statistic rejects H0 if ܨܨேିଵǡேሺ்ିଵሻିǤ
If reject H0 (the fixed effect model remains selected), then proceed to step (b), if accept H0 (selected common
effect model), then proceed to step (3).
b. Choose model between the fixed effect or random effect model using the Hausman test [2].
H0: Random effect model
H1: Fixed effect model
Test statistics:
݉ଵൌ
ෝᇱሾܸܽݎሺ
ෝሻሿିଵ
ෝሺʹሻ
ෝൌࢼ
ௗെࢼ
௫ௗ ሺ͵ሻ
Where ࢼ
ௗ is the estimator of the random effect model and ࢼ
௫ௗ is the estimator of the explanatory
variables parameter of the fixed effect model. The test statistic rejects H0, if ݉ଵ߯ሺǡఈሻ
ଶ with is the number
of explanatory variables.
3. Identify the model with the following stages:
a. Forming a spatial weighting matrix using queen contiguity, k-nearest neighbor, inverse distance, and
exponential distance. The weighting matrix used in this study is as follows:
x Queen Contiguity defines the value 1 if the regions share common side or vertex with the region of
interest and 0 otherwise.
x K-nearest neighbor defines the value of 1 if regions i and j are neighbors and 0 otherwise. This study
used two nearest neighbors.
x Inverse distance is a weighting matrix based on distance. The formula for calculating this matrix is:
ݓ ൌͳ݀
ൗሺͶሻ
x The exponential distance defines:
ݓ ൌ൛െ݀ൟሺͷሻ
b. Pesaran's Cross-sectional Dependency (CD) test is used to determine the spatial autocorrelation of the
response variables. If the response variable has a spatial autocorrelation, then the SAR model was used.
According to [10], Pesaran's CD test hypothesis is:
H0 : there is no spatial correlation between individuals
H1: there is a spatial correlation between individuals
The test statistics is:
020010-3
ܥܦൌඨʹܶ
ܰሺܰെͳሻቌ ߩ
ே
ୀାଵ
ேିଵ
ୀଵ ቍሺሻ
where:
T = number of time units
N = number of individual units
ߩ = correlation of residuals on i-th and j-th individual units
c. Calculating Moran Index test to determine the spatial autocorrelation in the explanatory variables. If the
response and explanatory variables have spatial autocorrelation, then the SDM model was used. The
hypothesis of the Moran Index test is:
x H0 : ܫൌͲ ; There is no spatial autocorrelation.
x H1 : ܫ്Ͳ (There is spatial autocorrelation)
Moran statistics test are [3]:
ܫൌܰ
ܹσσݓሺݖെݖҧሻ൫ݖെݖҧ൯
ே
ୀଵ
ே
ୀଵ σሺ
ݖെݖҧሻଶ
ே
ୀଵ ሺሻ
with
ܰ = number of individual units
ݖ = the observed value of a variable in the individual-i
ݖҧ = the average of the observations of a variable
ݓ = spatial weighting matrix that consists of i-th row and j-th column
ܹ = sum of the elements of the spatial weighting matrix
d. Autoregressive test AR(1) on response variable using ACF and PACF plots. The data defines the AR(1)
condition if the ACF plot has exponential decay (tails off) and the PACF plot has cut off after the first order
[9]. If the response variable has the AR(1) condition, then a dynamic effect is added to the model.
4. Analyzing dynamic spatial panel data using the Spatial Corrected Arellano-Bond estimation method.
5. Checking the model’s assumptions such as normality, homoscedasticity, and independence of the residuals.
6. Evaluation of the goodness of the model based on the coefficient of determination ሺܲݏ݁ݑܴ݀ଶሻ. The best model
has the highest score. The formula for calculating ݏ݁ݑܴ݀ଶ is:
ܲݏ݁ݑܴ݀ଶൌͳെܴܵܵ
ܶܵܵ ሺͺሻ
where RSS = sum of the squares regression and TSS = the sum of squares total.
7. Calculating the marginal effect of the best model. The marginal effect or spillover measures the magnitude of
the impact of changes in the response variable in a region due to changes in explanatory variables in other region
[15]. The marginal effect is calculated by deriving the response variable in the i-th region to the explanatory
variables in the j-th region which can be written as ߲ݕ߲ݔ
൘.
8. Interpreting the results and conclusions.
020010-4
RESULT AND DISCUSSION
Data Exploration
The graph of the percentage of poverty in Java from 2012 to 2018 is presented in Figure 1. The percentage of
poverty in Java tends to decrease every year. The percentage of poverty varies in the range of 5% to 16%. Banten and
DKI Jakarta have twice the poverty percentage compared to other provinces because the location of the province in
the center of government.
FIGURE 1. The Percentage of Poverty in Java Island 2012─2018
Figure 2 is a map of the distribution of the percentage of poverty in Java. Based on Figure 2, the regions with the
lowest poverty percentages are Banten Province and DKI Jakarta. The areas with the highest percentage of poverty
are Central Java, Sampang, Sumenep, Bangkalan, Probolinggo in East Java, and Kulon Progo, Yogyakarta. Figure 2
also shows indications of spatial dependencies because neighboring districts have relatively similar poverty
percentages. The relatively similar color indicates this.
Province
Year
020010-5
FIGURE 2. Poverty Map in Java from 2012 to 2018
Before making a regression model, the response and explanatory variables must have a linear relationship. Figure
3 shows that the data is not linear, especially the plot between MISKIN and PDRB variables, so it needs to be
transformed. This study uses the natural logarithm transformation on the MISKIN and PDRB variables. The pattern
of the data after the transformation becomes relatively linear which can be seen in Figure 4.
FIGURE 3. Scatter Plot of Variables Before Transformation
020010-6
FIGURE 4. Scatter Plot of Variables After Transformation
Multicollinearity is a condition when one explanatory variable has a strong relationship with another explanatory
variable. Multicollinearity examination is checked from the Variance Inflation Factor (VIF). Multicollinearity occurs
if the VIF value is more than ten [6]. Based on Table 2, only the RLS variable has multicollinearity because it has VIF
value more than 10. Figure 4 also shows that the correlation between the RLS variable and other variables was very
high, so the RLS variable is not included in the model. Meanwhile, the VIF value in other variables is less than 10,
which means no multicollinearity.
TABLE 2. VIF of Transformed Variables
Year
PDRB
PSD
AMH
PKP
PKU
HLS
RLS
PKM
2012
2.5746
4.3392
3.0707
6.1520
2.9342
2.1433
11.5458
1.2713
2013
2.4384
1.3585
2.0567
6.0940
3.1053
2.2532
9.4790
1.2907
2014
2.4147
3.2754
2.8249
5.3538
2.8812
2.2107
11.0025
1.3287
2015
2.3384
4.3414
2.2948
6.1486
1.7538
2.3339
11.6874
1.2255
2016
2.5458
2.8689
2.4852
5.9854
2.7458
2.6623
10.6515
1.2907
2017
2.8481
4.7396
3.1655
4.9889
3.1202
2.9261
10.5562
1.5557
2018
2.8846
3.7807
2.1300
3.8764
3.0916
2.7997
10.2523
1.7105
2012-2018
2.3711
2.5626
1.9033
3.7495
2.0269
1.9741
7.8847
1.1836
Model Identification
This study uses secondary data obtained from the results of BPS publications from 2012 to 2018. The unit of
observation in this study is 118 districts/cities on the island of Java. The variables used are shown in Table 1 as follows:
Panel Model Selection
Panel data can form three models: common effect, fixed effect, and random effect model. Panel data models were
selected using the Chow and Hausman tests [2]. The Chow test is a test to select common effect or fixed effect model.
Meanwhile, the Hausman test chooses random effect or fixed effect model. Based on Table 3, the decision of the
Chow and Hausman tests is rejecting H0, so the fixed effect model was chosen.
020010-7
TABLE 3. The Result of Chow Test and Hausman Test
Spatial Autocorrelation Test
Spatial autocorrelation test was conducted to determine the existence of spatial dependence on panel data. This
study examines the existence of spatial autocorrelation using Pesaran's Cross-sectional Dependency (CD) test and
Moran's Index. The Lagrange Multiplier (LM) test was used to test the spatial dependence on the response variable.
The LM test is used when the number of time units (T) is more than the number of individual units (N). However,
Pesaran's CD test is used if the number of time units (T) was less than the number of individual units (N) [7]. This
study has 118 districts or cities as individual units with a time unit of 7 years. Therefore, the test that will be used is
Pesaran's CD test. The CD test statistic obtained was 56,198 and the p-value was ʹǤʹൈͳͲିଵ. Therefore, the poverty
panel data in Java Island has a spatial dependence on the response variable. It shows that poverty in a city in Java is
influenced by explanatory variables and poverty in other cities.
The Moran index aims to examine the spatial dependence on explanatory variables. Table 4 shows the Moran
index of each explanatory variable. Based on Table 4, the variables PDRB, PSD, AMH, PKP, PKU, and HLS have a
significant Moran Index for all weighting matrices at the 5% significance level. That means these variables indicate
the spatial dependencies on the cities in Java Island. Meanwhile, the PKM variable has p -value more than 5%
significance level, meaning that there are no spatial dependencies of the percentage of health center. Because the
response variable and some of the explanatory variables have spatial dependencies, the spatial Durbin model was used.
TABLE 4. Moran Index for Each Variable
Variable
Moran Index
Queen
Contiguity
2-Nearest
Neighbor
Distance
Inverse
Exponential
Distance
PDRB
0.3936*
0.3403*
0.1360*
0.0994*
PSD
0.5145*
0.5445*
0.1965*
0.1998*
AMH
0.5009*
0.7165*
0.2157*
0.2785*
PKP
0.4532*
0.4807*
0.1246*
0.0611*
PKU
0.6128*
0.5667*
0.1978*
0.0969*
HLS
0.3203*
0.3705*
0.0678*
0.0306*
PKM
0.0365
-0.0766
-0.0124
-0.0185
*significant at 5% significance level
First Order Autoregressive Test
Before forming a dynamic model, the data needs to be tested to see the influence of dynamic effects. The test
can be used is the first order autoregressive AR(1) test by looking Autocorrelation Function (ACF) and Partial
Autocorrelation Function (PACF) plots. Based on Figure 5, it can be seen that exponential decay of the ACF plot (tails
Test
Test Statistics
p-value Selected models
Chow
298.260
Ǥൈି
Fixed effect model
Hausman
35.236
Ǥൈି
Fixed effect model
020010-8
off) and the PACF plot has been cut off after the first order. This condition is called AR(1). Thus, the percentage of
poverty in a year is influenced by the percentage of poverty in the previous year. Therefore, dynamic spatial panel
model was used.
FIGURE 5. Plots of ACF and PACF
Dynamic Spatial Panel Modeling
This study uses secondary data obtained from the results of BPS publications from 2012 to 2018. The unit of
observation in this study is 118 districts/cities on the island of Java. The variables used are shown in Table 1 as follows:
This study used dynamic spatial panel data model. Model estimation used the Spatial Corrected Arellano-Bond
method. The selected panel model is fixed effect model and spatial model is the Durbin spatial model (SDM) with
additional dynamic effects. The general form of the model is
ݕ௧ ൌ߬ݕǡ௧ିଵߩࢃଵݕ௧࢞௧ࢼࢃଶ࢞௧ࢽߤߝ௧ ሺͻሻ
The next step is to determine the spatial weighting matrix by comparing the coefficients of determination. The
results of estimating model parameters using a standardized row weighting matrix can be seen in Table 5. The higher
the ܴଶ value, the better model in explaining the diversity. Based on Table 5, the highest value is the model with the
two nearest neighbor matrix, equal to 0.8614. It means covariates of poverty used were able to explain the diversity
in the models of 86.14%.
TABLE 5. Dynamic Spatial Panel Model Parameter Estimator
Coefficient
Spatial Weighting Matrix
Neighbor
step queen
Two Nearest
Neighbors
Distance
Inverse
Exponential
Distance
߬
0.7043*
0.5823*
0.7194*
0.7313*
ߩ
0.7795*
0.5069*
1.8908*
1.7840*
ߚ
ln(PDRB)
-0.0518
-0.0563
-0.0446
-0.0466
PSD
0.0001
-0.0002
0.0005*
0.0004*
AMH
-0.0015*
-0.0019*
-0.0015*
-0.0020*
PKP
-0.0001
-0.0014
-0.0015
-0.0014
PKU
0.0000
0.0023*
0.0015
0.0012
HLS
-0.0052
-0.0057
-0.0094
-0.0105
PKM
0.0182*
0.0384*
-0.0024
-0.0020
020010-9
TABLE 5. Dynamic Spatial Panel Model Parameter Estimator
Coefficient
Spatial Weighting Matrix
Neighbor
step queen
Two Nearest
Neighbors
Distance
Inverse
Exponential
Distance
ߛ
ln(PDRB)
0.1226*
-0.0219
0.9893*
0.7883*
PSD
0.0001
-0.0004
0.0133*
0.0172*
AMH
0.0037*
0.0031*
0.0050*
0.0034
PKP
-0.0004
-0.0008
0.0103*
0.0111*
PKU
0.0000
-0.0014
-0.0123*
-0.0105*
HLS
0.0206
0.0329*
-0.1011*
0.0059
ܲݏ݁ݑܴ݀ଶ
0.7551
0.8614
0.3004
0.0682
*significant at 5% significance level
Based on Table 5, the coefficient of ߬ൌͲǤͷͺʹ͵ shows that the percentage of poverty is influenced by the
percentage of poverty in the previous year. This statement can be interpreted if the percentage of poverty increases by
one percent in a district/city, then will increase the percentage of poverty by ݈݊ሺͲǤͷͺʹ͵Ψሻ in the following year. Then,
the coefficient of ߩൌͲǤͷͲͻindicates a spatial dependence on the percentage of poverty between districts/cities. The
percentage of poverty will increase by ݈݊ሺͲǤͷͲͻΨሻpercent in a district/city, if the percentage of poverty in the two
nearest neighbors of the district/city increases by one percent.
The variables that significantly affect the percentage of poverty at the 5% significance level are the AMH, PKU,
and PKM variables. The literacy rate (AMH) is the proportion of the population aged 15 years and over who can read
and write. The literacy rate is related to education. A well educated person tends to have a better income than an
uneducated person. The literacy rate significantly affects poverty in Java [12]. Another significant variable is the
percentage of the productive age population (PKU). The productive age population is the population aged 15-64 years.
The population of this age plays an essential role in education, government, economy, social, hea lth, and others. The
increase in the productive age population should be in line with the improvement in the quality and quality of its
human resources. Another variable that affects poverty is the percentage of health center (PKM). PKM variable and
the percentage of poverty have a positive relationship because the estimated coefficient is positive. However, it cannot
be interpreted that the more health centers in an area, the poverty will increase. It is better to distribute the ratio of
health centers per population so that each region has sufficient health centers and adequate medical personnel.
Table 6 shows the marginal effects in the direct, indirect, and total effects of each explanatory variable. Based on
Table 6, the PDRB variable has the highest direct and indirect effect. The direct effect of the PDRB variable is 0.0688,
which means if PDRB in a region increases by one billion rupiahs, then poverty in that region will decrease by 0.0688
percent if the other values are constant. Meanwhile, the indirect effect means that if the PDRB in one area increases
by one billion rupiahs, then poverty in other areas will decrease by 0.0848 percent if the other variables are constant.
The marginal effects of another variable are presented in Table 6.
TABLE 6. Marginal Effect of The Explanatory Variable
Variable
Direct Effect
Indirect Effect
Total Effect
PDRB
-0.0688
-0.0848
-0.1537
PSD
-0.0003
-0.0008
-0.0011
AMH
-0.0013
0.0038
0.0024
PKP
-0.0016
-0.0026
-0.0043
020010-10
TABLE 6. Marginal Effect of The Explanatory Variable
Variable
Direct Effect
Indirect Effect
Total Effect
PKU
0.0022
-0.0007
0.0015
HLS
0.0014
0.0543
0.0558
PKM
0.0428
0.0349
0.0778
Residual Assumption
This study uses secondary data obtained from the results of BPS publications from 2012 to 2018. The unit of
observation in this study is 118 districts/cities on the island of Java. The variables used are shown in Table 1 as follows:
In the regression model, the assumption normality, homoscedasticity, and independence of the residuals must be
required. First, residual normality can be examined exploratively by looking at histograms and plot of normal
quantiles. Figure 6 shows the distribution of the data against the expected normal distribution because it tends to form
a straight line.
FIGURE 6. Plot of Normal Quantiles of Residuals
The assumption of homogeneity of residual can be checked by using a plot between the predicted response
variables and the residuals. That assumption is fulfilled because the residual bandwidth is relatively the same for all
predicted values (Figure 7).
FIGURE 7. Plot between Predicted Value and Residual
The results of the examination of the model assumptions are presented in Table 7. The assumption of normality
of the residuals can be checked using the Kolmogorov-Smirnov test, while the homogeneity of residual variance used
the Breusch–Pagan test. Because of the p-value more than 5% significance level for both tests, the assumptions
normality, and homogeneity of residuals are fulfilled.
Normal Quantile
Residual
y-hat
Residual
020010-11
TABLE 7. Model assumption
Assumption
Test Statistics
p-value
Conclusion
Normality of residual
0.0389
0.1633
fulfilled
Homoscedasticity
13.2482
0.0663
fulfilled
The assumption of residual independence can be checked using the Durbin-Watson test. Because of ݀ݑሺே்ǡሻ ൌ
ͳǤ͵ͻ൏ൌͳǤͶͶ͵൏Ͷെ݀ݑሺே்ǡሻ ൌʹǤ͵ͳ, then do not reject H0. That means that the residuals are independent,
so it can be said that the assumption of independence is fulfilled.
CONCLUSION
The best dynamic spatial panel model on poverty percentage data in Java Island from 2012 to 2018 is the dynamic
Durbin spatial panel model with a fixed effect using two nearest neighbors matrix. The variables that affect the
percentage of poverty include the literacy rate, the percentage of the productive age population, and the percentage of
health center per 1000 population. The model also shows that the percentage of poverty is influenced by the percentage
of poverty in the previous period. In addition, there is a spatial dependence on the percentage of poverty between the
cities in Java Island and the explanatory variables.
REFERENCES
1. [BPS] Badan Pusat Statistik, Data dan Informasi Kemiskinan Kabupaten/Kota Tahun 2020 (Badan Pusat
Statistik, Jakarta, 2020).
2. H. Baltagi, Econometrics Ed ke-5 (Springer, New York, 2011)
3. M. Fischer, and J. Wang, Spatial Data Analysis: Models, Methods and Techniques (Springer Science &
Business Media, Berlin, 2011)
4. D. Fitri, W. D. Safitri, and C. Chairunnas, Analisis pengaruh penduduk usia produktif yang tidak bekerja,
zakat, pertumbuhan ekonomi, dan indeks pembangunan manusia terhadap tingkat kemiskinan kabupaten/kota
di Provinsi Aceh, Ekobis Syariah, 1(2): 1-10 (2021)
5. R. Giovanni, Analisis pengaruh PDRB, pengangguran dan pendidikan terhadap Tingkat kemiskinan di Pulau
Jawa tahun 2009-2016, Economics Development Analysis Journal, 7(1):23-31 (2018)
6. N. Gujarati, and D. C. Porter, Basic Econometrics Ed ke-5 (The McGraw-Hill Companies Inc, New York,
2009)
7. R. E. D. Hoyos, and V. Sarafidis, Testing for cross-sectional dependence in panel-data models, The Stata
Journal 6(4): 482–496 (2006)
8. A. Jonaidi, Analisis pertumbuhan ekonomi dan kemiskinan di Indonesia, Jurnal Kajian Ekonomi, 1(1):140-
164 (2012)
9. C. Montgomery, C. L. Jennings, and M. Kulahci, Introduction to Time Series Analysis and Forecasting (Wiley
Interscience, Canada, 2008)
10. M. H. Pesaran, General diagnostic tests for cross section dependence in panels University of Cambridge,
Faculty of Economics, Cambridge Working Papers in Economics No 0435 (2004)
11. D. W. Puspita, Analisis determinan kemiskinan di Provinsi Jawa Tengah, JEJAK Journal of Economics and
Policy, 8(1): 100-107 (2015)
12. J. N. I. Sari JNI, and I. Nuraini, Analisis faktor yang mempengaruhi tingkat kemiskinan provinsi di Pulau
Jawa, Jurnal Ilmu Ekonomi JIE, 4(2): 309-323 (2020)
13. Sunandi, I. M. Sumertajaya, A. Djuraidah, A, H. Wigena, and A. Fitrianto, Dynamic spatial Durbin modeling
(Case: Percentage of poor population in Indonesia 2010-2019) In Journal of Physics: Conference Series
1863(1): 012046 IOP Publishing (2021)
14. W. R. Tobler, A computer movie simulating urban growth in the Detroit region, Economic Geography, 46(1):
234-240 (1970)
020010-12
15. S. H. Vega, and J. P. Elhorst, On spatial econometric models, spillover effects, and W, 53rd Congress of the
European Regional Science Association: "Regional Integration: Europe, the Mediterranean and the World
Economy" 27-31 August 2013 (European Regional Science Association/ERSA, Palermo, 2013)
020010-13