URBAN-RURAL DIFFERENCES IN COSTS OF LIVING AND THEIR IMPACT ON ...

URBAN-RURAL DIFFERENCES IN COSTS
OF LIVING AND THEIR IMPACT
ON POVERTY MEASURES
Abuzar Asra
Asian Development Bank
This note demonstrates empirically the importance of urban-rural price
differences and inflation figures in poverty analysis. Using data from the
National Socio-Economic Survey (Survei Sosial Ekonomi Nasional, widely
known as Susenas), it shows that the urban-rural food price differential during
the period 1987–1996 was 13–16%, not 28–52% as implied by the ‘official’ food
poverty lines. The urban-rural poverty comparisons and the components of
change in simulated poverty estimates presented here therefore differ from those
based on the ‘official’ figures. They indicate that migration to urban areas
between 1987 and 1996 accounts for a significant part of the observed decline in
poverty. The paper concludes that it is essential to use accurate urban-rural cost
of living differences in deriving aggregate, urban and rural poverty estimates.
INTRODUCTION
Studies of welfare in Indonesia have been hindered by a lack of appropriate price
indices. For example, to investigate real income changes among different groups,
estimates of price increases experienced by each group are required. Similarly, to
compare the standards of living of people in different localities, appropriate price
ratios among localities are needed. However, sufficient data (of reasonable
quality) are not yet available. This has generated research conclusions that are
either tentative or misleading as a basis for policy decisions.
The importance of having a proper price index for comparisons of costs of
living between urban and rural areas has been underlined by Ravallion (1992)
and Booth (1993).1 Ravallion examined the problems encountered in comparing
urban and rural poverty using the urban and rural poverty lines developed by
Biro (now Badan) Pusat Statistik (BPS, the Central Statistics Agency). He argued
that the urban-rural differential of about 70% in these poverty lines for 1984–1987
was far above actual urban-rural cost of living differences. This meant that
migration of someone previously living above the poverty line in a rural area to
an urban area where he or she enjoyed a higher actual standard of living that

---

was nevertheless below the official urban poverty line would result in an
increase in recorded aggregate poverty. In addition, urban-rural poverty
comparisons would not reveal the true situation. This led Ravallion to question
the methodology used by BPS in calculating its poverty lines. Likewise, Booth
(1993) and Asra and Virola (1992) argued that the use of a cost of calories method
(a variant of the food-energy-intake method that is heavily dependent upon the
unit price of calories) in deriving the BPS poverty lines until 1993 generated non-
comparable urban and rural poverty lines.
It is important that data on the geographic distribution of poverty should
not be distorted by an unrealistic relationship between urban and rural poverty
lines. Perceived regional differences in poverty can have a significant influence
on policy in areas such as migration. Most obviously, Indonesia’s transmigration
program has been based on the notion that poverty can be reduced by shifting
people from poor areas to better off—or potentially better off—regions.
Conversely, to the extent that spontaneous rural-urban migration is wrongly
believed to be associated with increasing poverty, there are likely to be attempts
to prevent or restrict it, such as refusal by urban authorities to issue population
identity cards to migrants from rural areas. In turn, this will prevent individuals
from moving to areas where they judge that they will have better income earning
opportunities, thus harming rather than helping poverty alleviation.
OVERALL POVERTY AND FOOD POVERTY
In its first publication of poverty figures, BPS (1984) defined two poverty lines:
batas miskin (‘poor line’, henceforth referred to as the ‘overall poverty line’ OPL)
and batas sangat miskin (the ‘very poor line’). The latter in fact refers to the level of
income needed to cover expenditure on the food component of the expenditure
basket reflected in the OPL, and will be referred to henceforth as the ‘food
poverty line’ (FPL). Note that since 1984 BPS has not reported figures for ‘food
poverty incidence’, although in some of its publications ‘food poverty lines’
(meaning the food components of the overall poverty lines) are presented. Thus
the term ‘official food poverty line’ as used here refers to the food component of
the BPS overall poverty line, and ‘official food poverty incidence’ refers to food
poverty incidence measures derived from these BPS ‘food poverty lines’.
The 1998 poverty lines were derived using a much smaller sample survey
(10,000 households) than the usual Susenas, whose sample size is around 65,000
2

---

households. Thus, in contrast with the 1996 poverty data, the 1998 survey does
not allow the computation of poverty lines by province. This small survey was
conducted to meet the urgent demand for poverty monitoring of the social
impact of the recent financial crisis. The standard Susenas survey was conducted
in early 1999, and estimates of poverty incidence based on the results of this
survey will probably be published by the end of 1999.
The 1998 poverty lines developed by BPS result in a ratio of urban to rural
overall poverty lines of 1.33 and of the urban to rural food poverty lines of 1.25.
The food bundle method used to derive the poverty lines allows for different
food consumption patterns (i.e. consumption of different quantities of food
items) in urban and rural areas (BPS 1999), whereas the methodology used until
1993 allowed for different calorie consumption patterns. The reason for using
different sets of quantity weights for urban and rural areas is to reflect the
specific characteristics of each area, so that the poverty line is ‘location specific’.
Whether focused on food or calorie consumption, however, the approach
used by BPS results in a loss of comparability across areas: that is, spatial
consistency is lost (Asra 1998: 22), as the independently derived urban and rural
poverty lines reflect different food consumption patterns.2 In other words, there
is a trade-off between specificity and consistency. As Ravallion and Bidani (1994:
77) suggest the measurement choice rests ultimately on the purpose of the
‘poverty profile’ being constructed.
Differences in urban and rural food poverty lines that far exceed actual
urban-rural cost of living (COL) differences will provide a misleading picture of
the distributional impact of development, which usually involves urban sector
expansion. In a dualistic economy, one of the ways the poor benefit from
development is through expansion of job opportunities in the modern—
predominantly urban—sector, in addition to increases in productivity in the
traditional—predominantly rural—sector (Fields 1980). In defining poverty lines
for urban and rural areas, therefore, one should ensure that they take into
account differences in the COL across these areas. This suggests the need for
properly determined ratios of urban to rural prices.
Price data have been collected in Indonesia since well before the 1960s
(Sastrosumarto 1995). For urban areas before 1979, prices of only 62 items were
3

---

recorded from several markets in Jakarta. Now, the number of items has reached
about 353, with coverage of 44 cities (27 provincial capital cities and 17 other
large cities). This urban consumer price survey is the basis for deriving the
widely used Consumer Price Index (CPI). In rural areas, a consumer price survey
that includes more than 500 rural markets in 26 provinces (with retail traders as
respondents) is carried out. It covers 138 food items and food materials and 163
non-food items.3 Together with the results of the rural producer price survey,
where farmers are the respondents, the results of the rural consumer price
survey are used to derive the farmers’ terms of trade (the ratio of the index of
prices received by farmers to the index of prices paid by farmers). Data from
these surveys, in addition to those from the Susenas and BPS’s COL surveys,
could be used to compute urban and rural price indices (and the corresponding
urban and rural inflation rates) if desired.
Recent studies show that estimates of poverty level are heavily dependent
upon the inflation rates used (Sigit and Sudarti 1999: 25–6, and Frankenberg,
Thomas and Beegle 1999: 15). This note presents ratios of urban to rural food
prices, and urban and rural food price indices, using data from the 1987, 1993
and 1996 Susenas tapes. It also gives empirical examples of how urban-rural
poverty comparisons, aggregate poverty estimates and the sectoral
decomposition of poverty changes would have been altered if appropriate food
price indices had been employed to derive urban and rural poverty lines. It is
hoped that the note may encourage the data users, in particular those responsible
for analyzing poverty statistics, to understand fully how poverty lines and
incidences are constructed before using them in poverty analysis. In addition, it
is expected that this exercise will lead to greater awareness of the urgent need for
additional and more appropriate price indices.
DATA
The data used in this exercise are food items, so the price indices derived are
food price indices; the share of food in total expenditure was about 48 percent in
urban areas and about 63 percent in rural areas in 1996. Susenas records the
value of expenditure on non-food items but not their quantity, so their implicit
prices cannot be derived. In principle, with additional effort to find appropriate
prices for these items—for example, by using the prices from BPS’s urban and
rural consumer price surveys—an overall price index might be produced, but
4

---

unfortunately these surveys do not cover all the Susenas non-food items. In the
future, as the importance of using better and more appropriate price indices
becomes more widely appreciated, this shortcoming should be overcome. In any
case, given the lack of overall price indices that include food and non-food items,
it can be argued that using food price indices for poverty analysis remains a valid
approach, on the grounds that food is the most essential consumption item.
On average, about 160 food items are used here in deriving the price
indices; some food items that were not consumed in both urban and rural areas
in the years being compared are not used. Susenas records the value of
expenditure and quantity consumed for each food item, allowing the derivation
of implicit prices. The quantities consumed and the implicit prices are then used
to calculate both urban and rural food price indices.
For the purpose of comparison, urban and rural non-food price indices
based on a limited number of non-food items were also calculated, using data
from BPS’s rural and urban consumer price surveys. Of the 50 non-food items
processed in the rural consumer price survey, 32 items are also covered in the
urban consumer price survey.4 Thirty-one non-food items were included in the
computation.5
In deriving the alternative sets of food poverty lines presented below,
‘official food poverty line’ estimates are used. The percentage distribution of
population by monthly per capita expenditure class, estimated separately for
urban and rural areas and published in the Susenas, is employed to estimate
food poverty incidence figures based on the ‘official’ and derived food poverty
lines. The alternative estimates of food poverty lines and food poverty incidence
presented here are not offered as the only ‘correct’ figures, but merely as
illustralive of a methodological issue.
METHODOLOGY
There are two ways of deriving the implicit price of each food item from the
Susenas data. The first is by dividing the value of expenditure on a particular
food item by its total quantity. The second is by doing this at the household level
and then taking the average of implicit prices paid by all households consuming
the item. Statistically, the second procedure is superior, but for the sake of
5

---

practicality the first procedure is used. In calculating the price indices, Laspeyres’
formula was used.6
The formula used to show changes in food prices over time is:
j
j
j
p q
∑ t o
p t
j
I =
=
W
∑
o
j
j
j
p q
p
∑ o o
o
where:
pj is price and qj is quantity for food item j; t and o denote time t
and the base year, respectively; and Woj is the share of food item j in total
food expenditure in the base year.
The ratios of urban to rural food prices (food price indices) were also
derived using a Laspeyres’ formula:
∑ j j
p q
.
u
r
I = j
∑
,
j
j
p q
.
r
r
j
where Pj is price and Qj is quantity for food item j; and u and r denote urban and
rural, respectively. This is the ratio of the cost of the rural food consumption
basket valued at urban prices to its value at rural prices.7 It can be re-expressed
as
j
pu
j
I =
w
.
∑
,
j
r
p
j
r
where
j
w is the share of expenditure on food item j in total rural food
r
expenditure.
The impact of differences in food prices and food price inflation on urban-
rural poverty comparisons, aggregate poverty, and the sectoral decomposition of
changes in poverty is highlighted by the re-estimation of the poverty incidence
figures. First, new sets of food poverty lines are derived using the ‘official’ food
poverty line estimates and the urban-rural food price ratios. Then the poverty
incidence figures are obtained using the POVCAL program (Chen, Datt and
Ravallion, undated), which estimates (among other things) the Lorenz curve, the
Gini index, poverty incidence and other indices as suggested by Foster, Greer
6

---

and Thorbecke (1984). The Lorenz curve plots the cumulative expenditure or
income of individuals or households accruing to the cumulative percentage of
individuals or households. The deviation of the curve from the 45° diagonal line
(showing perfect equality) indicates the extent of inequality: the closer it is to the
diagonal line the lower the inequality. The Gini index is a summary measure of
the inequality level of an income or expenditure distribution; it ranges between
zero and one, indicating perfect equality and perfect inequality respectively. it
can be calculated from the Lorenz curve or using various formulae (see Kakwani
1980, for instance). Two alternative specifications of the Lorenz curve are used—
the General Quadratic and the Kakwani Beta model—and the program
determines which specification fits the data better by comparing the sum of
squared errors over the part of Lorenz curve bounded by the headcount index of
poverty. The preferred estimates of poverty incidence are chosen depending on
which specification of the Lorenz curve fits better.
Following Ravallion and Huppi (1991), the change in aggregate poverty
can be decomposed into the intrasectoral effect (changes in urban and rural
poverty at the initial population shares), the intersectoral effect (changes in
poverty arising from population shifts from rural to urban areas), and the
interaction between these two. The decomposition is conducted using the
formula introduced by Ravallion and Huppi (1991), as follows:
r
r
P1 − P0 = (P1
P 0 )n0
(P1
P0 )n0
(n1
n0 P
) 0
(P1
P 0 )(n1
n0 )
u −
u
u +
r −
r
r + ∑
i −
i
i
+ ∑ i − i
i −
i
i=u
i=u
where P is poverty measures and n is population shares for each of two dates (0,
(0,1), and two areas or sectors (i = u for urban and r for rural).
The first two terms on the right comprise the intersectoral effects of the
changes in urban and rural poverty at the initial population shares. The third
term is the intersectoral effect (change in poverty arising from population shift),
and the last term is the effect of interaction between sectoral changes and
population shifts.
A similar but superior analysis is conducted later by employing both
urban-rural price ratios and food price inflation rates simultaneously. Although
7

---

the other poverty indices as suggested by Foster, Greer and Thorbecke (1984)
were computed8, only estimates of poverty incidence are presented here.
RESULTS
Ratios of Urban to Rural Prices
Ratios of urban to rural food prices and (unweighted) non-food prices are
given in table 1,9 together with the ratios of BPS’s urban and rural poverty lines
(both overall and food), for comparison. It is interesting to note that the poverty
line ratios declined during 1987–96, bringing them closer to the ratios of urban to
rural (food and non-food) prices. The previously mentioned change in BPS
methodology for deriving poverty lines from focusing on food consumption, is
partly responsible for the decline in the ratios of the urban-rural poverty lines.
Table 1 Ratios of Urban to Rural Prices and Urban to Rural Poverty Lines, 1987–1996
Ratio of BPS’s urban to rural
Year
Food
Non-Fooda
poverty lines
Overallb
Foodc
1987
1.13
1.26
1.69
1.52
1993
1.16
1.19
1.58 (1.53)
1.40 (1.50)
1996
1.16
1.12
(1.39)
(1.28)
a Unweighted.
b Taken from Asra (1998). Figures in brackets are based on BPS’s new (food bundle) methodology
for deriving its poverty lines (BPS 1994). The 1998 figure is 1.33 (BPS 1999).
cTaken from Asra (1995). Figures for 1987 are calculated from Asra (1997, table 4). Figures in
brackets are based on the new BPS methodology. The 1998 figure is 1.25 (BPS, 1999).
Between 1987 and 1996, the overall food price differential between urban
and rural areas, as indicated by the ratios in table 1, ranged from 13 to 16%.
These differentials were substantially smaller than those implied by the ratios of
the ‘official’ urban to rural food poverty lines.10 For instance, in 1987 food prices
in urban areas were 13% higher than in rural areas, whereas the ratio of the
urban to rural food poverty lines implied a differential of 52%. The urban to rural
overall poverty line ratio implied an even higher differential of 69%. The non-
8

---

food price differentials were also lower than the ratios of urban to rural food and
overall poverty lines in each of the years shown. In short, the urban-rural
differentials in the BPS poverty lines and, in particular, the food poverty lines,
were considerably in excess of the price differentials for both food and non-food
commodities between these areas.
Poverty Measurement and Analysis Using Spatially Consistent Food Price
Ratios
This section analyses the impact on urban-rural poverty comparisons of using
the urban-rural food price ratios shown in the first column of table 1. The aim of
this exercise is to estimate urban and rural poverty incidence using the
appropriate comparable food price indices, to compare these derived results with
those based on the ‘official’ food poverty levels, and recalculate the aggregate
poverty level.
The derived food poverty lines are computed using the ratios of urban
and rural food prices given in table 1. The exercise is done for two cases: case I
uses the ‘official’ rural food poverty lines as a basis, while case II uses the
‘official’ urban food poverty lines:
• case I derives an urban poverty line by multiplying the ‘official’ rural
poverty line by the urban-rural price ratio from table 1; e.g. for 1987
case I multiplies Rp9,710 (the ‘official’ rural poverty line) by 1.13 (the
urban-rural price ratio) to derive an urban poverty line of Rp10,972;
• case II derives a rural poverty line by dividing the ‘official’ urban
poverty line by the urban-rural price ration; e.g. for 1987 case II divides
Rp14,760 (the ‘official’ urban poverty line) by 1.13 (the urban-rural
price ratio) to derive a rural poverty line of Rp13,062.
Table 2 shows the comparison between the ‘official’ and the derived food
poverty lines. As expected, in case I, the urban food poverty lines are lower than
the corresponding ‘official’ poverty lines, while in case II, the rural food poverty
lines are higher than the corresponding ‘official’ food poverty lines.
9

---

Table 2 Official and Derived Food Poverty Lines
(Rp per capita per month)
Urban
Rural
Year
Official
Derived Case I
Official
Derived Case II
1987
14,760
10,972
9,710
13,602
1993
23,303
18,068
15,576
20,089
1996
29,681
26,908
23,197
25,587
Sources: The 1993 and 1996 figures are taken from BPS (1994: 21) and BPS (1999: 5), while the 1987
figures are recalculated from Asra and Virola (1992: table 3). The 1993 figures are based on the
new methodology described in the text.
Alternative sets of poverty incidence figures based on ‘official’ food
poverty lines and derived poverty lines were obtained using the POVCAL
program (table 3). As they are based on food poverty lines rather than overall
poverty lines (in which non-food living costs are also taken into consideration),
the poverty incidence measures derived here reflect the ‘very poor line’ referred
to earlier, rather than the more usual ‘poor line’, which measures overall poverty
incidence based on consumption of both food and non-food items.
Table 3 shows that in 1987, the rural poverty incidence based on the
‘official’ food poverty line was higher than poverty in urban areas. The figures
based on the official overall poverty line show the reverse (Asra 1998: 5). Both
‘official’ and derived estimates show that in 1987-96 food poverty was higher in
rural than in urban areas, but the derived estimates show a much greater
difference. All estimates show a decline in food poverty incidence during this
period. Case II suggests a sharper trend, however—especially for rural areas,
where food poverty incidence fell from around 35% in 1987 to about 9% in 1996.
Aggregate poverty estimates for 1987, 1993 and 1996 are also presented in
table 3. Case I, based on the derived urban food poverty lines, yields food
poverty incidence figures a little lower than those based on the ‘official’ food
poverty lines. But case II, based on the derived rural food poverty lines, results in
far higher estimates of food poverty incidence in the earlier years, and a far more
rapidly declining trend. This illustrates clearly the sensitivity of poverty
measures to urban–rural price differences.
10

---