• TABLE OF CONTENTS
HIDE
 Front Cover
 Front Matter
 Title Page
 Table of Contents
 List of Tables
 Foreword
 Acknowledgement
 Introduction
 A disaggregated model of national...
 Taiwan household expenditure...
 Provincial-level data from...
 Conclusion
 Reference
 Back Cover






Title: Structural changes in the demand for food in Asia
CITATION THUMBNAILS PAGE IMAGE ZOOMABLE
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/UF00085375/00001
 Material Information
Title: Structural changes in the demand for food in Asia
Alternate Title: Food, agriculture, and the environment discusson paper ; 11
Physical Description: Book
Language: English
Creator: Huang, Jikun.
Bouis, Howarth
Publisher: International Food Policy Research Institute
Place of Publication: Washington, D. C.
Publication Date: March, 1996
Copyright Date: 1996
 Record Information
Bibliographic ID: UF00085375
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: 34963955 - OCLC

Table of Contents
    Front Cover
        Front Cover
    Front Matter
        Page i
    Title Page
        Page ii
    Table of Contents
        Page iii
    List of Tables
        Page iv
        Page v
        Page vi
    Foreword
        Page vii
    Acknowledgement
        Page viii
    Introduction
        Page 1
    A disaggregated model of national food demand
        Page 2
        Page 3
    Taiwan household expenditure surveys
        Page 4
        Page 5
        Page 6
        Page 7
        Page 8
        Page 9
        Page 10
        Page 11
        Page 12
        Page 13
        Page 14
    Provincial-level data from China
        Page 15
        Page 16
        Page 17
    Conclusion
        Page 18
        Page 19
        Page 20
        Page 21
        Page 22
    Reference
        Page 23
        Page 24
    Back Cover
        Page 25
Full Text













Structural Changes in
the Demand for Food in Asia


Jikun Huang and
Howarth Bouis


20)20
VISION








































"A 2020 Vision for Food, Agriculture, and the Environment" is an initiative of
the International Food Policy Research Institute (IFPRI) to develop a
shared vision and a consensus for action on how to meet future world food
needs while reducing poverty and protecting the environment. It grew out
of a concern that the international community is setting priorities for
addressing these problems based on incomplete information. Through the
2020 Vision initiative, IFPRI is bringing together divergent schools of
thought on these issues, generating research, and identifying recommendations.

This discussion paper series presents technical research results that encom-
pass a wide range of subjects drawn from research on policy-relevant
aspects of agriculture, poverty, nutrition, and the environment. The discus-
sion papers contain material that IFPRI believes is of key interest to those
involved in addressing emerging Third World food and development prob-
lems. These discussion papers undergo review but typically do not present
final research results and should be considered as works in progress.






Food, Agriculture, and the Environment Discussion Paper 11


Structural Changes in

the Demand for Food in Asia


Jikun Huang and Howarth Bouis



















International Food Policy Research Institute
1200 Seventeenth Street, N.W.
Washington, D.C. 20036-3006 U.S.A.
March 1996















Contents



Foreword vii
Acknowledgments viii
A Disaggregated Model of National Food Demand 2
Taiwan Household Expenditure Surveys 4
Provincial-level Data from China 15
Conclusions 18
References 23















Tables



1. Per capital annual food consumption, Taiwan, 1940-92 5

2. Indexes of real income, expenditure, and real food price
levels for food groups, 1971, 1981, and 1991 5

3. Annual per capital food consumption, by region and
expenditure group, Taiwan, 1981 6

4. Annual per capital food consumption, by region and
expenditure group, Taiwan, 1991 6

5. Annual per capital food consumption, by year, occupation,
and expenditure group, Taiwan 7

6. Change in per capital food consumption between 1981 and 1991,
for income quintile 3 and the national average, Taiwan 7

7. Nonlinear FIML parameter estimates of a dynamic AIDS
model, Taiwan 10

8. Budget shares and expenditure elasticities evaluated at the
sample mean, Taiwan, 1981-91 11

9. Price elasticities evaluated at the sample mean, Taiwan, 1981-91 11

10. The effects of urbanization and occupation changes on annual
per capital food consumption in Taiwan between 1981 and 1991 11

11. The effects of demographic changes on annual per capital
food consumption in Taiwan between 1981 and 1991 12

12. Structural, income, and price factors contributing to differences
in per capital food consumption between cities and villages,
Taiwan, 1981 13

13. Structural, income, and price factors contributing to differences
in per capital food consumption between towns and villages,
Taiwan, 1981 13




















14. Structural, income, and price factors contributing to changes
in per capital food consumption in villages in Taiwan between 1981
and 1991 14

15. Annual per capital food consumption, rural and urban, China, 1991 15

16. Annual per capital food consumption by income group, China, 1991 16

17. Tests of differences in the consumption pattern between rural
and urban populations, China 16

18. Estimated parameters for the complete demand system,
slope dummy model, China 17

19. Estimated parameters for the complete demand system, intercept
dummy model, China 18

20. Estimated parameters for meat and the meat product demand
subsystem, slope dummy model, China 19

21. Estimated parameters for meat and the meat product demand
subsystem, intercept dummy model, China 20

22. Expenditure and own price elasticities of demand for the
commodity group evaluated at the overall sample mean, China 20

23. Meat expenditure and own price elasticities of demand for the
commodity group evaluated at the sample mean, China 21

24. The effects of urbanization on per capital annual food consumption
in China 21

25. The effects of urbanization on per capital annual consumption of
meat, fish, and their products in China 22















Foreword


Many Asian countries are expected to undergo structural transformations in their economies
and rapid urbanization over the next 25 years. The changes in tastes and lifestyles engendered
by urban living are likely to have significant influences on food demand-influences perhaps
as strong as the well-documented influences of household incomes and food prices. Changes
in marketing systems and occupational changes, closely linked with increasing GNP per
capital, also may influence the demand for food.
In this paper, estimates presented for Taiwan and China demonstrate that structural
changes in food demand (as distinguished from changes due to income and price effects) have
been significant factors driving the rapid changes in dietary patterns seen in East Asia over the
past three decades. Because most previous demand studies have ignored the possible influence
of structural shifts, which are highly correlated with increases in per capital income over time,
the effects of income on food demand have been overestimated. Relatively little is known
about the specific reasons underlying these structural shifts. Thus, it is difficult to form
judgments as to the points at which, in the process of economywide structural adjustment,
these structural changes in food demand will begin, accelerate, slow down, and perhaps stop.
The crucial question for demand projections in higher-income countries of Asia is the
determination of when these structural shifts will decelerate, at which point the upwardly
biased estimates of income elasticities for meat, fish, and dairy products will begin to
overestimate future demand. In lower-income countries, such as India and Indonesia, where
meat consumption is presently quite low because structural transformation is in an early stage,
time series data for meat consumption will not reflect a possible impending upward structural
shift in demand. Thus, existing income elasticities will underestimate the demand for meat if
these structural changes in food demand do indeed materialize.
This report establishes that structural changes can have large influences on food demand
patterns and so points toward the need for further research to understand the underlying factors
that account for these demand shifts.

Per Pinstrup-Andersen
Director General, IFPRI














Acknowledgments


The authors gratefully acknowledge the help of Wen Cher and Hwang-Jaw Lee in making the
Taiwan data available to us. We also thank Julian Alston and Scott Rozelle for helpful
comments.




















Sn the whole, direct per capital consumption of
cereal as food has declined over the past
three decades in the rapidly growing economies of
Japan, Korea, and Taiwan, while meat, fish, and
dairy consumption has increased dramatically. Typi-
cally, economists have explained such changes in
Asian food consumption patterns primarily as result-
ing from increases in disposable income and changes
in food prices (for example, Ito, Peterson, and Grant
1989; Capps et al. 1994).
There is no doubt that household income and
food prices strongly influence food consumption
patterns. This fact is perhaps as well substantiated
empirically as any relationship in the economics lit-
erature. Nevertheless, in projecting food demand
patterns over the long run, particularly in economies
undergoing the rapid structural transformation and
urbanization expected to occur in many countries in
Asia over the next 25 years, changes in tastes and
lifestyles also may be important influences on food
demand.
Unfortunately, methodologies for measuring the
effects of such changes in tastes and lifestyles have
not been as well developed as those for measuring
income and price effects. Moreover, because these
changes in tastes and lifestyles are strongly corre-
lated with increasing gross national product (GNP)
per capital, it is difficult to separate the two effects
empirically in time-series estimations.
A case in point is consumption of rice in Japan,
which has declined from 131 kilograms to 74 kilo-
grams per capital between 1962 and 1992 (FAO
1994). There is no denying the negative correlation
between rice consumption and the rapid increase in
Japanese disposable income that occurred during


this period. The question raised here is whether this
correlation also implies causality, so that projected
future increases in income are a good indicator of
future shifts in demand. That is, as Japanese incomes
continue to increase, will Japanese rice consumption
continue to decline to 60 kilograms, and then to 50
kilograms per capital, and so on, before leveling off?
At what point will per capital consumption of rice
stabilize?
There are a number of reasons to think that there
may be structural shifts (as distinguished from in-
come and price effects) in food demand patterns as
populations move from rural to urban areas:

1. There may be a wider choice of foods avail-
able in urban markets.'
2. Urban residents are more likely to be exposed
to the rich variety of dietary patterns of for-
eign cultures.
3. Urban lifestyles may place a premium on
foods that require less time to prepare, for
example, if employment opportunities for
women improve and the opportunity cost of
their time increases.2
4. Urban occupations tend to be more sedentary
than rural ones. Persons engaged in more
sedentary occupations require fewer calories
to maintain a given body weight.
5. Urban residents typically do not grow their
own food. Thus, their consumption choices
are not constrained by the potentially high-
cost alternative of selling one food at farm-
gate prices (say, rice) to buy another food


'Some evidence of this phenomenon for China is provided in Huang and Rozelle 1995.
2Some evidence of this for Sri Lanka is provided in Sahn and Alderman 1988.










(say, bread) at retail prices (a choice faced by
semisubsistence producers).3
While changes in food demand patterns that can-
not be attributed to increases in household incomes
and changes in food prices may first be noticed in
urban areas, as structural transformation proceeds to
a more advanced level, these same changes in food
demand patterns eventually may occur in rural areas
as well. At some point, market availability and life-
styles in urban and rural areas become virtually in-
distinguishable.
Because structural shifts in food demand pat-
terns have been ignored for the most part in previous
Asian food demand studies,4 the primary objective
of this paper is simply to establish their importance
empirically, using data from Taiwan and China. It is
left to subsequent studies to attribute such shifts to
the several explanatory factors (and perhaps others)
outlined above.
After providing a more rigorous definition of an
hypothesized shift in food preference patterns as
societies urbanize and agriculture commercializes,
household survey data for Taiwan for two years,
1981 and 1991, are analyzed. These data permit
comparison of food consumption patterns in farm
and nonfarm households after differences in family
income are controlled, as well as comparison of food
consumption in rural areas, towns, and cities. An
analysis of provincial-level data for China, disaggre-
gated by rural areas, small cities, and large cities, is
then presented and conclusions are drawn.

A Disaggregated Model of
National Food Demand
In this section a framework is developed for under-
standing the effects on national aggregate demand


for food staples brought about by the shifts described
in (4) and (5) above. Then the effects of lifestyle
changes described in (1), (2), and (3) are incorpo-
rated into this framework in the context of demand
for nonstaple foods.


Demand for Food Staples

As semisubsistence producers move into production
of commercial crops and into nonagricultural occu-
pations, decisions governing household-level supply
of and demand for food become less closely linked.
Observed reactions to changes in market prices and
income may change substantially because of how
income is earned, even though underlying prefer-
ence functions may remain unaltered.
To see this, assume that price and income elastici-
ties of demand are to be estimated using time series
data for aggregate national per capital consumption of
a particular food staple, for a country where semisub-
sistence production of that food staple is common. For
simplicity, assume two production technologies for
this staple, a large-scale, commercial technology and
a small-scale, semisubsistence technology.5
All households are classified into one of four
types, with 0 denoting the share of total population,
where

0subsis = small-scale, semisubsistence producers
of the staple food,


Ocommer = large-scale, commercial producers of
the staple food (both small- and large-
scale producers of the staple food may
produce other agricultural products as
well),


3Thus, one can expect that semisubsistence rice farmers will eat above-average amounts of rice, and corn farmers will eat
above-average amounts of corn, and so on, whereas it is impossible to make a priori judgments about the staple food preferences of
urban residents relative to national norms.
Often there are urban-rural differences in the retail prices of foods that also will account for a part of the difference in the levels
of consumption of specific foods between urban and rural areas. However, the effects of such price differentials (perhaps due to
transportation costs) on food demand already are taken into account in the existing food demand literature, using economic models
that do not include structural shifts in demand.
4Bouis (1991) and Huang and David (1993) are exceptions.
5Individual farm sizes and the share of total production marketed by individual farms, of course, will range between these two

extremes.












0othrur = other households in rural areas that are
not engaged in production of the staple
food,6 and
urban = urban residents.

Letting Q denote aggregate national per capital
consumption, and q denote per capital consumption of
the pertinent household group, in any particular year,

Q = Osubsis qsubsis + Ocommer qcommer
+ Oothrur qothrur + Ourban urban. (1)

Certainly, in general, qsubsis, qcommer, qohrur, and
qurb,, will not be equal, both because income levels
will differ among these groups and because rural
retail food prices may be lower than urban retail food
prices. Standard time-series estimation recognizes
such differences. Aggregate national prices and in-
comes are weighted averages for various population
groups. An often implicit assumption, however, is
that the demand functions across household groups
are similar, that is,


qsbsis = f(y, p, z), (2a)
qcommer = f2(y, z), (2b)
qohrur = f3(, P, z), (2c)
urban = f(y, z), (2d)

where y is per capital income, p is the retail price of
the staple food, and z is a vector of other relevant
variables.
However, household economics theory suggests
that how income is earned influences the entire vec-
tor of group-specific demand parameters (although
not necessarily the food preference functions them-
selves), implying thatfi, f2, f3, and f4 may be quite


dissimilar. For example, for specific urban occupa-
tions, y might reasonably be treated as exogenous in
equation (2d), but certainly y is endogenous in equa-
tion (2a). One implication is that a semisubsistence
farm household's consumption of a specific food
staple may be dramatically reduced, for example, by
a decision to migrate to an urban area, apart from
differences in retail prices between urban and rural
areas and earning power. At least two reasons could
account for such a reduction.
First, semisubsistence producers of a food staple
avoid retailing and other marketing costs when they
consume out of own production. Apart from greater
risks associated with growing nonfood crops, such a
cost saving may be an important additional incentive
for semisubsistence farmers to continue growing sta-
ples, and to resist commercialization and specializa-
tion. In effect, a decision to migrate to an urban area,
or even to reduce staple production below household
"requirements," is at the same time a decision to pay
a substantially higher price for that staple at the
margin, thus affecting consumption.
Second, assume that two persons with the same
demand preferences and levels of income and facing
the same prices differ only in the activity levels
necessary for earning that income. Energy require-
ments will be greater for the person engaged in the
more active occupation.
Under a reasonable set of assumptions, more
active persons will spend more for food than less
active persons; more important, diets of those who
are more active will be proportionately more staple-
based, since staples are relatively inexpensive
sources of calories (Bouis forthcoming). If rural oc-
cupations require a greater energy expenditure than
urban occupations, one would expect that rural
populations would consume more calories and more
food staples, controlling for prices and income
(some empirical evidence of this is provided in
Bouis 1991 and Ravallion 1990). Again, how in-
come is earned affects demand.7


6Landless agricultural laborers are included in this group, although, to the extent that they receive in-kind wages in the form of the
staple food, they may behave like semisubsistence producers. In many countries, members of producer households will also earn
income as agricultural laborers. Again, such categorizations are simplifications and are intended to make a point without loss of
generality.
70ther plausible but empirically less tractable reasons can be given for arguing that producers of a food will eat more of that food.
Aside from the intrinsic satisfaction of eating what they have grown themselves, farmers know the quality of their own produce and
are familiar with ways to prepare it. Even though alternative foods may become available, habits change slowly.










It is possible that, controlling income and any
observed differences in retail prices across geo-
graphical regions,8

qsubsis > qcommer > qothrur > urban" (3)

Over time, as aggregate per capital income in-
creases in developing countries and as their econo-
mies undergo structural transformation, 0subis and
0othrur will decline (they are negatively correlated
with income) and qcomer and urban will increase
(they are positively correlated with income).9 One
would expect, then, to observe a negative correlation
between Q and Y over time, as structural transforma-
tion takes place. This correlation is independent of
the influence of income per se on demand, and it
measures the effect of changes in the way that in-
come is earned. This, in turn, leads to an upwardly
biased (in absolute value) estimate of the income
elasticity.10 Although, for a time, this biased estimate
may provide accurate predictions of aggregate de-
mand as structural transformation proceeds, at some
point this transformation process runs its course, and
food consumption patterns will diverge from those
predicted using time-series income elasticity esti-
mates.

Demandfor Nonstaple Foods
The result just discussed for equation (3) provides a
plausible explanation for declining rice consumption
throughout Asia, in addition to possible income and
price influences. Incorporating effects (1), (2), and
(3), outlined in the introduction, simply reinforces
this result, implying increased consumption of
wheat, meat, fish, and fruits.
Typically, baked wheat products are more read-
ily available in urban markets than in rural areas.
They provide some variety in staple food consump-
tion and are less labor-intensive to prepare for meals.


In addition, a greater variety of meat, fish, dairy
products, and fruits will be available in urban mar-
kets than in rural areas, encouraging habitual con-
sumption of these foods. Adults who have recently
migrated from rural to urban areas may resist signifi-
cant changes in accustomed dietary patterns, but
their children may be less resistant. Because adult
energy requirements may be lower in urban areas
(due to less strenuous occupations), hunger in urban
areas may be more easily satiated (for a given total
expenditure on food). Therefore, a higher percentage
of calories may be derived from higher-cost nonsta-
ple foods.
This section has provided several arguments for
expecting structural shifts in food demand patterns
in Asia. The next two sections will attempt to mea-
sure their importance empirically, using data sets
from Taiwan and China.


Taiwan Household
Expenditure Surveys
An Overview of Changing Food
Consumption Patterns in Taiwan
First, average diets may change rapidly in countries
experiencing rapid economic growth and structural
transformation. In the 30 years between 1959-61
and 1989-91, per capital rice consumption in Taiwan
declined by one-half (Table 1). Consumption of
meat (including pork, chicken, and beef) quadrupled.
Fruit consumption increased five times and fish con-
sumption doubled. Thus, substitution of calories ob-
tained from nonstaple food sources for staple food
sources has been substantial.
Table 2 shows indexes of real per capital income
and food price levels for these food groups for 1971,
1981, and 1991 (1981=100), using the consumer
price index as a deflator. Incomes have increased
substantially; prices of wheat and meat have fallen


8That qcommer > qohrur assumes that the price effect due to the ability to "buy" own production at favorable prices is stronger than
the energy expenditure effect. Commercial producers presumably are using mechanized technologies or hiring in labor. Their activity
levels may not be substantially different from those of urban residents. Such an assumption is not crucial to the arguments being made.
9At some later stage of development, especially as rural population densities decline, Oco,,mer may also decline.
l'Aggregate domestic production may increase or decrease, depending on the profitability of commercial production of the staple
relative to other commercial crops. Related to this, staple food prices could increase, decrease, or remain constant, depending on
government trade and buffer stock polices.











Table 1-Per capital annual food
consumption, Taiwan, 1940-92
Sweet
Period Rice Wheat Potato Meat Fish Fruit
(kilograms/capita/year)
1940-44 109 0 91 11 10 27
1949-51 133 7 66 13 12 16
1959-61 137 22 62 16 23 20
1969-71 136 25 24 25 33 43
1979-81 105 24 4 40 38 72
1989-91 68 29 2 62 45 108
1992 64 29 2 66 42 100
Source: Taiwan, Council for Agricultural Planning and Development,
various years.


Table 2-Indexes of real income, expenditure,
and real food price levels for food
groups, 1971, 1981, and 1991
Per Capita Real Food Price
Year Income Expenditure Rice Wheat Meat Fish Fruit
(1981 = 100)
1971 41 58 n.a. n.a. n.a. n.a. n.a.
1981 100 100 100 100 100 100 100
1991 213 192 99 79 77 149 134
Sources: Taiwan, Department of Agriculture and Forestry 1981,
1991; Taiwan, Department of Budget, Accounting and Sta-
tistics 1992.
Note: N.a. means not available.


by one-fifth, while fish and fruit prices have risen by
one-third to one-half. To what extent do increases in
income and declining food prices explain the dra-
matic transformation of Taiwanese diets shown in
Table 1, as compared with changing tastes and life-
styles? Conventional wisdom, of course, holds that
rising incomes and declining food prices explain
most of the change.
An ideal data set for measuring structural shifts
in food demand patterns would record foods con-
sumed, prices, income by source, and standard
demographic information for a large number of
families before and after these families migrated
from rural to urban areas. Such a longitudinal data
set would record this information across two or more


generations. To the best of the authors' knowledge,
such complete data are unavailable, however.
A next best alternative (not involving observa-
tions for the same households over time) would be
two national cross-sectional household surveys
taken several years apart. Cross-sectional survey
data are available for Taiwan for 1981 and 1991,
consisting of 11,886 observations in 1981 and
12,734 observations in 1991. These surveys contain
household-level information on (1) age and gender
of family members, (2) total expenditures on spe-
cific food and nonfood items, (3) income earned and
occupations of various household members, and (4)
geographic location, among other variables.
A major shortcoming of the data set for this
analysis was absence of data on food quantities or
prices. Prices for specific food items were obtained
from published sources (Taiwan Department of Ag-
riculture and Forestry 1981, 1991), which provided
county- and region-specific variation in prices. As
outlined later in the paper, the dependent variables in
the regression estimations are budget shares for vari-
ous foods, so that food quantity data are not required
for these estimations. Prices for rice, wheat flour,
chicken, eggs, and fruit are available at the county
level. Prices for other foods are available at the
regional level for Taipei, Taichung, Tainan, and
Kaohsiung. Weighted average prices for major meat
products (pork, beef, mutton, and chicken), six ma-
jor fish products (sea bream, marlin and sailfish,
tuna, cuttlefish, striped prawn, and shrimps), and
three major fruits (banana, pineapple, and citrus)
were used as the prices of meat, fish, and fruit,
respectively, in the regression analysis. County and
provincial-level food budget shares for these disag-
gregate, individual foods for 1981 were used as
weights in computing both the 1981 and 1991 prices
for the aggregate food groups."
Tables 3 and 4 give per capital food consumption
levels for 1981 and 1991, respectively, disaggre-
gated by villages, towns, and cities (that is, moving
from the least to the most urbanized settings) and
income quintile. Table 5 provides similar informa-
tion disaggregated by farming and nonfarming occu-
pations for 1981 and 1991. In constructing these


SlHousehold-level food quantity data are required for construction of Tables 4, 5, 6, 10, 11, 12, 13, and 14. These were obtained by
dividing average household expenditures for a specific food (from the household surveys) by national per capital availability (from
published food balance sheets) to obtain an average national-level price for each food. The household-level expenditure data for
individual foods were divided by national prices to obtain a household-level estimate of quantity consumed.












Table 3-Annual per capital food
consumption, by region and
expenditure group, Taiwan, 1981

Region/Income' Rice Wheat Meat Fish Fruit


(kilograms/capita/year)
86.9 28.6 46.8 42.9 105.5
83.3 16.9 29.2 25.9 55.2
86.5 24.3 41.6 37.8 89.0
87.7 31.7 51.1 46.4 115.2
88.9 37.4 59.0 55.1 144.9
90.6 47.4 71.5 68.4 182.4
100.6 21.9 39.6 34.0 75.9
96.9 15.5 28.8 23.0 46.7
101.5 21.9 40.0 34.2 76.1
103.6 28.2 48.0 43.7 97.4
104.0 32.6 57.6 50.3 127.0
105.9 34.0 65.3 61.9 162.6
114.4 17.9 32.0 28.3 52.7
107.3 12.8 26.2 23.3 36.3
121.9 21.1 35.5 31.5 63.0
126.0 32.9 43.9 40.9 92.2
132.8 34.0 57.6 47.7 112.7
142.8 42.1 78.1 50.1 162.2
99.4 23.4 40.1 35.8 80.5
100.0 14.3 27.4 23.7 42.6
100.5 22.7 39.4 35.0 78.2
97.5 30.9 49.2 44.9 107.1
97.7 35.9 58.6 53.1 137.0
97.7 44.5 70.9 65.7 177.1


Table 4-Annual per capital food
consumption, by region and
expenditure group, Taiwan, 1991

Region/Income" Rice Wheat Meat Fish Fruit


City
1
2
3
4
5
Town
1
2
3
4
5
Village
1
2
3
4
5
Average
1
2
3
4
5


(kilograms/capita/year)
35.4 70.0 45.5 122.3
9.8 33.6 21.1 36.5
18.9 48.4 29.3 61.8
23.8 59.0 37.1 87.3
31.7 68.8 44.6 112.2
46.4 79.2 52.3 156.5
27.7 60.3 37.5 98.3
10.7 39.2 22.3 38.0
16.4 49.4 29.2 66.5
21.8 55.4 34.2 81.4
28.7 63.1 39.2 102.8
41.8 70.2 45.0 136.2
18.4 56.2 35.6 92.5
9.1 39.1 22.6 46.0
13.4 47.5 29.2 64.5
16.6 55.6 35.0 83.9
21.2 63.1 40.0 109.5
31.7 68.0 46.4 156.9
28.9 63.5 40.4 107.2
9.7 38.7 22.4 42.4
15.5 48.4 29.2 64.8
21.0 56.5 35.3 83.9
28.9 65.7 41.8 108.3
43.6 75.5 49.6 150.5


Source: Taiwan Household Expenditure Surveys 1981.
Note: To compute food consumption levels of various groups, an
implicit aggregate price for each commodity is estimated,
based on food expenditure and food balance data. This price
is used to deflate the food expenditure to derive food con-
sumption by income group and region.
'The ranges of per capital expenditure in Taiwanese dollars (NT$ in real
1981 prices) for five groups of households are: 1: < 30,000; 2: 30,000-
45,000; 3: 45,000-60,000; 4: 60,000-90,000; and 5: > 90,000.



tables, all households were placed in one of five
income quintiles, based on real income, with 1981
used as the base.
In comparing food consumption between vil-
lages, towns, and cities, or between farming and
nonfarming occupations, within identical income
groups (say income quintile 3) for a specific year,
note that rice consumption is lower in more urban-
ized areas and occupations, while wheat, meat, fish,
and fruit consumption is somewhat higher (egg con-
sumption is about equal between urban and rural
areas and therefore is not shown in the tables). These


Source: Taiwan Household Expenditure Surveys 1991.
Note: To compute food consumption levels of various groups, an
implicit aggregate price for each commodity is estimated,
based on food expenditure and food balance data. This price
is used to deflate the food expenditure to derive food con-
sumption by income group and region.
aThe ranges of per capital expenditure in Taiwanese dollars (NT$ in real
1981 prices) for five groups of households are: 1: < 30,000; 2: 30,000-
45,000; 3: 45,000-60,000; 4: 60,000-90,000; and 5: > 90,000.



differences presumably are due to structural differ-
ences in tastes and lifestyles between geographic
locations and occupations.12
When food consumption is compared across in-
come quintiles for a given year and geographic loca-
tion or occupation, income elasticities appear to be
high for all foods except rice. Table 6 shows the
difference in consumption levels between 1981 and
1991 for income quintile 3 by geographic location
and by occupation. These differences in food con-
sumption are due to some combination of price ef-
fects and structural changes within geographic loca-


12Urban-rural retail price differentials may explain some of the difference in consumption levels between urban and rural areas.
However, rural consumption is lower for several foods for which the rural price can be expected to be lower than the urban price.


City
1
2
3
4
5
Town
1
2
3
4
5
Village
1
2
3
4
5
Average
1
2
3
4
5













Table 5-Annual per capital food consumption,
by year, occupation, and
expenditure group, Taiwan

Region/Incomea Rice Wheat Meat Fish Fruit
(kilograms/capita/year)
1981
Nonfarming 93.2 25.5 42.6 38.5 90.3
1 92.0 15.6 28.4 24.6 48.7
2 93.2 23.2 40.2 36.0 82.1
3 93.4 31.3 49.9 45.5 109.7
4 94.7 36.5 58.4 53.9 139.3
5 96.0 45.2 69.9 66.4 178.7
Farming 120.9 16.2 31.3 26.5 46.8
1 112.0 12.3 26.1 22.4 33.7
2 133.6 20.6 35.9 30.8 60.2
3 139.4 27.4 42.3 38.6 81.2
4 142.3 27.9 60.8 41.2 102.7
5 146.8 23.2 99.3 44.0 128.8
1991
Nonfarming 65.3 30.7 64.5 41.0 110.1
1 64.6 10.0 39.2 22.6 40.6
2 67.9 16.5 48.4 28.9 64.8
3 66.9 22.0 56.6 35.0 84.3
4 65.4 29.7 65.9 41.9 108.1
5 62.8 44.5 75.9 49.7 151.8
Farming 84.2 16.6 55.8 36.1 86.3
1 69.0 9.3 37.8 22.1 45.1
2 76.3 12.7 48.4 30.0 64.7
3 81.3 15.8 55.8 36.6 81.8
4 90.7 20.2 63.5 40.8 110.3
5 107.6 24.3 65.6 46.7 120.9
Source: Taiwan Household Expenditure Surveys 1981, 1991.
Note: To compute food consumption levels of various groups, an
implicit aggregate price for each commodity is estimated,
based on food expenditure and food balance data. This price
is used to deflate the food expenditure to derive food con-
sumption by income group and region.
aThe ranges of per capital expenditure in Taiwanese dollars (NT$ in real
1981 prices) for five groups of households are: 1: < 30,000; 2: 30,000-
45,000; 3: 45,000-60,000; 4: 60,000-90,000; and 5: > 90,000.


Table 6-Change in per capital food consumption
between 1981 and 1991, for income
quintile 3 and the national average,
Taiwan

Location/Occupation Rice Wheat Meat Fish Fruit
(kilograms/capita/year)


Income quintile 3
City
Town
Village
Nonfarming
Farming
National


-26.1
-34.5
-47.9
-26.5
-58.9
-28.1


National average
(all income groups) -30.8


-7.9
-6.4
-16.3
-9.3
-11.6
-9.9


+7.9
+7.4
+11.7
+6.7
+13.5
+7.3


+5.5 +23.4


-27.9
-16.0
-8.3
-25.4
+0.6
-23.2


+4.6 +26.7


Source: Tables 3, 4, and 5.


tion or occupation between 1981 and 1991, roughly
controlling for differences in real income. Note that
rice consumption declines substantially between
1981 and 1991, even though the rice price changed
little. The decline is much higher for farming than
for nonfarming populations. This suggests a consid-
erable structural shift in demand.
Meat prices declined between 1981 and 1991,
while fish and fruit prices increased. Thus, the posi-
tive changes for meat and the negative changes for
fish and fruit in Table 6 are to be expected. Price
effects and the hypothesized positive structural
shifts when income is controlled reinforce one an-
other for meat. Price effects outweigh the hypothe-
sized positive structural shifts for fish and fruit.
Nevertheless, note that fruit consumption increased
among those in farming occupations, despite a sub-
stantial increase in price. Apparently, structural
shifts for this group outweighed price effects.
Having provided an intuitive overview of possi-
ble structural influences that have caused food con-
sumption patterns to change in Taiwan between
1981 and 1991, these influences are documented and
measured econometrically.


Model Specification

An almost ideal demand system (AIDS) is used as
the basic modeling framework (Deaton and Muell-
bauer 1980). The AIDS model in the budget share
form is expressed as


wi = A, + B, log(X/P) + Eri log(pj), (4)

for i,j = 1, ... n, where w, is the budget share of the
i* commodity, Xis total consumption expenditure, p
is commodity price, P is a price index defined by


log(P) = a + -kAk log(pk)
+ 1/2 YEkYrkj log(pk) log(p), (5)


and a0, A,, B,, and r, are parameters to be estimated.
The effects of nonincome and nonprice factors
(structural shifts) on food consumption can be intro-
duced into the AIDS equations by allowing any sub-
set of parameters (designated here as vector Z) to
depend on these structural variables.
To incorporate the effects of the Z variables into
the AIDS, it is assumed that the parameters A, and Bi











(but not the parameters r#) of the demand system in
equations (4) and (5) depend linearly on the zip's:

Ai = a, + saiszs, (6)

and
B; = bi + ,biz, (7)

where Z= (z,,. .,z, a,, aZ , b,, and b,1 are parame-
ters to be estimated.
The adding-up restrictions for the demand sys-
tem, equations (4)-(7), require

Eiai = 1, (8a)

Eiais = Eib;s = 0, for s = 1, . ., m, (8b)


Zibs = 0,



,jry = 0.


(8c)



(8d)


The commodities included in the system are
rice, wheat, meat, fish, fruit, other foods, and all
nonfoods, which are aggregated.
In applying the model to the data for Taiwan, Z
is a vector with nine elements: four dummies (town,
city, occupation, and d91 x village, the last repre-
senting "urbanization" and other structural changes
over the period 1981-91 within rural areas; d9 = 1
for 1991 survey data and d91 = 0 for 1981 survey
data), and five demographic variables (family size
and various age composition variables). Since, the
dependent variables are in budget shares and the
error-covariance matrix is singular, one of the shares
is dropped during estimation.
The expenditure (eiy), uncompensated price (e,),
and compensated price (ce,) elasticities are derived
as follows:


eiy = 1 + (b+ + Esbi;z,) / wi,

ei = -86y + r# /w; (bi + EYb;izs)
[ai + EsaisZs + Ykrkj log(pk)] / Wi,


The homogeneity restriction is


jrni = 0,


and the cross-equation symmetry restrictions can be
imposed as

rij = rjifor i j. (10)

To specify a stochastic structure for equations (4)-
(7), using h to index household-level information,
the error term, Eih, may be added to equation (4).
The demand model presented above is a non-
linear system. Assuming that sih follows a multivari-
ate normal distribution, this nonlinear system of
equations can be estimated by the full information,
maximum likelihood method (FIML) (Amemiya
1977), with the imposition of the homogeneity and
symmetry restrictions in equations (9) and (10).


cej = ei + wi eiy,
where 56 is the Kronecker delta.


(11)


(12)



(13)


In equations (11)-(13), expenditure and price
elasticities of demand vary as the level of urbaniza-
tion, occupation, family structure, and the other fac-
tors change to the extent that coefficients associated
with the z, variables are statistically significant from
zero.
Mathematically, the impact of each structural
factor z, on the commodity demand can be derived as
follows.3 First, differentiate equation (4) with re-
spect to the mth zi, holding all other variables con-
stant, which gives

dwi = aim dZm (bi + Es.mbiszs) [Yjajs log(pj)]dz,
[bim Ejaj, log(pj)] dz,2 + bi,


13The effects of various structural variables on demand for specific commodities can be evaluated, first, by taking the first-order
derivatives of equations (11) through (13) with respect to zi(and after appropriate manipulations, calculating elasticitiess of elasticities
of demand") or, second, by calculating the total differentiation of equation (4), given that equations (5) through (10) hold with respect
to each zi, keeping income and price constant (for example, at the sample means).











[logX- ao Yj(aj + Z.,,iajz,)log(pj)

EiEjrij log(pj)log(pj)] dz.,. (14)

Because equation (14) controls for the income
and price changes, dividing equation (14) by w, im-
plies the percentage change of ih commodity con-
sumption (q;) due to the different level of z,,. That is,

dw, / wi = dq, / qi

= {ai,, dz, (bi + Y,,mbi,z,)

[2jajs log(pj)]dz,

[bi,, jajs log(pj)]dz,,2 + bi,
[logX ao Yj(aj + Y.majszs)

log(pj) i;jrij
log(pj)log(pj)]dzm} / wi. (15)



Econometric Estimations
The detailed estimation results are shown in Table 7.
Expenditure and own-price elasticities, calculated
from the Table 7 coefficients, are presented in Ta-
bles 8 and 9, respectively. Table 7 shows that almost
all of the coefficients on variables representing
structural shifts are significantly different from zero
for each food. The net effects of these variables on
observed differences in consumption between geo-
graphic locations and observed changes in consump-
tion over time are discussed later in this section.

Income and Price Elasticities. The expenditure
elasticities themselves, for any given food group, do
not vary greatly between villages, towns, and cities.
The greatest variation across geographic location
occurs for rice. As expected, all expenditure elastici-
ties are positive, even for rice, for which per capital
consumption declined significantly between 1981


and 1991. With the exception of rice, all expenditure
elasticities fall in a range between 0.5 and 1.0. The
elasticities for meat and fish, in particular, are lower
than those estimated by Capps et al. (1994) using
time-series data that did not take structural shifts in
demand into account.
With the exception of rice and meat, own-price
elasticities are above 1.0 (in absolute value). In that
expenditure elasticities are relatively low, this sug-
gests generally strong cross-price substitution ef-
fects, except for meat.14

Magnitudes of Structural Shifts in Demand. While
measured structural shifts in demand are signifi-
cantly different from zero in a statistical sense, such
shifts may or may not be important from a policy
perspective. As seen in Table 1, the diets of Taiwan-
ese consumers changed substantially between 1981
and 1991. How much of this change was due to price
and income effects and how much was due to struc-
tural factors? Tables 10-14 provide information on
the magnitude of shifts in demand due to various
structural factors calculated from the estimated re-
gression coefficients.
Table 10 provides estimates of differences in
consumption of various foods, after controlling for
income, price, and demographic differences (1) be-
tween villages, towns, and cities in 1981, (2) within
villages between 1981 and 1991, and (3) between
farming and nonfarming occupations in 1981. In
general, rice consumption declined and fruit con-
sumption increased with greater urbanization. The
percentage changes for meat, fish, wheat, and other
foods are positive but lower than for fruit.
Table 11 presents demographic effects such as
size of household and ages of household members,
controlling urbanization and price and income ef-
fects. As families become smaller, controlling age
composition effects, there is a tendency to eat some-
what less rice and more wheat and nonstaple foods.
Changes in age composition shown in Table 11
affect consumption in at least two ways. First, be-
cause young children are physically small and peo-
ple over 50 years of age are likely to be less active,


14Real chicken prices fell substantially between 1981 and 1991, more than pork and beef prices fell, and chicken consumption
increased more than consumption of pork and beef. The estimated low price response for meat in the aggregate may, to some extent,
reflect shortcomings in not having household-level price data and having to use country- and provincial-level prices instead (see
footnote 11). It may also be that the now wealthy consumers of Taiwan have a relatively strong desire to eat meat. They may be willing
to substitute various types of meat within this aggregate group in response to changing prices.














Table 7-Nonlinear FIML parameter estimates of a dynamic AIDS model, Taiwan

Commodity (i)

Parameters Rice Wheat Meat Fish Fruit Other Foods
(kilograms/capita/year)


ai (base)

ai (town)

ai (city)

ai (village x d91)a

ai (occupation)

ai (family size)

ai (percent < 7 years)

ai (18-40 years percent)

ai (percent 41-50 years)

ai (percent > 50 years)

bi (base)

bi- (town)

bi (city)

bi- (village x d91)a

bi- (occupation)

bi (family size)

bi (percent < 7 years)

bi -(percent 18-40 years)

bi (percent 41-50 years)

bi (percent > 50 years)

ril (rice)

ri2 (wheat)

r3 (meat)

ri4 (fish)

ris (fruit)

ri6 (other food)

ao


-0.2996
(-17.61)
0.0668
(17.97)
0.1041
(25.46)
0.1235
(23.43)
-0.0526
(-15.93)
-0.0462
(-16.60)
0.0817
(9.82)
0.1053
(13.78)
0.0766
(9.92)
0.0568
(7.68)
-0.0512
(-30.46)
0.0150
(21.23)
0.0243
(32.20)
0.0259
(25.47)
-0.0118
(-18.47)
-0.0105
(-19.03)
0.0184
(11.17)
0.0223
(14.53)
0.0150
(9.43)
0.0102
(6.74)
0.0208
(4.83)
0.0254
(18.19)
-0.0024
(-1.28)
-0.0010
(-0.65)
-0.0111
(-7.59)
-0.0239
(-6.86)
16.1333
(18.19)


0.0275
(2.85)
-0.0182
(-6.62)
-0.0197
(-7.23)
-0.0165
(-4.85)
-0.0042
(-1.71)
-0.0062
(-4.65)
-0.0035
(-0.83)
-0.0116
(-2.89)
-0.0231
(-5.38)
-0.0195
(-4.55)
0.0036
(3.88)
-0.0035
(-6.65)
-0.0039
(-7.48)
-0.0027
(-3.93)
-0.0005
(-1.12)
-0.0010
(-3.82)
-0.0005
(-0.58)
-0.0015
(-1.80)
-0.0036
(-4.17)
-0.0031
(-3.61)


-0.0058
(-5.32)
-0.0149
(-15.29)
-0.0009
(-1.63)
0.0050
(5.51)
-0.0132
(-9.56)
16.1333
(18.19)


-0.1332
(-8.36)
-0.0098
(-2.31)
0.0009
(0.19)
-0.0343
(-4.81)
0.0123
(3.68)
-0.0011
(-0.42)
-0.0365
(-4.65)
-0.0140
(-1.77)
-0.0143
(-1.82)
-0.0128
(-1.55)
-0.0259
(-14.66)
-0.0021
(-2.38)
0.0005
(0.48)
-0.0088
(-5.32)
0.0029
(4.17)
-0.0003
(-0.57)
-0.0053
(-3.21)
-0.0001
(-0.07)
-0.0015
(-0.90)
-0.0019
(-1.10)




0.0523
(24.57)
-0.0123
(-9.70)
-0.0139
(-8.94)
-0.0231
(-7.62)
16.1333
(18.19)


0.1924
(14.28)
0.0240
(5.38)
0.0189
(4.06)
-0.0246
(-3.89)
0.0163
(4.57)
0.0158
(5.97)
-0.0356
(-4.56)
-0.0152
(-1.98)
0.0013
(0.17)
-0.0045
(-0.56)
-0.0291
(-15.74)
0.0044
(4.85)
0.0031
(3.24)
-0.0072
(-5.12)
0.0039
(5.37)
0.0036
(6.46)
-0.0056
(-3.49)
-0.0006
(-0.39)
0.0019
(1.20)
-0.0007
(-0.40)







-0.0346
(-27.77)
-0.0147
(-15.28)
-0.0503
(-23.15)
16.1333
(18.19)


0.2737
(14.95)
-0.0405
(-6.45)
-0.0455
(-7.24)
-0.0582
(-8.30)
0.0219
(5.33)
-0.0051
(-1.82)
-0.0361
(-3.93)
-0.0357
(-4.09)
-0.0244
(-2.77)
-0.0360
(-3.91)
0.0053
(2.49)
-0.0088
(-7.43)
-0.0102
(-8.66)
-0.0135
(-9.83)
0.0051
(6.33)
-0.0003
(-0.48)
-0.0065
(-3.55)
-0.0057
(-3.23)
-0.0035
(-1.96)
-0.0055
(-2.96)









-0.0156
(-8.77)
0.0321
(14.21)
16.1333
(18.19)


0.3339
(10.08)
-0.0799
(-7.23)
-0.0690
(-6.26)
-0.1019
(-7.47)
0.0020
(0.22)
0.0006
(0.10)
-0.0921
(-4.10)
0.1618
(8.36)
0.0877
(4.39)
0.0276
(1.34)
-0.0304
(-6.92)
-0.0149
(-7.02)
-0.0113
(-5.27)
-0.0226
(-8.06)
0.0009
(0.48)
0.0003
(0.22)
-0.0213
(-4.81)
-0.0303
(7.81)
0.0171
(4.25)
0.0062
(1.50)











-0.0363
(-5.19)
16.1333
(18.19)


Notes: Sample size: 24,233. The complete almost ideal demand system (AIDS) is estimated using nonlinear full information, maximum likelihood
(FIML) measure with the imposition of the homogeneity and symmetry restrictions. The parameter estimates converge for FIML after 22
iterations, using converge = 0.001 as the convergence criteria. The numbers in parentheses are t-values.
ad91 is a dummy variable for 1991.














Table 8-Budget shares and expenditure elasticities evaluated at the sample mean, Taiwan,
1981-91

Sample Mean of Budget Share Expenditure Elasticities

Commodity Mean Village Town City Mean Village Town City

Rice 4.2 6.4 4.0 3.0 0.173 0.285 0.065 0.129
Wheat 1.1 1.0 1.2 1.2 0.768 0.895 0.739 0.721
Meat 6.6 7.3 6.5 6.3 0.569 0.591 0.547 0.571
Fish 5.6 6.1 5.4 5.5 0.605 0.578 0.633 0.603
Fruit 3.8 3.5 3.8 3.9 0.807 0.919 0.794 0.752
Other food 14.0 15.1 14.2 1.3 0.795 0.828 0.775 0.788
Nonfoods 64.6 60.6 65.0 66.9 1.192 1.217 1.199 1.173
Source: Taiwan Household Expenditure Surveys 1981, 1991.



Table 9-Price elasticities evaluated at the sample mean, Taiwan, 1981-91

Uncompensated Own Price Elasticity Compensated Own Price Elasticity

Commodity Mean Village Town City Mean Village Town City

Rice -0.609 -0.804 -0.614 -0.390 -0.612 -0.786 -0.612 -0.386
Wheat -1.514 -1.578 -1.503 -1.489 -1.505 -1.569 -1.492 -1.481
Meat -0.243 -0.317 -0.230 -0.200 -0.206 -0.274 -0.194 -0.164
Fish -1.638 -1.599 -1.656 -1.652 -1.604 -1.564 -1.622 -1.619
Fruit -1.412 -1.438 -1.415 -1.397 -1.381 -1.405 -1.385 -1.367
Other food -1.259 -1.237 -1.259 -1.276 -1.148 -1.127 -1.149 -1.172
Nonfoods -1.641 -1.701 -1.651 -1.598 -0.870 -0.964 -0.872 -0.813





Table 10-The effects of urbanization and occupation changes on annual per capital food
consumption in Taiwan between 1981 and 1991
Structural Changes

Move from Village to Move from Village to Village Difference Switch from Farming to
Commodity Town in 1981 City in 1981 between 1981 and 1991 Nonfarming in 1981

Percentage changes (percent)
Rice -11.9 -22.8 -23.3 -13.6
Wheat 7.5 19.0 -23.9 14.1
Meat 5.0 4.5 10.7 3.2
Fish 3.2 9.9 16.9 6.0
Fruit 24.3 37.6 45.4 18.5
Other food 2.5 -1.5 10.5 1.5
Absolute changes (kilograms)
Rice -13.6 -26.0 -26.6 -15.5
Wheat 1.3 3.4 -4.3 2.5
Meat 1.6 1.5 4.4 1.0
Fish 0.9 2.8 4.8 1.7
Fruit 12.8 19.8 23.9 9.8












Table 11-The effects of demographic changes on annual per capital food consumption in Taiwan
between 1981 and 1991
Hypothetical Structural Changes
Family Size 100 Percent 100 Percent 100 Percent 100 Percent
Commodity Doubles 1-7 Yearsa 18-40 Yearsa 41-50 Yearsa Over 50 Yearsa
(kilograms/capita/year)
Percentage change
Rice 1.1 -27.0 -5.2 6.7 9.7
Wheat -8.9 -13.2 -22.8 -35.6 -35.8
Meat -3.0 -23.2 -10.9 -7.0 -7.6
Fish -5.1 -21.0 -11.8 -10.7 -4.7
Fruit -8.3 -14.9 -11.2 -14.2 -22.8
Other food -2.0 2.1 5.5 2.1 -3.1
Absolute change (kilograms)
Rice 0.9 -22.4 -4.3 5.6 8.1
Wheat -2.4 -3.5 -6.0 -9.3 -9.4
Meat -1.5 -12.1 -5.7 -3.7 -4.0
Fish -1.9 -8.0 -4.5 -4.1 -1.8
Fruit -7.9 -14.1 -10.6 -13.4 -21.5
aChanges are calculated assuming that the entire population falls within the indicated age category.


aggregate per capital consumption of various foods is
expected to decline if these age groups represent a
higher percentage of the total population. Second,
there may be generational differences in tastes, due
either to recent secular influences (for example, me-
dia advertising) or changing tastes and nutritional
requirements with age (for example, a decline in
milk consumption). Older generations appear to pre-
fer rice to wheat compared with younger genera-
tions. Younger generations prefer more fruit than
older generations.
Tables 12 and 13 provide a disaggregation be-
tween villages and towns and between villages and
cities, respectively, of several factors determining
observed differences in consumption of food in
1981: (1) urbanization, (2) occupation, (3) family
size and age structure, (4) income, and (5) prices.
For rice, wheat, and fruit consumption, structural
factors account for a higher percentage of the differ-
ences in consumption between geographic areas than
income and price influences. For meat and fish con-
sumption, income and price factors affect differ-
ences in consumption more than structural influ-
ences.
A disaggregation of villages in 1981 and 1991 is
presented in Table 14 with a pattern similar to that in
Tables 12 and 13; the structural factors strongly
influence consumption of rice, wheat, and fruit, and
price and income effects strongly influence meat
consumption. However, the income and price effects


are individually strong for most foods, but influence
demand in opposite directions so that their joint in-
fluence tends to be much smaller.



Summary
Analysis of the household-level data for Taiwan for
1981 and 1991, which allows disaggregation by ur-
ban and rural areas and by occupation, for a time
period during which there was rapid economic de-
velopment and dramatic changes in the composition
of diets in Taiwan, provides strong empirical support
for the hypothesis that demand for food is substan-
tially influenced not only by growth in family in-
come and price changes, as might be expected, but
also by differences in urban and rural lifestyles, the
development of more advanced marketing systems,
and occupational changes that are closely linked
with increasing GNP per capital. Because urbaniza-
tion is expected to proceed rapidly in a number of
developing countries over the next several decades,
projections of future global food supply and demand
balances need to take such structural changes into
account. While analysis of the Taiwan data serves a
useful purpose in initial attempts to understand the
magnitude and nature of such shifts, it is far more
important from a global perspective to document and
understand these phenomena for countries with
larger populations such as China.








Table 12--Structural, income, and price factors contributing to differences in per capital food consumption between cities and
villages, Taiwan, 1981

Specific Structural Factors
Income,
Age Structure Sum of Price, and Price and
Observed Structural Other Income Other
Commodity Change Urbanization Occupation Family Size Total 1-7 Years 18-40 Years 41-50 Years Over 50 Years Factors Factors Factors Factors
Percentage
change (percent)
Rice -24.0 -22.8 -6.6 -0.4 -2.5 -1.6 -0.3 -0.3 -0.3 -32.3 8.3 7.7 0.6
Wheat 59.8 19.0 6.5 1.7 0.7 -0.6 -1.6 1.6 1.3 27.9 31.9 43.2 -11.4
Meat 46.2 4.5 1.5 0.6 -1.2 -1.0 -0.7 0.3 0.3 5.4 40.9 34.2 6.6
Fish 51.6 9.9 2.9 1.0 -1.1 -0.9 -0.8 0.5 0.2 12.7 38.9 36.2 2.7
Fruit 100.2 37.6 8.9 1.6 0.1 -0.6 -0.8 0.6 0.8 48.2 52.0 45.1 6.9
Absolute
change (kilograms)
Rice -27.4 -26.0 -7.5 -0.5 -2.9 -1.8 -0.3 -0.4 -0.3 -36.8 9.4 8.8 0.6
Wheat 10.7 3.4 1.2 0.3 0.1 -0.1 -0.3 0.3 0.2 5.0 5.7 7.7 -2.1
Meat 14.8 1.5 0.5 0.2 -0.4 -0.3 -0.2 0.1 0.1 1.8 13.0 11.0 2.0
Fish 14.6 2.8 0.8 0.3 -0.3 -0.3 -0.2 0.1 0.1 3.6 11.0 10.2 0.8
Fruit 52.8 19.8 4.7 0.9 0.0 -0.3 -0.4 0.3 0.4 25.4 27.4 23.8 3.6




Table 13-Structural, income, and price factors contributing to differences in per capital food consumption between towns and
villages, Taiwan, 1981

Specific Structural Factors
Income,
Age Structure Sum of Price, and Price and
Observed Structural Other Income Other
Commodity Change Urbanization Occupation Family Size Total 1-7 Years 18-40 Years 41-50 Years Over 50 Years Factors Factors Factors Factors
Percentage
change (percent)
Rice -12.0 -11.9 -4.5 -0.1 -2.6 -1.8 -0.2 -0.3 -0.3 -19.1 7.2 1.7 5.4
Wheat 22.3 7.5 4.5 0.6 1.0 -0.7 -1.2 1.6 1.2 13.5 8.8 19.7 -10.9
Meat 23.8 5.0 1.1 0.2 -1.1 -1.2 -0.6 0.3 0.3 5.1 18.6 14.6 4.1
Fish 20.1 3.2 2.0 0.4 -1.0 -1.0 -0.6 0.5 0.2 4.6 15.6 16.9 -1.3
Fruit 44.0 24.3 6.2 0.6 0.2 -0.7 -0.6 0.6 0.8 31.2 12.9 21.2 -8.3
Absolute
change (kilograms)
Rice -13.7 -13.6 -5.2 -0.1 -2.9 -2.0 -0.2 -0.4 -0.3 -21.9 8.2 2.0 6.2
Wheat 4.0 1.3 0.8 0.1 0.2 -0.1 -0.2 0.3 0.2 2.4 1.6 3.5 -1.9
Meat 7.6 1.6 0.3 0.1 -0.4 -0.4 -0.2 0.1 0.1 1.6 6.0 4.7 1.3
Fish 5.7 0.9 0.6 0.1 -0.3 -0.3 -0.2 0.1 0.0 1.3 4.4 4.8 -0.4
Fruit 23.2 12.8 3.2 0.3 0.1 -0.3 -0.3 0.3 0.4 16.4 6.8 11.1 -4.4







Table 14-Structural, income, and price factors contributing to changes in per capital food consumption in villages in Taiwin
between 1981 and 1991

Specific Structural Factors
Income,
Age Structure Sum of Price, and Price and
Observed Structural Other Income Other
Commodity Change Urbanization Occupation Family Size Total 1-7 Years 18-40 Years 41Years41 ears Over 50 Years Factors Factors Factors Factors
Percentage
change (percent)
Rice -30.9 -23.3 -1.8 -1.0 3.5 1.8 0.3 0.3 1.0 -22.6 -8.3 27.5 -35.8
Wheat 2.8 -23.8 1.8 3.7 -3.1 0.7 1.9 -1.6 -4.0 -21.4 24.2 86.3 -62.1
Meat 75.6 10.7 0.4 1.3 0.9 1.2 0.9 -0.3 -0.9 13.3 62.3 57.0 5.3
Fish 25.8 16.9 0.8 2.4 1.0 1.1 1.0 -0.5 -0.5 21.1 4.7 55.8 -51.1
Fruit 75.5 45.4 2.5 3.6 -1.6 0.6 0.9 -0.6 -2.6 49.9 25.6 88.6 -63.0
Absolute
change (kilograms)
Rice -35.3 -26.6 -2.1 -1.2 4.0 2.1 0.4 0.4 1.1 -25.9 -9.4 31.4 -40.8
Wheat 0.5 -4.3 0.3 0.7 -0.6 0.1 0.3 -0.3 -0.7 -3.9 4.4 15.5 -11.1
Meat 24.2 3.4 0.1 0.4 0.3 0.4 0.3 -0.1 -0.3 4.2 20.0 18.2 1.8
Fish 7.3 4.8 0.2 0.7 0.3 0.3 0.3 -0.1 -0.2 6.0 1.3 15.8 -14.5
Fruit 39.8 23.9 1.3 1.9 -0.9 0.4 0.5 -0.3 -1.4 26.2 13.6 46.7 -33.1











Provincial-level Data from China

An Overview of Food Consumption
Patterns in China
For 1991, provincial-level information on quantities
consumed of various foods, expenditures for these
foods, and per capital income were obtained from
China's State Statistical Bureau (SSB 1992). For
each of 28 provinces, data are disaggregated by rural
areas, small cities, and capital cities, giving 89 ob-
servations. Similar data for 1989 are available from
SSB (1990), but only for urban areas; these data
yield an additional 57 observations.
Average consumption levels in China are shown
in Table 15. Similar to the data for Taiwan, grain
consumption is higher in rural areas and consump-
tion of various nonstaple foods is higher in urban
areas. Again, the primary question to be addressed is
the extent to which these differences can be ex-
plained by (1) differences in incomes and prices and
(2) apparent structural shifts in demand.
In Table 16, food consumption data for income
groups 5, 6, and 7 for rural areas are compared with


Table 15-Annual per capital food
consumption, rural and urban,
China, 1991
Rural Urban
Food Commodity 1 2 Small Capital
(kilograms)
Grain 199.4 196.5 138.7 118.6
Meat, fish, and their
products
Red meat 12.2 13.5 23.9 27.4
Pork 11.2 10.9 16.8 19.0
Beef and mutton 1.0 1.2 3.7 3.5
Poultry 1.3 1.4 3.4 4.9
Eggs 2.7 3.0 5.8 9.5
Milk 1.2 1.3 2.5 10.6
Fish 2.2 2.6 6.4 8.7
Edible oil 5.7 5.8 6.2 7.2
Vegetables 127.0 123.3 121.3 126.6
Fruit 6.8 7.7 32.8 38.0
Note: Figures for rural are weighted averages published by SSB
(1993); all other figures (rural2, small city, and provincial capital
city) are sample means based on provincial aggregate data (SSB
1992, 1993).


income groups 2, 4, and 7 for urban areas. Such a
comparison provides a rough control for income.
Grain consumption is 80 to 100 kilograms higher
in rural areas.15 Meat (pork, poultry, mutton, and beef)
consumption is from 3 to 8 kilograms higher in urban
areas. Vegetable consumption is higher in rural areas,
except at very high income levels.

Model Specification
In specifying the demand equations to be estimated,
the small sample size made it necessary to reduce the
number of Z variables used for China, as compared
with the analysis for Taiwan. Options were (1) to
include separate intercept variables for rural/ur-
banl/urban2 (labeled as the "intercept dummy"
model in the tables; urbann" refers to small cities
and "urban2" refers to large, capital cities) or (2) to
allow income elasticities to vary for rural/urban l/ur-
ban2 (labeled as the "slope dummy" model in the
tables). The Wald chi-square statistics in Table 17
show that all combinations of the null hypotheses of
equality between intercepts and slopes for rural/ur-
banl/urban2 are rejected, whether homogeneity and
symmetry conditions are imposed or not.
An AIDS model similar to that specified in the
previous section was also applied to China's provin-
cial-level data set. However, because of the small
sample size and the high collinearity between the
explanatory variables, the model either failed to con-
verge or converged (when assigned a value for ao, a
priori), with few of the estimated parameters being
statistically significant. Therefore, a linear approxi-
mation (LA) of AIDS-LA/AIDS; log(P) in equa-
tion (5) is replaced by Eiwi log(p,)-was applied. An
iterative, seemingly unrelated regressions procedure
was used.
Both the intercept dummy and slope dummy
models were estimated, using a two-step procedure.
First, an LA/AIDS model was used to estimate de-
mand (budget shares) for various aggregate food
groups, including an aggregate "meat" group, which
includes pork, beef and mutton, poultry, fish, and
eggs. The results from this first-stage estimation
were then used to estimate demand (budget shares)
for various component foods for the aggregate meat
group.


15It is questionable whether direct grain consumption is as high as 215 kilograms per capital in rural areas. This represents an energy
intake of 2,050 calories per day before adding calories from nonstaple foods. It may be that some grain is used as livestock feed.













Table 16-Annual per capital food consumption by income group, China, 1991

Per Capita Food Consumption
Region/Income
Group Average Income Grain Red Meat Poultry Fish Edible Oil Vegetables
yuann) (kilograms)
Weighted average
Rural 709 199.5 12.2 1.3 2.2 5.7 127.0
1 234 179.5 7.6 n.a. n.a. 4.0 89.1
2 550 201.5 11.6 0.7 0.6 5.4 127.7
3 693 211.4 12.7 1.6 2.6 6.1 138.7
4 890 215.0 14.5 1.4 3.3 6.5 146.8
5 1,193 215.4 15.8 2.5 5.8 7.2 150.7
6 1,703 213.5 17.5 n.a. n.a. 7.4 146.8
7 2,721 215.2 20.1 n.a. n.a. 8.5 146.0
Urban 1,713 127.9 22.2 4.4 8.0 6.9 135.3
1 1,007 120.1 16.5 2.7 6.1 5.6 117.2
2 1,240 122.5 18.8 3.3 6.8 6.1 119.7
3 1,439 123.4 19.8 3.7 7.2 6.3 125.3
4 1,671 126.5 21.4 4.3 7.8 6.6 129.2
5 1,951 128.9 23.0 4.8 8.1 7.0 135.6
6 2,283 138.8 24.8 5.3 8.9 7.6 144.8
7 2,957 148.7 28.3 6.4 9.0 8.5 158.9
Source: SSB 1992, 1993.
Notes: Figures for poultry and fish consumption in the rural area are computed based on aggregate provincial-level data with income levels of
546, 696, 836, and 1,202 yuan for rural income groups 2, 3, 4, and 5, respectively. n.a. means not available.


Table 17-Tests of differences in the consumption pattern between rural and urban populations,
China

Wald Chi-Square Statistics

Degrees Without Homogeneity and With Homogeneity and Critical Value
Hypothesis of Freedom Symmetry Symmetry (5 percent)
Unrestricted model
Intercepts
Rural = Urban = Urban2 14 143.18 118.44 23.68
Slope for Ln(X/P*)
Rural = Urbanl = Urban2 14 141.66 115.89 23.68
Intercept and slope
Rural = Urban 1= Urban2 28 312.92 329.47 41.34
Intercept dummy model
Rural = Urbanl = Urban2 14 166.18 208.23 23.68
Urbanl = Urban 2 7 36.77 86.61 14.07
Slope dummy model
Rural = Urbanl = Urban2 14 164.56 205.55 23.68
Urban = Urban 2 7 37.91 87.17 14.07
Note: Urbani and urban2 are small and large or capital cities, respectively.


The formula for the expenditure elasticities (e,y) Econometric Estimations
is the same as for the true AIDS model, but the price
elasticity formula for LA/AIDS is defined as


e# = 8y + r /wli (bi + Ysbiszs)

[wj + IkWk log(pk) (ekj + 86)] 1/wi.


In1e irs-stag~ e estimal ions aric pIesIIlntU 111 aui e
18 and 19, and the second-stage estimations are pre-
sented in Tables 20 and 21. Income and price elas-
(16) ticities computed for the slope dummy model esti-


.. .-...... ;. T blat










Table 18-Estimated parameters for the complete demand system, slope dummy model, China
Budget Share (Wi)
Beverages and
Variable Grain Meat Edible Oil Vegetables Fruit Other Foods Tobacco
Ln(Pgm) 0.050 -0.035 -0.003 -0.052 0.010 0.007 -0.002
(5.59) (-2.74) (-0.83) (-6.48) (1.67) (1.10) (-0.26)
Ln(Pmeatandproduct) -0.035 0.198 -0.004 0.055 0.016 -0.004 -0.076
(-2.74) (5.73) (-0.46) (3.15) (1.08) (-0.28) (-5.12)
Ln(Poi) -0.003 -0.004 0.022 -0.014 -0.005 -0.005 0.004
(-0.83) (-0.46) (5.47) (-2.64) (-0.92) (-1.17) (1.00)
Ln(Pvegetabes) -0.052 0.055 -0.014 0.025 -0.002 -0.002 -0.016
(-6.44) (3.15) (-2.64) (1.63) (-0.19) (-0.19) (-1.60)
Ln(Pfrit) 0.011 0.016 -0.005 -0.002 0.010 -0.015 -0.014
(1.67) (1.08) (-0.92) (-0.19) (0.63) (-1.63) (-1.72)
Ln(Potherfood) 0.007 -0.004 -0.005 -0.002 -0.015 0.041 -0.001
(1.10) (-0.28) (-1.17) (-0.19) (-1.63) (3.64) (-0.18)
Ln(Pbeveragesand tobacco) -0.002 -0.076 0.004 -0.016 -0.014 -0.001 0.061
(-0.26) (-5.12) (1.00) (-1.60) (-1.72) (-0.18) (4.92)
Ln(Pmisceiianeous) 0.023 -0.150 0.005 0.005 0.004 -0.020 0.044
(2.84) (-8.27) (1.00) (0.49) (0.04) (-2.20) (4.45)
Ln(X/P*) -0.055 -0.031 -0.005 0.131 0.010 0.005 0.061
(-3.58) (-1.18) (-0.78) (8.22) (0.94) (0.40) (3.69)
Urbanl x Ln(X/P*) -0.017 0.015 -0.003 -0.009 0.007 0.003 -0.005
(-8.30) (4.38) (-3.09) (-4.12) (4.49) (1.53) (-2.34)
Urban2 x Ln(X/P') -0.019 0.024 -0.002 -0.009 0.008 0.004 -0.010
(-8.27) (6.23) (-2.32) (-4.02) (4.72) (2.33) (-4.08)
Ln(family size) 0.185 0.034 -0.007 0.054 -0.042 -0.086 -0.064
(5.80) (0.63) (-0.55) (1.67) (-1.85) (-3.35) (-1.19)
Year dummy (1991) -0.004 0.028 0.002 0.004 -0.001 -0.004 -0.017
(-0.89) (2.91) (0.08) (0.65) (-0.27) (-0.86) (-3.27)
Constant 0.261 0.386 0.040 -0.705 0.043 0.203 -0.129
(2.15) (1.91) (0.822) (-5.52) (0.48) (2.07) (-1.03)
R2 0.926 0.783 0.608 0.700 0.759 0.627 0.456
Notes: System R2 = 0.999. The model is estimated using Zellner's seemingly unrelated regression with the imposition of the homogeneity and
symmetry restrictions. Regional dummies are not shown. The numbers in parentheses are t-values.


nations are reported in Table 22 (first-stage estima-
tions) and Table 23 (second-stage estimations).
The grain expenditure elasticity in rural areas is
estimated to be 0.25 and that in urban areas to be
below 0.10. Except for the miscellaneous category,
all other expenditure elasticities are quite high, ap-
proaching 1.0 or higher, and are similar between
rural, urban, and urban2 areas.
The aggregate meat expenditure elasticity is es-
timated to be about 0.9. The pork, poultry, and fish
expenditure elasticities are estimated to be in the 1.0
to 1.4 range. The milk expenditure elasticity is sub-
stantially higher and the expenditure elasticities for
beef and mutton and eggs are substantially lower.
All own-price elasticities for individual foods have
reasonable magnitudes, with the exception of that
for beef and mutton.
The primary objective of this analysis, however,
is not to estimate price and income effects, but to
measure apparent differences in food demand pat-


terns between urban and rural areas, after controlling
for income and price effects. These results are shown
in Tables 24 and 25.
Table 24 indicates that grain, edible oil, vegeta-
ble, and beverage and tobacco consumption is higher
in rural areas, after differences in income and food
prices are controlled. Meat, fish, dairy, and fruit
consumption is higher in urban areas. Table 25 gives
a disaggregation of the 5.8 and 9.3 kilogram differ-
ence in per capital intake of meat, fish, and dairy
products between rural and urban and urban2 areas,
respectively.
What are the implications of structural shifts of
this size for projections of demand for meat, fish,
and dairy products? A very rough calculation can be
made by taking 7.5 kilograms as a midpoint for the
5.8 and 9.3 kilograms just mentioned, due to struc-
tural shifts in moving from rural areas to small cities
and large cities, respectively. If the urban population
were to increase from one-quarter of the population










Table 19-Estimated parameters for the complete demand system, intercept dummy model,
China
Budget Share (Wi)

Beverages and
Variable Grain Meat Edible Oil Vegetables Fruit Other Foods Tobacco

Ln(Pgr,) 0.050 -0.032 -0.003 -0.054 0.012 0.008 -0.002
(5.65) (-2.57) (-0.88) (-6.69) (1.80) (1.15) (-0.27)
Ln(Pmeatand product) -0.032 0.200 -0.004 0.057 0.015 -0.004 -0.076
(-2.57) (5.60) (-0.42) (3.25) (0.98) (-0.29) (-5.08)
Ln(Poi,) -0.003 -0.004 0.022 -0.014 -0.005 -0.006 0.004
(-0.88) (-0.45) (5.49) (-2.62) (-0.91) (-1.18) (0.99)
Ln(Pvegetables) -0.054 0.057 -0.014 0.025 -0.001 -0.002 -0.016
(-6.69) (3.25) (-2.62) (1.54) (-0.11) (-0.22) (-1.64)
Ln(Ppui,) 0.012 0.015 -0.005 -0.001 0.010 -0.015 -0.014
(1.80) (0.98) (-0.91) (-0.12) (0.60) (-1.60) (-1.70)
Ln(Potherfood) 0.008 -0.004 -0.006 -0.002 -0.015 0.041 -0.001
(1.15) (-0.29) (-1.18) (-0.22) (-1.60) (3.66) (-0.18)
Ln(Pbeverages and tobacco) -0.002 -0.076 0.004 -0.016 -0.014 -0.001 0.061
(-0.27) (-5.08) (0.99) (-1.64) (-1.70) (-0.18) (4.93)
Ln(Pmiscetaneous) 0.022 -0.150 0.005 0.006 0.0004 -0.021 0.044
(2.77) (-8.24) (1.00) (0.53) (0.04) (-2.22) (4.33)
Ln(X/P') -0.065 -0.022 -0.006 0.129 0.013 0.007 0.055
(-4.22) (-0.83) (-0.93) (7.85) (1.17) (0.52) (3.34)
Urbanl -0.095 0.080 -0.013 -0.041 0.034 0.013 -0.030
(-8.78) (4.31) (-3.04) (-3.59) (4.36) (1.50) (-2.55)
Urban2 -0.106 0.127 -0.011 -0.044 0.040 0.023 -0.056
(-8.72) (6.15) (-2.17) (-3.48) (4.58) (2.29) (-4.29)
Ln(family size) 0.176 0.031 -0.006 0.068 -0.044 -0.086 -0.070
(5.61) (0.58) (-0.51) (2.08) (-1.94) (-3.37) (-2.10)
Year dummy (1991) -0.004 0.026 0.0003 0.005 -0.001 -0.004 -0.017
(-0.91) (2.80) (0.11) (0.90) (-0.32) (-0.87) (-3.34)
Constant 0.328 0.355 0.045 -0.727 -0.002 0.196 -0.094
(2.67) (1.70) (0.88) (-5.48) (-0.37) (1.94) (-0.73)
R2 0.929 0.783 0.607 0.692 0.757 0.626 0.460
Notes: System R2 = 0.999. The model is estimated using Zellner's seemingly unrelated regression with the imposition of the homogeneity and
symmetry restrictions. Regional dummies are not shown. The numbers in parentheses are t-values.


to two-thirds of the population (over a specified
period), this would result in a 3-kilogram increment
in demand for these products (40 percent of 7.5
kilograms, after population growth, income, and
prices are controlled). Total national per capital con-
sumption of meat, fish, and dairy products presently
is about 30 kilograms per capital annually, so that this
is about a 10 percent increase.'6
Ten percent may not seem like a large additional
increase. However, as the Taiwan data demonstrate,
it is crucial to note that substantial structural shifts in
food demand are occurring within rural and urban


populations as well. In this section of the paper, only
the difference between two moving targets has been
measured. If, as can be expected, structural shifts
occur within rural and urban populations, the total
additional demand may be much larger than 10 per-
cent.



Conclusions
Empirical estimates presented for Taiwan and China
support an hypothesis that structural changes in food


16Comparing the results for demand for meat, fish, and dairy for China and Taiwan, the observed difference in consumption between
rural and urban areas for 1989-91 is about 40 kilograms for China (see Table 15). Thus, 7.5 kilograms, the estimate used here for the
structurally related shifts in demand for China between rural and urban areas, accounts for about 20 percent of this 40-kilogram
difference. For Taiwan, the difference in meat and fish consumption (not including dairy) between villages and cities in 1981 is
estimated to be just under 30 kilograms (see Table 12). Between 5 and 6 kilograms or about 20 percent of this difference is attributed
to structural shifts.










Table 20-Estimated parameters for meat and the meat product demand subsystem, slope
dummy model, China
Budget Share (Wi)
Variable Pork Beef and Mutton Poultry Fish Eggs
Ln(Ppork) -0.224 -0.097 0.075 0.075 0.098
(-4.07) (-2.94) (2.92) (2.63) (3.15)
Ln(Pbeefandmutton) -0.097 0.120 0.006 0.028 0.027
(-2.94) (3.54) (0.32) (1.23) (1.15)
Ln(Ppoutry) 0.075 0.006 -0.034 -0.032 -0.032
(2.92) (0.32) (-1.45) (-1.87) (-1.69)
Ln(Pfish) 0.075 0.028 -0.032 -0.061 -0.052
(2.63) (1.23) (-1.87) (-2.14) (-2.22)
Ln(Peggs) 0.098 0.027 -0.032 -0.052 -0.024
(3.15) (1.15) (-1.69) (-2.22) (-0.74)
Ln(Pmilk) 0.073 -0.085 0.017 0.042 -0.017
(3.08) (-4.97) (1.11) (2.62) (-0.95)
Ln(Xmea/P*) 0.005 -0.054 0.013 0.053 -0.046
(0.19) (-2.36) (0.94) (2.55) (-2.11)
Urbanl x Ln(XmeadP*) -0.034 0.036 0.005 0.014 -0.019
(-3.34) (4.10) (1.01) (1.79) (-2.32)
Urban2 x Ln(Xmeat/P') -0.043 0.033 0.010 0.012 -0.016
(-3.78) (3.34) (1.65) (1.42) (-1.73)
Ln(family size) 0.035 0.145 0.011 0.140 -0.363
(0.36) (1.73) (0.21) (1.90) (-4.65)
Year dummy (1991) -0.004 0.028 0.015 -0.008 0.004
(-3.11) (2.52) (2.04) (-0.81) (0.40)
Constant 0.752 -0.164 0.022 -0.246 0.790
(3.93) (-1.00) (0.21) (-1.69) (5.12)
R2 0.750 0.734 0.666 0.611 0.506
Notes: System R2=0.990. The model is estimated using Zellner's seemingly unrelated regression with the imposition of the homogeneity and
symmetry restrictions. Regional dummies are not shown. The numbers in parentheses are t-values.


demand can be quite significant factors, driving the
rapid changes in dietary patterns seen in East Asia
over the past three decades. Because most previous
studies have tried to explain these dramatic changes
in diets primarily in terms of rising incomes and
changing food prices, relatively little is known about
the specific reasons causing these structural shifts.
Thus, it is difficult to form judgments as to the points
at which, in the process of economywide structural
adjustment, these structural changes in food demand
will begin, accelerate, then slow down, and perhaps
stop.
How this lack of information will affect policy is
perhaps most easily understood for rice. Almost all
cross-sectional evidence on demand for rice indi-
cates that the income elasticity is either close to zero
or positive as it substitutes for less-preferred food
staples with rising incomes. Yet per capital consump-
tion of rice has declined dramatically over time in a
few countries, including Japan and Taiwan. These
two empirical observations could be reconciled if
rice prices had risen dramatically compared with
substitute foods, but that has not occurred. There-


fore, significant downward structural shifts in de-
mand likely explain a substantial portion of the ob-
served decline.
Assuming continued sustained economic
growth, at what level of income or structural trans-
formation or both will per capital rice consumption
begin to decline in China, India, and Indonesia (to
name three countries with large populations that ac-
count for a high proportion of world rice consump-
tion)? Once the decline begins, will rice consump-
tion eventually fall by as much as 50 percent, as has
happened in Japan and Taiwan, or perhaps more?
Might such a decline not occur? It is interesting to
note that rice consumption in Korea did not fall
much between 1973 and 1992, declining from 140
kilograms to 126 kilograms per capital (FAO 1994).
Two very different future scenarios are possible
for demand for rice in Asia to the year 2020, one in
which per capital consumption of rice declines dra-
matically and another in which per capital consump-
tion falls marginally. Existing estimates of income
elasticities for rice provide little insight into which
scenario might occur.











Table 21-Estimated parameters for meat and the meat product demand subsystem, intercept
dummy model, China
Budget Share (Wi)

Variable Pork Beef and Mutton Poultry Fish Eggs

Ln(Ppork) -0.239 -0.095 0.076 0.079 0.103
(-4.39) (-2.93) (2.94) (2.82) (3.31)
Ln(Pbeefandmullon) -0.095 0.128 0.008 0.027 0.018
(-2.93) (3.83) (0.45) (1.19) (0.78)
Ln(Ppo,,ry) 0.076 0.008 -0.034 -0.033 -0.034
(2.94) (0.45) (-1.44) (-1.93) (-1.79)
Ln(Pflsh) 0.079 0.027 -0.033 -0.064 -0.051
(2.82) (1.19) (-1.93) (-2.21) (-2.19)
Ln(Peggs) 0.103 0.018 -0.034 -0.051 -0.020
(3.31) (0.78) (-1.79) (-2.19) (-0.60)
Ln(Pmak) 0.076 -0.087 0.017 0.041 -0.016
(3.26) (-5.09) (1.11) (2.54) (-0.90)
Ln(Xmear/P') -0.008 -0.046 0.016 0.054 -0.046
(-0.32) (-2.21) (1.20) (2.82) (-2.29)
Urbanl -0.117 0.146 0.018 0.055 -0.091
(-3.38) (4.92) (0.97) (2.08) (-3.21)
Urban -0.157 0.143 0.035 0.055 -0.086
(-3.81) (4.04) (1.65) (1.72) (-2.57)
Ln(family size) 0.030 0.194 0.011 0.156 -0.418
(0.31) (2.33) (0.22) (2.11) (-5.32)
Year dummy (1991) -0.045 0.034 0.015 -0.005 -0.001
(-3.36) (3.03) (2.07) (-0.52) (-0.07)
Constant 0.810 -0.282 0.012 -0.284 0.891
(4.10) (-1.69) (0.11) (-1.89) (5.60)
R2 0.750 0.748 0.666 0.613 0.514
Notes: System R2=0.991. The model is estimated using Zellner's seemingly unrelated regression with the imposition of the homogeneity and
symmetry restrictions. Regional dummies are not shown. Numbers in parentheses are t-values.


Table 22-Expenditure and own price elasticities of demand for the commodity group evaluated
at the overall sample mean, China
Expenditure Elasticity Uncompensated Own Price Elasticity
Budget
Commodity Share Mean Rural Urbani Urban2 Mean Rural Urbani Urban2

Grain 0.106 0.134 0.246 0.092 0.073 -0.437 -0.448 -0.432 -0.430
Meat 0.280 0.914 0.841 0.928 0.967 -0.279 -0.268 -0.281 -0.288
Edible oil 0.032 0.733 0.786 0.700 0.719 -0.306 -0.310 -0.304 -0.305
Vegetables 0.130 2.315 2.460 2.253 2.242 -1.103 -1.136 -1.089 -1.087
Fruit 0.066 1.311 1.225 1.340 1.360 -0.864 -0.859 -0.866 -0.867
Other food 0.079 1.124 1.089 1.125 1.153 -0.486 -0.484 -0.486 -0.487
Beverages and tobacco 0.105 1.721 1.833 1.701 1.639 -0.527 -0.543 -0.524 -0.520
Miscellaneous 0.202 0.241 0.162 0.291 0.262 -0.388 -0.373 -0.399 -0.393
Note: The elasticities are computed based on the model presented in Table 18.


An analagous dilemma presents itself in project-
ing demand for meat, fish, and dairy products. It
would seem clear that income elasticities for these
products are substantially above zero. However, ex-
isting income elasticity estimates from time-series
data may be upwardly biased for countries such as
Japan and Taiwan, measuring both the positive ef-
fect of income on demand for these products and


structural shifts in demand, structural shifts that are
positively correlated with increases in per capital
gross domestic product.
Again, the crucial question for demand projec-
tions in these economically more advanced countries
is at what point these structural shifts will slow down
and perhaps stop, at which point the upwardly biased
estimates of income elasticities for meat, fish, and











Table 23-Meat expenditure and own price elasticities of demand for the commodity group
evaluated at the sample mean, China
Expenditure Elasticity Uncompensated Own Price Elasticity
Budget
Commodity Share Mean Rural Urbanl Urban2 Mean Rural Urbani Urban2

Pork 0.471 0.954 1.011 0.936 0.916 -1.462 -1.480 -1.456 -1.449
Beef and mutton 0.103 0.689 0.453 0.824 0.788 0.239 0.296 0.206 0.214
Poultry 0.107 1.176 1.129 1.179 1.222 -1.329 -1.325 -1.329 -1.333
Fish 0.141 1.454 1.391 1.489 1.482 -1.478 -1.472 -1.481 -1.480
Eggs 0.149 0.595 0.676 0.545 0.564 -1.089 -1.104 -1.080 -1.084
Milk 0.029 2.071 2.057 1.960 2.196 -2.086 -2.085 -2.079 -2.095
Note: The elasticities are computed based on the model presented in Table 20.



Table 24-The effects of urbanization on per capital annual food consumption in China
Intercept Dummy Model Slope Dummy Model

Commodity Rural to Urbant Rural to Urban2 Rural to Urbant Rural to Urban2
(kilograms/capita/year)
Grain -63.2 -70.1 -58.3 -64.2
Meat, fish, and their products 5.8 9.3 5.7 8.9
Edible oil -1.4 -1.2 -1.4 -1.1
Vegetables -21.2 -22.9 -23.0 -24.9
Fruit 8.3 9.8 8.2 9.6
Other food 1.8 3.1 1.8 3.0
Beverages and tobacco -5.0 -9.3 -4.4 -8.5
Notes: The effects of urbanization on food consumption are measured as the consumption differences between rural and urban consumers after
controlling for the income and price effects. In the above analysis, consumption changes are computed based on a hypothesis of moving a
consumer from a rural to an urban area, holding his total expenditure (X) and prices constant at the levels that he or she faced in the rural area.


dairy products will begin to overestate future de-
mand. Interestingly, in countries such as India and
Indonesia, where meat consumption is presently
quite low because structural transformation is in an
early stage, time-series data for meat consumption


will not reflect a possible impending structural shift
in demand. Thus, existing income elasticities will
give underestimates of demand for meat, if these
structural changes in food demand do indeed materi-
alize.













Table 25-The effects of urbanization on per capital annual consumption of meat, fish, and their
products in China

Intercept Dummy Model Slope Dummy Model

Commodity Rural to Urbani Rural to Urban2 Rural to Urbani Rural to Urban2
(kilograms/capita/year)
Neutral effect
Pork 3.1 4.9 3.0 4.8
Beef and mutton 0.3 0.6 0.3 0.5
Poultry 0.4 0.6 0.4 0.6
Fish 0.7 1.2 0.7 1.1
Eggs 0.9 1.4 0.8 1.3
Milk 0.4 0.6 0.4 0.6
Total 5.8 9.3 5.7 8.9
Biased effect
Pork -1.5 -2.5 -1.2 -2.0
Beef and mutton 2.2 2.1 1.5 1.3
Poultry 0.3 0.5 0.2 0.3
Fish 1.1 1.0 0.8 0.6
Eggs -1.7 -1.7 -1.0 -0.9
Milk -0.3 0.7 -0.3 0.7
Net effect
Pork 1.7 2.5 1.8 2.7
Beef and mutton 2.5 2.6 1.8 1.8
Poultry 0.7 1.1 0.6 0.9
Fish 1.8 2.2 1.5 1.7
Eggs -0.9 -0.3 0.2 0.4
Milk 0.1 1.2 0.1 1.3
Total 5.8 9.3 5.7 8.9
Notes: The impact of urbanization on individual meat consumption is measured as the difference between rural and urban consumption, with income
and price effects controlled. In the table, consumption differences are separated into two components: a "neutral" impact and a "biased" impact
The former assumes that the impact of urbanization is equal for all component foods in the aggregate meat, fish, and their products group (the
first-stage estimations), while the latter reflects the adjustments (substitutions) among component foods due to urbanization (the second-stage
estimations). The net impact is the sum of these two effects.
















References


Amemiya, T. 1977. The maximum likelihood esti-
mator and the nonlinear three-stage least squares
estimation in the general nonlinear simultaneous
equation model. Econometrica 45 (4): 955-968.

Bouis, H. E. 1991. Rice in Asia: Is it becoming a
commercial good?: Comment. American Jour-
nal ofAgricultural Economics 73 (2): 522-527.

Forthcoming. A food demand system
based on demand for characteristics: If there is
"curvature" in the Slutsky matrix, what do the
curves look like and why? Journal of Develop-
ment Economics.

Capps, O., R. Tsai, R. Kirby, and G. Williams. 1994.
A comparison of demands for meat products in
the Pacific Rim region. Journal of Agricultural
and Resource Economics 19 (1): 210-224.

Deaton, A. S., and J. Muellbauer. 1980. An almost
ideal demand system. American Economic Re-
view 70 (3): 312-326.

FAO (Food and Agriculture Organization of the
United Nations). 1994. FAO Agrostat-PC, food
balance sheets and population domains, 1994,
Version 3.0. Rome. Computer disk.

Huang, J., and C. David. 1993. Demand for cereal
grains in Asia: The effect of urbanization. Agri-
cultural Economics 8(1): 107-124.

Huang, J., and S. Rozelle. 1995. Market develop-
ment and food demand in rural China. Food
Consumption and Nutrition Division Discussion
Paper 4. International Food Policy Research In-
stitute, Washington, D.C.


Ito, S., E. W. F. Peterson, and W. R. Grant. 1989.
Rice in Asia: Is it becoming an inferior good?
American Journal ofAgricultural Economics 71
(1): 32-42.

Ravallion, M. 1990. Income effects on undernutri-
tion. Economic Development and Cultural
Change 38 (3): 489-515.

Sahn, D., and H. Alderman. 1988. The effects of
human capital on wages, and the determinants of
labor supply in a developing country. Journal of
Development Economics 29 (2): 157-183.

SSB (State Statistical Bureau of China). 1990. Ur-
ban household income and expenditure survey
for China, 1989. Beijing: State Statistical Press.

1992. Urban household income and ex-
penditure survey for China, 1991. Beijing: State
Statistical Press.

1993. Rural household survey yearbook
of China, 1991. Beijing: State Statistical Press.

Taiwan, Council for Agricultural Planning and De-
velopment. Various years. Food balance sheets.

Taiwan, Department of Agriculture and Forestry.
1981. Monthly agricultural prices and costs.

Taiwan, Department of Budget, Accounting and Sta-
tistics. 1991. Monthly agricultural prices and
costs.

Various years. Household income and
expenditure survey.








Food, Agriculture, and the Environment Discussion Papers


1. Alleviating Poverty, Intensifying Agriculture, and Effectively Managing
Natural Resources, by Per Pinstrup-Andersen and Rajul Pandya-Lorch, 1994

2. Sociopolitical Effects of New Biotechnologies in Developing Countries,
by Klaus M. Leisinger, 1995

3. Africa's Changing Agricultural Development Strategies: Past and Present
Paradigms as a Guide to the Future, by Christopher L. Delgado, 1995

4. A 2020 Vision for Food, Agriculture, and the Environment in Sub-Saharan
Africa, edited by Ousmane Badiane and Christopher L. Delgado, 1995

5. Global Food Projections to 2020: Implications for Investment, by Mark W.
Rosegrant, Mercedita Agcaoili-Sombilla, and Nicostrato D. Perez, 1995

6. A 2020 Vision for Food, Agriculture, and the Environment in Latin America,
edited by James L. Garrett, 1995

7. Agriculture, Trade, and Regionalism in South Asia, by Dean A. DeRosa and
Kumaresan Govindan, 1995

8. Major Natural Resource Management Concerns in South Asia, by Gerard J.
Gill, 1995

9. Agriculture, Technological Change, and the Environment in Latin America:
A 2020 Perspective, by Eduardo J. Trigo, 1995

10. Overcoming Malnutrition: Is There an Ecoregional Dimension?, by Manohar
Sharma, Marito Garcia, Aamir Qureshi, and Lynn Brown, 1996

11. Structural Changes in the Demand for Food in Asia, by Jikun Huang and
Howarth Bouis, 1996









Jikun Huang, formerly a research fellow at IFPRI, is a professor and director at the Center for Chinese
Agricultural Policy, Institute of Agricultural Economics, Chinese Academy of Agricultural Sciences.
Howarth Bouis is a research fellow in IFPRI's Food Consumption and Nutrition Division.























































IFPRI 1200 SEVENTEENTH STREET, N.W.* WASHINGTON, D.C. 20036-3006 U.S.A. 1-202/862-5600
il I FAX: 1-202/467-4439 E-MAIL: ifprl@cgnet.com WEB: http://www.cgiar.org/ifpri




University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs