Measuring Inequality with Multiple Indexes: An Evidence from Empirical of Thai Household Survey. Paponpat Taveeapiradeecharoen Abstract Economists have developed many measuring index of inequality so as to create the most accurate measurement. There have been substantially increased in tools for doing so. This research are conducted by using many of those tools of inequality to decompose the inequality index into two parts. First is the evaluation of income distribution within group (or within cluster) and between group. In this sense, we are able to understand the inequality of income distribution of Thailand from a different perspective. Index we applied1 are Atkinson, Generalized Entropy Families, Extended Gini Coefficient and Stratification Index. The primary aim is to identify the inequality of income distribution in subgroups for Gender Gap, Age, Region and Rural-Urban. Data used in this working-paper is collected by National Statistic Office (NSO) of Thailand. We focus mainly on the distribution of income in 2015 where all observation 33,869 were calculated. We found that by using maximum BGI developed by Elbers, Lanjouw, Mistiaen and Ozler (2005) showed that the city of Songkhla are largely not equal when compare income both Rural and Urban group, whereas Phuket has the widest range of different in income distribution in gender gap (male and female). Cities in Bangkok Metropolitan Regions which are extremely high correlated to the proportion of GDP growth, only Nonthaburi has a wide gap of inequality between male and female in gender decomposition. The additional empirical results are illustrated in appendix B.
1
Lecturer at Mae Fah Luang University, Office of Border Economy and Logistics Study (OBELS) Researcher.
1. Introduction and Data Configuration According to many recent research on inequality index has proved that the gap of income distribution in Thailand was steadily decreased over time during 1992 until 2011 by using Gini coefficient see more details S.W. Paweenawat, R, McNown (2014). However, the consideration of decomposing the sub-group by using modern inequality indexes such as stratification index or extended Gini by Yitzhaki and Lerman (1991). In order to understand the current circumstances about inequality index of Thailand. It is therefore crucial for us to decompose the sub-clusters in both within group and between groups of data. The data is household survey2 across regions from all over Thailand. The duration of years are during the 2007-2015 biyearly. In 2015, observations are simulated in total number of 33,869 households from five regions, Bangkok Metropolis3, Central, North, Northeast and South. For further details about cities in each region, the reader are referred to the appendix A. Before getting in the computational process for each measurement, it is worth to discuss the transformation of the matrix first. Let yij be the individual income of i in group j . Like mention in above about how inequality index can be computed separately between within group and between group, then we need to stack data in order to match individual income to each group. The group we would decompose are Gender Gap (Male and Female), Area (Rural and Urban)4, Regions (Bangkok Metropolis, Center, North, Northeast and South). To illustrate this more clearly, the data can be organized in the following:
yi j y k data i 1 ... ... yn
j j k k ...
...
Where n number of the last observation in each group, j , j , j are clusters between group and { j, k , ...} are noted as sub-group or within cluster. Please note that every sub-group from each group do not have the same dimension, it depends on how the data is collected5. The reason we manage to stack data like above is due to the different dimension on each group and sub-group. For instance, the frequency of income observations collected from capital city of Thailand (Bangkok) is higher in the number and average of data which means we cannot stack all data and compute it at once. We need to process it separately. 2. A Brief of Selected Measurement 2.1 Atkinson’s index Atkinson’s index is obtained using functional form by underlying between social welfare function approach and welfare weight6. Suppose each household’s form of unity function is:
yi(1e) U ( yi ) a b 1 e 2
(1)
We as the OBELS team would like to thank you to National Statistical Office for providing the available data of this household survey. As this generosity, we are able to squeeze the best result from them and to create a new potential evidence to the better upcoming policy in Thailand. 3 The reason we separate the Bangkok metropolis as one region is due to the highest frequency of 8 percent from all survey. 4 All household survey data were collected and identified about the area of observer, municipal=urban and non-municipal area=rural. 5 See more detail about each sub-cluster at Appendix. 6 See also Atkinson (1970) On the Measurement of inequality.
Where e = epsilon which we prefer not to be “1”. If e[0,1] is satisfied, then the unity function can be written:
U ( y) ln( y)
(2)
e is a measure of the level of inequality aversion, on the other words, it is referred as the sensitivity to transfers at a different income rank. If e is higher means the more weights are transferred from the higher level of income distribution to the lower. To put it simply, we would set the higher epsilon when we think most households in specific country are developing country due to higher number of low to medium income level. The atkinson’s index can be derived by two steps. First is to obtain the equalized social wealfare where everyone in the observation has the same level of income distribution, We .
yi(1e ) 1 n We a b U ye U ( ye ) n i 1 i 1 1 e For every income ye , n
(3)
1
1 n 1e ye yi1e n i 1
(4)
Second is to contrast between the equalized income distribution and mean income . The Atkinson’s index can now be derived as follow: 1
1 n y 1e 1e ye Atkinson ' s index 1 1 i n i 1 Where Atkinson ' sindex (0,1)
(5)
2.2 Entropy Measuring Families Entropy measurement is based on the probability of distribution theory by assuming that every event’s probabilities in every specific set of those events can be calculated. Let us suppose that there are n events numbered 1, 2, 3, … n. which of those are expected to have their own probability of occurrence p1 , p2 , p3 ,..., pn , these number must be non-negative due to the impossibility of occurrence from data set, and
p
i
1.
The concepts are simple. If events numbered “1” is more likely to be frequent ( p1 is closed to 1), then this information is relatively low h( p1 ) . Conversely, if this event has low probability to happen then we got a higher of h( p1 ) . Another assumption is that if two events are statistically independent, probabilities of event 1 and event 2 would be pi p j , therefore the condition met the requirement and will be written as:
h( pi p j ) h( pi ) h( p j ) where i j
(6)
Now, the one function above satisfies the decreasing information content property. The individual event now can be written:
h( pi ) ln(
1 ) pi
(7)
The average content of information (finalized entropy), of circumstance H ( pi ) is done by using sum
n
of all individual event
h( p ) , now the formula becomes: i 1
i
n
n
i 1
i 1
H ( p) pi h( pi ) pi ln(
1 ) pi
(8)
The interpretation from the equation above is to measure the probability of different varieties of occurrence are equal or not. On other words, the smaller of the differences in probabilities as pi becomes closer to 1/n. Most of the inequalities index in the past were able to analyze inequality only the data itself, where the source or the type of how data were collected are not in the part of the process until Shorrocks (1980) and among others, proving that the Generalized Entrypy Class can be rewritten as follow to evaluate the between group and within groups separately.
GE
n y 1 fi ( i )c 1 c(c 1) i 1
for c 0,1
(9)
According to Cowell, F, A (2009). If c 0 , the measurement will also be known as Theil-L or Mean Log Deviation indexes.
MLD
y 1 n ln i n i 1
(10)
If c 1 the measurement will be known as Theil’s index (T) which evaluate the distinction between the maximum and the average content of information of events.
T ln(n) H ( p)
(11)
Rewrite the equation above and it becomes, n 1 T p ln(n) ln( ) n i 1 i
Now transform pi into the term of income inequality measurement by replace pi
(12)
yi n
y i 1
i
Where yi is individual income of household number i, and n is number of observation, then the finalized Theil index is:
T
y 1 n yi ln( i ) n i 1
(13)
2.3 Extended Gini Coefficient Gini coefficient is the most popular among others for measuring inequality using income distribution. The logic behind this is because this measurement satisfies both of statistical and social welfare axioms. - Mean Independence: the gini coefficient is not changed when all household’s income is doubled. - Population size independence: on other words, ceteris paribus, which means that if the population size changes. The measurement remains unchanged.
-
Pigou-Dalton Transfer Sensitivity: inequality index means calculating how income is distributed by telling us in number, by this it means that if the income is to be transferred from higher to lower income. The inequality index must be decreased too. Symmetry: if two individual observation with different income is swap, the final index is unchanged. From Elbers C., Lanjouw P., Mistiaen J.A., Ozler B. (2005) stated the maximum of betweengroup inequality as 100% of total inequality share mean cannot be possible in realistic possibility. Under only two circumstance which the statement above is not true, however, is whether each household constitutes a separate “group” or there were no difference of per capital income between all individual households within each category. It is therefore advisable to generalize the index from a different category to be identical by using the following formula.
-
Rb '
-
BGI total inequality Rb max imum BGI max imum BGI
(14)
Where BGI is between group inequality index, Maximum BGI is the maximum possible BGI which can strictly calculated by using Elbers and Alii method (2005). As Cowell and Jenkins (1995) noted that the between group inequality can never be exceeded total inequality, the same notation goes to Rb cannot be higher than Rb’. The interpretation from Rb and Rb’ is that the closer of Rb’ get to Rb the less inequality inter-group become. The reason is simple, if Rb’ get closer to Rb, it proves that the between inequality is significantly near to the value of total inequality.
-
2.4 Stratification Index The development of Yitzhaki and Lerman7 (1991) for the stratification of inequality index. The objective for creating this index is to identify the distribution of between group and within group of data. To put it in simpler words, suppose there are two groups in provided data, showing that the number of books that each manufacturing cluster can produce. By comparing this, means that we are analyzing the divergence of how much manufacturer’s productions are. Other inequality index can only point out the probability of inequality such as Atkinson’s index. However, for stratification from YL it is able to separate each class of each individual within group and between groups are. First, let yij be the data where i individual data of group j . Assume there are L j groups.
L
j
A , Pj
Lj A
where Pj is straightforward for itself as number proportion of group j from all
observation. The concept of stratification procedure is to find the covariance between yi among members of group j only. The following is the example of stratification of group j in the case there are only two clusters:
CO j
Cov j [( Fj Fnj ), y ] Cov j ( Fj , y )
, where 1 CO j 1
(15)
Formula above represents the covariance over group j which determined by the difference between the ranking of a member from within each individual in group j . Hence Fj ( yij )
7
We will denote this index as YL index hereafter.
Rank ( yij ) (Lj )
is the
cumulative distribution of individual i within group j only. On the other hand, Fnj ( yij )
Rank ( yij ) ( A Lj )
is the same value but for individual that is not in group j . After the covariance measurement, we can move on to the next step by using the decomposable stratification index. For the sake of brevity, we will not present the full details of how this specific index can be reformulated to be decomposable, interested authors are referred to Income Stratification and Income Inequality (1991). The decomposable stratification index can be written as follow:
G S j G j S j G j CO j ( Pj 1)
2Cov( y j , Fj )
(16)
y
Where S j is the data share of group j ;
y is the mean data of all observation; y j is the mean data in specific group j ;
CO j is the covariance mentioned above; The first component of the RHS is determined as within inequality proportions whereas the second term is represented as intra-group variability in overall ranks. On other words, how overall ranks will be changed due to the divergence on individual data alteration. The last term is Gini index across all between groups. After the decomposition process, as stated by YL’s work that this decomposition index is more complicated than the other index such as Theil. Speaking loosely, each component has no identical distribution to each other. For instance, if there is changes in CO j for only group j would leave the Gini and unchanged and affect only the impact of stratification or intra-group variability (second component). In addition to this, if there is alteration in covariance means and means of rank of specific cluster, it leave overall gini index unchanged and only last term of RHS would be affected. 3. Empirical Results Table 3.1 Inequality in HFC implied by Grouping Decomposable Atkinson's index. Decomposed Decomposed Factors Groups e=0.20 e=0.25 8 0.003 0.149 0.004 0.187 Age Child
e=0.30 0.005 0.226
e=0.35 0.006 0.265
2.2%
97.8%
2.2%
97.8%
2.1%
97.9%
2.1%
97.9%
Male
0.000
0.152
0.001
0.190
0.001
0.229
0.001
0.269
Female
0.3%
99.7%
0.3%
99.7%
0.3%
99.7%
0.3%
99.7%
Buddhism
0.000 0.1%
0.152 99.9%
0.000 0.1%
0.190 99.9%
0.000 0.1%
0.229 99.9%
0.000 0.1%
0.269 99.9%
0.006 4.2%
0.146 95.8%
0.008 4.1%
0.184 95.9%
0.000 0.1%
0.229 99.9%
0.000 0.1%
0.269 99.9%
Teenage Adult Middle Adult Eldery Gender Religion
Christian Islam Others Region
BKKMETRO Central North
8
The clustering are done by using general psychology where 1-12=child, 13-19=teenage, 20-39=adult, 40-64=middle adult and above 65=eldery.
Table 3.1 Inequality in HFC implied by Grouping Decomposable Atkinson's index. Decomposed Decomposed Factors Groups e=0.20 e=0.25
e=0.30
e=0.35
Northeast South Municipal Area Hour works Cities
Urban
Rural
0.006 4.2%
0.146 95.8%
0.009 4.4%
25 groups
0.014
0.139
(2-12)
9.4%
90.6%
77 groups
0.012
0.141 92.0%
8.0% Source: author’s calculation by household survey.
0.184 95.6%
0.009 4.1%
0.222 95.9%
0.011 4.0%
0.018
0.176
9.4%
90.6%
0.015
0.178
7.9%
92.1%
0.261 96.0%
0.022
0.213
0.026
0.250
9.3%
90.7%
9.3%
90.7%
0.018
0.215
0.021
0.253
7.9%
92.1%
7.8%
92.2%
Seven groups are decomposed from the Atkinson’s index based on the income distribution in 2015. Four of epsilons are chosen so as to investigate how sensitive of weight when transferring them from higher income per month to lower income group. These results quite speak for itself as the more epsilon increase, means that lower income dominates and have more effect to overall inequality index. Despite the fact mentioned above, some decomposed clusters remain unchanged when epsilons are altered. For example, Age, Gender and Region. With given e = 0.2 over all inequality goes up to 0.152 (15.2%) and this counts as 99.7% of overall Atk’s index. When e increase to 0.35 Atk rises to 0.269. In addition to this, there are 99.7% of overall inequality came from within-group (Male and Female) rather than 0.3% of between-group index. This could lead to a possible result that gender gap during this globalizational era are narrower compared to the past century. However, the decrease between gender gap thus lead to the increase between the same sex within group instead. The further estimate to confirm this should be the Gender decomposition of income during the previous period should be calculated. On the other hand, the most sensitive to the weight changed in this index is Region cluster. Although, there is only moderately changes in Atkinson’s index during epsilon 0.2-0.25, but it is clear that the transmission from e 0.25-0.3, the share of decomposition between-group and within-group raises from 95.9 – 99.9% (almost 100%), this implied that lower income distribution somehow overlapped with the middle and high income distribution and thus the more sensitive of each cluster to overall index. Table 3.2 Theil's index and the mean logarithmic deviation (MLD) Decomposed Factors Age
Theil
Decomposed Groups
MLD
Between
Within
Between
Within
Child
0.002
0.750
0.892
1.413
Teenage
0.3%
99.7%
38.7%
61.3%
Male
0.017
0.735
0.016
1.399
Female
2.2%
97.8%
1.2%
98.8%
Buddhism
0.001
0.751
0.001
1.414
Christian
0.1%
99.9%
0.1%
99.9%
0.028 2.0%
1.387 98.0%
Adult Middle Adult Eldery Gender Religion
Islam Others Region
BKKMETRO
0.033
0.719
Central
4.4%
95.6%
North
Table 3.2 Theil's index and the mean logarithmic deviation (MLD) Decomposed Factors
Theil
Decomposed Groups
MLD
Northeast South Rural
0.034
0.718
Urban
4.5%
95.5%
0.036 2.6%
1.379 97.4%
groups9
0.07136 9.5%
0.68058 90.5%
0.08201 5.8%
1.333219 94.2%
0.06209 8.3%
0.6898567 91.7%
0.06135 4.3%
1.353888 95.7%
Municipal Area Hour works
(2-12) Cities
77 groups10
Source: author’s calculation by household survey.
Theil and MLD index are generalized entropy families, where Theil is calculated using alpha = 1, MLD’s alpha = 0 in equation (10) The most prominent result is that the difference between these two indexes in Age clustering, among these groups, MLD tells us that just above half of measures are from within, where the rest 38.7% resulted from between. One possible logic behind this approach is that the mean logarithm process which we replace alpha as zero led to the extremely low value in an extreme distribution where the income from child and teenage are really low from the given data set. Therefore, the deviation is changed a lot based only on alpha. Both Table 3.1 and 3.2 are the representative for inequality index which loose assumption about considering of the ranking from each individual from clustered group. For example, if we are talking about the economist who can publish many articles online. Then we can evaluate inequality index of those article that get published by calculating atkinson’s index or generalized entropy families. Suppose there are three clusters, Atkinson and generalized entropy families never evaluate the individual rank, which means that rank of economists within group and overall population are missed. Until the work of stratification index developed by Yitzhaki S., Lerman. R. (1991). To put it simply, other index are evaluated using “how different of each individual data within, between group and overall”, Conversely, for stratification index by YL are based on “How similar of each individual within cluster, between group and overall”. It seems awkward at the first place but it provides huge benefit to evaluate inequality index and we are able to identify the type of distribution in specific group of data. The following table are based on the stratification index calculation. Table 3.3 Extended Gini with stratification index by YL. (Gender Gap Decomposition). Sgini Gender Cluster Male Female
Mean income
Population Share
Income Share
Group Gini
Group Stratification
10,946.51
45.8%
57.6%
0.667
-0.008
9,577.91
54.2%
42.4%
0.587
0.012
1.00
1.00
0.621
Total Within inequality
0.620
Between inequality
0.002
Stratification
-0.001
BGI Maximization Source: author’s calculation by household survey.
9
Hour works are per day unit, started from two hrs/day till twelve hrs/day. For further details, see Appendix A.
10
0.425
To interpret table above, let’s start with the first column from the left, this is the Gender Gap clustered by stratification measurement between Male and Female, the average income for male is 10k whereas women earned in 2015 approximate 9.5k. Third column referred to number of observation in each cluster. 45.8 goes to male and the rest goes to female. The Group Gini is showing us within subgroup, how stratification about inequality demonstrate to us in a number. For example, there are 66.7 percent of male gender who owns most of income, whereas 58.7 percent in female cluster. The last column is group stratification, this provide a benefit to us about how inequality is measured by not looking at only the extended Gini index (col=5), but rather the distribution of each cluster’s data. To read this column, if stratification = 1 ( CO j 1 ) means that for specific of group j , there is no rank of individual data in that group duplicate in other groups. In this case, YL stated that it is an perfect strata or one specific horizontal layer are booked by one group only. The lower of CO j leads to the less perfect strata, in this sense, other group’s rank are in the same horizontal rank of group j . However if CO j drops until zero, this shows that the individual’s rank in group j is the same as when compare rank in overall population or comparing every data at once. On other words, it never form strata from the first place. In addition to this, as close as CO j get to -1, it proves that one cluster of j can be composed into two identical group if it reaches -1. In this case the divergence of rank within group j is higher than overall population. The stratification of overall observation is -0.001 which is really close to zero, in this case, we can conclude that most of the rank of individual data in group j is the same as the rank when compare everyone at once. Hence, the within group of extended Gini largely account in within inequality measurement, roughly 99.7%. For the sake of brevity we will not present every Sgini table here. Now we will move on to the maximum between group index which is implied and developed by Elbers, Lanjouw, Mistiaen and Ozler11 (results are on Appendix B, Figure B.1 and B.2). As described above that the more divergence between the Rb ' and Rb the higher inequality in different sub-clusters. In the figure B1, Rural and Urban are decomposed by extended Gini and plotted so as to see if there is any inequality between subclusters and within cluster itself. It is worth concluding that the highest inequality presented is Songkhla city which is located in South region in Thailand. Please note that in capital city (Bangkok) have no difference between Rb ' and Rb since this is not a big city and every part of the city is accounted as urbanized province. It is thus no different between sub-clusters. On the other hand, on the gender income gap between male and female is in figure B.2 (for further details at appendix B), the city which has most inequality between both sex is Phuket, follow by Yasothorn, Nongkai and Nonthaburi. Surprisingly, for Bangkok Metropolitan Region12, there is only one city that has excess inequality between both gender (Nonthaburi). Ranked third among all 77 cities in Thailand whereas the expectation for the rest provinces in BMR should be exponentially high relative to others. The possible calculation can be claimed by two reasons. First is the observation collected in each cities are not sufficient and cannot be representative to reflect all of inequality for each. Second, for the last century the educations are developed not for male side only, there are tremendous increase in the number of female who graduate a higher degree especially Master and PhD. Hence, this could lead to the narrower in income gap between both genders. 4. FURTHER VISION ON INEQUALITY DISTRIBUTION IN THAILAND. It’s pretty obvious that the data that I have used to calculate the inequality distribution is still small relatively to the number of population in each city. For instance, population in Bangkok has 11
Re-Interpreting Sub-Group Inequality Decompositions (2005). Bangkok Metropolitan Region contains 6 cities (Bangkok, Nakhonpathom, Nonthaburi, Pathumthani, SamutPrakan, Samutsakhon). 12
officially reported that there are 8.28 million people. However, the sample that we collected was only 2 thousand. Statistically speaking, this is not such an ideal sample to be the representative of all population. In order to tackle this problem efficiently, we need to design the structure process of how we approach the data. In addition to this and most importantly, the algorithm used to apply the complex model must be specified since the start. Otherwise, the system will constantly adjusted during the collecting period and thus lead to a greater problems. APPENDIX A. NO.
Cities
NO.Obs
Proportion
NO.
Cities
NO.Obs
Proportion
1
Angthong
424
1%
41
Phayao
246
1%
2
Aumnadcharoen
421
1%
42
Phetchabun
341
1%
3
Bangkok
2,795
8%
43
Phetchaburi
478
1%
4
Buengkan
297
1%
44
Phichit
306
1%
5
Burirum
376
1%
45
Phitsanulok
296
1%
6
Chachoengsao
506
1%
46
Phrae
427
1%
7
Chainat
413
1%
47
Phranakhonsi
476
1%
8
Chaipum
386
1%
48
Phuket
368
1%
9
Chanthaburi
364
1%
49
Prachinburi
411
1%
10
ChiangMai
404
1%
50
PrachuapKhirikhan
478
1%
11
ChiangRai
167
0%
51
Ranong
397
1%
12
Chonburi
604
2%
52
Ratchaburi
555
2%
13
Chumphon
343
1%
53
Rayong
551
2%
14
Kamphangphet
669
2%
54
Roiet
338
1%
15
Kanchanaburi
401
1%
55
Sakaeo
401
1%
16
Kanlasin
241
1%
56
Sakonnakhon
365
1%
17
Khonkaen
562
2%
57
SamutPrakan
667
2%
18
Krabi
405
1%
58
Samutsakhon
623
2%
19
Lampang
494
1%
59
Samutsongkhram
352
1%
20
Lamphun
466
1%
60
Saraburi
513
2%
21
Lay
212
1%
61
Satun
359
1%
22
Lopburi
437
1%
62
Singburi
524
2%
23
MaeHongSon
304
1%
63
Songkhla
447
1%
24
Mahasarakam
392
1%
64
Srisaket
400
1%
25
Mukdahan
373
1%
65
Sukhothai
388
1%
26
Nakhonnayok
345
1%
66
Suphanburi
262
1%
27
NakhonPathom
513
2%
67
SuratThani
382
1%
28
Nakhonrachasima
582
2%
68
Surin
423
1%
29
Nakhonsawan
436
1%
69
Tak
366
1%
30
Nakhonsithammarat
490
1%
70
Trang
411
1%
31
Nakonpanom
344
1%
71
Trat
362
1%
32
Nan
302
1%
72
Ubon
392
1%
33
Naratiwat
426
1%
73
Udon
312
1%
34
Nongbua
211
1%
74
Uthaithani
286
1%
35
Nongkai
345
1%
75
Uttaradit
298
1%
36
Nonthaburi
658
2%
76
Yala
232
1%
77
Yasothorn
415
1%
37 Pathumthani 734 2% Source: National Statistical Office Thailand (NSO).
APPENDIX B. (EMPIRICAL RESULTS) Table B.1 Extended Gini with stratification index by YL. (Age Decomposition). Sgini Age Cluster
Mean income
Population Share
Income Share
Group Gini
Child
8,416.080
32.2%
0.626
0.020
Teenage
8,794.664
13.3%
0.666
-0.052
Adult
10,811.731
30.8%
0.561
0.105
Mid adult
13,635.182
18.9%
0.634
-0.069
Eldery
10,948.849
4.8%
0.637
-0.055
Total
0.621
Within inequality
0.612
Between inequality
0.011
Stratification
-0.001
BGI Maximization Source: author’s calculation by household survey.
Group Stratification
0.578
Table B.2 Extended Gini with stratification index by YL. (Religion Decomposition). Sgini Religion Cluster
Mean income
Population Share
Income Share
Group Gini
Group Stratification
10,402.494
94.1%
95.1%
0.619
0.036
Christian
8,556.681
5.4%
4.2%
0.573
0.038
Islam
11,575.209
0.5%
0.6%
0.661
0.001
Others
25,270.417
0.0%
0.1%
0.766
0.000
Buddhism
Total
0.621
Within inequality
0.621
Between inequality
0.001
Stratification
0.000
BGI Maximization Source: author’s calculation by household survey.
0.110
Table B.3 Extended Gini with stratification index by YL. (Region Decomposition) Sgini Region Cluster
Mean income
Population Share
Income Share
Group Gini
Group Stratification
BKK Metro
20,003.162
8.3%
32.7%
0.473
0.286
Central
9,428.967
35.6%
15.7%
0.599
0.020
North
8,756.437
18.3%
20.2%
0.658
-0.045
Northeast
9,526.886
21.8%
15.1%
0.657
-0.010
South
10,096.575
16.1%
0.626
-0.014
Total
0.621
Within inequality
0.603
Between inequality
0.035
Stratification
-0.017
BGI Maximization Source: author’s calculation by household survey.
0.584
Table B.4 Extended Gini with stratification index by YL. (Rural-Urban Decomposition) Sgini Area Cluster
Mean income
Population Share
Income Share
Group Gini
Group Stratification
Rural
6,879.66
36.8%
24.2%
0.667
-0.018
Urban
12,275.55
63.2%
75.8%
0.587
0.110
Total
0.621
Within inequality
0.607
Between inequality
0.031
Stratification
-0.016
BGI Maximization Source: author’s calculation by household survey.
0.353
Table B.5 Extended Gini with Stratification index by YL. (Cities Decomposition) City Cluster Bangkok
Mean income 20,003.16
Sgini Population Share 8.3%
Income Share 16.3%
Group Gini 0.163
Group Stratification 0.604
SamutPrakan
9,931.51
2.0%
1.9%
0.019
0.697
Nonthaburi
16,709.97
1.9%
3.2%
0.032
0.473
Pathumthani
11,128.38
2.2%
2.4%
0.024
0.649
Phranakhonsi
10,713.95
1.4%
1.5%
0.015
0.725
Angthong
8,935.89
1.3%
1.1%
0.011
0.512
Lopburi
12,470.86
1.3%
1.6%
0.016
0.690
Singburi
9,552.68
1.5%
1.4%
0.014
0.666
Chainat
8,646.85
1.2%
1.0%
0.010
0.663
Saraburi
9,894.80
1.5%
1.5%
0.015
0.649
Chonburi
11,840.07
1.8%
2.1%
0.021
0.555
Rayong
11,027.11
1.6%
1.7%
0.017
0.366
Chanthaburi
6,825.77
1.1%
0.7%
0.007
0.616
Trat
8,792.47
1.1%
0.9%
0.009
0.781
Chachoengsao
8,678.68
1.5%
1.3%
0.013
0.723
Prachinburi
9,894.73
1.2%
1.2%
0.012
0.643
Nakhonnayok
11,836.33
1.0%
1.2%
0.012
0.591
Sakaeo
9,609.36
1.2%
1.1%
0.011
0.570
Nakhonrachasima
11,724.71
1.7%
1.9%
0.019
0.669
Burirum
8,524.34
1.1%
0.9%
0.009
0.702
Surin
10,236.36
1.2%
1.3%
0.013
0.579
Srisaket
10,273.31
1.2%
1.2%
0.012
0.655
Ubon
10,337.22
1.2%
1.2%
0.012
0.513
Yasothorn
8,722.61
1.2%
1.1%
0.011
0.686
Chaipum
10,077.82
1.1%
1.1%
0.011
0.697
Aumnadcharoen
6,006.93
1.2%
0.7%
0.007
0.581
Buengkan
8,661.33
0.9%
0.7%
0.007
0.667
Nongbua
9,491.51
0.6%
0.6%
0.006
0.636
Khonkaen
10,318.10
1.7%
1.7%
0.017
0.599
Udon
8,330.07
0.9%
0.7%
0.007
0.704
Lay
10,711.60
0.6%
0.6%
0.006
0.610
Nongkai
8,740.24
1.0%
0.9%
0.009
0.595
Mahasarakam
11,293.93
1.2%
1.3%
0.013
0.606
Roiet
9,074.96
1.0%
0.9%
0.009
0.561
Kanlasin
8,830.76
0.7%
0.6%
0.006
0.641
Sakonnakhon
8,324.30
1.1%
0.9%
0.009
0.438
Nakonpanom
11,374.38
1.0%
1.1%
0.011
0.418
Mukdahan
8,056.51
1.1%
0.9%
0.009
0.705
ChiangMai
4,575.29
1.2%
0.5%
0.005
0.518
Lamphun
5,833.85
1.4%
0.8%
0.008
0.716
Lampang
10,312.02
1.5%
1.5%
0.015
0.643
Uttaradit
12,484.24
0.9%
1.1%
0.011
0.686
Phrae
8,107.15
1.3%
1.0%
0.010
0.653
Nan
10,977.67
0.9%
0.9%
0.009
0.682
Phayao
7,307.16
0.7%
0.5%
0.005
0.519
ChiangRai
8,405.59
0.5%
0.4%
0.004
0.707
MaeHongSon
12,593.25
0.9%
1.1%
0.011
0.510
Nakhonsawan
10,028.72
1.3%
1.3%
0.013
0.382
Uthaithani
9,516.79
0.8%
0.8%
0.008
0.611
Kamphangphet
4,041.00
2.0%
0.8%
0.008
0.615
Tak
12,089.25
1.1%
1.3%
0.013
0.569
Sukhothai
8,713.74
1.1%
0.9%
0.009
0.747
Phitsanulok
12,127.80
0.9%
1.1%
0.011
0.549
Phichit
11,425.45
0.9%
1.0%
0.010
0.710
Phetchabun
6,778.36
1.0%
0.7%
0.007
0.597
Ratchaburi
6,220.60
1.6%
1.0%
0.010
0.664
Kanchanaburi
5,040.00
1.2%
0.6%
0.006
0.546
Suphanburi
6,480.86
0.8%
0.5%
0.005
0.736
NakhonPathom
8,249.48
1.5%
1.2%
0.012
0.646
Samutsakhon
5,035.67
1.8%
0.9%
0.009
0.545
Samutsongkhram
5,769.28
1.0%
0.6%
0.006
0.626
Phetchaburi
8,689.62
1.4%
1.2%
0.012
0.674
PrachuapKhirikhan
7,638.50
1.4%
1.0%
0.010
0.566
Nakhonsithammarat
7,650.70
1.4%
1.0%
0.010
0.667
Krabi
10,802.91
1.2%
1.3%
0.013
0.698
Phangnga
10,710.54
0.9%
0.8%
0.008
0.677
Phuket
11,687.46
1.1%
1.3%
0.013
0.522
SuratThani
9,775.00
1.1%
1.0%
0.010
0.632
Ranong
8,746.93
1.2%
1.0%
0.010
0.607
Chumphon
11,703.59
1.0%
1.0%
0.010
0.667
Songkhla
11,825.67
1.3%
1.5%
0.015
0.606
Satun
10,645.08
1.1%
1.1%
0.011
0.590
Trang
8,932.23
1.2%
1.0%
0.010
0.664
Phatthalung
7,226.18
1.1%
0.8%
0.008
0.629
Pattani
6,502.24
1.5%
0.8%
0.008
0.527
Yala
18,196.23
0.7%
1.2%
0.012
0.798
Naratiwat
11,449.08
1.3%
1.2%
0.012
0.699
Total
0.621
Within inequality
0.584
Between inequality
0.061
Stratification
-0.024
BGI Maximization Source: author’s calculation by household survey.
0.614
Figure B.1
Rural and Urban Decomposed using extended Gini coefficient by YL calculation Yasothorn Yala Uttaradit Uthaithani Udon Ubon Trat Trang Tak Surin SuratThani Suphanburi Sukhothai Srisaket Songkhla Singburi Satun Saraburi Samutsongkhram Samutsakhon SamutPrakan Sakonnakhon Sakaeo Roiet Rayong Ratchaburi Ranong PrachuapKhirikhan Prachinburi Phuket Phranakhonsi Phrae Phitsanulok Phichit Phetchaburi Phetchabun Phayao Phatthalung Phangnga Pattani Pathumthani Nonthaburi Nongkai Nongbua Naratiwat Nan Nakonpanom Nakhonsithammarat Nakhonsawan Nakhonrachasima NakhonPathom Nakhonnayok Mukdahan Mahasarakam MaeHongSon Lopburi Lay Lamphun Lampang Krabi Khonkaen Kanlasin Kanchanaburi Kamphangphet Chumphon Chonburi ChiangRai ChiangMai Chanthaburi Chaipum Chainat Chachoengsao Burirum Buengkan Bangkok Aumnadcharoen Angthong
0
0.1
0.2
0.3
0.4
Rb
Rb'
0.5
0.6
0.7
0.8
Figure B.2
Gender Gap decomposed using extended Gini coefficient by YL calculation Yasothorn Yala Uttaradit Uthaithani Udon Ubon Trat Trang Tak Surin SuratThani Suphanburi Sukhothai Srisaket Songkhla Singburi Satun Saraburi Samutsongkhram Samutsakhon SamutPrakan Sakonnakhon Sakaeo Roiet Rayong Ratchaburi Ranong PrachuapKhirikhan Prachinburi Phuket Phranakhonsi Phrae Phitsanulok Phichit Phetchaburi Phetchabun Phayao Phatthalung Phangnga Pattani Pathumthani Nonthaburi Nongkai Nongbua Naratiwat Nan Nakonpanom Nakhonsithammarat Nakhonsawan Nakhonrachasima NakhonPathom Nakhonnayok Mukdahan Mahasarakam MaeHongSon Lopburi Lay Lamphun Lampang Krabi Khonkaen Kanlasin Kanchanaburi Kamphangphet Chumphon Chonburi ChiangRai ChiangMai Chanthaburi Chaipum Chainat Chachoengsao Burirum Buengkan Bangkok Aumnadcharoen Angthong
0
0.02
0.04
0.06
0.08 Rb
Rb'
0.1
0.12
0.14
0.16
0.18
Acknowledgement Special thanks go to the National Statistic Office of Thailand for the generosity to provide the household survey of income. In addition, I send appreciation to Mae Fah Luang University and Office of Border Economy and Logistics Study (OBELS). References Atkinson A.B. (1970). On the Measurement of Inequality. Journal of Economic Theory, 2,pp.244263.
Shorrocks (1980). The Class of Addictively Decomposable Inequality Measures. Econometrica, Vol. 48 No. 3, Apr. 1980. Blackorby C., Donaldson D., Auersperg M. (1981). A new procedure for the measurement of inequality within and among population subgroups. Canadian Journal of Economics, 14, pp.665-685. Elbers C., Lanjouw P., Mistiaen J.A., Ozler B. (2005). Re-Interpreting Sub-Group Inequality Decompositions. World Bank, World Bank Policy Research Working Paper 3687, 42 p. Schechtman E., Yitzhaki S. (2008). Calculating the Extended Gini Coefficient from Grouped Data: A Covariance Presentation. Bulletin of Statistics & Economics, 2(S08), pp.64-69. Yitzhaki S., Lerman R. (1991). Income Stratification and Income Inequality. Review of Income and Wealth, 37(3), pp.313-29.
Cowell and Jenkin. (1995). How Much Inequality Can We Explain? A Methodology and an Application to the United States. Economic Journal. Vol. 105 issue. 429, 421-30. Cowell F.A. (2000). Measurement of Inequality. In Atkinson A.B., Bourguignon F.(Eds.) Handbook of Income Distribution. Elsevier, Vol. 1, pp.87-166.
Cowell, F, A. (2009). Measuring Inequality. Oxford University Press. S.W. Paweenawat, R. McNown (2014). The determinants of income inequality in Thailand: A synthetic cohort analysis. Journal of Asian Economics 31-32, pp.10-21.