Jurnal Bina Praja 10 (1) (2018): 1-12 Jurnal Bina Praja

e-ISSN: 2503-3360 | p-ISSN: 2085-4323

Accreditation Number 735/AU2/P2MI-LIPI/04/2016

http://jurnal.kemendagri.go.id/index.php/jbp/index Stochastic Dominance Analysis for The Poverty in Tabalong Regency 2013 and 2017

Ahmadi Murjani* Statistics of Tabalong Regency Jalan Jaksa Agung Soeprapto No. 82 Tanjung, Tabalong Regency, South Kalimantan, Indonesia

Received: 20 January 2018; Accepted: 18 April 2018; Published online: 12 May 2018

DOI: 10.21787/jbp.10.2018.01-12

Abstract Although its issue could be addressed from of various perspectives over the years, poverty is the object of the government’s policy programs to be alleviated since the Indonesian independence. With the advancement of technology and science in the recent decades, the availability and the completeness of poverty data in Indonesia getting better. In fact, the policy-makers can assess the effectivity of their public-oriented programs easier by effectively utilizing the complete and up-to-date poverty data. However, in various approaches to poverty measurement, the Indonesian poverty data should be accompanied by another approach. This paper aims to evaluate the change of welfare of particular region over a period of several years by using the stochastic dominance method. This method also incorporates the price level impact into its poverty assessment through the extrapolated CPIs. To conduct the measurement, the National

Socioeconomic Survey of Tabalong regency data, as well as the Tanjung city’s inflation in the period from 2013 to Therefore,2017, are employed. this result Thecould results also be indicate the additional that the input welfare for inthe 2017 poverty is better alleviation’s than in evaluation 2013 since in the order distribution to provide in a 2017solid conclusionstochastically about dominates the welfare the changes.distribution in 2013 at the first and second order at any possible level of poverty lines.

Keywords: Tabalong, Stochastic Dominance, Poverty, Welfare, Cumulative Distribution Function

I. Introduction The concept of poverty could be addressed Poverty is a complex problem and can be in many ways. On one standing, it can be seen addressed in many ways because it is highly mixed as a relative subject while others see it as an with every aspect of living; for instance, social, absolute substance (Statistics Indonesia (BPS), economic, culture, etc. In fact, the various efforts 2016). Suyanto (2001) translated poverty not have been being done to overcome the poverty only considers on materials but also abstract problem since the independence of Indonesia. One subjects such as fragility, powerless, etc. While the of the important aspects that should be underlined World Bank (in Statistics Indonesia (BPS), 2016) is the availability of the poverty data, the accurate mentioned poverty as the deprivation of well-being, data to be emphasized. It is mandatory since the MacPherson & Silburn (1998) simplify it with the policy-making process needs the valid and reliable notion of the basic lack of surviving. There is no data. A good poverty data can be utilized to evaluate doubt that the poverty concept differs as well as its the effectiveness of the government’s policy on the measurement. poverty, comparable time to time and interregional, In Indonesia, poverty is examined by Statistics and also able to put the targeting poor people as the Indonesia (hereafter, BPS). In line with Lanjouw’s statement (1998), Indonesia has been joining programs (Statistics Indonesia (BPS), 2016). majority countries to use expenditure approach subjects to receive the benefit of poverty alleviation in calculating poverty. BPS (2009) explained that

* Corresponding Author © 2018 Ahmadi Murjani Phone : +62 852 4888 8473 This work is licensed under the Creative Email : [email protected] Commons Attribution-NonCommercial- 1 ShareAlike 4.0 International License. Indonesia has been providing the poverty data covering the period of 1970 to 1996 (using the old method) and the period of 1996 to 2017 (using the theHuppi CPI deflated into the the urban base and year’s the poverty rural areas, line withand new method). In the recent years, BPS also provided thenmodified multiplied consumer by pricethe purchasing index (CPI) power and differed parity. the poverty data twice a year since 2011. There were some reasons why they examined the The poverty data series can be observed poverty alleviation progress in 1984 and 1987 on the World Bank website (through the World using a different poverty approach compared with Development Indicators (WDI)) and BPS’ website (in Indonesia’s poverty measurements. Two of reasons the Poverty and Inequality section). Furthermore, were about the changing the relative price of WDI also provides the poverty rate using the dollar- goods and services, and rupiah’s devaluation over a-day approach for Indonesia as a comparison. periods. These backgrounds illustrated the poverty BPS provides not only the percentage of poor calculation using real data, rather than the nominal people data on its website but also other poverty indices such as poverty depth index and poverty severity index. Furthermore, those indices are also domesticfigures. Eventually, poverty lines as it isin possible, the different the construction time frame available in time series comprising provinces in andof similar economic research situation by is essential. utilizing the predefined Indonesia. Thus, these series seems appropriate to Indeed, the research conducted by Ravallion draw the poverty alleviation progress in Indonesia. and Huppi in 1991 is likely to be presented in the recent periods in order to provide further measurement in Indonesia, the poverty data comprehensive analysis of poverty comparison over synthesizedAs an officialby BPS statistic stands of among the povertysome the periods in Indonesia. However, there are limited considerations from researchers in Indonesia. published papers focusing on this comparison Marbun & Suryahadi (2009), Susilowati (2010), especially in Indonesia. Ravallion and Huppi utilized and recently Khomsan et al. (2015), delivered some the previously-introduced method by Atkinson (1987) so-called the Stochastic Dominance analysis not only the poverty calculation method but also on the poverty in their research. highlightingfindings in theirthe low research rate of that poverty highly measurement concerning Ultimately, this paper examines the poverty or produced by BPS. Thus, this paper aims to measure welfare comparison to justify whether the poverty indeed declines (increasing welfare) over the periods by using Tabalong regency of South Kalimantan povertythe wealth lines change as well overas the specific poverty period rates produced. using the methodIn regards that considers for some theprevious flexibility issues of highlighting a range of Moreover, this paper uses the Stochastic Dominance the uncertainty in the poverty calculation due to the province, Indonesia data in the period 2013 to 2017. chosen poverty line, Atkinson (1987) proposed the and also the sensitivity of the domestic poverty lines new method to measure poverty based on the change toanalysis the poverty to see rate the produced.robustness of the poverty figure After revealing the results, there will be some the per capita expenditures from the cumulative concluding remarks regarding the poverty in distributionof welfare reflected function. from The thestrength stochastic of this ordering method ofis that the welfare change could be examined in a range are three sections remaining for this paper: method, of some poverty thresholds; therefore, the chosen result,Tabalong and regency conclusion. in the period 2013 to 2017. There poverty lines could accommodate various suggested poverty lines. Hence, this method describes the A. Tabalong’s Regency Poverty Figures sensitivity of the chosen poverty line. However, as a weakness of this method, the exact poverty rate Kalimantan, Tabalong was the second highest could not be advised because the analysis runs in Among 13 cities and regencies in South a range of thresholds. So, this method could only was one position under Hulu Sungai Utara regency describe the welfare change regardless the quantity regency for the poverty rate by 6.35% in 2016, it of the changes. absolute number, Tabalong regency contained the This paper elaborates the long-ago research sixth-highestthat possessed poor the people poverty (BPS rate Provinsi by 6.76%. Kalimantan In the conducted by Ravallion & Huppi (1991) as the Selatan, 2017). Tabalong’s regency poverty data can background. They investigated the poverty changes be obtained from BPS-Statistics of South Kalimantan in the period of 1984 and 1987 using the National Province and BPS-Statistics of Tabalong Regency, Socioeconomic Surveys (Susenas) in Indonesia. They available from 1996 to 2017. Furthermore, the focused on constructing the new poverty rate using poverty lines are available in the period 2002-2017. the purchasing power parity approach, decomposing Figure 1 depicts the trends of poverty rates those measurements, and comparing them. and poverty lines in Tabalong regency in 1996 to Formulated by Foster, Greer, and Thorbecke 2017 and 2002 to 2017 respectively. It is clearly in 1984, known as FGT indices, Ravallion and seen that in between 1996 and 1999 there was a

Jurnal Bina Praja 10 (1) (2018): 1-12 2 Figure 1.

The Trends of Poverty Rates (%) and Poverty Lines (Rupiah) in Tabalong Regency, 1996-2017, and 2002-2017 Source: BPS

highly-impacting shock which was an economic BPS has been presenting the poverty rate data crisis.significant After increasethat, the poverty in the povertyrate declined rate gradually. due to a annuallyurban and (BPS, the 2016a). rural areas. Moreover, since 2003, Starting in 2008, Tabalong’s poverty rate touched BPS (2016a) has been adopting the basic-need approach since initially calculated poverty. The measurement of the poverty rate comprised of 52 of single digit number (8.13%) until 2017 (6.09%). food commodities in the urban and the rural areas There was no significant increase in term of poverty while for the non-food commodities covered 51 development,rate fluctuation there since is a 2008. clear Thetrend poverty shown in rate Figure was items in the urban areas and 47 items in the rural 1,below from 10%2002 constantly.to 2017 poverty In term lines of inclined poverty in line’s the areas. Using this approach, the poverty is addressed same manner, increased gradually with relatively as a lack of the economic ability to satisfy those basic same slopes. needs for food and non-food measured from the Focusing on the slopes, the declining of poverty expenditure; eventually, these thresholds become rates in Tabalong regency can be divided into two a poverty line. Poor people are those who possess the per capita per month expenditure below the 1999 to 2007, the section that depicted the years poverty line. aftersections. the economic The first sectionshock in occurred 1997-1998. in theThe period slope of the trend was steep, showing a high progress 1) FGT Indices of poverty rate reduction. In these years, poverty After the poverty line is set, further calculation rates were in double-digit numbers. In the second of the household data can be done to develop a section, the period 2007 to 2017, the poverty rates known poverty measurement so-called Foster- the slope of the poverty rates trend in these years, Greer-Thorbeckepoverty figure. BPS (FGT) has Indices been (1984), utilizing given the by well- thewere poverty only in reductiona single digit seems (under slow, 10%). shown Based by theon

α 1 n zy− Ρ= i flatter slope of the trend line. α ∑ yz<  (1) i  B. Theoretical Background nz BPS initially calculated Indonesia’s poverty rate data in 1984. At once, the calculation covered years

1976 to 1981 by utilizing Susenas consumption where P 0 or module data. After that, once in every three years, 1 or α is FGT index (α=0 translated as P BPS provided the poverty rate data divided by the 2 or poverty poverty rate/incident, α=1 translated as P severity index), n is the number of samples, yi is per poverty gap ratio, α=2 translated as P

Stochastic Dominance Analysis for the Poverty in Tabalong Regency 2013 and 2017 Ahmadi Murjani

3 capita income household i (household weight for xx z ()x− y dF y ≥− () x y dF y (4) ∫∫00AB( ) ( ) is the poverty line. Instead of using an income for yindividual, BPS employed is need the to expendituredraw per capita data figure), to substitute and income. for all x ≤ z. With poverty line z, the poverty gap with expenditure y is expressed as 2) Stochastic Dominance Analysis in Poverty ∗ gzy( ,) =−= ( zy )+ max( zy − ,0) =− zy (5) correlations that connect between two distributions. The applicationStochastic dominanceof the stochastic is defined dominance as a varies set of over many disciplines, one of which is in poverty where equation (5) draws a poverty gap area analysis. The most common application in the 2 underneath the DxB () when poverty line equal to poverty analysis is examining income distribution 2 z; therefore, the total area become Dz(). When B and income inequality (Davidson, 2006). Initially B stochastically dominates A at poverty line z, in the introduced on the poverty analysis by Atkinson second-order dominance, Dz22()≥ Dz (). It can be (1987), this method is then utilized by many AB researches. concluded that the poverty gap in A always bigger The bottom line for the connection between than in A when B stochastically dominates A at the poverty and the stochastic dominance is on the poverty line z. Based on these two-times ordering, it poverty indices (FGT). Davidson & Duclos (2000) is clear that when B stochastically dominates A, the illustrated in the following explanations and poverty rate as well as the poverty gap in B always equations. smaller than A, it can be said that the welfare in B is Having two distributions of expenditures better than in A (taken from, i.e., Susenas data), then arranged that for the next order, the condition also holds (the . Davidson and Duclos also notified into two cumulative distribution functions (CDFs) poverty severity condition is formed at the third order). named FA and FB with the property of a non-negative number. Here, expenditures are the manifestation of 1 3) The Sensitivity of the Poverty Line to the welfare. Set D() x= Fx () and A A Poverty Previously, the notion B stochastically χ ss−1 dominates A would be addressed as the welfare DAA(χ ) = D( y) dy (2) ∫0 in B is better than in A, depicted by the CDFB and the CDFA. Let’s take the example from the research s conducted by Ravallion & Huppi (1991) in Indonesia for any integer s ≥ 2 and consider DB (χ ) as follow is comparable. For any order of s, Ds(x) can be Figure 2 shows that, in general, the welfare in expressed as Indonesia in 1987 was better than in 1984 because stochastically the monthly consumption per person χ s −1 in 1987 dominated the other in 1984. It is also D s (χχ) =1 ( − y) dF( y ) (s −1!) ∫0 can be seen, when the 1984 and 1987 CDF curves (3) spread in some distances, the chosen poverty line Distribution B can be addressed dominate the would yield the lower poverty rate in 1987. When distribution A stochastically at the order s when those CDF curves are united, or even crossing each ss DxAB()≥ Dx () for all xR∈ . Suppose the poverty line is z > 0, then B is stochastically dominates A at other (as can be seen in Figure 3), the poverty ss order s up to the poverty line z if DxAB()≥ Dx () for all x ≤ z. Connected with the previous notion, it is concluded that FxAB()≥ Fx () for all x ≤ z. In other words, the poverty rate is always bigger in A than in B, as long as the poverty line does not exceed z. This notion is also known as the First-order stochastic dominance. Second-order stochastic dominance of A by B up to poverty line z expressed as Dx22()≥ Dx () AB Figure 2. Distribution of Monthly Per Capita Consumption implied in in Indonesia, 1984 and 1987

Source: Ravallion & Huppi (1991)

Jurnal Bina Praja 10 (1) (2018): 1-12 4 might be different with the stochastic dominance; however, the main idea relatively similar, forming a

In Indonesia, Ravallion & Huppi (1991) investigateddominance figure. the poverty reduction progress by employing the stochastic dominance method onto Susenas data years 1984 and 1987. Additional adjustments were applied, especially on the price level in general and the rural areas’ price level prediction. In addition, they also examined the decomposition of changes in the poverty measures into growth and redistribution effects. They Figure 3. The Cross of Poverty Incidence Curves concluded, in general, the poverty had declined in Indonesia from 1984 to 1987. Source: Madden & Smith (2000) The research in Indonesia utilizing the stochastic dominance analysis on the welfare area continued. Ravallion (1992) underlined the line set now is so-called sensitive to the poverty other research’s evidence that explains the less rate produced, because the poverty rate for G(x) responsiveness of the energy intakes to the poor’s either can be higher or lower than F(x). The close income. The energy intakes then translated into position between two CDF lines can describe the undernutrition. By using the stochastic dominance sensitiveness of the chosen position poverty line. In method on the Susenas 1984 and 1987 data, he found the method section, there will be a particular step to that the aggregate undernutrition in Indonesia was determine the dominance regarding this sensitivity. responsive to the changes in prices and incomes. z , it is can a Madden & Smith (2000) employed the be said that G(x) stochastically dominates F(x); and stochastic dominance analysis when assessing the at povertyBased line on Figurez , F(x) 3, stochastically at the poverty dominates line G(x). b poverty in Ireland in the period 1987 to 1994. They So, both situations are called restricted dominance utilized two kinds of poverty lines, absolute and because it depends on the particular poverty line. relative, to measure the poverty rate by using the Hence, the chosen poverty line is sensitive to the poverty. they found that poverty in Ireland decreased in such To determine which distribution dominates period,Household in case Budget absolute Survey poverty data in Ireland. lines are Eventually, used. In aggregately, the areas beneath the CDFs need to addition, when they employed the relative poverty be calculated as mentioned in equation 4. When lines, there was stochastic dominance on the second order for expenditure data and the third order for income data. the dominance does not hold at the first order, domination. Not only for examining the changes of welfare the situation in Figure 3 is called second-order over some periods, the stochastic dominance C. Stochastic Dominance in Poverty: analysis could also be imposed on the simulation’s Empirical Applications study. Makdissi & Wodon (2002) applied the method when simulating the impact of the marginal tax 1) Application in the Poverty Analysis reform on the poverty in Bolivia, 1999. They utilized Atkinson (1987) introduced the application the so-called Consumption Dominance Curve to of the stochastic dominance in poverty by drawing investigate the robustness of poverty reduction the cumulative percentage below different poverty over different poverty measures and poverty lines levels against the unit of the percentage of the after the marginal tax reform was imposed on two commodities. They found that the increase in prices 1979, and 1982. He decomposed the possible of medicines and the lower rate for interprovincial povertyofficial povertymeasurements line in using the United such illustration. States for 1974, transportation would reduce any poverty measure Chosen poverty line can be critical in of the second order, showing that the chosen determining the exact value of a poverty (welfare). particular order was essential in their analysis. Thus, with the inappropriate poverty line, the While the existed theory put the assumption conclusion can be ambiguous. Foster & Shorrocks that the continuity holds in incomes of households. (1988) underlined such problem in their research. Araar (2006) explained in his research about the To overcome the problem, they suggested a method discontinuity nature in real data. He compared that examines the partial ordering of various Burkina Faso’s data in 1994 and 1998 and using poverty indices. This method yielded poverty the stochastic dominance analysis for assumed orderings and welfare dominance. The calculation discrete data. His research’s result suggested that

Stochastic Dominance Analysis for the Poverty in Tabalong Regency 2013 and 2017 Ahmadi Murjani 5 tax family income. In a further test, they employed the Kolmogorov-Smirnov (KS) test to statistically dominancethe poverty analysis rate figures proved from that 1994 there and 1998were hadno the stochastic dominance condition. Their study, robustnot changed conclusions significantly regarding sincethe poverty the stochasticchanges. somehow, underlined a new type of KS test they the changes in households’ income structure and compared with the previous KS type test. povertyEstudillo, reduction Sawada, in three & villages Otsuka in (2008) the Philippines assessed utilizedArndt, and Hussain, finally yieldedSalvucci, theTarp, similar & Østerdal result (2016) were focusing their research on the the headcount ratio reduced over the periods, they comparison of the poverty measurements using investigatedusing the data the from sensitiveness 1985 to 2004. of the After chosen finding poverty that the census data and the small-area estimation. In line. The result shows that the poverty indeed declined robustly after implementing the stochastic method was utilized in Mozambique. For the census dominance analysis. data,this study,1997 and the 2007 first-order were the stochastic chosen periods dominance while The stochastic dominance utilization was 1996/1997 and 2008/2009 were for the survey’s measure the progress of the Human Development stochastic dominance presents robust comparisons Indexemployed (HDI). by They Pinar, used Stengos, the HDI & Topaloglou data from the (2013) United to acrosssource manyperiods. backgrounds They concluded of populations. that the first-order Nations Development Program in the period of 1975 to 2000. Their research eventually suggested 2) Application in the Other Disciplines the different rankings of countries based on the As the complementary documentation on how stochastic dominance test for the robust result. the stochastic dominance, and to show the further application of a newly stochastic dominance test on method on the other topics, some brief literature by employingDavidson the & Luxembourg Duclos (2013) Income studied Study (LIS) the reviewsdevelopments are presented and modifications in an ascending for the respected period. that comprises USA data year 2000, the Netherland year 1999, the UK year 1999, Germany year 2000, utilizes a so-called almost second-degree stochastic and Ireland year 2000. They tried to overcome dominanceTzeng, Huang, to rank& Shih all (2013) observed presented distributions a paper in that the the classical problem for stochastic dominance context of decision making. test assumption which was the continuity of On the other utilization, Al-Khazali, Lean, & distributions. All in all, they suggested using the Samet (2014) employed the stochastic dominance empirical likelihood methods in dealing with the continuous distribution of data. the dominance of the Islamic stock indexes to the conventionalin the field of stock finance indexes. which Their conducted study examined to assess the data in the two periods, 1996-2012 and 2001- of PeruvianExamining adults the that ordinal living variables in Lima was to the the adults focus 2006. Their study, eventually, indicated that the livingof study elsewhere by (2013). in Peru. The The study data’s compared source thewas data the Islamic investment exhibited a better performance Peruvian National Households Survey 2001 that covers 16,515 households. The results suggested that in this case, with the ordinal variables, the orderin the globalstochastic financial dominance crisis periods. in their paper about stochastic dominance method could be also imposed portfolioDupačová optimization & Kopa with (2014) some employed constraints. the first-The by deriving the ordinal-variable equivalent for the additive social welfare functions. employed in the area of discrete distribution Levine, Muwonge, & Batana (2014) examined withstochastic respect dominance to a scenario. method Further, was Daskalakis, specifically the multidimensional poverty using the survey Deckelbaum, & Tzamos (2015) studied the revenue data in Uganda. The data, presented in 2000/2001 maximization for the multiple-good monopoly and 2005/2006, were obtained from the Uganda condition. They examined the purchasers’ Bureau of Statistics and ORC Macro respectively. distribution using the stochastic dominance method. The weighted-stochastic-dominance analysis It is evident that the stochastic dominance method was imposed to robustly prove the declining in not only has been developing but also be utilized in multi-dimensional poverty in Uganda. The study the wider scope of disciplines. contrasted the poverty rate in Uganda using the Alongside many kinds of literature examining monetary approach with the multi-dimensional the poverty and/or welfare progression in various poverty which revealed the big margin between the places and times using the stochastic dominance two measurements. analysis, and for the motivation to assess the poverty Donald & Hsu (2016) utilized the data from reduction progress in Tabalong from the other perspective, this paper aims to robustly determine and 1986 to explore the welfare changes due to a whether the poverty in Tabalong regency move to a the Canadian Family Expenditure Survey in 1978

Jurnal Bina Praja 10 (1) (2018): 1-12 6 better condition. Against the chosen relative poverty where CPIt-1 CPI2013 lines presented by BPS, this paper also investigates the value of CPI2014 and the years beyond can be the sensitiveness of a range of poverty lines to the extrapolated =using equation = 100 in (8).the baseThus, year. when Thus, all poverty rate. the CPI

with the from ratio 2013 of CPI to in2017 2017 have to thebeen CPI obtained, in the base the II. Method per capita per month expenditures can be deflated capita expenditures are expressed in form of natural A. Data Selection logarithmyear 2013 (ln)(100).. To simplify the numbers, the per To perform the poverty comparison analysis The next step is forming a cumulative using the stochastic dominance method, the paper distribution function (CDF) for both years. For

2017; in particular, for Tabalong regency locus. For directly create two weighted CDFs for the per capita collected the Susenas data in the period 2013 and the exact number, the total samples are 550 and 557 persimplicity, month thisexpenditure paper employs in the observed the STATA periods. 13 to Once created, the CDFs are then drawn by STATA for 2013 and 2017 respectively. To include the price the per capita households’ expenditures are stochastic dominance condition. levels’ influence on the households’ expenditure, in oneThe figure. last step This is stepto test is the to dominance assess visually of those the deflated with the Tanjung city CPI’s. This paper set description, some poverty indicators are included in to 2017. Two-sample Kolmogorov-Smirnov test (as this2013 paper as the such base as povertyyear (CPI=100). line data, To headcount complete ratio the CDFs statistically. The test will be done from 2013 employed to check whether two nearby distribution isseen different on Estudillo, or not. Sawada, The difference and Otsuka between (2008)) two is (Po), and inflations. B. Analysis Formulation distributions can be translated as the dominant This section elaborates the process of position of a distribution on another. In case the assessing the poverty analysis in Tabalong regency. stochastic dominance is done to investigates the dominance of a CDF over another to prove that the the Tanjung city CPIs. In the stochastic dominance welfare has progressed robustly, then at the order analysis,The first stepthe ofdata the shouldanalysis be is thein extrapolationthe real value, of s, the null hypothesis of Two-sample KS can be meaning that the per capita consumptions should expressed as absorb the impact of price changes over the years. year, in this case, the CPI in Tanjung city will be set

Further, the year 2013 is chosen for the base HFF0 =−=AB0 (9) same level, the extrapolation process is done by the equal to 100. By maintaining the inflation rate at the where FA is the CDF for distribution A, and FB is the basic inflation formula CDF for distribution B in two different years. For CPI− CPI the sensitiveness of the chosen poverty line, a range Inf = tt−1 x 100% t  (6) CPIt −1 of deflated poverty lines in 2013 and 2017 will be attach of the CDFs figure to check whether the where Inf t, t incident in Figure 3 occurs. is the in period , is the in period . CPIt CPI t CPIt-1 CPI t-1 III. Results and Discussion is the inflation rate in the period (year) (6) can obtain the CPI values by rearranging into A. Tabalong’s Poverty Lines Progression The inflation rate is expressed in percent. Equation

CPI x 100 in Tabalong Regency increased around Rp100,000. CPI = t (7) As canFrom be 2013seen to in 2017, Table nominally, 1, at the the initial poverty year lines of t −1 100 + Inft

and nextobservation, period with the thepoverty exception line was for 2016set at poverty Rp330,764 line and increased by approximately 5% to 6% in the the increment of poverty lines in Tabalong regency CPI x(100+ Inf ) (grew 10.23% from the previous year). Nominally, CPI = tt−1 (8) t 100 (with the exception of 2016 poverty line that  in each year in a range of Rp18,000 to Rp23,000

increased Rp37,745 significantly).

Stochastic Dominance Analysis for the Poverty in Tabalong Regency 2013 and 2017 Ahmadi Murjani 7 Table 1.

Poverty Lines (PLs) in Tabalong Nominal Numbers (IDR), 2013 to 2017 Annual NPLs’ Nominal NPLs’ Nominal Po Inflation Year PLs CPI Change Increase (%) (%) (NPLs) (%) (IDR)

2017 430,129 5.77 23,460 6.09 121.46 2.4

2016 406,669 10.23 37,745 6.35 118.61 2.18

2015 368,924 5.19 18,187 6.59 116.08 6.69

2014 350,737 6.04 19,973 6.21 108.80 8.8

2013* 330,764 Na Na 6.15 100.00 5.29

Source: Statistics Indonesia (BPS), Author’s calculation.

* Inflation in Tanjung 2013 was obtained from the local Cost Living Survey, BPS provided Tanjung’s inflation starting from 2014.

automatically sorted. The graphical analysis is the annual nominal poverty lines (NPLs) can be important to determine initially whether the 2 In Table 1, the fluctuation of inflation and CDFs are crossing or not. Furthermore, the relative distance between two curves also can be observed. examined. When in 2014 inflation 8.8% occurred, To magnify the picture due to a large digit of per the NPLs increased by 6.04%. However, when a capita per month expenditure, the conversion to ln relatively-lower inflation happened in 2016 by has been done. The poverty lines, in an approximate 2.18%, the NPLs grew significantly by 10.23%. manner, are also converted. aThis poverty finding line could concerning not be putabout into the our exception concluding in Figure 4 depicts the two adjacent CDFs for per thenote year about 2016. the relationshipThus, examining between the inflationrelationship and capita per month expenditures in Tabalong regency, could bring dubious conclusions. between inflations and the poverty lines directly welfarefor the periodin Tabalong of 2013 regency and 2017.has increased. There are This some is B. Stochastic Dominance Analysis: The notes can be taken regarding figure 4. In general, the Graphical Inspection hand side in 2017. There is no notable crossing CDFs inshown this bycase the which movement can bringof CDF a in robust 2013 toconclusion the right- adjacent CDFs using its weight because in Susenas about poverty reduction progress. Because the CDF individualSTATA weight 13 provides is given. an Also, easiness the CDFs to draw can two be in 2017 is mainly at the right-hand side of 2013 CDF,

Figure 5. CDF Figure 4. (ln) Magnified s of Per CapitaCDF per Month Expenditures CDFs of Per Capita per Month Expenditures (ln) in in Tabalong Regency 2013 and 2017 (Per Capita per Month Tabalong Regency 2013 and 2017 Expenditures ≥ 15 and value ≥ 0.985) Source: STATA processing Source: STATA processing

Jurnal Bina Praja 10 (1) (2018): 1-12 8 In Figure 6, the upper part of the two adjacent

is easier to determine whether the CDF2017 fully dominatesCDFs can bethe examined. CDF at After the upper being side magnified, of those it curves. After both distributions are sorted, there 2013 is evidence that CDF2017 always dominate CDF at every level of the poverty line. Therefore, based 2013 on this condition, the distribution in 2017 is called robustly stochastically dominates the distribution in

This graph inspection, eventually, produces the initial2013 at conclusion every possible that the level welfare of chosen in 2017 poverty increased lines.

However, to legitimate the conclusion about thecompared stochastic with thedominance welfare conditionof the distribution in 2013. in Figure 6. CDF (ln) Magnified s of Per Capita per Month Expenditures of stochastic dominance can be performed. The in Tabalong Regency 2013 and 2017 (Per Capita per Month CDF2017 to can the bedistribution considered in stochastically 2013, the second-order dominates Expenditures ≥ 15.6 and CDF value ≥ 0.999) 2017 Source: STATA processing the CDF at the second order if the total area beneath the CDF is bigger than the area beneath 2013 there is a conclusion that CDF in 2017 stochastically the CDF (in the poverty sense, the poverty gap 2017 2013 words, there is a better welfare in 2017 compared Calculating the area under the CDFs is considerably dominates the CDF in 2013 at the first order. In other sensitiveof the distribution to the outliers; in 2013 so is the bigger highest than data in 2017). from poverty in 2017 can be robustly proven based on both distribution are taken out. Therefore, the theto 2013. picture. Moreover, the conclusion about the lower calculation result of the total area beneath both However, as described in Figure 4, for the value CDFs given by the table 2. of ln per capita per month expenditures more than Table 2 shows the integral calculation for the 15, there is a closing CDFs. To examine the incidence total area under both CDFs. It is obvious that the of the crossing CDFs or when the difference between total area beneath the CDF is bigger than the total ln per capita per month expenditures in 2017 to area for the CDF 2017 2013 2017 is above CDF ), we can there is a conclusion that the distribution in 2017 put our focus on the value when ln per capita per (2.428 vs. 1.976). Eventually, 2013 month2013 is expendituresnegative (CDF are more than 15. Figure 5 the second-order. stochasticallyThe next dominatesanalysis that the can distribution be done toin 2013analyze at capita per month of expenditures after being sorted the sensitivity of chosen poverty line to the poverty basedprovides on theCDFs detailed and bigger figure than when 15. the value of ln per At a glance, in Figure 5, there is no incident that the CDF crosses the CDF2017. In this position, andis by 2017. utilizing The magnified-picture analysis can be seen of figurefrom Figure 4 with 7. the there is no undoubting opinion that can prove the companion of converted poverty lines (PLs) in 2013 2013 and 2017, there are four points of the curves’ distribution in 2017 at the value of poverty line In the converted poverty lines in 2013 moredistribution than 15. in However, 2013 stochastically at the CDF dominatesvalues ranging the utilized as the threshold for poverty criteria, CDF intersection. When the poverty line in 2013 is because the CDFs lines are so near. Therefore, the seen when B is in the lower position compared to STATAfrom 0.999 now tois set1 seem at such to needvalue to as be a boundary.more magnified Ain (this 2017 can dominates be also translated the CDF inas 2013.the lower This poverty can be

Table 2.

Integral Calculation for the CDF of ln per Capita per Month of Expenditures in 2013 and 2017 Using Tabalong Regency Data Number of Minimum Maximum Year Integral Mean Points Value Value

2013 549 2.4280763 12.31781 15.85112 13.6287

2017 556 1.976205 12.40699 15.59193 13.77924

Source: STATA calculation

Stochastic Dominance Analysis for the Poverty in Tabalong Regency 2013 and 2017 Ahmadi Murjani 9 Table 3.

Descriptive Statistics and Hypothesis Testing Result for the Two-Sample KS Test Using Tabalong Regency Data, 2013 and 2017 Obs. Obs. with without Std. Variable Observations Min Max Mean missing missing deviation data data

lncapita2013 550 0 550 12.318 16.178 13.633 0.648

lncapita2017 557 0 557 12.407 16.417 13.784 0.578

D 0.125

p-value 0.000***

alpha 0.01

Source: XLSTAT calculation.

*** Indicates the D statistics is significant at α = 1%. rate in B compared to A since B dominates A). When the poverty line in 2017 is set as the boundary, the same situation also happens. Point D is lower than povertyconsiderations, data to produce the findings a solid of base this for paper the (aspoverty well C, meaning that poverty rate in 2017 is lower than alleviationas the method) policy cantaken complement by the policymakers. the BPS’ official set, the distribution in 2017 always dominates the in 2013. Eventually, whichever the poverty line is IV. Conclusion Investigating the welfare change in the reality because of no crossing incidence for the CDFs at the possibledistribution range in of 2013. poverty The lines. analysis All in all, is quitethe welfare clear the period is the motivation of this paper. This paper investigatesthat poverty therates progress (headcount of the ratio) poverty fluctuated reduction over of the chosen poverty lines. in 2017 is better than in 2013 with the insensitivity (i.e. welfare development) in Tabalong regency in C. Stochastic Dominance Analysis: Hypothesis Testing extrapolated2013 to 2017. CPIs, By usingthe progress the per of capita welfare per change month Further analysis is the hypothesis testing of hasexpenditures been examined. from Susenas data deflated by the dominance. One distribution is said to dominate The stochastic dominance method indicated another when the null hypothesis is cannot be that the welfare in Tabalong regency has robustly accepted. By using the hypothesis in equation 9 and improved. This approach included the purchasing power impact (by adjusting the data in 2017 with the extrapolated CPIs) to the per capita per month astwo-sample well as theKS test,hypothesis the result testing can be results seen in fortable the 3. expenditures of all samples. Distribution in 2017 two-sampleTable 3KS presentstest for two the adjacent summary CDFs statistics in the observation. The test indicates that the P value is sense,stochastically the conclusion dominated should the distribution have been inabsolute; 2013 at cannot be accepted; therefore, both distributions moreover,the first and with the thesecond additional order, this examination means in thefor realthe lower than 1%, proving that the null hypothesis upper part of both CDFs. the distribution in 2017 robustly dominated the In a range of possible poverty lines, the are assumed significantly different. In conclusion, robustness also stands still, the chosen poverty lines are then said to be insensitive to the poverty. There sensitivedistribution to the in 2013; poverty. the welfare has increased from was no crossing curve of CDFs that supports the 2013The to 2017 poverty and the headcount chosen poverty ratio linesin Tabalongwere not conclusion of the method (the welfare change was easier to be recognized). respectively, meaning that there was a decline in In addition, the two-sample KS test also theregency poverty in 2013 rates and in 2017Tabalong. were In 6.15% line withand 6.09%those proved that both distributions are statistically BPS’ data, this paper also supports the conclusion different and the distribution in 2017 stochastically that the welfare indeed inclined in the observed stochastic domination of the distribution in 2017 dominates 2013. As an additional note, the period. Eventually, regardless some researchers’ Jurnal Bina Praja 10 (1) (2018): 1-12 10 on Economics and Computation - EC ’15 (pp. restricted because of no evidence that showing the 449–450). New York, New York, USA: ACM Press. crossingto the distribution incident of both in 2013 CDFs. cannot be considered Various methods of the stochastic dominance Davidson, R. (2006). Stochastic Dominance analysis have been being developed by many http://doi.org/10.1145/2764468.2764539(Departmental Working Papers). McGill researchers. This paper presents the simple application of the stochastic dominance method on Retrieved from https://ideas.repec.org/p/ the poverty analysis. For further development, this mcl/mclwop/2006-19.htmlUniversity, Department of Economics. method could also be mixed with the simulation Davidson, R., & Duclos, J.-Y. (2000). Statistical analysis; for instance, the impact of a policy on the Inference for Stochastic Dominance and for poverty/welfare at one particular time. the Measurement of Poverty and Inequality. In the various concepts and calculation Econometrica, 68 methods of the poverty threshold, the poverty org/10.1111/1468-0262.00167 rate analysis should not be taken as the only (6), 1435–1464. http://doi. measurement of the poverty alleviation policy Restricted Stochastic Dominance. Econometric evaluation. The policymakers should also underline Davidson,Reviews R.,, 32 &(1), Duclos, 84–125. J.-Y. http://doi.org/10.108 (2013). Testing for the other methods that accommodate the diversity of poverty analysis. Thus, this paper stands as an Donald, S. G., & Hsu, Y.-C. (2016). Improving the alternative as well as a complementary method to Power0/07474938.2012.690332 of Tests of Stochastic Dominance. assess the welfare change. So, the policymakers Econometric Reviews, 35 could produce a solid conclusion regarding the poverty dynamics over a period. (4), 553–585. http:// optimaldoi.org/10.1080/07474938.2013.833813 portfolios under risk and stochastic Dupačová,dominance J., & constraints. Kopa, M. (2014). European Robustness Journal of Acknowledgement Operational Research, 234 A good thought for all of my colleagues that supported me to construct this paper. I thank for (2), 434–441. http:// BPS for the excellence data and website; also, for Povertydoi.org/10.1016/j.ejor.2013.06.018 and Income Dynamics in Philippine BPS of Tabalong regency that gave me the access Estudillo,Villages, J. P.,1985–2004. Sawada, Y., Review & Otsuka, of Development K. (2008). to Susenas data. I also thank for the constructing Economics, 12(4), 877–890. http://doi. comments from the reviewers. I responsible for all of the errors in this paper. Orderingsorg/10.1111/j.1467-9361.2008.00458.x and Welfare Dominance. In V. References Foster,Distributive J. E., & Justice Shorrocks, and Inequality A. F. (1988). (pp. 91–110). Poverty Al-Khazali, O., Lean, H. H., & Samet, A. (2014). Berlin, Heidelberg: Springer Berlin Heidelberg. Do Islamic stock indexes outperform conventional stock indexes? A stochastic 6_6 dominance approach. Pacific-Basin Finance http://doi.org/10.1007/978-3-642-73816- Journal, 28, 29–46. http://doi.org/10.1016/j. Class of Decomposable Poverty Measures. Foster,Econometrica J., Greer, , J., 52 & Thorbecke, E. (1984). A Araar, A. (2006). Poverty, Inequality and Stochastic Dominance,pacfin.2013.09.003 Theory and Practice: Illustration Khomsan, A., Dharmawan,(3), 761–766.A. H., Saharuddin, http://doi. with Burkina Faso Surveys. CIRPEE Working org/10.2307/1913475 Paper No. 06-34 Indikator Kemiskinan dan Misklasifikasi OrangAlfiasari, Miskin Sukandar,. Jakarta: D., Yayasan & Syarief, Pustaka H. (2015). Obor Arndt, C., Hussain, A. M.,. http://doi.org/10.2139/Salvucci, V., Tarp, F., & Indonesia. Østerdal,ssrn.938555 L. P. (2016). Poverty Mapping Based Lanjouw, J. O. (1998). Demystifying Poverty Lines. Retrieved from http://siteresources. from Mozambique. Journal of International worldbank.org/PGLP/Resources/ Developmenton First-Order, Dominance28 with an Example LanjouwDemystifyingPovertyLines.pdf Levine, S., Muwonge, J., & Batana, Y. M. (2014). A Atkinson, A. B. (1987).(1), On the 3–21. Measurement http://doi. of Poverty.org/10.1002/jid.3200 Econometrica, 55(4), 749–764. http:// Uganda. Journal of Human Development and CapabilitiesRobust Multi-dimensional, 15 Poverty Profile for Daskalakis, C., Deckelbaum, A., & Tzamos, C. (2015). Strongdoi.org/10.2307/1911028 Duality for a Multiple-Good Monopolist. MacPherson, S., & Silburn,(4), 369–390. R. (1998). http://doi.org/1 The meaning In Proceedings of the Sixteenth ACM Conference and0.1080/19452829.2014.897310 measurement of poverty. In J. Dixon & D.

Stochastic Dominance Analysis for the Poverty in Tabalong Regency 2013 and 2017 Ahmadi Murjani 11 Poverty: A Persistent Global Pendapatan. Jakarta: Statistics Indonesia Reality (pp. 1–19). London & New York: (BPS). Retrieved from https://www. Routledge.Macarov (Eds.), Madden, D., & Smith, F. (2000). Poverty in Ireland, 1987-1994: A Stochastic Dominance Approach. kemiskinan-ketenagakerjaan-dan-distribusi-bps.go.id/publication/2009/03/04/ The Economic and Social Review, 31 pendapatan.htmlbfe286bb431babaad062642f/analisis- 187–214. Retrieved from http://hdl.handle. Statistics Indonesia (BPS). (2016). Penghitungan (3), dan Analisis Kemiskinan Makro Indonesia Makdissi, P., & Wodon, Q. (2002). Consumption Tahun 2016. Jakarta: Statistics Indonesia (BPS). dominancenet/2262/60131 curves: testing for the impact of Retrieved from https://microdata.bps.go.id/ indirect tax reforms on poverty. Economics Letters, 75 Statistics of South Kalimantan Province (BPS org/10.1016/S0165-1765(01)00599-7 Provinsimikrodata/index.php/citations/346 Kalimantan Selatan). (2017). Jumlah Marbun, D., & Suryahadi,(2), 227–235. A. (2009). http://doi. Kriteria Penduduk Miskin Menurut Kabupaten/Kota Kemiskinan Konsumsi: Praktik di Indonesia di Provinsi Kalimantan Selatan Tahun 1996- dan Beberapa Catatan. Jurnal Analisis Sosial, 2016. Retrieved from https://kalsel.bps.go.id/ 14 akatiga.org/index.php/jurnal-analisis-sosial/ penduduk-miskin-menurut-kabupaten- item/468-jas-vol-14-no-2-menelusuri-(2), 13–30. Retrieved from http://www. kota-di-provinsi-kalimantan-selatan-dynamictable/2017/06/20/973/jumlah- kriteria-kemiskinan-perspektif-masyarakat- tahun-1996-2016.html sipil untuk Mengukur Kemiskinan. Forum Penelitian Measuring human development: a stochastic Susilowati,Agro S.Ekonomi H. (2010)., 28 Pendekatan(2), 91–105. Skala http://doi. Ekivalensi Pinar,dominance M., Stengos, approach. T., & Topaloglou,Journal of N.Economic (2013). org/10.21082/fae.v28n2.2010.91-105 Growth, 18(1), 69–108. http://doi. Suyanto, B. (2001). Kemiskinan dan Pemberdayaan Masyarakat Miskin. Masyarakat, Kebudayaan Ravallion, M. (1992). Does Undernutrition Respond Dan Politik, XIV(4), 25–42. Retrieved from toorg/10.1007/s10887-012-9083-8 Incomes and Prices? Dominance Tests for Indonesia. The World Bank Economic Review, 6 http://web.unair.ac.id/admin/file/f_19997_ wber/6.1.109 Almostjr3.pdf Second-Degree Stochastic Dominance. Ravallion,(1), M., 109–124. & Huppi, http://doi.org/10.1093/M. (1991). Measuring Tzeng,Management L. Y., Huang, R. ScienceJ., & Shih,, P.-T.59(5), (2013). 1250–1254. Revisiting Changes in Poverty: A Methodological Case http://doi.org/10.1287/mnsc.1120.1616 Study of Indonesia during an Adjustment Period. The World Bank Economic Review, 5(1), Ordinal Variables: Conditions and a Test. Yalonetzky,Econometric G. (2013). Reviews Stochastic, 32 Dominance with Statistics Indonesia (BPS). (2009). Analisis 57–82.Kemiskinan, http://doi.org/10.1093/wber/5.1.57 Ketenagakerjaan dan Distribusi (1), 126–163. http:// doi.org/10.1080/07474938.2012.690653

Jurnal Bina Praja 10 (1) (2018): 1-12 12