3 link;
In“: 3‘; 1. ,3.

'1, Y r. m‘
v.7..-'u.n3v- ‘

‘ 'r:
. .ww Hf"
. 1.1.1;-

_, y
‘ -;. .\I1\'r
firm 1.3..)
:

Rh.“

.0. In

m
H: ..‘
omnwinre

,,
,...... 5.
:orv- .. ”nun”:
n- ,

» u

, -
....u..

9'
..".......,.
~3‘J- .' ,.. . . '..."" . W,
3! 1 I n " 1'..',-~y,~. I
,H....,.,
.:v;'.?“;"."r' ”3'

’7 2"... . “fr-'1'...“ .
...»..:..u
I: l

. . "
‘ . n .3 1
,".,,.f".'.',." .H .' ..T.,',.",,'
nu... .,, u. , ..-....,..,..

. hw-

 

!' “7H"?
nu...“

,n 1., - :r-v-
’1'

1
1.
<. s: 1-1

1
‘4‘
.4 ,.
. 1" . . 3 _‘
:ijfi‘é 2:“:suf :4?
3r H a :3;‘...
..

§,f€‘:....~.!,.~

«.1
3-3:}

5 m
:1 3:1: 1....
n,"

1
Sn)“;

”,5.

n
.2 .,, . .
1

:‘1—‘3‘1..!:.

',.’"'..:"“
14m
\

AV -‘<.~i~t«

'0'. ..
‘rvgm

am
.11!

.,. .
. 1.1; 5.. .\
”2:1"53-

.
"nu1,._...‘lw
'IJ'I” 'WIJI‘J
i’ ‘I 1-." 1'. r1. 2.”
5'7

Lu
1... ..
.k\.~~ "‘1'

1:5,. 1"

11.4- all

.1—5". 15.13}.
‘2

..1h:.r p...

3 f1” 6 u 4 3
Had-Mi‘afr... ‘ .
‘0

, :3??? '
.2532; .33: €31). “if!

34;.“
5‘

V.

is.-
J'W-

'.:.1...1Jk .
. V,‘....::L
. mum;
:'-.\~m4f~
.-1r

.n.

an.
L

,. . “0,1:

..,,
v.
_r;.'

‘vtr

ilk:
412:1":

’1

,

-.~ ...q

'15- “fit
‘K .

, ”a
.f.‘.":‘-’,

 

z
with: ”-3.1.-
191

duh-vani-

fir. ‘

. ,1 13%...1

IVERSI ITY UBRARIE

H l Limit“

 

 

 

 

 

 

 

 

 

IHUWHHHHWNW“

31293 010

This is to certify that the

thesis entitled

A REGIONAL ANALYSIS OF THE EFFECTS OF FOOD
SUBSIDY POLICIES ON HOUSEHOLD CONSUMPTION
PATTERNS IN MOROCCO

presented by

Jennifer Bard Wohl

has been accepted towards fulfillment
of the requirements for

Master degree in Science

 

 

jkL/f $48“

Major profes )1]

Date April 14, 1992

 

0-7639 MS U is an Affirmative Action/Equal Opportunity Institution

 

UBRARY
Michigan State
University

 

 

 

PLACE IN RETURN BOX to remove We checkout from your record.
TO AVOID FINES return on or before due due.

   
 

DATE DUE DATE DUE DATE DUE

F i

-| l-
[—1
ES
F i i
E ii I J

msu le An Affirmetlve ActiorVEquel Opportunity Institution
ammo-m

 

 

 

 

 

 
 

 

 

 

 

 

 

 

E;

 

 

 

 

 

 

 

 

 

i:

 

 

 

 

 

 

 

 

 

 

#

 

A REGIONAL ANALYSIS OF THE EFFECTS OF FOOD SUBSIDY POLICIES
ON HOUSEHOLD CONSUMPTION PATTERNS IN MOROCCO

by

Jennifer Bard Wohl

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

MASTER OF SCIENCE

Department of Agricultural Economics

1992

ABSTRACT

A REGIONAL ANALYSIS OF TEE EFFECTS OF FOOD SUBSIDY POLICIES
ON HOUSEHOLD CONSUNPTION PATTERNS IN NOROCCO

by

Jennifer Bard Wohl

since the early 19708 the Government of Morocco (60M)
has used consumer price subsidies to transfer income to the
poor and to keep urban wages poor. The fiscal crisis the
subsidies has engendered has led to pressure for their
removal. This thesis uses a demand model with a functional
form similar to the AIDS to estimate regional elasticities
of demand. The elasticities are used to discuss the
regional consumption effects of removing the subsidies.
Heckman's two-stage procedure is used to correct for the
sampling bias introduced when households report zero
consumption. The elasticities were found to vary by region.
Barley is found to an inferior good in the two urban areas
and a necessity in the rural areas. The inverse Mill's
ratio was found to be significant in only 2 of 10 demand

equations.

ACKNOILEDGNENTS

I would like to thank my committe members John Staatz,
Ching-Fan Chung, and Dirck Stryker for their careful reading
and insightful comments on earlier drafts of this thesis,
and for their patience, patience, and more patience as I
wrote this thesis.

I would also like to thank Lisa Schwartz, Cynthia
Phillips, Jim Sterns, Cynthia Donovan, and Kim Aldridge for
their support and encouragement, and for their patience,
patience, and more patience as I wrote this thesis.

Thanks also to Zanne Newman for an endless supply of
magic wands, thesis monsters, and other sundry thesis tools
that, albeit belatedly, worked wonders.

Finally, the biggest thanks to my mom and dad, Audrey

and Herb Wohl, for always accepting my collect phone calls.

iii

TABLE OF CONTENTS

CHAPTER I: INTRODUCTION . . . .
Importance of Study . . .
Hypotheses . . . . . . . .
Methods . . . . .
Organization of the Thesis

CHAPTER 2: THEORETICAL AND CONCEPTUAL
Introduction . . . . . . . . . . . .
Goals of Subsidy Programs . . .
Targeted versus Untargeted Subsidies
Explicit vs Implicit Subsidies . . .
Costs of Subsidies . . . . . . . . .
Forms of Subsidies . . . .

Unrestricted food price subsidie
Food stamps . . . . . . . . . .
Rationing . . . . . . . . .
How Subsidies Affect Consumption
Conceptual Framework . .
Adding Up . . .
Homogeneity . . .
Symmetry . . . .
Weak Separability .
Conclusion . . . . . .

CHAPTER 3: METHODS . . . . . . . . .
Introduction . . . . . . . . . .
Demand Systems vs. Single Demand
Flexible Functional Forms . . .
The AIDS . . . . . . . .

Restrictions . . .
Adding-up . .
Homogeneity .
Symmetry . . .
Limitations . . . . .
Income Elasticities
Practical Aspects . . .
Data Requirements
Stone's Index .
Software Requirements

Other Modeling Issues . .
Unit Values . . . . .
Non-Consumption . . .

How to Correct for

eeeeeeeeeeeeetljee

a

Sampling Design . . . . . . . .
Weighting . . . . . . . . . . .
Conclusion . . . . . . . . . . . . .

iv

2

eeeeeeeeeeeeeerfee

p.

eeeeeeeeeeeeeeeOee

Censored S
Heckman's Two-Stage Procedure

:5

eeeegeeeeeeeeeeeeeeemot

o
a

O O % O O O O O O O O O O O O O O O O O O

H

O O O m O O O O O O O O O O O O O O O O O O

CHAPTER 4: DESCRIPTIVE STATISTICS
Description of Study Areas .

General Description of Morocco

Study Area of Marrakech
Sidi-Kacem . . . . . .
Casablanca . . . . . .
Descriptive Statistics about
Morocco . . . . .
Food Shares . .
Cereals .

Meat . . .

Oils . . .

Beverages . . .
Conclusions of this Section

CHAPTER 5: ANALYTIC DISCUSSION OF RESULTS
Description of Variables in the Model

Dependent Variables . .
Independent Variables .
Unit Values . . .
Total Expenditure

Squared Expenditure Term
Inverse Mill's Ratio .

Household Size . .

Dummy Variables .

Interaction Terms
The General Model . . .

Judging Which is the Best Model

Discussion . . . . . .
Inverse Mill's Ratio
Cereals . . . . .

Soft Wheat

Hard wheat

Barley . .

Oils . . . . .
Meats . . . . .
Sugar . . . . .
Other Studies . . .

CHAPTER 6: CONCLUSIONS AND POLICY
Policy Implications . . . .
Methodological Conclusions .

Unit Values . . . . . .
Zero Consumption . .
Model Specification .
GAUSS Program . . . .
Regional Analysis .
Further Research . . . .

Food Con ump

eeeeeeemeeeeee
eeeeeeerfeeeeee

O O O O O O O O O O O O O O O O
.
O O O O O O O O O O O O O O O O O O O O

INPLICATIONS

eeeeeeeOeeeeee

eee.eeee:eeeeee

65
65
65
67
68
68

68
7O
72
79
86
89
91

94
94
94
95
95
96
97
97
97
98
98
99
100
111
111
112
112
113
114
115
116
117
118

122
122
126
126
127
128
130
131
131

APPENDIX

APPENDIX

APPENDIX
APPENDIX

APPENDIX
APPENDIX

APPENDIX

APPENDIX

”me:

APPENDIX

APPENDIX

APPENDIX

2-1:

3-18

3-28

3-38

3-4:
3-5:

3-58

4-18

4-28

4-38

5-1!

BIBLIOGRAPHY

"ADDING-UP" RESTRICTION WRITTEN IN TERMS
OF BUDGET SHARE AND ELASTICITY . . . . .

OTHER DEMAND MODELS . . . . . . . . . .

TRANSFORMING THE AIDS COST FUNCTION INTO
8mm smzs O O O O O O O O O O O O O

CALCULATION OF EXPENDITURE AND PRICE
'WTICITIBB O O O O O O O O O O O O O O

GROUPE DE TRAVAIL RESULTS . . . . . . .
CLUSTERING AND STRATIFICATION . . . . .

SAMPLING METHODS USED FOR THE MOROCCAN
DAT; 8 3! O O O O O O O O O O O O O O O O

“I T mugs O O O O O O O O O O O O O O

HOUSEHOLD SIDE BY STUDY AREA AND INCOME
on“! I L: O O O O O O O O O O O O O O O O

CONVERSION FACTORS . . . . . . . . . . .

VARIANCE-COVARIANCE MATRIX WHEN MILL’S
RATIO INCLUDED . . . . . . . . . . . . .

PROBIT RESULTS FROM FIRST STAGE OF
HECXMAN PROCEDURE . . . . . . . . . . .

vi

134

136

141

142
150

151

156

159

160

161

162

164

169

Table

Table

Table
Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

4-1:
4- :

4-3a:

4-3b:

4-5b

4-6a:

4-6b:

4-7:

4-8a:

4-8b:

4-9:

4-10a:

4-10b:

4

11:

LIST OF TABLES

Regression Coefficients from Regressing
Unit Value on Total Expenditure and
Household Size . . . . . . . . . . . . .
Percentage of Sampled Population
Reporting Positive Consumption of
Products . . . . . .

Percentage of Sampled Population Living
Below the Poverty Level by Study Area .
Statistics on Total Expenditures and
Expenditures on Food . . . . . . . . . .
Average Per Capita Consumption of
Cereals by Study Area for Households
Consuming that Cereal (kg) . . . . . . .
Average Per Capita Consumption of
Cereals by Study area for all Households
Surveyed (kg) . . . . . . . . .
Percentage of Food Budget Spent on
Cereal Products by Consuming Households
Average Per Capita Consumption of Cereals
by Region and Income Quartile for All
Households Surveyed . . . . . . .
Average Per Capita Consumption of Cereals
by Region and Income Quartile for
Households Consuming that Cereal . . . .
Average Per Capita Consumption of Meats
by Study area for All Households
Surveyed (kg) . . . . . . . . .
Average Per Capita Consumption of Meats
by Study area for Households Reporting
Consumption of that Meat (kg) . . . .
Percentage of Food Budget Spent on Meat
Products by Consuming Households . . . .
Average Per Capita Consumption of Meats
by Region and Income Quartile for All
Households Surveyed . . . . . . . . . .
Average Per Capita Consumption of Meats
by Region and Income Quartile for
Households Consuming that Meat . . . .
Percentage of Food Budget Spent on Oil
Products by Consuming Households . . . .
Average Per Capita Consumption of Oils
by Province for all Households Surveyed
(kg) . . . . . . . . .
Average Per Capita Consumption of Oils
by Province for Consuming Households
(kg) . . . . . . . . . . . . . . . .
Percentage of Food Budget Spent on
Beverages by Consuming Households (kg)

vii

52

59
69

7O

74

74

75

76

77

8O

80

81

83

84

87

87

88

89

Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table

Table
Table

Table
Table
Table
Table
Table

Table
. Table

4-12a:

4-12b:

5-1:
5-2:

5- :
5-4:
5-5:
5- :
5-7:
5-8:
5-9:

5-10:
5-11:

5-12:
5-13:
5-14:
5-15:
A3-4:

A4-1:
A4-2:

Average Per Capita Consumption of
Beverages and Sugar for All Households
Surveyed (kg) . . . . . . . . . . .
Average Per Capita Consumption of
Beverages and Sugar for Consuming
Households (kg) . . . . . . . . . . . .
Cereal Elasticities . . . .
Soft Wheat Elasticities by Study Area
and Income Quartile . . . .
Hard Wheat Elasticities by Study Area
and Income Quartile . . . . . . .
Barley Elasticities by Study Area and
Income Quartile . . . . . . . . . . . .
Oil Elasticities . . . . . . . . . . . .
Meat Elasticities . . .

Chicken Elasticities by Study Area and

Income Quartile . . . . . . . ._.
Beef Elasticities by Study Area and
Income Quartile . . . . . . . . . .

Mutton Elasticities by Study Area and
Income Quartile . . . . . . . . . . . .
Sugar Elasticities . . . . . .
Probit Results from the First Stage of
Heckman's Two Stage Procedure . . . . .
Cereal Elasticities from Groupe de
Travail (1990) . . . . . . . . . .
Oil Elasticities from Groupe de Travail

(1990) O O O O O O O O
Meat Elasticities from Groupe de Travail
(1990) O O O O O O O O O O O O

Beverage Elasticities from Groupe de
Travail (1990) . . . . . . . . . . . . .
Regressions of Unit Values . . . . . . .
Unit Values by Study Area . . . .
Households Size by Study Area and Income
Quartile . . . . . . . . . . . . . . . .

viii

90

90
102
103
104
105
106
107
108
109

110
111

116
119
119
119
120
150
159

160

Figure 2-1:
Figure 2-2:

Figure 4-1:

LIST OF FIGURES

Tradeoffs in consumption between food and
non-food . . . . . . . . . . . . . . . 21

Tradeoffs in consumption between subsidized
food and other food . . . . . . . . . . 23

Map of Morocco . . . . . . . . . . . . . 66

ix

CHAPTER 1: INTRODUCTION

Since the early 1970s the Government of Morocco (GOM)
has used consumer price subsidies to transfer income,
ostensibly to the poor, and to keep urban wages low. The
windfall revenue from the phosphate boomz between 1973 and
1975 easily covered the costs of the consumer subsidies on
imported foods such as soft wheat, vegetable oil, sugar
powder, meat, and dairy products. By 1981, however, the
subsidies were a serious drain on public resources and were
contributing to a general fiscal crisis. The crisis started
in 1976 when the price of phosphates on the international
market crashed. Further contributing to the fiscal crisis
were a prolonged war in the Sahara Desert, increased costs
of imported oil, a series of severe droughts, and general
stagnation in the agricultural sector.

In 1981 the COM was forced to turn to the International
Monetary Fund (IMF) for assistance in improving the balance
of payments and restructuring the foreign debt. One of the

conditions of the IMF assistance was the gradual removal of

 

1 Tuluy and Salinger (p 65) argue that if one accounts
for the overvaluation of the Dirham, the subsidization of
soft wheat actually began as early as 1960. In the case of
sugar, the indirect effect of the overvalued currency was
that beginning in 1960, consumers were taxed less than if
one only considers the direct effects. Subsidization here
will refer to only direct intervention by the COM.

2 Morocco is the world's leader in exports of
phosphate.

2
consumer subsidies. Although the consumer subsidies on meat
and dairy products were eliminated by 1979, the consumer
prices of soft wheat, vegetable oil, and sugar are still
subsidized. Attempts by the COM in 1981 and 1984 to remove
the subsidies on these consumer goods were thwarted by
fierce riots during which many civilians were killed. The
severity of the unrest attests to the importance of the
subsidies to the population; the speed with which the
changes were rescinded shows the potential threat such
unrest can pose to the legitimacy of the GOM.

Historically, the subsidy program operated through a
series of rebates to participants in the marketing chain.3
In the case of soft wheat, for example, licensed traders
bought wheat from farmers at support prices, sold to the
mills at a fixed price and were reimbursed the difference,
plus transportation costs and a fixed margin. Traders could
also import grain from the world market at world market
prices. When world prices were lower (higher) than the
support price, the traders (government) reimbursed the
(traders) government the difference between the support
price and the world price, minus (plus) transportation costs
and a fixed profit margin. Millers were also rebated when
they sold flour to wholesalers and bakeries. This was the

major part of the subsidy. Wholesalers and millers sold

 

3 The following discussion of the subsidy program is
borrowed in part from Laraki (1988). For more details on
the system of subsidies operated see Laraki (1988).

3
their products to retailers and bakeries at fixed prices but
were not reimbursed for their processing and transportation
costs. Similarly, retailers sold to consumers at a fixed-
price but are not reimbursed for their costs.‘

In 1988 the GOM reformed the subsidy system. It
declared that it would subsidize only 1 million tons of soft
wheat per year. In the same year, it discontinued the
subsidy on soft wheat luxury flour (farine de luxe). Since
then it has subsidized only soft wheat national flour
(farine nationals). The GOM estimates regional supply and
demand based on consumption surveys from the Statistics
Division, seed needs, loses, stocks, production, and milling
needs. It then determines monthly import needs and
merchants are authorized to imported this amount. Rebates
are still issued to the various actors in the marketing
chain, and a the consumption price is subsidized. The
subsidies are financed by a variable levy on imports and
general taxes.

Since there is excess demand for goods at the
subsidized price, the subsidies are a rent to whoever can
get the subsidized goods. It is therefore unclear whether
the beneficiaries of the subsidy program are actually the
poor, or others who have access to subsidized flour, quota

holders in surplus areas, and traders who are able to

 

‘ The subsidy programs for vegetable oil and sugar
operated in a similar manner and are therefore not described
in detail here.

4
purchase soft wheat from farmers at below support price
market prices and claim they purchased it at the support
price. This is a question for further research.
While removing the subsidies may ease the financial
crisis, there are some poor who will be hurt by their

removal. In his PhD dissertation entitled £99g_ggn§nmptign

1,. ..g 49:.gl‘s ', j,. . . y: y r 0,: j. .
Reform, Laraki investigates how the financial costs of food
subsidies can be decreased without affecting the standard of
living and nutritional status of the poor. He estimates the
welfare, nutritional and budgetary effects of various price
policies in the urban and rural areas. His findings
indicate that because the current subsidy program is
untargeted, many of the benefits accrue to high-income
households. Yet because the poor spend a high proportion of
their income on food, price increases associated with the
removal of the current consumer subsidy program may have
severe negative welfare and nutritional effects on poor
households. He concludes that eliminating the current
subsidies on soft wheat, vegetable oil and sugar powder and
replacing them with subsidies on barley and hard wheat would
not have detrimental welfare effects but would result in
important savings for the government.

While Laraki's analysis distinguished between urban and
rural households, it did not distinguish between households

in different regions of the country. Morocco is

5
geographically diverse, with a fertile agricultural plain in
the northeast and arid regions to the south and east, with
the respective areas consuming different commodities.
Estimating responses to price and income changes by grouping
all low-income households together may blur these
distinctions. A regional analysis of the effects of
removing the consumer price subsidies will be the goal of

this research.

Importance of study

Disaggregated consumption parameters are needed to
trace the effects of price and income policies on the food
intake of the poor. For designing policies that target the
poor, knowledge of these parameters is essential (Timmer, et
al., 1983, p. 61). Because the consumption patterns are
different in urban and rural areas, eliminating or reforming
the consumer subsidies will affect the urban population
differently than it will affect the rural population. Price
policies may also affect the poor in rural Marrakech
differently than they affect the poor in rural Sidi-Kacem.
Unintended distributional consequences could result if there
are unknown differences between regions. A regional
analysis of the effects of various price and income policies
will help delineate some of the potential distributional

Consequences e

6

Furthermore, where significant regional variations in

consumption patterns exist, it may be necessary to implement

more than one food-subsidy program. These could include

non-price policies such as food stamps, ration cards,

supplemental feeding programs, or direct transfers.

Hypotheses

This thesis examines four geographical regions (rural

Marrakech, rural Sidi-Kacem, urban Sidi-Kacem, and urban

Casablanca) of Morocco to test the following hypotheses:

The estimated consumption parameters vary not only
by income quartile and by urban and rural area,
but also by geographical location. These
differences stem from varying production patterns,
access to ports, and availability of complements
and substitutes which contribute to the
establishment of habit, tradition, and taste in
consumption. Different preferences will result in

different responses to income and price changes.

Barley is the only inferior good in Morocco.
Barley is thus the only commodity suitable for
self-targeting. Subsidies of other staples will
leak to higher-income households. There may

therefore be better, more cost-effective ways of

The
tested:

7
reaching the poor than price subsidies on soft-

wheat flour, vegetable oil, and sugar.

If increasing the welfare of the poor is the goal
of a consumer price subsidy, subsidizing the
consumption of soft wheat may be effective,
despite its high cost, if a large percentage of
the poor consume soft wheat and have access to the
subsidized soft wheat flour. Given the
speculative activities in the marketing chain of
soft wheat, reallocating quotas by regional needs
may be effective in assuring access by the poor to

subsidized soft wheat.

following methodological hypotheses will also be

Since market prices are not available, unit values
are used as proxies for prices. Deaton has
described several problems associated with using
unit values as proxies for prices (these are
described in chapter 3), the main problem being
that unit values reflect quality choices. The
hypothesis tested will be that the unit values do

reflect quality choices and may not be used as

8
proxies for prices without correction for the

quality choice.

2. For many of the goods studied, several households
reported that they did not consume any of a
particular good during a reference period. As is
explained in chapter 3, whether or not the
households that reported zero consumption are
included in the sample, the resulting coefficients
are biased and inconsistent because of the nonzero
mean. Including the inverse Mill's ratio as an
independent variable in the regression corrects
for the nonzero mean of the disturbance term. The
hypothesis tested here is that the coefficient on
the Mill's ratio is significant.

Methods

The Statistics Division of the Ministry of Planning of
Morocco undertook a national household budget survey of
expenditure patterns of 14,520 households in 1984/85. This
study will use a sub-sample of 1027 households from this
survey to estimate price and income elasticities in four
regions (rural Marrakech, rural Sidi-Kacem, urban Sidi-
Kacem, and urban Casablanca). The elasticities will then be
used to assess the distributional consequences of various

price policies by geographical area.

9

I will use a functional form similar to the AIDS to
estimate the parameters of the model. This functional form
specifies the budget share as the dependent variable. This
allows for the calculation of elasticities across income
classes and regions without having to include interaction
terms as independent variables in the model.

In estimating the model, I will use Heckman's two-stage
procedure for correcting for the bias introduced when a
large percentage of the sampled population reports zero-

consumption of a staple commodity.

Organisation of the Thesis

Chapter 2 presents a theoretical and conceptual
framework for analyzing the effect of food subsidies on the
poor. The chapter discusses how price and non-price
consumer subsidies affect consumption. Chapter 3 discusses
. general modeling issues: how one selects a functional form
for modeling consumption behavior, problems associated with
using unit values in cross-sectional data when prices are
not available, biases resulting from using truncated samples
when a large proportion of the sample reported zero
consumption of a particular commodity, and biases in
standard errors when one uses a complex sampling design.
The chapter also describes the AIDS. Chapter 4 describes
the areas studied and presents descriptive statistics on

these areas to try to understand who the poor are, where

10
they live, and what they consume. From this information,
some preliminary hypotheses of how consumers will respond to
price and income changes are formulated. Chapter 5 presents
the empirical results from the model and an analytic
discussion of the parameters. The final chapter presents
the conclusions and policy implications of the analysis. It

also suggests areas for further research.

CHAPTER 28 THEORETICAL AND CONCEPTUAL FRAMEIORX

Introduction

A consumption subsidy is defined as any policy that
keeps the consumption price of a commodity below the price
that would prevail in the absence of intervention. It may
take the form of a law that legislates the retail price of a
commodity, or it may be an overvalued exchange rate that
makes imports inexpensive, in which case the consumption of
imported goods is subsidized. Food stamps that serve as
cash in the purchase of food products may be considered a
form of subsidy, as may direct distribution of food to the
poor, ration shops, or supplementary feeding programs. The
effectiveness, cost, and effect on consumption patterns of a
subsidy program depend on the goal of the subsidy, the type
of subsidy, the particular conditions of a country, and the
elasticity of demand for the subsidized good and its
substitutes. This chapter describes several subsidy
programs and discusses the advantages and disadvantages of
each. It also discusses the effect of a subsidy on
consumption and presents a conceptual framework for

analysis.

Goals of Subsidy Programs
The goals of a subsidy program may be varied. Five

objectives of food subsidies listed by Valdés (1988) are 1)

11

12
reducing malnutrition among low-income groups: 2) achieving
food security (i.e., reducing instability in food
consumption, particularly for some targeted segment of the
population); 3) redistributing income in urban areas: 4)
lowering food prices, thus enabling wages to be low relative
to the price of industrial goods; and 5) reducing domestic
price inflation. Different forms of subsidy may be
successful at reaching different objectives and unsuccessful
at meeting others. For this reason, it is often difficult
to say, a priori, that one form of subsidy is "better" than
another. The questions of "at what cost?" and "for what
purpose?" must always be considered when evaluating the
effectiveness of a subsidy program. For example, programs
that focus exclusively on mothers and preschoolers (groups
most often identified as "at risk") may be effective at
reducing malnutrition, but only at a high cost (Kennedy,
1988).

Although it is difficult to prejudge the effectiveness
of a particular type of subsidy, the evidence from several
country studies shows that the untargeted or poorly targeted
food subsidy pregrams are not the most effective means of
transferring income from better-off to poor households,
improving food consumption among the poor, or improving
their nutritional status (Pinstrup-Andersen, 1988).

However, this evidence must be seen in light of other

objectives of food subsidies such as fulfilling social

13
contracts to provide low-cost food to the general

population.

Targeted versus Untargeted Subsidies

Subsidies may be targeted or untargeted. Targeted
subsidies are subsidies directed at groups of the population
that are identified as "at risk," or particularly vulnerable
to malnutrition. Untargeted subsidies, on the other hand,
provide low-cost staple grains or basic commodities-to
households at below-market prices, regardless of income,
nutritional status, or any other criterion.

Targeting is accomplished by basing program
participation on income level, nutritional status,
geographic location, age (school-aged children, for
example), vulnerability to malnutrition (pregnant and
lactating women, for example, are particularly vulnerable to
malnutrition), or some other criterion. Examples of
targeted programs include "fair-price shops" located in
specifically chosen locations, food stamps issued to
families that meet specified criteria, and supplementary
feeding programs that reach school-aged children or pregnant
and lactating women. A subsidy may "self-target" if the
subsidized food is one which only the target group consumes.
Self-targeting toward the poor can be accomplished by
subsidizing the consumption of goods for which the income

elasticity of demand is negative. That is, as income

14
increases, the consumption of the good decreases. One of
the difficulties of this type of subsidy is identifying a
good with a negative income elasticity. Few consumption
studies have found purely "inferior" goods. The consumption
of staples tends to either stabilize or increase with rising
incomes (Kumar and Alderman, 1988). Also, low income
elasticity foods are often nontradables, so that
implementation of the subsidy is difficult.

One of the advantages of targeted subsidies, to the
extent that the targeting is effective, is that they are
more likely than untargeted subsidies to reach the intended
group. The disadvantage is that targeting can be difficult
and costly to achieve. It is often difficult to identify
"at risk" groups, geographic targeting will undoubtedly miss
some people and include some unintended, and programs such
as supplementary feeding programs may lead to reallocation
. of food within the household so that the net consumption of
the intended beneficiary may not change at all. In areas
in which a large segment of the population is needy,
targeting may be less necessary than when only a small
segment of the population is "at risk," since the "at risk"
group will more likely be reached by a general subsidy.
Finally, the high administrative costs of targeting may
outweigh any savings achieved by only subsidizing a select

group of households or individuals.

15

An example of an untargeted food subsidy is an
unrestricted food price subsidy: unlimited amounts of the
subsidized goods are, in principle, available at below-
market prices to anyone who chooses and is able to buy them.
Untargeted subsidy programs reach a wider population, but
they do not guarantee that the "at risk" group is reached.
In fact, because the price is fixed below market price, the
quantity demanded will exceed the quantity supplied. If
imports are not used to make up the difference, the
artificially low price will mean that non-price mechanisms
may be used to allocate goods. Long lines to purchase the
subsidized goods may be necessary, and the resale of goods
common. While lower-income people may be more able than
wealthier to afford to stand in line to obtain a subsidized
good, wealthier people may be able to hire people to stand
in line to obtain the goods. If one is able to purchase
large quantities, wealthier people, who may have access to
more disposable income, may also be able to buy large
quantities and resell the goods at a higher price to less
fortunate consumers, or divert subsidized grain to animal
feed. Limiting the quantity one is able to purchase reduces
this possibility. Resale is also limited if the subsidized

good is bulky to transport or highly perishable.

16
Explicit vs Implicit Subsidies

Given the above definition of a subsidy, a subsidy can
be either explicit or implicit. Explicit subsidies entail a
fiscal outlay by the government. They are generally
financed out of the general government treasury. As such,
it is difficult to pinpoint the exact source of financing.
It is also difficult to speculate about how the money would
have been spent if not on subsidies. If explicit subsidies
are financed by general taxes, those who pay taxes may not
be the ones who benefit from them. If taxes are
progressive, the redistribution of income from the wealthier
to the poorer members of the population pay for the
subsidies.

Contrary to explicit subsidies, implicit subsidies
require no fiscal outlay by the government. This does not
mean, however, that they have zero economic cost. Implicit
subsidies include an overvalued exchange rate, low fixed
producer prices that make the price of food products
inexpensive to the final consumer, export taxes, and forced

food procurement.

Costs of Subsidies
Although explicit subsidies are easier to account for
than implicit subsidies, both have costs that are often not
taken into account when considering the cost of subsidies to

the economy. For example, food subsidies place downward

17
pressure on wages since subsidies provide what may be
considered "in-kind" payments (Pinstrup-Andersen, 1988).
When considering the income embodied in a particular
subsidy, one must account for the income foregone with the
lower wages.

If the costs of food subsidies increase the budget
deficit, and the budget deficit is financed by printing more
money, food subsidies may lead to inflation, which erodes
the purchasing power of the poor to purchase goods and
services other than those subsidized. The net effect may
actually be detrimental to the poor.

High costs do not imply an ineffective program. One
needs to assess the size of the needy population, the scope
of the program, and the effectiveness of the program at
reaching established objectives. Furthermore, there are no
inherently superior programs. What works in one country may
not work in another because different programs have
different requirements in terms of administration and
infrastructure. For example, as will be discussed below,
food stamp programs may be effective if the appropriate
infrastructure exists to distribute food, but may be
disastrous if retailers have no confidence in the government

that coupons will be redeemed.

18
Forms of Subsidies5
Unrestricted food price subsidies

In theory, an unrestricted food price subsidy offers an
unlimited quantity of a good at a subsidized, below market-
level price to anyone who chooses to and is able to purchase
it. Coverage is maximized with this system, but generally
at a high cost. In reality, however, unrestricted subsidies
may be restricted subsidies if the subsidized good is not
freely available. Those for whom the subsidy is intended
may not benefit if they cannot get access to the goods
because of non-price rationing. The larger the percentage
of the population needing assistance, the less "leakage"
will result from this type of program. Also, the more the
good chosen is one that is "self-targeted", that is, a good
that the poor eat more of and the wealthy eat less of
(income elasticity negative), the less leakage this type of
program will have.

Food stamps
In a food stamp program, coupons are issued to
households, on the basis of some criterion such as income.
The coupons may then be used as cash to purchase food from
participating retailers. In order for this type of program

to be successful, however, eligible households must be

 

5 This section borrows heavily from Rogers (1988).
Only some of the forms of subsidy programs are covered here.
For more extensive coverage, see Rogers (1988).

19
identifiable. That is, if income is to be the determining
criterion, good records of income levels must be available.
It is also important that institutions and infrastructure be
in place to support a food stamp program. For example, it
is important to have several retail outlets, as well as
banks where coupons can be redeemed. It is critical that
the population have confidence in the government because
participating stores and banks need to know that the stamps
will be redeemable. For these reasons, food stamp programs
have been successful mostly in developed countries that have
the required infrastructure.

Advantages of the food stamp system are that it makes
use of the existing distribution system, which cuts down on
administrative costs, decisions about supply are
decentralized rather than centralized, and price distortions
are minimized. Some disadvantages are that a food stamp
. system may fail to reach people in areas not currently
served by the existing distribution system, and a certain
amount of administration is needed to identify eligible

households.

Rationing
A rationed subsidy is one in which the quantity
available of the subsidized good is restricted to some per
capita limit. The advantages of such program are that it

limits resale of the subsidized good since each individual

20
can only buy a limited quantity. Even if the poor do
resell, however, it adds to their real income and the
subsidy may then be considered as an implicit income
transfer. Rationing may also reduce leakage to wealthier
households since it may not be worth their while or money to
stand in line for a limited amount of the good. Another
advantage is that the costs of the program are predictable,
which makes financial planning easier. The disadvantages of
rationing are that a complex administration is required to
administer coupon books and keep track of rations. A high

degree of literacy among sellers is also necessary.

How Subsidies Affect Consumption
How subsidies affect consumption depends on the type of
subsidy and on which goods are subsidized. If the subsidy
serves as an income transfer, as in the case of food stamps,
the result is an outward shift of the consumer's budget
constraint. This is illustrated in figure 2-1, which shows

the tradeoffs in consumption between food and non-food.

21

 

 

 

 

 

 

 

 

 

 

and
c u1.uz
A
\ F

E
o \ L .
3 D mums
Figure 2-1: Tradeoffs in consumption between food

and non-food

An increase in the income available to spend on food
shifts the original budget line, AB, to the new budget line,
CD. The consumer can be on a higher indifference curve by
consuming bundle F, where the indifference curve is tangent
to the new budget line, rather than E, the old consumption
bundle. Although this particular illustration indicates
that the consumer will consume more of both food and non-
food, this is not necessarily the case. If food is
"inferior," the consumption of food will decrease as income
increases. By connecting all the points of consumer choice
for all levels of incOme changes, one can construct an
income-expansion path. Engel's law says that the income-

expansion path is steeper for low income consumers and

22

relatively flatter for higher income consumers. This
reflects the larger proportion of income spent on food by
lower income consumers. The income elasticity captures the
relationship between changes in income and consequent
changes in consumption, at a particular income level. As
indicated by the income-expansion curve, the income
elasticity is likely to vary by income level. It is also
likely to vary by consumer. The income-expansion path is
also predicated on a particular set of relative prices. If
the relative prices change as well, the income-expansion
path may be invalidated. The point is that one income
elasticity for an entire country is unlikely to accurately
capture the diversity among consumers or the variation in
price scenarios in different regions.

If the subsidy establishes a below-market price for a
commodity, the effect is a rotation of the budget line.
This is shown in figure 2-2.

23

 

 

 

 

 

 

 

 

 

 

ahflud
an

O

a

\\ 0
Fl
\ u2
3:
U1
0
O .. A «mane
Figure 2-2: Tradeoffs in consumption between subsidized

food and other food

Without intervention the budget line is represented by
AB and the consumption bundle is represented by E, where the
indifference curve is tangent to the budget line. When the
price of the good to be subsidized falls, the budget line
rotates to AC. The consumption of the subsidized good
increases to 0F', and the consumption of other food goods
drops to 06. This change reflects both an income and a
substitution effect. Lower prices cause real purchasing
power to increase if there is no decrease in money income.
This increase in purchasing power will cause the consumption
of all goods to increase (except inferior goods). 'The
change in relative prices also causes a shift toward the

less expensive good (the pure substitution effect). The

24
total response to the change in the price of the good (the
price elasticity) is a combination of the income effect and
the pure substitution effect. Unless the good subsidized is
an inferior good that is consumed in large quantities, the
own-price elasticity will be negative.

Income elasticities can give some guidance as to how
consumption will change with a change in a subsidy policy.
If the income elasticity of the subsidized good is negative,
the subsidy to the poor will be higher in both absolute and
relative terms than the subsidy to the wealthy. If it is
positive but less than one, the benefit to the poor will be
relatively higher but absolutely smaller than the benefit to
the wealthy. If the income elasticity is greater than one,
both the absolute and relative benefit will be higher for
the wealthy (Pinstrup-Andersen and Alderman, 1988).

The choice of good to subsidize can also affect
. consumption patterns. If the good is one already consumed
by the poor, such as wheat, rice, or other staples, the
subsidy will serve as an income transfer and will increase
the consumption of all goods. If the subsidy is on a good
rarely consumed by the poor, such as milk, beef, oil, or
some other luxury good, marginal rather than inframarginal
consumption is altered and the subsidy may cause shifts away
from the consumption of less expensive nutrients sources

(Kumar and Alderman, 1988).

25
conceptual Framework

The discussion above of how subsidies affect
consumption is based on neoclassical demand theory, which
suggests that a consumer derives utility from the goods
he/she consumes. The way in which consumption yields
utility to the consumer gives rise to the utility function.
The rational consumer will choose a consumption bundle in
order to maximize his/her level of utility. Since consumers
are restricted by their budgets, maximization will be
subject to a budget constraint.

One way of obtaining a consumer's demand equations is
by maximizing a utility function subject to a budget
constraint. This will result in a complete system of demand
equations, each of which is a function of prices and income.
The estimation of these demand equations is one of the main
objectives of modeling consumer behavior.

If the demands are the result of the maximization of a
utility function subject to a budget constraint, consumer
theory tells us that the demand equations will have certain
properties. These are "adding-up," "homogeneity,"

"symmetry," and "negativity."

Adding Up
The "adding-up" criterion expresses the notion that
consumers spend all of their disposable income. This will

be true for the expenditure on the entire consumption bundle

26
as well as on sub-groups of consumption. It implies, for
example, that the value of the quantity of food purchased
sums to the total expenditure on food. Adding up is

expressed as:

qukax . (2'1)

where:
p, - the price of the kth good
q; - the quantity of the kth good

x - total expenditure

By rewriting 2p“; in terms of derivatives, one can look at
how changes in income may affect changes in quantity.
Differentiating (2-1), the budget constraint, with respect
to income and rewriting in terms of elasticities and budget

shares this may by written as:

zwkefl (2'2)

where:
tn = the share of the budget spent on good k

ex = the expenditure elasticity of demand for good k

This is the Engel aggregation condition. To see how this is

obtained, see Appendix 2-1. We also know that:

2“?“- (2‘3)

1

The Cournot aggregation condition is obtained by
differentiating (2-1) with respect to price. Written in

terms of budget shares and elasticities this yields:

2 wken+w1=0 (2'4)

where:
eu_- the cross-price elasticity between goods k and i

vn - the budget share spent on good i

Homogeneity
If one multiplies both income and all prices by a
constant, 8, the linear budget constraint that the consumer
faces remains unchanged. If the budget constraint remains
unchanged then the corresponding demand for each good also
remains unchanged. Demand is therefore also homogeneous of

degree 0 in prices and income. This may be expressed by:

q1(OX.6p) =q1(X.p) (2-5)

28
where:

qi - the demand for good i

This property implies that only relative prices matter in
consumption decisions. This is expressed in terms of

elasticities as:

zen-rekeo (2-6)

where:
en_- the cross-price elasticity between goods k and 1

ex - the income elasticity of demand for good k

This restriction says that the sum of the own- and cross-
price elasticities plus the income elasticity for any

commodity is zero.

Symmetry
Symmetry implies that the cross-price derivatives of
the Hicksian demands are symmetric:

5133017?) . 511101,?)
6P1 5P3

 

(2‘7)

29
where:
1n(u,p) - the Hicksian demand for good i as a function of
utility and prices, i.e., the demand that results from
the minimizing the expenditure needed to reach a given

level of utility
u - consumer utility

This result comes from the relationship between the cost
function and demands. The Hicksian demands result from
differentiating the cost function twiCe. Young's theorem

tells us it will also be symmetric.

Weak Separability
The estimation of an entire system of demand equations

can easily become intractable. For n commodities, the
estimation procedure would require calculating n income
elasticities and n2 own- and cross-price elasticities, a
total of n(n+1) parameters. The classical restrictions
. described above reduce this number of parameters to (nz-+11-

2)/2. Even so, if n is large, degrees of freedom problems
may result. The assumption of weak separability of
preferences is one way of dealing with this situation. Weak
separability of preferences suggests that one can divide the
set of n commodities into mutually exclusive subsets without
losing any generality. This assumption implies that the
marginal rate of substitution between two commodities within
a subset be independent of the quantity consumed of a good

in another subset. This allows us to model the consumption

30
of a set of commodities independently of the commodities
consumed in other groups. In the discussion above about how
subsidies affect consumption, for example, weak separability
allows one to talk about how a change in the price of a food
commodity will change the consumption of all foods, without
reference to how the consumption of non-food goods will
change. If one assumes weak separability between food and
non-food, the change in the price of the food good should
not affect the consumption of the non-food goods. Assuming
weak separability reduces the number of parameters to be
estimated. As Deaton and Muellbauer (1980b, p. 125) point
out, by assuming weak separability "detailed commodity
expenditure can be related to group outlay and prices alone.
This has obvious econometric advantages since it is possible
to find an explanation for behavior through a much smaller

number of variables."

Conclusion

Income and price elasticities can aid in analyzing the
anticipated effects of different subsidy programs on the
consumption patterns of different groups. Goods with high
positive income elasticities, for example, are not as
suitable as goods with low income elasticities for reaching
the poor since upper-income consumers consume more of the
good and therefore receive more of the subsidy than low-

income consumers.

31
The estimation of income and price elasticities is
based on neoclassical demand theory, which assumes that
consumers derive utility from the goods they consume and
that consumers maximize their utility subject to a given

budget constraint. Their estimation will be the goal of the
rest of this thesis. '

CHAPTER 3: METHODS

Introduction

Designing effective food policy requires a
comprehensive understanding of how consumers will be
affected by a prospective policy. Knowledge of how
consumers respond to income and price changes is therefore
imperative. Yet estimating income and price elasticities
via econometric analysis of household budget data is a
highly subjective endeavor. The choice of functional form
for modeling consumer behavior can dramatically alter the

estimates.

Demand Systems vs. Single Demand Equations

There are basically two ways of modeling the demand for
commodities: the complete systems approach and the single
equation approach. The complete systems approach entails
estimating the set of demand equations that result from
allocating total expenditure among a group of commodities.
Under the assumption of separability, this implies that the
consumption of all goods consumed need not be modeled, but
rather that the consumption of a complete subset of goods be
modeled together. Whether modeling a subset or the entire
consumption basket of a consumer, the sum of expenditures on
the goods in that category must sum to total expenditure on

that category. The complete systems approach allows for the

32

33
interdependence of demand for various commodities in the
group.

In contrast, in the case of the single-equation
estimation where the equations are not part of a system of
equations, one models the consumption of one commodity at a
time without reference to the interrelationships between the
goods. The single-equation approach is not based on the
maximization of utility constrained by the budget
constraint. Rather, an ad hoc specification of the'demand
functions is used to render the estimation problem
manageable. The double logarithmic functional form is a
commonly used model of this type. In modeling the
consumption of a single commodity, this approach may
suffice. Yet it still must be recognized that the link
between theory and empirical work is quite weak. If one is
looking only at the demand for a single commodity, neither
adding-up nor symmetry may be validated since both require
cross-equation restrictions. Homogeneity is the only
testable restriction.‘ In examining a whole group of
commodities using the double logarithmic specification, it
is impossible for adding up to be satisfied.7 Most

researchers have been acutely aware of this problem but have

 

‘ Adding-up, homogeneity, and symmetry are described in
detail in Chapter 2.

7 To see how adding-up is not satisfied in the double-
logarithmic specification, see Deaton and Muellbauer (1980b,
p 17-18).

34
traditionally chosen to ignore it because, as Deaton and
Muellbauer (1980b) point out, when modeling a limited number
of commodities, the specification of the functional form of
the remaining commodities remains unstated.

The choice between estimating an entire system of
demand equations and estimating single demand equations
involves several tradeoffs. Because of the cross-equation
restrictions embodied in the symmetry condition, the
estimation of a complete system will require non-linear
estimation. This may render this approach more costly than
the single equation estimation technique. Furthermore, the
statistical problems of multicollinearity and nonspherical
error structure for complete systems are more burdensome
than under the single equation approach (Raunikar and Huang,
1987, p.24). The advantages of the systems approach are
that the postulates of consumer theory are directly testable
.and the generation of empirical results is consistent with
the economic theory of consumer behavior. Yet the
theoretical superiority of the systems approach may be
outweighed by the high estimation costs.

When employing a complete demand system one must assume
that the functional form is the same for all goods, while
single equations allow different functional form for

different commodities.
There are two ways of formulating demand systems: 1)

specify a functional form for the direct or indirect utility

35
function or for the cost function from which the demand
equations can be derived and 2) specify the functional form
of the demand equations directly and impose restrictions on
them to ensure that they conform to the theory. As
mentioned above, all equations in the system are of the same
functional form since cross-equation restrictions must be
imposed. The problems with the first approach are that the
number of known and well-behaved utility functions is very
limited, and the derivation of demand equations is not A
always possible. There may be doubts about the suitability
of adopting a particular specification, and in many cases,
the derived demand relations turn out to be nonlinear in
their parameters. For these reasons, economists have
sometimes preferred to work with an arbitrary but manageable
functional form for demand, imposing constraints which
insure their theoretical plausibility (Capps and Havlicek,
1987).

Complete demand systems include the translog system,
the Rotterdam system, the addilog system, the constant
elasticity of demand system (CEDS), the linear expenditure
system (LES), and the AIDS. The AIDS is discussed below;

the others are discussed in Appendix 3-1.

Flexible Functional Forms
One of the problems associated with the single-equation

approach such as the double logarithmic equation is that the

36
choice of the function is ad hoc. Flexible functional forms
circumvent this problem by approximating the "direct utility
function, the indirect utility function or the cost function
by some specific functional form that has enough parameters
to be regarded as a reasonable approximation to whatever the
true, unknown function may be" (Deaton and Muellbauer,
1980b, p 74). Thus the functional form specified is an
approximation to any true function. This approach to
consumption modeling offers a more general framework of
analysis, hence the name ”flexible" functional form. This
is the basis for the translog model, the generalized
Leontief model, and the AIDS.

A second-order Taylor approximation to a true utility
function should give rise to demand functions satisfying all
the theoretical requirements. If the empirically estimated
demand functions do not satisfy the theoretical
requirements, one must question whether the demand theory
accurately represents behavior. But the empirical
estimation is based on a particular shape of the utility
function that may or may not be the correct utility
function. Therefore, one does not know if the theory is
wrong or if the approximation is inaccurate (Deaton and
Muellbauer, 1980b).

One of the advantages of the flexible functional form

offered by the AIDS is that its generality allows one to

37
easily impose and test the homogeneity and symmetry

conditions. The adding up is satisfied automatically.

Th. AIDS

Duality theory tells us that the consumer who maximizes
utility subject to a budget constraint is simultaneously
minimizing the cost of attaining that utility level. It is
therefore true that if we obtain a set of demand equations
from a particular cost function or indirect utility
function, these demands correspond to a certain utility
function. This will be true as long as the cost function
exhibits the following properties: it must be homogeneous of
degree one in prices, increasing in utility, nondecreasing
in prices, and concave in prices.

The AIDS starts from this premise and specifies a
particular functional form for a cost function. It then
_ employs the duality theory to arrive at a complete set of
demand equations. The AIDS is based upon the Working and
Laser model for Engel curves, which proposes that the value
shares are a function of the logarithm of total

expenditures. This may be expressed as:

WQ=ai+BilOQX' (3-1)

38

where:

w; is a budget share for good i
x is total expenditure

a1 and B1 are parameters

A cost function that gives rise to this set of Engel curves

is:

logc(u.p) = (1-u) 1096! (p) +ulogb(p) (3-2)

where :
c.(u,p) is the cost function;
jp is the vector of prices;

u. is a utility level;
at(p) and b(p) are linear homogeneous concave functions.

'The specific functional forms of log a(p) and log b(p) were
chosen by Deaton and Muellbauer such that the resulting cost
function has enough parameters to be a an approximation to

any arbitrary cost function. The functions they have chosen

are:
loga (p) =a°+2 aklogpp-g—z: z: yzjlogpklogpj (3‘3)
and
logb(p) =1oga (p) +BH pf.” (3‘4)
where:

a, 3, and y are parameters

39

The AIDS cost function may then be written as:

. (3'5)
logc ( u.p) my; aklogpk+%; z: yijlogpklogpj‘I-upon pf"

Equation 3-5 "has sufficient parameters that at a selected
point the first- and second-order derivatives with respect
to p and u can be set equal to those of an arbitrary cost
function" (Johnson, et al., 1986, p.6). Equation 3-5 is a
second-order Taylor expansion to any cost function. Hence
the flexible-form attribute of the AIDS. From this cost
function, the Hicksian demand functions can be obtained by
applying Shepherd's Lemma (see Appendix 3-2). The Hicksian
demands, expressed in terms of the budget share (w1) will be

of the form:

 

alO C(UI )- _. k -

where:

=_p1qi (3-7)

40

and

Y11=€%(Yij+7;i) (3-8)

At equilibrium, c(u,p) is equal to total expenditure, x.
Thus, equation 3-5 can be solved for u in terms of p and x.
This expression for u is then substituted into equation 3-6

to yield the Marshallian demand functions:

w1=a1+£ yulogpj-rflilogi-E} (3-9)

where P is the price index defined by:

logP=a°+¥ aklogpk+-%; z: ykjlogpklogpj (3'10)

Restrictions
d' -u
As is true for all complete demand systems, the
"adding-up" criterion is satisfied by construction in the

AIDS. The adding up restriction requires:

11 R

ganja gyifo :91“) (3-11)

41
This implies that only n-l income coefficients need to be
estimated since the the estimates of the nth may be derived
from the above condition.
We know from the section on adding-up in Chapter 2 that

adding-up will be satisfied as long as

2; w1=1 (3‘12)

Even if the system is estimated as a set of single equations
using OLS, this will be true since the observed expenditure
on individual goods, summed over all goods in the category,

will equal total observed expenditure on that category.

W
Homogeneity, on the other hand, is not satisfied

automatically but may be "tested" by imposing the

restriction

;71j=0 (3‘13)

and then using F ratios for each equation to test whether or
not the results are significantly different from the
unrestricted case. Once the assumption of homogeneity has
been made, it is only necessary to estimate n-l price

parameters. The nth may than be calculated using 3-13.

42

Symmetrx
Symmetry in the AIDS requires that

731:711 (3-14)

Since this requires cross equation restrictions, it may only

be tested using non-linear methods.

Limitations
W

The formulae for calculating the income and price

elasticities from the AIDS are given in Appendix 3-3. The
I31

income elasticities are given by (Eu) . If B is positive,
the income elasticity will be greater than unity (luxury
good) for all income classes; if the B is negative, the

_ income elasticity will be less than unity (necessity or
inferior good) for all income classes. A good cannot be a
luxury for one income class and a necessity or inferior good
for another. .

Whether or not this is a serious limitation depends on
whether you would expect the goods in question to be
necessities for one group and luxuries for another. In
developing countries where a some basic foods are luxuries
for some income groups but necessities for others, this may

be a serious limitation.

43

Similarly, whether the good is a luxury, necessity, or
inferior good, demand always becomes more income inelastic
as income rises. To see that this is true, assume the good
is a luxury item, which implies that the B coefficient is
positive and that the budget share is increasing with
incomes. The income elasticity will always be greater than
unity but will diminish as the share increases. For a
necessity or inferior good, the B coefficient will be
negative but the budget share will decline as income rises.
Thus the income elasticity will become more inelastic with
higher income levels.

The only means of rectifying this problem is by using a
squared income term. The transcendental logarithmic demand
function (TL) as described by Swamy and Binswanger (1983)
uses the same functional form as the AIDS except that the
squared income term is included. This is justified by the
authors by dropping the assumption of perfect aggregation
over consumers.a If aggregation is an issue, it is here
that judgement is necessary in making the tradeoff between
maintaining perfect aggregation or increasing the
flexibility of the functional form. In this study,
aggregation is not necessary since the data are at the

household level. Adding a squared income term in this case

 

3 Perfect aggregation refers to "the conditions under

which it is possible to treat aggregate consumer behavior as
if it were the outcome of the decisions of a single
maximizing consumer" (Deaton and Muellbauer, 1980b, p. 148).
This is an issue when using aggregate market data.

44

does not detract from the theoretical soundness of the
model.

Deaton and Muellbauer (1980a, p.315) point out that
"the flexible functional form property of the AIDS cost
function implies that the demand functions derived from it
are first-order approximations to any set of demand -
functions derived from utility-maximizing behavior." While
the cost function does represent an approximation to any
cost function, it is itself a particular cost function, thus

its generality is not as global as was originally argued.

Practical Aspects

W

When using the AIDS to model consumer behavior, the
first decision to be made is the group of commodities for
which the estimates will be made. The separability
assumptions allow one to estimate parameters for a separate
group (such as food) based on system estimations. The key
here is to define an expenditure category narrowly enough so
that the number of products is manageable but broad enough

so that weak separability holds.9

Stone's Index
The price index (3-10) of the AIDS is one of the points

which makes the model fairly intractable since it makes the

 

9 Weak separability is defined in Chapter 2.

45
estimation non-linear. In order to avoid the non-linearity,
most researchers use the "Stone's index" (P’) as suggested

by Deaton and Muellbauer:

logP‘=2 wklogpk (3'15)

The Stone's index is easy to calculate and may be an
adequate approximation to the true index as long as prices
are closely collinear. Since using the Stone's index makes
the estimation procedure linear, it is highly tempting to
adopt this strategy. It must be noted, however, that
estimation problems arise because the use of the Stone's
index means that wg is on both sides of the equation. When
the dependent variable (wk) is correlated with the error
term, the classic assumptions of linear regression are
violated. Most researchers chose not to address this issue,
yet it is important to recognize that the parameters
estimated using the Stone's index will be biased. If one is
doing non-linear estimation anyway, it is advisable to use

the original index (3-10).

Software Requirements
If one does not want to test the symmetry condition and
if the Stone's index is used as a proxy for the true price
index, the estimation of the AIDS becomes linear. The system

may be estimated equation by equation using OLS. Thus one

46
can use any software package which offers OLS, subject to
the data handling abilities of the software.

If, on the other hand, one does want to test the
symmetry condition, or if the true price index is used, then
non-linear estimation is necessary and a programming
language such as GAUSS, FORTRAN, C, or any software which
can handle non-linear estimation must be used.

A functional form that is similar to the linear AIDS
will be used to estimate the system of demand equations for
the Morocco data. The modifications that will be made to

the AIDS are discussed in Chapter 5.

Other Modeling Issues

In addition to questions of which functional form to
use to model consumer behavior, there are some anomalies in
modeling that must be addressed. They include, but are not
limited to, using unit values as proxies for price when
using cross-sectional data, reported non-consumption of
goods by households, the effect of sampling design on the
reliability of standard error estimates, and weighting
regressions when a complex survey design has been used.

These issues are discussed in this section.

Unit Values
In demand analysis using cross-sectional household

budget surveys, it is traditionally assumed that all

47
households face the same prices. (Cox and Wohlgenant cite
Allen and Bowley, Prais and Houthakker, George and King).
This assumption is based on the notion that prices remain
unchanged during a short period of time. For this reason,
most cross-sectional surveys do not even collect data on
prices. It is likely, however, that there is price
variation in cross-sectional data due to spatial and
seasonal fluctuations in price. The former results if the
sample is distributed over a large geographical region that
exhibits variable transportation and other distributional
costs; the latter results if cross-sectional surveys are
conducted over a period of a year or more such that
capturing seasonal effects is possible.

The result of the assumed lack of price variability has
been an adherence to estimating Engel curves. However,
Polinsky (as cited in Cox and Wohlgenant) suggests that if

there is true price variation, then the failure to capture
price effects could lead to misleading and biased results.

Since most cross-sectional data sets do not contain
information on prices, but do contain data on expenditures
and quantities, many researchers use a unit value
(expenditure divided by quantity) as a proxy for price
(Timmer and Alderman, 1979). As Deaton points out "armed
with expenditures and with quantities, the temptation to

divide one by the other is irresistible." (Deaton, 1988b,

p.419).

48

Deaton has outlined several problems associated with

using unit values as proxies for price:

1)

2)

Consumers choose the quality of their purchases, and
unit values reflect this choice.

The awkward transition from theory to application
gives rise to this problem. Neo-classical consumer
demand theory describes the demand for a homogeneous
good with a single price. In application,
non-homogeneous goods are aggregated to make the
estimation problem tractable. For example, meat is not
a homogeneous product with a single price; there are
several different kinds of meat, and different
qualities and cuts within the sub-categories of meat.
Unit values will reflect the quality chosen. Since
consumers chose the quality of the goods they consume,
”the regression of quantity on unit value is a
regression of one choice variable on another, and runs
all the usual risks of possible lack of identification,
simultaneity bias, and interpretational ambiguity
(Deaton, 1988b, p 420)."

Quality choice may itself reflect the influence of
prices, as consumers respond to price changes by
altering both quantity and quality.

If market prices rise, the unit values will
presumably rise by less because consumers will move
away from the more expensive qualities and types toward

the less expensive. Consumers could thus adjust

consumption toward lower-priced, lower-quality goods,

49
keep total expenditure the same, and keep the total
quantity consumed the same. In this case, unit values
would not change although market prices had risen. The
consequence is that if unit values are used instead of
prices when regressing quantity on price, a quantity
difference will be associated with a smaller change in
unit value than the change in price had prices been
used. The elasticity will therefore be larger for the
unit value than for the price: using unit values will
exaggerate the price elasticity.

3) Measured unit values are contaminated by errors of
measurement in expenditures and in quantities and are
likely to be spuriously negatively correlated with
measured quantities.

Unit values are calculated by dividing expenditure
by quantities. If the quantities are measured with
positive (negative) error, that is, the measured
quantity is higher (lower) than the true quantity, the
unit value will be too low (high). There will be a

spurious correlation between quantity and unit value.

From the viewpoint of demand analysis, spatial and
temporal unit value variation is desirable because it can
help to capture price effects (Cox and Wohlgenant, p.908).
However, variation in unit values due to the quality effects
just described is more problematic. "A preliminary step in
the analysis of a cross-sectional data set from which prices

are derived by dividing observed expenditures by quantities

50
is to determine whether prices are correlated with income.
If they are, one needs to determine an indicator price, by
region and season, that reflects a constant quality"
(Alderman, 1986, p.59). In other words, one needs to
determine if the quality elasticity (the elasticity of unit
value with respect to total household expenditure) is zero.
If it is, one can assume that the unit value index moves
proportionately with the market price. If the quality
elasticity is positive, unit values will move less than.
proportionally with prices, the shortfall depending on the
size of the quality elasticity as well as on the overall
price elasticity. If the quality elasticity is positive, an
"indicator" price needs to be calculated which "purges" the
unit value series of the quality effects.“

In order to test for the quality elasticity, Deaton
assumes that all households in a cluster, or a unit of
observation, face the same "true" prices because they are
surveyed at the same time and are in a small geographical
area (usually a village). Thus, any differences in unit
values observed within a cluster must be due to the quality
differences described above. Regressing unit value on total
expenditure and household size indicates the extent of the

quality effect. One would expect higher-income households

 

m The method for "purging" the unit value series of
the quality effects is described in Deaton (1988a) as well
as in Laraki (1988). Since the quality elasticity was found
to be zero for this case, the method is not described here.

51
to buy higher quality goods, keeping household size
constant.
The Groupe de Travail pour l'Estimation des
Elasticities de Demande des Produits Agro-Alimentaire (1990)

tests for the quality elasticity by estimating:

logV=cO+cD,,,log(D/N) +cvmlog (U/N) +cLL (3-16)

where:

V is unit value

D/N is per capita income

U/N is adult equivalents divided by number of household
memebers11

L is a vector of dummy variables representing region
co, cum, cum, and cL are parameters

The results are presented in Appendix 3-4. They
suggest that the quality elasticity, cmm,is significantly
different from zero in only 6 out of 13 cases.

Adding a dummy variable for the different clusters
would indicate the importance of income on unit values
within each cluster. Regressing deviations from cluster
means for the above equation yields the same result as
including a dummy variable for every cluster (Deaton, 1988a,
p.6). The results of the regression presented in table 3-1

indicate no significant within-cluster effects of income on

 

n "Adult equivalent" is defined on page 97.

52
unit values. It should be noted that this analysis is
limited to the extent that income is homogeneous within a
cluster, making it difficult to assess the quality effect by

looking at deviations from cluster means.

Table 3-1: Regression Coefficients from Regressing Unit

Value on Total Exggnditure and Household Size

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dependent Variable Independent Variables

Unit Value Log of Per Capita Log of Per Capita
Expenditure Adult Equivalent

0.022 0.097

Soft Wheat 0.024 0.147
0.022 0.091

Barley 0.018 0.076
0.021 0.081

Mutton 0.006 0.004
0.006 0.024

Beef 0.005 0.013
0.004 0.014

0.012 0.045

Table Oil -0.006 0.016
0.003 0.014

Olive Oil 0.018 -0.029
0.007 0.027

Butter 0.019 -0.067
0.010 0.041

Tea 0.001 0.055
0.010 0.040

0.014 0.055

Sugar 0.007 0.058
0.005 0.022

Fresh Milk 0.002 0.009
0.002 0.008

 

Stendard errors are below coefficients

 

53

Based on these results and those of the Groupe de
Travail, one can conclude that the substantial variation
exhibited in the unit value data reflect actual price
variation: the unit values do not seem to be contaminated by
the quality effects described here. The analysis will
therefore use unit values as proxies for price. This thesis
does not, however, test for the measurement error described

above.

Non-Consumption

One problem that arises in cross-sectional surveys is
that of the "limited dependent variable." Limited dependent
variables occur when the data on a dependent variable are
limited to a certain range. There are two types of samples
with limited dependent variables: truncated samples and
censored samples. A sample is truncated when the
independent variables are not available for dependent
variables outside the given range. Common examples of
truncated samples include negative income tax experiments in
which information on only a sample of households with income
below some threshold is provided, or experiments to
determine the earnings of low achievers for which
information on only individuals with intelligence scores
below a specified level are sampled. Censored samples, on
the other hand, are samples for which the dependent variable

is in a certain range but the sample also include

54
observations on the independent variables when the dependent
variable is outside the range.

The classic treatment of the censored regression model
is the tobit model, developed by Tobin (1958) using the
example of the consumption of durable goods. Most .
households will not purchase a car, or other durable goods,
during a reference period although this does not necessarily

mean they would never purchase a car. For non-durable
goods, if the goods are highly disaggregated, it is likely
that some of the households will not have consumed the good
during the reference period. The problem with the resulting
censored sample is explained concisely by Kmenta using the
example of durable goods: "since expenditure on durable
goods is not continuous, purchases are not made until the
'desire' to buy the good exceeds a certain level. However,
we cannot observe desires, only expenditures, and those will
be nonzero only if the good is actually purchased.
'Negative' expenditures, corresponding to various levels of
desire below the threshold level, cannot be observed, and
all households with no purchases are recorded as showing
zero expenditure" (Kmenta, 1986, p.560).

The durable goods example can be extended to include
food items which may not have been purchased during the
reference period.

The problem of using a censored sample in regression

analysis is explained as follows:

55

Yf-a+pag+e§ (i=1,2,...,n) (3'17)

be a regression for which all basic assumptions are
satisfied. Let the dependent variable, Y‘, represent the
consumption of some commodity, and x represent income. For
the households that consumed positive quantities of the
commodity, Y' is the actual quantity. For households that
did not consume any of the commodity, Y” represents an index
of the "desire" to purchase the commodity.

For households that did not consume the good, Y' is not
observed and is recorded as zero. Thus, instead of
observing‘Yf, we actually observe Y3,*which is defined as

Y1=Yf if v1*>o,

-o if sto.

Equation 3-17 then becomes

Y3=a+Bzg+ei (3-18)

where Y1 is truncated at zero and e is truncated at -(a +
BX). This means that the lower tails of Y3 and 51 are both
cut off and the probabilities are piled up at the cut-off
point. The implication of this is that the mean of Y; is
different from that of Yf'and that of £1 is different from
sf, which is zero. This is true whether the points for

which Yg=0 are or are not included in the sample (i.e., this

56
is true whether the sample is truncated or censored).
Limiting the range of the values of the dependent variable
thus leads to a nonzero mean of the disturbance and to

biased and inconsistent least squares estimators.

HQH_LQ_QQII2QI.IQI.§.§£D§QI§Q.§§EDL§
Tobin originally proposed maximum likelihood technique

to deal with the limited dependent variable problem. This
technique is called "Tobin's probit" or "tobit".

Heckman proposed another way of dealing with the
problem of censored data. Suppose that of all n
observations there are m (m<n) observations for which Yf >

0. The regression equation for these observations is

Yg=a+ng+ei (i=1,2,...nu (3-19)

where Y; ands1 are truncated normal variables. Since the
regression is only for those observations for which Yf > 0,

the conditional expectation of Y1 given Y,‘ > 0 is

(3-20)
E(Y1|Y;>O)=a+BX1+E(£1|Y;>O)=a+BX1+E(61|e;>-a-BX1)

(i = 1,2...,m)

Given that ef%N(0,oz), the mean of the corresponding

truncated variable, 61, is

57

 

 

.E(eAe§>-a-BAQ)=011 (3'21)
where
f( ¢+pX1)
Al. (3-22)
“Eff—X!)

and f(*) represents the density and F(*) the cumulative
distribution function of a standard normal variable. 11 is
called the inverse of the Mill's ratio. To allow for the

non-zero mean of £1, the equation

Y3=a+ng+oli+e1 (i=1,2,...,nn (3-23)

is run on the m observations for which Yf > 0. The error
term in equation 3-23 then has a mean of zero.

If 11 were observable, this equation would have a
disturbance with zero mean and the least squares
coefficients would be unbiased. Since 1.1 is not observed,
it must be estimated. The estimation procedure is as
follows: probit analysis is run using a dichotomous variable
for consumption and non-consumption as the dependent
variable, and the explanatory variables of the model as the
independent variables. The results are used to construct (a

+ pxgr/o. 11 can then be calculated using equation 3-22.

58
Once A; is estimated, equation 3-23 can be estimated using
the least squares method using the m observations for which
Yi>0. The estimators of a and B are consistent and

asymptotically normal.

W

Although the problem of "non-consumption" is less of a
problem the higher the percentage of surveyed units having
positive consumption, the case of fairly disaggregated food
groups, such as those used here, warrant the correction for
censored sample bias. Table 3-2 presents the percentage of
the sampled population that reported positive consumption of

the various commodities.

59

Table 3-2: Percentage of Sampled Population Reporting Positive
C22===ption of Products

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Product Percentafl
Mutton r 40
Beef 54
Chicken 46
Table Oil 89
Olive Oil 43
Butter 46
Coffee 73
Tea 94
_§ugar 98

Fresh Milk 30 I

Soft Wheat 86 I
Hard Wheat 30
Berle 39

 

 

The approach taken here to correct for the censored
sample bias will follow Heckman (1976, 1979), and to correct
for the resulting heteroskedasticity. The following steps
will be taken:

1. Probit analysis will be run using as a dependent
variable the dummy variable for consumption/non-
consumption. The independent variables will be
those used in the model of demand.

2. The predicted values of the dependent variable
will be calculated and used in equation 3-22 to
estimate the inverse of the Mill's ratio.

3. The inverse of the Mill's ratio will be used in
the demand equations as an independent variable to
correct for the missing variable bias.

60
Sampling Design

Standard errors calculated using standard statistical
packages assume simple random sampling (SRS) in which every
element of the population has an equal chance of being
selected for inclusion in the sample. When simple random
sampling is not used to sample a population, the formula
used for the standard errors from SRS is no longer valid.
For the consumption survey used here, the sampling methods
were a combination of clustering and stratification.12
Stratification, and in particular proportionate
stratification as employed in the data set used here, leads
to more precision in the survey estimates than would SRS.
Clustering, however, decreases the precision relative to
SRS. Although the standard errors reported here may over-
or under-state the true standard errors because of the
sampling technique, they are probably close to accurate
since the decrease in precision resulting from clustering is
offset by the over precision that results from
stratification. Therefore, no correction for sampling is

made to the standard errors.

 

u Clustering and stratification and their implications
for the estimation of standard errors are discussed in
Appendix 3-5. Appendix 3-6 gives a detailed discussion of
the sampling procedures used to sample the Moroccan
population for the consumption survey.

61
Weighting

Because of the complex survey design involved in
sampling the Moroccan population, households are not given-
equal weighting. Instead, each household is assigned an
extrapolation coefficient, or weight, which may be used to
extrapolate the sample to the population level. In
calculating means, it is imperative that each household be
weighted by its coefficient to provide a consistent estimate
of the population mean. There is disagreement, however, on
whether or not to use the weights when doing regression
analysis. Standard texts either ignore the issue or suggest
that no weighting be used.

Buse and Chavas (1990) argue that the decision of
whether or not to use weights in the regression depends on
the explicit assumptions that undergird the economic model.
They compare the assumptions of three models: a simple
linear model, weighted least squares, and a weighted average
coefficient where individual coefficients are calculated for
each stratum and a weighted average is computed for the
overall coefficient. Weighted least squares weights the
regression by a diagonal matrix whose diagonal elements are
the sampling weights. This would be an appropriate model if
the heteroskedasticity of the error term is characterized by
the specific function obtained from the sampling weights,
that is, if the standard deviation of the error term is

proportional to the inverse of the weighting matrix. If

62
not, then the coefficients obtained are not minimum variance
estimators. In general, there is no reason to believe that
the standard deviation of the error term will be
proportional to the inverse of the weighting matrix, thus
using weights in the regression is inappropriate. This
argument justifies the unweighted regression analysis used

here.

Conclusion

The application of consumer theory to applied demand
analysis still poses several obstacles for researchers. The
unobservability of a consumer's utility function, and the
restrictions on demand equations that theory requires are
just a few of the problems. Thus the bridge between
consumer theory and application is still quite weak. As
Deaton and Muellbauer (1980b, p 80) believe "It is our view
_ that a careful transition from theory to application is a
fundamental prerequisite of sound econometric work, and we
find much current empirical work profoundly unsatisfying
because of the lack of such a basis."

The AIDS is a model of consumer behavior which
addresses some of these issues. It "gives an arbitrary
first-order approximation to any demand system; it satisfies
the axioms of choice exactly; it aggregates perfectly over
consumers without invoking parallel linear Engel curves; it

has a functional form which is consistent with known

63
household-budget data: it is simple to estimate, largely
avoiding the need for non-linear estimation: and it can be
used to test the restrictions of homogeneity and symmetry
through linear restrictions on fixed parameters" (Deaton and
Muellbauer, 1980a, p 312). Yet several limitations of the
model should be highlighted. Although the model gives a
first-order approximation to any demand system, the
functional form for the cost function used to derive the
demands is a particular cost function, the choice of which
remains ad hes: a good cannot be a luxury for one income
class and a necessity or inferior good for another; and
demand always becomes more inelastic as income rises.

The approach in this analysis will be to estimate
demand using a functional form similar to the AIDS. Each
equation will be estimated using OLS. This approach has the
advantage over the double log specification that regional
analysis is possible without introducing interaction terms.
The number of coefficients to be estimtated is thus smaller
with this approach than with the double log. The income
variable will be proxied by total expenditure, rather than
total expenditure on the foods in the system, to make the
estimation of elasticities more straightforward. Since unit

values for non-food goods are not available, it is

64
impossible to calculate the Stone's index. As such, the
income term will not be deflated by a price index.13

Several other issues arise when modeling consumption-

using cross-sectional data. These include reported
zero-consumption, using unit values as proxies for prices,
using survey weights in regression analysis, and sampling
design. The following procedures will be used to deal with
these issues:

1) Heckman's two-staged procedure will be used to
correct for the censoring bias of reported zero-
consumption:

2) because the quality elasticity was found in most
cases to be negligible, unit values will be used
as proxies for prices:

3) survey weights will not be used in the regression
because of the lack of evidence that the
heteroskedasticity of the error term is related to
the sampling weights: and

4) the design effects is assumed to be negligible

since the sampling procedure used both
stratification and clustering.

 

” It should be noted that when the expenditure term is
not deflated by a price index, the resulting expenditure
elasticity will be contaminated by price effects that have
not been eliminated through deflating.

CHAPTER 4: DESCRIPTIVE STATISTICS

Description of Study Areas
The 1984/85 consumption survey divided Morocco into

seven regions, with both rural and urban areas in each
region. Each region has many provinces. This study uses a
subsample of three provinces - Sidi Kacem (rural and urban),
Casablanca (urban), and Marrakech (rural). These provinces
were chosen because although they are all within the.
agricultural area of Morocco, they represent diverse parts

of the country. A description of each area is given here.

General Description of Morocco
Morocco is located in the northwestern corner of

Africa. Four mountain ranges (the Rif, the Middle-, High-
and Anti-Atlas) divide the country into two distinct areas -
to the east of the Atlas mountains is mostly desert area
with little agricultural production: the area bounded by the
Atlantic Ocean and the four mountain ranges defines an area
where most of the agricultural production of the country
takes place (see map on next page). Within the agricultural
area, a further distinction can be made: modern irrigation
and agricultural development have been largely confined to
the broad plains and plateaus defined by the Gharb, Chaouia,
Sais, Abda, Doukkala, Haouz, Souss, and Basse Moulouya. On

the other hand, traditional agriculture, which receives

65

 

66

 

3:

 

ES a [Son em

Egh e ..IOts— 0.
it».
EIIEsO‘f-TK I ..i
eels-It: (......essv. .0.“ i ll

it“s: I ii 8“. I

NOZNOUJ

A .
II"'|.

 

                  

.....

. ... ..

N . OED-=3 .

\ Sentence « s
. . s

.
v

.. are e. . . ...
f. , r . _
. (is.

nllllnlell-«sl

 

 

Map of Morocco

Figure 4-1

67

little or no government assistance, is largely situated in
the area delimited by Kenitra, Fes, Marrakech, and Safi.
Traditional agriculture occupies 70% of Morocco's cultivated
land and involves over 90% of farmers. The irrigated
sector, which is situated on the best soils, comprises only
10-15% of the arable lands but contributes over 85% of all
commercial agricultural production (Payne, 1986).

The four areas studied are within the traditional

agricultural area but none is in an irrigated area.

Study Area of Marrakech

Marrakech is located in the Tensift region of Morocco
in the center of the country. The Tensift is the weakest
part of the country economically, with per capita income of
3011 Dirham (DH) in 1984/85,“ 17 percent lower than the
national average. Rural Marrakech has one of the lowest per
capita expenditures in the country (2451 DH). This area
produces most of the nation's barley. Barley is the
traditional domestically grown food crop. It covers half of
the cultivated cereals acreage and accounts for 40-50
percent of total grains production (Tuluy and Salinger, p
13). Marrakech also produces olive oil, the traditional
edible oil: Marrakech and Fee together cultivate two-thirds

of the area planted to olive trees.

 

“ In 1984 the official exchange rate was 8.81 DH/S;

the equilibrium exchange rate was 9.50 DH/$ (Tuluy and
Salinger, 1988).

68
Sidi-Kacem
Sidi-Kacem is located in the Northwestern part of
Morocco, a part of the country which is in the fertile
agricultural plain. This area produces both soft and hard
wheat. The Northwest is one of the most economically
favorable areas but also has one of the highest disparities

in income between the poorest and the wealthiest households.

Casablanca
Casablanca is the largest city in Morocco and is

located along the central part of the Atlantic coast. For
this reason, Casablanca probably has access to more imported
goods than do the other study areas. Urban Casablanca has a
higher per capita expenditure per year than any other part
of the country: per capita income is 1.2 times higher than
the national average (Laraki, p49). Consequently, cereal
products become less important in the food budget than meat

(Laraki).

Descriptive Statistics about Food Consumption in Morocco
Analyzing the effect of various food policies on the
consumption patterns of the poor requires an understanding
of who the poor are, where they live, and what they consume.
This section uses basic statistics on Morocco and some

statistics from the subsample of the 1984/85 consumption

69
survey to give a preliminary overview of the situation of
the poor.

World Bank unpublished data suggest that in 1984,
nearly 35 percent of the population (of which 23 percent
resides in the rural areas) lived below the absolute poverty
level (World Bank, 1986). Absolute poverty in that year was
defined to be per capita income of less than 2376 DH
(current 1984) in urban areas, and less than 1533 DH
(current 1984) in rural areas.

Using total expenditure“ as a proxy for income, table
4-1 indicates the percentage of the sampled population
living below the absolute poverty level, based on the above
definition of absolute poverty.

 

 

 

 

 

Table 4-1: Percentage of Sampled Population Living
‘ Below the Poverty Level by Study Area

Study Area Percent

Urban Sidi-Kacem 19

N=105

Rural Sidi-Kacem 15

N=220

Marrakech 18

N=385

Casablanca 14

N=317

 

 

 

 

1“
Note: N is the number of households surveyed in each
area.

 

” Total expenditure includes home-production valued at
the local market price and the opportunity cost of renting
for those who own their home.

70
Although the percentage of the population that lives

below the poverty level is smaller for the sampled
population than for the total population, it is still clear
that the poor are not concentrated in one geographical area.
In order to evaluate the effects of various policies on the
poor, it is helpful to know if there are any common
consumption patterns among them. Some of the basic

consumption data will help identify common traits.

Food Shares
As seen in table 4-2, even though rural Sidi Kacem has
the lowest per capita total expenditure, households in
Marrakech spend the least on food.

Table 4-2: Statistics on Total Expenditures and
Expenditures on Food

 

 

 

 

 

 

 

 

 

 

 

 

_
Tot Par Cap % of Total % of
Study Area Annual Annual Expand. Total
, Expand. Expand. Devoted to Expand.
Par Cap on Food Food Spent on
(Dirham) in Lowest Food by
Income Lowest
Quartile Income
(Dirham) Quartile
Urban 5778 1848 49 55
Sidi-Kacem (4045) (474) (13) (12)
N=105
Rural 2885 1394 60 61
Sidi-Kacem (2631) (279) (13) (13)
N=220
Marrakech 2983 1310 58 60
N=385 (2197) (260) (12) (12)
Casablanca 6454 2137 49 57
N=317 (7239) (584) (14) (14)
m E
Note: N 1s t e number of house 0 s surveyed in each area.

Figure in parentheses is the standard deviation.

71

It is a little misleading to compare total expenditure
across zones without correcting for regional differences in
the cost of living.16 However, inflation during the study
period was less than 4 percent, and cost-of-living indeces
are not available for the rural areas. Table 4-2 shows that
although absolute expenditures on food are higher in urban
Sidi Kacem and Casablanca than in rural Sidi Kacem and
Marrakech, the percentage of total expenditures devoted to
food is higher in the rural areas than in the urban areas.
This corroborates Engel's law which states that as incomes
rise, the percentage of income spent on food declines.

The study subsample has an average per capita total
expenditure of 4087 DH, slightly higher than the national
average for that year of 3623 DH. Similarly, both rural
study areas and both urban study areas have per capita total
expenditures above the national average for rural and urban
areas (4915 DH urban, 2637 DH rural).

From these data, it appears that one can distinguish
urban from rural groups since per capita expenditure, per
capita expenditure on food, and share of total expenditure
devoted to food are similar by rural/urban breakdown.

However, a look at the percentage of the food budget

 

“ Although prices are important in determining
consumption patterns, they are not included in this chapter.
Unit values are included in Appendix 5- -1 and will be
incorporated into the regression analyses.

72
allocated to different foods and quantities consumed shows

some differences by study area, especially for cereals.

Cereals

Cereals constitute a major part of the diets of
Moroccans in terms of both calories and expenditures.
Nationally, cereals accounted for nearly 30% of the total
food budget (Direction de la Statistique): in the study
sample the average was 21%.

Few households reported consuming grain of hard wheat,
soft wheat, or barley. Although households may purchase
grain or use home-grown grain, they probably reported their
consumption in the form in which it was used or consumed
(flour, most likely) (Direction de la Statistique). There
is also likely multicollinearity between the products of one
grain. For these reasons, no analysis of individual grain
products will be undertaken for this study. All cereal
figures reported here have been converted into grain
equivalent.17

Table 4-3a shows that among consuming households the

consumption of soft wheat is higher than the consumption of

 

" Conversion factors are in Appendix 4-3. The
products listed there indicate which products are made from
the different grains. The most commmon forms of consumption
reported in the survey are the following:

soft wheat: flour and bread
hard wheat: flour and grain
barley: flour, semoule, and grain

73
hard wheat in all study areas except Casablanca. However,
table 4-3b shows that when figures are based on overall
consumption, per capita consumption of soft wheat is higher
than that of hard wheat, even in Casablanca. Soft wheat is
also the cereal consumed by the highest percentage of
households in all four study areas. Almost all urban
households consume some soft wheat (98%): about 80% of the
households in the rural areas consume soft wheat. Barley is
heavily consumed in Marrakech, although the consumption of
hard wheat and soft wheat in Marrakech is also high.

The percentage of the food budget spent on hard and
soft wheat products in table 4-4 is clearly delineated by
rural/urban lines: the households in the rural areas spend a
higher percentage of the food budget on cereals than do
those in the rural areas. Only a negligible portion of the
food budget is devoted to barley in all areas except
Marrakech, where it is almost as important as the other
cereals. When the figures are converted to an overall
average (by multiplying the figures in table 4-4 by the
percentage consuming), soft wheat becomes the most important

budgetary item for all four study areas.

Tabla 4-3a:

74

Average Per Capita Consumption of
Cereals by Study Area for Households

Consuming that Cereal (kg)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

( Study Area Hard Wheat Soft Wheat Barley

1 urban Sidi-Kacem 136 163 8

r N-105 (126) (109) (4.3)

i [40%] [98%] [13%]

3 Rural Sidi-Kacem 169 183 32

Q N=220 (121) (147) (28)

I [62%] [76%] [3%]

i Marrakech 124 155 141

I N=385 (88) (147) (137)

. [13%] [80%] [72%]

i Casablanca 116 85 ’ 16 !

N-317 (105) (77) (21) l

; M [25:] [98%] [33%] I

Notes: - e n u.er o ouse o -s surveye- in
each area.

- Figure in parentheses is the standard
deviation. _

- Figure in brackets is the percentage of the
population that reported positive consumption
of the good.

Tabla 4-3b: Average Per Capita Consumption of
Caraals by Study area for all Households
Surveyed (kg)
Study Area Hard Wheat Soft Wheat Barley
Urban Sidi-Kacem 54 160 1
N-105 (50) (107) (.56)
Rural Sidi-Kacem 105 139 .96
N=220 (75) (112) (0.8)
'Marrakech 16 124 102
E N=385 (11) (118) (99)
iCasablanca 29
' (26L

 

 

 

 

area.
Figure in parentheses is the standard deviation.

75

 

 

 

 

  

 

   
 

 

Table 4-4: Percentage of Food Budget Spent on
Cereal Products by Consuming
Households
Study Area Soft Wheat Hard Wheat Barley
Urban Sidi-Kacem 11 8 1
N=105 (8) (5) (0.5)
Rural Sidi-Kacem 15 18 6
N=220 (11) (12) (7)
Marrakech 15 15 13
(11) (11) (10)
Casablanca 10 1

 

 

   
 

 

 

(8) (2)
ouseho s surveye 1n eac

area.
Figure in parentheses is the standard deviation.

Table 4-5a and 4-5b indicate that cereals are probably
necessities for most households in all four areas, as the
consumption of all three cereals increases less than
proportionally as income increases. These tables also
illustrate the importance of soft wheat even for the lowest
income quartiles. Even though the per capita quantity
consumed increases as income increases, the percentage of
low-income households consuming soft wheat is high.

For hard wheat, on the other hand, the percentage of
low-income households that consume any hard wheat is low
relative to the percentage of high-income households that
consume hard wheat. This is reflected in the low overall
per capita consumption of hard wheat among the low-income
groups. Both the quantity consumed and the percentage of

households consuming increase as incomes increase.

76

 

Tabla 4-5a: Average Per Capita Consumption of
Cereals by Region and Income
Quartile for All Households Surveyed
(k l
i Income
i Study Area Quartile Hard Wheat Soft Wheat Barley
j Urban 1 18 146 0.7
; Sidi-Kacem (13) ¥(107) (0.3)
' N=105 2 44 169 1.2
. (31) (136) (0.9)
g 3 57 171 1.8
i (67) (98) (0.7)
E 4 96 153 0.7
g (70) (85) (0.2)
i Rural 1 46 113 1.1
éSidi-Kacem (39) (70) (0.6)
i N=220 2 70 132 2.7
( #(59) (82) (2.1)
g 3 143 129 0.5
g (91) (92) (0.2)
§ 4 165 186 -
i 1122) (169) -
TMarrakech 1 11 62 59
i N=385 (8) (47) (53)
; 2 16 99 86
i (12) (71) #157)
- 3 23 131 115
i (15) (99) (96)
; 4 14 203 148
i (10) (187) (156)
; Casablanca 1 7 54 5
N=317 (6) (50) m
2 14 79 6
(10) (as) (7) J,
i 3 54 83 6 i
' (49) (71) (8)
i 4 42) 110 4
Notes: - N is the number of households surveye-

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

each area.

 

 

n

Figure in parentheses is the standard

deviation.

 

77

 

 

 

     

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

    

 

 

 

 

Iebls 4-Sb: Average Per Capita Consumption of Cereals by'lagion and Income Quartile for
Households that Cereal )
Income
A!“ _92.-.r_%.__ 3!“ Wt- 803 “wt __B_-=_l.-z___i
Urban ' 79 152 9
Sidi-tacam (56) (111) (4)
N-105 4412321 [9611 [81]
2 112 169 8
(80) (136) (6)
(391] [10021 (L
3 155 171 8
(180) (98) (3)
[3721 LOOL 1231]—
4 155 159 9
(113) (89) (2)
[621] [96%] [81] ;
Rural 1 114 149 27
Sidi-Kacam (97) (92) (14)
H-220 (4011 4176!] [411
2 121 ' 165 45
(101) (103) (35)
[58!] [801] [61]
8 196 189 13
(124) (135) (6)
[731] [68!] [411
4 211 238 -
(118) (217) -
[781] I782]
Marrakech 1 113 92
N-385 (84) (70)
[101) 4167!]
2 113 119
(82) (85)
[141] [831]
J 3 127 160
(84) (121)
[182] [gall
4 140 226
(100) (208)
[101] [901]
Casablanca 1 57 55
fl-317 (50) (51)
[121] [981]
2 68 81
(49) (69)
[212] [98114i
i 3 142 85 18
‘ (128) (72) (24)
1 [382] [9821 [33!]
i 4 139 115 14
l (97) (94) (19)
:_ __ _ a ___ ,__________-M___._,____________ [2721
Notes: - N is the number of households surveyed in each area.

- Figure in parentheses is the standard deviation.

Figure in brackets is the percentage of the sampled population that reported
positive consumption of the good.

78

Thus, if hard wheat is subsidized, the subsidy will
benefit wealthier households more since they consume more
hard wheat than do the poor, and the percentage of wealthy
households that consume hard wheat is higher than the
percentage of poor households that consume hard wheat. A
subsidy on hard wheat will also leak more than does the
subsidy on soft wheat because there is a larger increase in
the consumption of hard wheat as incomes increase than the
increase in soft wheat. The data indicate that the soft
wheat subsidy may actually be reaching the poor, given the
high percentage of households that consume soft wheat.

In urban Sidi Kacem the consumption of soft wheat does
not increase very much with income, and even declines for
the last quartile; the quantity consumed of hard wheat
increases across income quartiles both overall and for
consuming households only (except in the upper income
quartiles for overall consumption). The subsidy on soft
wheat probably reaches the poor better than would a subsidy
on hard wheat.

In Casablanca, consumption of all wheats is
significantly lower than in the other areas. This probably
reflects a switch into meats, the consumption of which is
generally higher, even for the poor, in Casablanca than in
other areas. Consumption of cereals still increases with

income in Casablanca, but the amount consumed is much lower.

79

The unit values of soft and hard wheat are higher in
Casablanca than they are in the other study areas, probably
because the other areas are more predominantly cereal
growing areas. The relative price of meat is thus lower in
Casablanca (the nominal unit values of meats are about the
same in all four study areas). This may explain the higher
meat consumption in Casablanca.

In this sample, barley is important only in Marrakech
both in terms of quantity and percentage of households
consuming any. The consumption of hard and soft wheat in
Marrakech is lower than the consumption of barley, but still

important.

M33;

Heat is an important part of the Moroccan diet, even
among the poor. Meats occupy the largest portion of the
food budget (23%) in urban areas and are second only to
cereals in rural areas (19%). Mutton, beef, and poultry are
the most important meats; together they account for 83% of
expenditures on meat. Tables 4-6a and 4-6b show the per
capita consumption of mutton, beef, and chicken for all
households surveyed and for only those reporting positive
consumption of meats. Table 4-7 shows the percentage of the

food budget devoted to these meats.

TIbl. C-CIS

80

Average Per Capita Consumption of Keats by

Study area for All

Households Surveyed (kg)

 

 

 

 

 

study Area mutton Beef Chicken
Urban Sidi-Kacem 4.1 5.6 4.8
N=105 (4.4) (6.3) (4.2)
Rural Sidi-Kacem 2.2 1.2 3.8
N=220 (3.0) (0.8) (2.1)
Marrakech 2.9 1.3 1.5 I
N=385 (5.0) (1.0) (2.1)

 

Casablanca

 

 

7.9

 

5.0

 

area.

Figure in parentheses is the standard deviation.

lel. C-Gb:

Average Per Capita Consumption of Heats by
study area for Households Reporting

Consumption of that neat (5g)

 

  

 

  

 

 

 

 

 

 

 

 

study Area mutton Beef Chicken I
Urban Sidi-Kacem 11.4 8.1 7.2
N=105 (12.2) (9.2) (6.4)
[36%] [69%] [66%]
Rural Sidi-Kacem 7.6 3.0 6.7
N=220 (10.5) (1.9) (7.1)
[29%] [41%] [56%]
Marrakech 7.3 3.4 6.3
N-385 (12.4) (2.8) (8.6)
[40%] [37%] [24%]
Casablanca 11.9 10.0 8.2
N=317 (12.0) (11.6) (6.9)
[38%]
Note: - N is e n er o

each area.
Figure in parentheses is the standard
deviation.
Figure in brackets is the percentage of the
sampled population that reported positive

consumption of the good.

 

Table 4-7:

81

Percentage of rood Budget Spent on Meat
Products by Consuming households

 

  

 

  

 

 

 

Study Area Mutton Beef Chicken
Urban Sidi-Kacem 9 11 6
N=105 (7) (11) (3)
Rural Sidi-Kacem 10 8 11
N=220 (5) (4) (7)
Marrakech 10 8 7
N=385 (9) (5) (5)
Casablanca 10 13 6

 

 

 

 

Figure in parentheses is the standard deviation.

Among those households that consume beef, mutton, and

chicken (table 4-6b), mutton is the most important meat in

terms of quantity in all four study areas.

But because

fewer households eat mutton than beef (except in Marrakech),

the overall per capita consumption of beef is higher than

other meats in the urban areas (table 4-6a).

Chicken

becomes the most important meat in rural Sidi Kacem, and

mutton is still the most important meat in Marrakech (table

4-6a).

Among households consuming meats, beef is the most

important expenditure item in the urban areas: mutton is

more important in the rural areas.

The consumption of meats

seems to be clearly delineated by urban/rural lines. The

only clear provinceal difference is that chicken appears to

be slightly more important in the food budget in rural Sidi

82
Kacem compared to other areas, and compared to the

consumption of other meats in that area.

Table 4-8a:

83

Per Capita Consumption of Meats by

Region and Income Quartile for All

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Households Surveyed (kg)
Income
I Study Area jusrti 1e Hutton Beef Chicken
Urban 1 0.3 1.2 1.1
Sidi-Kacem (0.1) “(0.8) (0.5)
N'105 2 1.2 2.7 3.5
(0.7) (0.8) (0.6) n
3 4.3 4.8 5.3 l
(3.3) 9(3.4) (5.2)
4 9.3 12.7 8.3 n
(7.7) (10.4) (5.1)
Rural 1 0.3 0.4 1.2 n
aSidi-Kacem (0.1) (0.2) (0.5)
. “'220 2 0.9 1.3 2.2 n
(0.3) (0.7) (1.2)
3 1.4 1.5 4.3
(0.8) (0.8) (2.4)
4 6.2 1.7 7.8
| (7.0) (1.1) (8.1) .
[Marrakech 1 0.3 0.6 0.5
4 N=385 (0.1) (0.3) (0.3)
2 1.2 0.7 1.1
(0.8) (0.4) (0.8)
3 3.1 1.2 1.4
(2.9) (0.6) (1.6)
4 6.9 2.4 2.8 1
(10.5) (1.6) (3.6)
Casablanca 1 0.2 2.4 2.9
N-317 (0.1) (2.0) (2.7)
2 2 O 6 6 O 4 4 O 9
(3.6) (5.1) (4.2)
3 3.7 7.3 5.6
(3.6) (5.3) (4.2)

 

 

 

 

 

each area.
Figure in parentheses is the standard
deviation.

 

14.5

 

 

6.7

 
  

 

84

 

 

 
   

 

 

 

 

 
  

 

 

 

 

 

 

 

 

 

  

 

 

 

 

 

 

Table 4-8b: Awerage Per Capita.Consumption oflleats by'lagion and Income Quartile for
Consuming Households (ks)
Income
Study’drea Quartile Itnnzmi leaf Chicken
Urban 1 2.8 2.6 2. 5
Sidi-Kace- (1.2) (1.8) (1. 3)
l-105 (12!) (461] (422]
2 5.0 3.9 5. 3
(3.2) (2.0) (3. 3)
[231] [69!] [65!)
3 7.3 5.9 6.5
(5.6) (4.6) (6.3)
[59!) [74!] [8211
4 18.5 14.9 11.3
(15.3) (12.2) (7.0)
[50!) [851] [73!]
Rural 1 1.8 1.7 3. 4
Sidi-laces (0.6) (0.9) (1.4)
l-220 [16!] (222] (362)
2 3.6 3.1 4. 4
(1.2) (1.6) (2. 3)
(25!) [42!] [512)
3 5.5 2.8 6. 3
(3.0) (1.5) (3. 6)
[252] [54!] [661)
4 13.0 3.9 10.8
(14.5) (2.6) (11.2)
[48!) [442] [722)
Marrakech 1 2.0 1.7 2. 7
l-385 (0.7) (0.9) (1. 3)
[15!] [341) [20!)
2 3.1 2.1 4.2
(2.1) (1.1) (3.1)
[40!) (352) [252]
3 6.0 3.4 7.0
(5.7) (1.9) (7.1)
[511] [341] [22!]
4 12.5 5.4 10.0
(19.0) (3.6) (12.8)
[55!) [442] [282)
Casablanca 1 1.9 3.4 5. 0
8'317 (1.2) (2.9) (4. 7)
[102] [702] [571]
2 6.3 7.9 7. 5
(8.7) (6.3) (6. 5)
[411] [811] [651]
3 9.6 8.8 8. 8
(8.9) (6.4) (6. 5)
[382) [831] [641]
4 17.3 17.5 11.5
(13.0) (17.2) (8.0)
[671) [831] [581]
FEW” _
surveye n eac ares.

 

 

 

 

- Figure in parentheses is the standard deviation.

Figure in brackets is the percentage of the sampled population that reported
positive consumption of the good.

 

85

The information in tables 4-6a and 4-6b is broken down
by income quartile in tables 4-8a and 4-8b. In all of the
study areas, for all households surveyed as well as only for
those reporting positive consumption, the consumption of the
three meats increases substantially with incomes. Among
consuming households, the highest income quartile consumes
as much as nine times the quantity of meat as the lowest
income quartile. In most cases, the percentage of
households consuming a given meat also increases as income
increases. The increase in the number of consuming
households is not substantial, however, for beef and chicken
in Casablanca, where the number of households consuming
these meats is high even among the lowest-income quartiles.

Geography may not play a role in determining the
consumption of meats. Both rural Sidi Kacem and Marrakech
households have similar, low per-capita consumption of all
meats, urban Sidi Kacem households consume slightly more
meat, and Casablanca, which has the highest per capita total
expenditure, consumes the most.

The consumption of mutton, beef, and chicken increases
with increasing incomes despite the virtually constant unit
values across income quartiles. (This is also true for
0118; the case is mixed for cereals.) This hints that the
quality issue described in chapter 3 may not be very
important. In other words, as income rises, households

still want to consume more in quantity (and more households

86
are coming into the market) and are not making dramatic
shifts into higher quality meats. Unit values could be
constant, however, if food prices were higher in poor areas,
so that the constant unit values reflect higher-quality
consumption by the wealthy.

These data, as well as the data on other commodities,
should be interpreted with care. Appendix 4-2 shows that
the number of household members declines as income rises.
Yet one would expect that the wealthier households would
serve relatives and friends in addition to immediate family
members. The survey may not have captured all consuming
individuals of a household. The per capita quantities may

be thus be more overstated as incomes rise.

9.115

The two main oils consumed in Morocco are olive oil,
the traditional oil, and vegetable 011, most of which is
imported. Butter is also consumed. For some households,
butter is a substitute for oils, for others, it is a
complement. There are not large differences in the
consumption of butter or olive oil, in terms of either
percentage or quantity consumed, with the exception of olive
oil being more important in Marrakech and butter being

slightly less important in Casablanca.

87

 

 

 

 

 

 

 

 

Table 4-9: Percentage of Food Budget Spent on 011
Products byConsuming Households

:Study Area Vegetable Olive oil Butter “

f 011

:Urban Sidi-Kacem 5 4 3

1 N=105 (3) (3) (2)

‘ Rural Sidi-Kacem 6 6 5

i N=220 (3) (6) (4)

{ Marrakech 4 8 4

1 N=385 (2) (5) (4)

x

‘ Casablanca 4 4 2

f N=317 (2) (4) ("by f (2)
No e: N s e n uoer o' ouse o os surveye- On eac

area.

Figure in parentheses is the standard deviation.

Table 4-10a:

Average Per Capita Consumption of Oils by

Province for all Households Surveyed (kg)

 

 

 

  
 

 

   
 

 

  
   

 

       
  
    

   
 
 

“N=317
Note: N
area.

 

 

(1.4)

 

os surveye-—

Vegetable Olive Oil Butter 1

Study Area 011 1
Urban Sidi-Kacem 17.8 1.5 2.2
N=105 (11.6) (1.1) (2.0)
Rural Sidi-Kacem 12.7 1.5 0.9
N=220 (8.2) (1.4) (0.8)
9 Marrakech 7.9 6.3 0.8
; N-385 (7.1) (5.5) (0.9)

?Casablanca 18.0 1.6 1.8 :

(2.7)

Figure in parentheses is the standard deviation.

88

Table 4-10b:

Average Per Capita Consumption of Oils by

Province for Consuming Households (kg)

 

 

 

 

 

 

 

 

'l Vegetable Olive Oil Butter H

i Study Area Oil

‘Urban Sidi-Kacem 19.6 6.9 3.4

1 N-lOS (12.8) (5.1) (3.2)

! [91%] [22%] [64%]

iRural Sidi-Kacem 13.7 6.3 .

. N=220 (8.8) (6.2) (2.7)

i [93%] [23%] [29%]

! Marrakech 9.8 8.8 3.0

‘ N=385 (8.8) (7.7) (3.7)

w [81%] [72%] [25%]

[Casablanca 18.9 7.0 2.4 i
N-3l7 (18.1) (6.1) (3.6) I

[95%] [23%] [76%] J

No e: - N is t e n u-er o‘ ouse o is surveye. in

each area.

- Figure in parentheses is the standard
deviation.

- Figure in brackets is the percentage of the

population that reported positive consumption
of the good.

Vegetable oil is the most heavily consumed oil in all
of the areas, which is to be expected given the relative
prices of vegetable oil and olive oil. Despite the
disparity in prices, in Marrakech, olive oil is still
important in consumption. Both olive oil and vegetable oil
are consumed by a large percentage of households in
Marrakech; in the other study areas, vegetable oil is the
most important both in terms of the number of households
consuming and in terms of quantity. Butter is consumed by
76% of the households in Casablanca, although the per capita

quantity consumed among consuming households is the lowest

of all four areas.

89
2:231:99:
Sugar is included in the beverages category because of

the complementarity between sugar and both coffee and tea.

Per capita consumption of sugar in Morocco is 30 kg per

annum, a major source of calories in the Moroccan diet

(compared to a world average of 21 kg per capita and a

developing-country average of 12 kg per capita) (Tuluy and

Salinger, p.20).

For consuming households and overall, both

CasablanCa and Marrakech have higher per capita consumption

of sugar (about the same level) than do urban Sidi Kacem and

rural Sidi Kacem.

consumption of tea.

consumption of coffee.

They also have higher per capita

Casablanca also has the highest

One thing to note about the

beverages is the high percentage of the population that

consumes coffee, tea, and sugar.

 

 

 

 

 

 

 

 

 

 

 

Table 4-11: Percentage of Food Budget Spent on Beverages
by Consuming Households (kg)
Study Area Coffee Tea Fresh Sugar
Milk
L Urban Sidi-Kacem 3 3 2 4
]N=105 (2) (2) (1) (3)
Rural Sidi-Kacem 3 .3 5 7
N=220 (2) (2) (4) (4)
Marrakech 3 5 4 10
N=385 (2) (3) (3) (5)
Casablanca 2 4 2 6
N-317 <2) <3) <2) (4)
Note: N 18 t e n er of ouse o s surveye 1n eac

area .

Figure in parentheses is the standard deviation.

90

Table 4-12a: Average Per Capita Consumption of Beverages

and Sugar for All Households Surveyed (kg)
_——1

 

 

  

 

  

 

  

 

  

 

  

 

 

 

 

 

   
   

Study Area Coffee Tea Fresh Sugar
Milk
N-105 (1.1) (0.8) (1.0) (15)
Rural Sidi-Kacem 1.1 .1.1 18.2 27
N=220 (1.2) (1.0) (18.2) (16)
Marrakech 1.0 1.7 10.8 35
N-385 (1.1) (1.3) (9.0) (23)
Casablanca 1.7 1.9 0.8 35
N-317 (2.3) (1.6) (0.7) (30)
No e e n er 0 ouseholds surveye: In each

area.
Figure in parentheses is the standard deviation.

Table 4-12b: Average Per Capita Consumption of Beverages

and Sugar for Consuming Households (kg)

 

 

 

 

 

Study Area Coffee Tea Fresh Sugar
Milk

Urban Sidi-Kacem 1.7 1.0 21 25
N-105 (1.3) (1.0) (13) (16)
[81%] [78%] [8%] [96%]

Rural Sidi-Kacem 1.5 1.1 33 27
N=220 (1.6) (1.0) (33) (16)
[72%] [97%] [55%] [99%]

Marrakech 1.3 1.7 25 35
N=385 (1.4) (1.3) (21) (23)
[76%] [98%] [43%] [99%]

Casablanca 2.5 2.0 21 36
N=317 (3.4) (1.7) (17) (31)
[98%]

 

 

 

 

 

 

each area.

- Figure in parentheses is the standard
deviation.

- Figure in brackets is the percentage of the
population that reported positive consumption
of the good.

91
Conclusions of this Section

Barley seems to be important only in Marrakech, with
other areas spending only about 1% of the food budget on
barley. The quantity of barley consumed in other areas is
also small, while the quantity consumed in Marrakech is high
(141 kg per capita for households that consume it). This is
not surprising considering that Marrakech is the largest
barley producing province in Morocco. Seventy percent of
those surveyed in Marrakech reported positive consumption of
barley, compared with less than 28% in the three other
provinces. Furthermore, more than half of those reported
that on average 95% of their consumption of barley came from
home production.

Cereals appear to be important for all income groups,
although less so for upper-income quartiles. As incomes
increase, the quantity of all cereals increases while the
percentage of the food budget spent on cereals declines.
Both soft wheat and hard wheat are important for the lowest-
income quartile in rural Sidi Kacem and Marrakech. The
quantity consumed of all cereals (both overall and among
consuming households) is low relative to consumption levels
in other areas.

Subsidizing any of the cereals is likely to benefit
wealthier households because the wealthy consume more
cereals than do the poor. In terms of increasing the

calories consumed by the poor, soft wheat seems more logical

92
to subsidize than hard wheat because the poor eat more of it
than they do hard wheat (except in Marrakech and Casablanca,
where its about the same), and the rate of increase in
consumption with rising incomes is not as fast as hard
wheat.

The subsidy on sugar seems likely to affect the
consumers in Marrakech and Casablanca more than in the other
areas, given their high per-capita consumption of sugar.
Although sugar is a major source of calories in the Moroccan
diet, the validity of subsidizing sugar on nutritional
grounds must be questioned.

Finally, the subsidy on vegetable oil is probably not
effective in reaching the poor in Marrakech, given the high
per-capita consumption of olive oil and the relatively low
per capita consumption of vegetable oil. It may be
increasing the consumption of vegetable oil among the poor
in other areas, given the high percentage of low-income
households that consume vegetable oil.

Although the descriptive data give some indication of
consumption patterns by study area and income quartile, how
households respond to price and income changes will be more
useful in terms of predicting behavior. Elasticity
estimates will follow the discussion of methods in the next
chapter. The estimates will be compared to the estimates
from the Groupe de Travail to see if there are regional

differences in the elasticity estimates as well as to see if

93
a slightly different methodological approach alters the

estimates.

CHAPTER 58 ANALYTIC DISCUSSION OF RESULTS

This chapter presents the general model corresponding
to the methods described in chapter 3, and an analysis of
the elasticities of demand that are derived from it. The
policy implications drawn from this discussion follow in

chapter 6.

Description of Variables in the Model
Dependent Variables

Since the models used to estimate income and price
elasticities are based on the AIDS, the dependent variable
for each commodity is the budget share allocated to that
commodity. The budget share is calculated by dividing
annual household expenditure on a commodity by the total
annual household expenditure. Expenditure on a commodity is
the unit value multiplied by the quantity consumed.

The commodities modeled are: soft wheat, hard wheat,
barley, olive oil, vegetable oil, butter, beef, mutton,
chicken, and sugar. These commodities were chosen because
they are the most basic staples for Moroccan households.
These broad commodity groups were constructed by aggregating
more detailed consumption data. Soft wheat, for example,

comprises soft wheat flour, bread made of soft wheat, soft

94

95
wheat grain, and other product made from soft wheat, all

converted to their grain equivalents.“

Independent Variables

The specification of the AIDS inspired the choice of
the independent variables. The AIDS specifies the budget
share as a function of own price, cross prices of all goods
in the system, and total real expenditure, all in
logarithmic form. Although the models presented here do not
constitute a true AIDS, own price, cross prices, and total
expenditure were included as independent variables. Other
than own price, only the prices of goods assumed to be
either substitutes or complements to the good being modeled
were included. This is justified on the assumption that the
coefficients on the prices of other goods are zero.
Household size was also hypothesized to be an important

determinant of consumption.

W

Because prices were not available, unit values served
as proxies for prices. A unit value is defined as total
expenditure on a good divided by the quantity purchased of
that good. If the unit value was missing because either the

quantity or expenditure was missing, the average unit value

 

“ The coefficients used to convert goods to their

grain equivalents are contained in Annex 4-3

96
for the cluster (5 households) was used. This is based on
the assumption that all households in a cluster face the
same prices. Chapter 3 describes the tests used to justify
the use of unit values as proxies for prices without

correcting for the quality effects or measurement error.

Winn:
Total expenditure was defined as the sum of household

expenditures on all goods in one year. Although the AIDS
specifies the total expenditure variable as total
expenditure on all goods in the system, total expenditure on
all goods consumed, not just on those being modeled, is used
here. The use of total expenditure on all goods makes it
possible to calculate expenditure elasticities directly from
the estimated coefficients. This eliminates the need to
call upon separabilty assumption since it implies that the
coefficients on all other goods that make up the total
expenditure are assumed to be zero.

Total expenditure was not deflated by a price index
because the data are cross-sectional and regional price
indices are not available, and it was impossible to

calculate the Stone's index.19

 

” See p.63 for discussion of why it is not possible to
calculate the Stone's index in this case.

97
W
A squared expenditure term was also included to allow
goods to be a necessity or inferior good for one income.
quartile and a luxury item for another (see Chapter 3).
When the squared income term was not significantly different

from zero, the term was dropped.

W

In addition to prices and total expenditurer the
inverse Mill's ratio was included as an independent variable
to account for the censoring problem described in Chapter 3.

The inverse Mill's ratio is defined as

.f a+BX1
1- ( ° ) (5-4)
1

o+BX;)

F( O

 

 

where f(*) represents the density and F(*) the cumulative

distribution function of a standard normal variable.

11911591152151.8122

"Adult equivalents," which convert the number of all
household members into an "equivalent" adult-number, is used
to account for the effect of family structure on
consumption. The number of adult equivalents was
calculated using the following coefficients: head of

household =1, other adults=0.7, children less than 15 years

98
old-0.5.2° Including this variable allows one to discuss
the effect of household expenditure on consumption, holding

household size constant.

W125

Dummy variables for the study areas of Casablanca, and
rural and urban Sidi Kacem were also included to account for
regional differences in consumption. .Theoretically, if the
dummy variables are entered linearly and not as part of an
interaction term, the dummy variables should only shift the
intercept term; the slope coefficients should not change.
Multicollinearity between the dummy variables and other
parameters in the model, however, may affect other parameter
estimates when the dummy variables are included.

A dummy variable for home production was also included.
The dummy variable had a value of one if the household had

any home production of the good, a zero otherwise.

111W

The elasticities will vary by study areas since the
budget shares vary by study area, and the budget shares are
part of the formula for the elasticities. The elasticities

may also vary if the s10pe coefficient on the price or

 

“ Although it may be more appropriate to count all
adults equally, the data used here prohibit using different
coefficients since the raw data from which the adult
equivalents were calculated was not available.

99
income term varies by study area. Interaction terms were
included to allow the slope coefficients to vary by study
area. When the interaction terms were not significant, they

were dropped from the equation.

The General Model
The most general form of the model, when all of the

above independent variables are included is:

wi '3 “1+; Ygglogp3+pul°g (X) +pgz[1°g (X) 12+a%%;_ (5'5)

"’ “1109611": 512193;: (110210ng :2: tuD‘logx "' w:

where:
«Q - the share of total expenditure allocated to the
ith good
p] - the unit value of the jth good
x - total household expenditure
(XJ ) Inverse Mill's ratio
0(x’fl)
CU - number of consuming units in the household
1% - dummy variable for Casablanca
Dz== dummy variable for rural Sidi Kacem
[h = dummy variable for urban Sidi Kacem
In - dummy variable for home production of good i

“u 71]: $11: 512: 0: “1: 511: 912: 513: C1: C2: C3: 1'1: 1'2: 1:3: W

are parameters

100
Judging Which is the Best Model

For each commodity, several variations of the above
model were tested. Each included own price, income,
household structure, and dummy variables for study areas.
Prices of substitutes and complements, a squared income
term, and interaction terms were also included in subsequent
regressions, but were dropped when their coefficients were
not significant.

Although there is no one criterion with which to judge
which model most accurately reflects reality, several
criteria can be used to make an overall assessment. The
most important is the theoretical soundness of the model.
The analyst should have strong a_prigzi reasons to suspect
that one variable is a determinant of another.

The standard errors of a parameter's estimate can be
used to formulate t-tests of the significance of a variable.
Economists commonly test the significance of variables at
the 1 or 5 percent level: the 5 percent level will be used
here. A five percent significance level means that if the
sample were repeated an infinite number of times, there
would be 5 percent chance that the coefficient would not be
within 2 standard deviations of the original estimate.

When the model includes the inverse Mill's ratio as an
independent variable, the standard errors must be adjusted
to account for the fact that the variable is an estimate and

is not directly observed. The formula for the standard

101
error when the inverse Mill's ratio is used is included in
Appendix 5-1.

The R2 is commonly used to assess how much of the
variation in the independent variable is attributable to the
variation in the independent variables. When the inverse
Mill's ratio is used, however, this statistic is invalid.
The "information accuracy" will be used instead to assess
the model's fit (Theil, Chung,and Seale, 1989). This

statistic is defined as:

. w
In: ' wih ‘ wih 4' wihlog'fig (5'6)
ih

where:
'wm is the observed budget share for the ith good for

household h
'wm is the predicted budget share for the ith good for

household h.

If wulperfectly predicts'wm, I“ will be zero: the
better the fit of the model, the closer will be I“ to zero.

The estimated elasticities are presented in tables 5-1
through 5-10. The first-stage probit results used to

calculate the elasticities are presented in Appendix 5-2.

102

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

‘l’file 5-1: Cereal Elasticities
Product Study Total Price Price Price Mill's Info
Area Expand Soft Barley Hard Ratio Accuracy
Wheat Wheat 51gnif7‘
1 0.54* -1.52" 0.14* 0.33' no 0.0037 F
Soft Wheat (0.02) (0.02) (0.03) (0.02)
2 0.68* -1.30* 0.089 0.19*
(0.02) (0.01) (0.02) (0.01)
3 0.67* -1.32* 0.09“ 0.20*
(0.02) (0.01) (0.02) (0.01)
4 0.42* -1.67* 0.19' 0.42*
(0.03) (0.02) (0.04) (0.03)
1 -5.28* 1.98* no 0.017
Barley (0.33) (0.50)
2 0.13' -0.59*
(0.05) (0.07)
3 0.64* -0.83'
(0.02) (0.03)
4 -2.59* 0.70*
(0.19) (0.29)
1 0.18* -0.69* yes 0.0296
hard Wheat (0.07) (0.03)
2 0.48* -0.74*
(0.03) (0.02)
3 0.22' -1.39*
(0.03) (0.07)
4 0.53 -0.65*
(0.05) (0.03)
_
Notes:
Study.Areas

1 8 urban Sidi Kacem
2 8 rural Sidi Kacen
3 I Marrakech

4 I Casablanca

* indicates significant at least at the 5% level.

a "yes“ if significant at least at the 10% level, “no" if not.
figures in parentheses are standard errors

 

 

tfile 5-2:

 

  

Soft "bait Elltticitlel

income

103

 

Price

  
       

Price

Area and lncoaa Quartile

   

 

   

 

 

 

 

 

 

 

 

 

 

     
 

 

       
 

 

    
 

 

      
 

 

     
 

 

      
 

ce Quartile Expand Soft Barley
Wheat
1 1 0.58’ ~1.37* 0.10. 0.23'
(0.03) (0.01) (0.02) (0.01)
2 0.59* -1.42* 0.12* 0.27'
(0.02) (0.02) (0.02) (0.02)
3 0.50* -1.60* 0.17‘ 0.38*
(0.03) (0.02) (0.03) (0.02)
4 0.41* -1.90* 0.25* 0.57' I
(0.04) (0.03) (0.05) (0.04)
2 1 0.73* -1.22* 0.06* 0.14* I
(0.02) (0.01) (0.01) (0.01)
2 0.70' -1.27* 0.08* 0.17* l
(0.02) (0.01) (0.02) (0.01)
3 0.66* -1.33* 0.09* 0.21*%
(0.02) (0.01) (0.02) (0.01)
4 0.56* -1.50* 0.14* 0.32*
(0.02) (0.02) (0.03) (0.02)
3 1 0.70* -1.25* 0.07‘ 0.16*
(0.02) (0.01) (0.01) (0.01)
2 0.66* -1.31* 0.09' 0.20*
(0.02) (0.01) (0.02) (0.01)
3 0.66* -1.34* 0.09* 0.22*
(0.02) (0.01) (0.02) (0.01)
4 0.67* -1.39* 0.11* 0.24*
(0.02) (0.01) (0.02) (0.02)
4 1 0.49* -1.46* 0.13' 0.30“
(0.04) (0.02) (0.03) (0.02)
2 0.41* -1.63* 0.17* 0.40'
(0.03) (0.02) (0.04) (0.03)
3 0.39* -1.73* 0.20' 0.46*
(0.03) (0.03) (0.04) (0.03)

 

 

 

7 Notes:

    
 

Study'Areas

0.33*

 

<0.)

 

, ‘09",-

 

.200”

 

(0.06)

 

0.30*

1 8 urban Sidi Kacela
2 8 rural Sidi Kacen
3 8 Marrakech

4 I Casablanca

 

        
 

* indicates significant at least at the 5% level.

Tdale 5-3: Ilard ll'oeat Elasticities Area lid Inca urti la
Expand
0.51* -o.s1*
(0.04) (0.02)
0.19* -o.69*
(0.07) (0.03)
0.27- -o.72*
(0.04) (0.03)
-0.20* -0.54*
(0.08) (0.04)
0.55* -0.77* I
(0.03) (0.02)
0.51* -0.75* I
(0.03) (0.02)
0.51* -0.75*
(0.02) (0.02)
0.37* -0.68*
(0.03) (0.03)
0.52* -1.24*
(0.02) (0.04)
0.26* -1.37*
(0.04) (0.06)
0.15* -1.42*
(0.23) (0.07)
-0.36* -1.68*
(0.06) (0.12)
0.53* -0.65*
(0.06) (0.03)
0.51* -0.64*
(0.07) (0.03)
0.62* -0.72*
(0.03) (0.02)
0.33*
(0.06)

104

  
 
 

 

 

 
  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1 I urban Sidi Kacem
2 I rural Sidi Kacen
3 I Marrakech

4 I Casablanca

* indicates significant at least at the 5% level.

105

     
 
 
 

    

 

 

 

 

 

 

 

 

 

 

 

 

 

 

T‘le 5-4: Iarl Elasticities Area Id Inc. hrti la
income
Area Quartile Expand Carley
1 1 ~2.25* 0.54* I
(0.17) (0.26)
2 -4.22* 1.47'
(0.28) (0.42)
3 -0.24* 3.38*
(0.49) (0.74)
4 -10.31* 4.36*
(0.60) (0.91)
2 1 -0.02 -0.52*
(0.05) (0.08)
2 0.41* -0.72*
(0.03) (0.05)
3 -2.75* 0.73'
(0.20) (0.30)
4 . .
3 1 0.71* -0.86*l
(0.02) (0.02)
2 0.68* -0.85*
(0.12) (0.02)
3 0.66* -0.84*
(0.02) (0.03)
4 0.46* -0.74*
(0.03) (0.04)
4 1 -1.36* 0.12
(0.12) (0.19)
2 -1.86* 0.35
(0.15) (0.23)

 

 

  

 

 

 

 

 

Notes:
Study'Area
1 I urban Sidi Kacem
2 I rural Sidi Kacen
3 I Marrakech
4 I Casablanca

* indicates significant at least at the 5% level.

106

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

tfila 5-5: Oil Elasticities
__
Product Study Total Price Price Price Will's Info
Area Expand Veg Olive Butter Ratio Accuracy
Oil Oil Signif?
1 0.58* -0.79' no -0.0005
Vegetable (0.03) (0.24)
Oil
2 0.66* -0.83*
(0.03) (0.20)
3 0.52* -0.76‘
(0.04) (0.28)
4 0.44* -0.72*
(0.04) (0.32)
Olive Oil 1 0.02 -1.02* no 0.0274
(0.62) (0.24)
2 0.45 -1.01*
(0.35) (0.14)
3 0.54 -1.01*
(0.29) (0.11)
4 0.04 -1.02'
(0.61) (0.23)
Cutter 1 1.40* -0.89* yes 0.0014
(0.49) (0.17)
2 1.20* -0.95*
(0.25) (0.09
3 1.23* -0.94*
(0.29) (0.10)
4 1.48* -0.87*
(0.59) (0.20)
EEK —
Duns:
Study Area

1 I urban Sidi Kacea
2 I rural Sidi Kacen
3 I Marrakech

4 I Casablanca

' indicates significant at least at the 5% level.

a "yes" if significant at least at the 10% level, “no" if not.
figures in parentheses are standard errors

 

107

 
  
 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

   
   
  
   
  
   
 

  
  

  
  

Tfile 5-6: last Elasticities
=
Product Study Total Price Price Price Hill's Info :
Area Expand Chicken Beef Hutton Ratio Accuracy
Signif?
1 0.63* -0.37* -1.44* no 0.0048
Chicken (0.01) (0.01) (0.02)
2 0.85* -0.75* 0.27*
(0.01) (0.01) (0.03)
3 0.77. .Oaw 0.102.
(0.01) (0.01) (0.05)
4 0.68* ~0.44* -0.59*
(0.01) (0.01) (0.01)
Beef 1 0.86* -1.36* 0.26 no 0.0092
(0.09) (0.14) (0.14).
2 0.86' -1.38' 0.28
(0.09) (0.15) (0.15)
3 0.85* -1.38* 0.28
(0.09) (0.15) (0.15)
4 0.89* -0.80* 0.20
(0.07) (0.11) (0.11)
Mutton 1 1.08* 0.39' -1.05* no 0.007
(0.04) (0.02) (0.01)
2 1.18* 0.31* -1.04*
(0.04) (0.01) (0.01)
3 1.17' 0.31* -1.04*
(0.04) (0.01) (0.01)
4 1.07' -1.05*
(0.04) (0.01)
Notes:
Study'Area

1 I urban Sidi Kacem
2 I rural Sidi Kacem
3 I Marrakech

4 I Casablanca

' indicates significant at least at the 5% level.
a "yes" if significant at least at the 10% level, “no" if not.
figures in parentheses are standard errors

108

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Twle 5-7: Chicks: Elasticities F E Area Id inc. urtile
Study Income Total Price Price
Area Ouartile Expand Chicken Beef
1 1 0.64* -0.37* -1.42*
(0.01) (0.01) (0.03)
2 0.67' -0.44* -1.28*
(0.01) (0.01) (0.02)
3 0.65* -0.40' -1.38*
(0.01) (0.01) (0.02)
4 0.57* -0.268 -1.67* I
(0.01) (0.01) (0.03)
2 1 0.87' -0.78* 0.24*
(0.01) (0.01) (0.03)
2 0.86' -0.75* 0.26'
(0.01) (0.01) (0.03)
3 0.85' -0.75' 0.27'
(0.01) (0.01) (0.03)
4 0.83' :0.71' 0.31'
(0.01) (0.01) (0.04)
3 1 0.76' -0.58* 0.45'
(0.01) (0.01) (0.05)
2 0.78* -0.62* 0.40*
(0.01) (0.01) (0.05)
3 0.74* -0.65* 0.38*
(0.01) (0.01) (0.04)
4 0.74* -0.56* 0.47'
(0.01) (0.01) (0.06)
4 1 0.74’ -0.56* -0.47*
(0.01) (0.01) (0.01)
2 0.68* -0.45* -0.59*
(0.01) (0.01) (0.01)
3 0.70' -O.49* -0.55*
(0.01) (0.01) (0.01)
4 0.52' ~0.17* -0.89*
(0.01) (0.01) (0.01)
“I

 

 

 

 

Notes:

 

Study'Area
1 I urban Sidi Kacem
2 I rural Sidi Kacen

 

 

 

3 I Marrakech
4 I Casablanca

' indicates significant at least at the 5% level.

109

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Tfile 5-8: leaf Elasticities - . Area all In urti le
Study lncoaa Total Price Price
Area Ouartile Expand Seef Nutton ;
1 1 0437*' -1.35* 0.29 '
(0.08) (0.14) (0.13)
2 0.85' -1.40* 0.29*
(0.10) (0.16) (0.15)
3 0.84* -1.42* 0.30
(0.10) (0.17) (0.16)
4 0.88‘ -1.31* 0.23*
(0.07) (0.12) (0.12)
2 1 0.87' -1.35* 0.25
(0.07) (0.14) (0.13)
2 0.89' -1.29* 0.21
(0.07) (0.12) (0.11)
3 0.84' -1.42* 0.31
(0.10) (0.17) (0.16)
4 0.86. -1.49' 0.35
(0.09) (0.20) (0.19)
3 1 0.87* -1.35* 0.26
(0.08) (0.14) (0.14)
2 0.84* -1.42' 0.30
(0.10) (0.17) (0.16)
3 0.86* -1.37* 0.27
(0.09) (0.15) (0.14)
4 0.85* -1.39* 0.28
(0.09) (0.16) (0.15)
4 1 0.89' -1.29' 0.21
(0.07) (0.12) (0.11)
2 0.91* -1.25' 0.18
(0.06) (0.10) (0.10)
3 0.89* -1.29* 0.21
(0.07) (0.12) (0.11)
4 0.87' -1.30' 0.22'
(0.07) (0.12) (0.11)
otes:
Study’Area

1 I urban Sidi Kacen
2 I rural Sidi Kacen
3 I Marrakech

4 I Casablanca

' indicates significant at least at the 5% level.

110

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Tble 5-9: ”to: Elasticities . Area III Inca tartila
Study lncoaa Total Price Price
Area Quartile Expand Hutton Chicken ;

1 1 1.14* -1.07* 0.49'
(0.07) (0.01) (0.02)

2 1.15* -1.05* 0.38*

(0.04) (0.01) (0.02)

3 1.048 -1.06* 0.43*

(0.04) (0.01) (0.02)

‘ as”. '1.05' 0.36.

(0.02) (0.01) (0.02)

2 1 1.45* -1.07* 0.48*
(0.07) (0.01) (0.02)

2 1.22* -1.04* 0.31'

(0.04) (0.01) (0.02)

3 1.15. -1.04* 0.28'

(0.03) (0.01) (0.01)

‘ 1a”. '1a“. 0.28.

(0.03) (0.01) (0.01)

3 1 1.38' -1.05* 0.39'
(0.06) (0.01) (0.02)

2 1.27* -1.05* 0.40“

(0.05) (0.01) (0.02)

3 1.14* -1.04* 0.29*

(0.03) (0.01) (0.01)

4 1.04. -1.04* 0.27*

(0.02) (0.01) (0.01)

4 1 1.47‘ -1.09* 0.66*
(0.09) (0.02) (0.03)

2 1.11* -1.04* 0.32*

(0.03) (0.01) (0.02)

3 1.05* -1.06* 0.42*

(0.04) (0.01) (0.02)

4 0.85* -1.06* 0.418

__,___ (0.23)h _(0.01)v7 (0.02) 7‘
Notes:
Study Area

1 I urban Sidi Kacen
2 I rural Sidi Kacea
3 I Marrakech

4 I Casablanca

' indicates significant at least at the 5% level.

111

 

 

 

 

 

 

 

 

 

 

 

Tile 5-10: r Elasticities
Product Study Total Price Price Nill's Info
Area Expand Sugar Tea Ratio Accuracy
Signif?
1 0.19' -0.26 no 0.0012
Sugar (0.08) (0.40)
2 0.58* -0.62*
(0.04) (0.21)
3 0.70* -0.72*
(0.03) (0.15)
4 0.43* -0.48
(0.06) (0.28)
Notes:
Study'Area
1 I urban Sidi Kacea
2 I rural Sidi Kacen
3 I Narrakech
4 I Casablanca

* indicates significant at least at the 5% level.
a I'yes" if significant at least at the 10% level, “no" if not.
figures in parentheses are standard errors

Discussion
Inverse Mill's Ratio

The inverse Mill's ratio was significant in only two
demand equations. The Mill's ratio was included, however,
even when its coefficient was insignificant. Dropping the
Mill's ratio substantially altered the estimates of the
other coefficients in the equations, and in some cases made
the coefficients unreasonable. The insignificance of the
inverse Mill's ratio is probably due to correlation between
the Mill's ratio and the other variables in the equations,
since the Mill's ratio is itself a function of those
variables. Since a large proportion of households reported
zero consumption of most of the goods (see table 3-2), the

censoring problem is probably more important than the

112 ,
multicollinearity that arises by including the inverse

Mill's ratio.

Cereals
§Q£L_Eh§Q§

Soft wheat is a necessity in all four study areas but
is least sensitive to changes in income in Casablanca
(expenditure elasticity of 0.42). Casablanca has the
highest total annual expenditure per capita of all four
study areas and thus as income increases households may
switch out of cereal consumption into the consumption of
meats. The regional expenditure elasticities are inversely
related to their respective income levels (see table 4-2).
Casablanca has the highest total annual expenditure per
capita and the lowest expenditure elasticity for soft wheat.
Marrakech and rural Sidi Kacem have similar low total annual
expenditure per capita and similar high expenditure
elasticities.

Consumers in all four study areas are quite sensitive
to the price of soft wheat (own price elasticities in all
four study areas are greater than 1 in absolute value):
households are the most sensitive in Casablanca and urban
Sidi Kacem. Since these are both urban areas, households in
these may be more sensitive to price changes since they may
have access to more variety of foods in the markets. Both

barley and hard wheat are substitutes for soft wheat,

113
although this relationship was not captured in the equations
for barley or hard wheat.

A breakdown of the elasticities by income quartile
within each study area (table 5-2) shows that the wealthy
households in all four areas are more sensitive to changes
in soft wheat prices than are the poor. The difference
between the own-price elasticity of the lowest income
quartile and the highest is most pronounced in urban Sidi
Kacem and Casablanca. The wealthy also have smaller income
elasticities of demand for soft wheat than do the poor in
all study areas. The response to income changes is almost

the same among the poor and the wealthy.

W

Hard wheat is a necessity in all four areas. It is
also a normal good in all four areas, i.e., as the price
goes up, consumption declines. But the elasticities vary
not only by urban/rural breakdown but also by study area.
Households in Casablanca have the highest expenditure
elasticity: those in urban Sidi Kacem have the lowest.

What these figures do not capture, however, is how
consumption changes within a study area by income. Table 4-
8b shows that as income increases within the study areas,
not only does per capita consumption of hard wheat increase,
but the number of households consuming hard wheat also

increases. Table 5-3 shows the breakdown of elasticities by

114
income quartile within each study area. Households in all
study areas except Marrakech become less sensitive to own-

price changes as income increases.

Earls)! _

Barley is an inferior good in Casablanca and urban Sidi
Kacem but a necessity in the both rural areas. In rural
Sidi Kacem, barley has an income elasticity of 0.13. The
only area where barley is a strong necessity is Marrakech.
Barley is the main cereal consumed in Marrakech, the study
area with the highest per capita consumption and the highest
percentage of the population reporting positive consumption
of barley. Unlike the consumption of soft wheat, the
consumption of barley is not clearly delineated by
urban/rural distinction, but rather seems to be
distinguished by particular area of the country.

Although barley is a normal good in rural Sidi Kacem
and Marrakech, it is a Giffen good in urban Sidi Kacem and
Casablanca. This is probably because the income effect is
so strong (highly negative income elasticities in the urban
areas) and the substitution effect is probably weak.

The breakdown of elasticities by income quartile within
each zone (table 5-4) shows barley is an inferior group for
the four income quartiles in all study areas except
Marrakech, the second quartile in rural Sidi Kacem, and the

fourth quartile in Sidi Kacem where the elasticities could

 

115
not be calculated. This means that a subsidy on barley or
an income enhancement will successfully increase the
consumption of barley by the poor. Since barley is a
necessity for all income quartiles in Marrakech and a normal
good, a price decrease on barley would lead to increased
consumption of barley by all income groups. Since Marrakech
is the poorest study area in the subsample, this could be an
effective targeting commodity.

The own-price elasticities of barley in urban Sidi
Kacem and Casablanca indicate that barley is a Giffen good
for all four income quartiles. This is highly suspect since
Giffen goods are rare, especially for the low-income
quartiles. These data should therefore be interpreted with

caution.

Oils

Vegetable oil and olive oil are both necessities in all
study areas: butter is a luxury in all study areas. It was
not possible to estimate significant expenditure
elasticities for olive oil in any study area, but households
appear to be sensitive to the price of olive oil, with
little variation by study area. Olive oil, vegetable oil,
and butter were not found to be substitutes or complements.

The first stage (probit) of the Heckman model (table 5-
11) for vegetable oil has some interesting implications. It

suggests that the price of olive oil is an important

116

determinant of whether or not a household consumes vegetable
oil. The higher the price of olive oil, the more likely a
household is to consume vegetable oil. But once a household
has decided to consume vegetable oil, the price of olive oil
is not a significant determinant of how much vegetable oil
it consumes. This suggests that households would prefer to
consume olive oil, but its price may be too high to be

considered a substitute for vegetable oil.

Table 5-11: Probit Results from the first Stage of
Heckman's Two Stage Procedure

Dependant Variable: Dummy variable: 1 if household
consumed vegetable oil, 0 if not.

 

 

 

 

Independent Variable Coefficient] Standard Srror
Constant -3.04 1.35
Price of vegetable oil 0.26 0.47
Price of olive oil 1.33 0.37
Total expenditure -0.11 0.11
Household size 0.44 0.13
Dummy - Casablanca 1.20 0.25
Dummy - rural Sidi Kacem 0.48 0.18

, Dummy - urban Sidi Kacem 0.72 0.25

 

 

Meats
Chicken and beef are necessities in all four study
areas; mutton is a luxury in all study areas. There is very
little difference between the expenditure elasticities in
the four study areas, confirming the hypothesis presented in

chapter 4 that expenditure elasticities would not vary

117
significantly by study area. Surprisingly, beef is a
complement to chicken in the urban areas but a substitute in
the rural areas. These relationships hold even when
elasticities are calculated by income quartile in each zone
(table 5-8). Beef and mutton are substitutes in all four
study areas. Households in rural Sidi Kacem are more
sensitive to the own price of chicken than households in
other areas: chicken and beef are strong complements in

urban Sidi Kacem.

Sugar

Sugar is a necessity in all four study areas. However,
the expenditure elasticity varies significantly by study
area. In urban Sidi Kacem, for example, the elasticity is
only 0.19, while it is 0.70 in Marrakech. This is not
surprising considering that households in Marrakech spend a
larger percentage of their food budget on sugar than do
households in other study areas. Households in Marrakech
also have higher per capita consumption of sugar than
households in all other study areas except Casablanca. The
own-price elasticity also varies by study area with urban
Sidi Kacem being the least sensitive to own-price changes (-
0.26) and Marrakech being very sensitive (-0.72). Tea was
not found to be a complement to sugar although sugar would

probably have been a complement to tea.

118
Other Studies

The Groupe de Travail Pour l'Estimation des Elasticites
de Demande des Produits Agro-Alimentaire (1990), undertook a
study of food consumption in Morocco using the same data set
used in this study. The study compared the results from a
double-log specification of demand with those from a
modified AIDS for the same commodities studied here.
Although the estimated coefficients were not highly
sensitive to the specification, the inverse Hill's ratio was
not included in the final analyses. Given the large number
of households that reported non-consumption of several goods
(see table 3-2), it is critical to account for the resulting
bias when the censored sample is used. Furthermore,
although elasticities were calculated by urban/rural
distinctions using the results from the double-log
specification, the elasticities were not reported by
specific area, nor were the elasticities by area reported
for the AIDS.

The elasticities from the double-log specification by

urban/rural breakdown are reported in table 5-12 through 5-
15.

119

 

 

 

 

Tole 5-12: Cereal Eleeiciit fro C - 7- a: Treveil (1990) 7
[ Price Soft Price
: Product Nodal Total E - - meat Berle Wheat
' Nerd West Dolble Log .537 -1.004
; AIDS .49? -.839
; Soft Wheat Dolble Log .468 -1.535 .304
'1 AIDS .530 -1.211 .188
i Barley Dotble Log .372 -.861
ll AIDS a, .574 -.705 If

 

 

 

 

 

Tdsla 5-13: . Oil Elastiicitee fro S . Travail (1990)

    

Price Vegetable

 

 

Oil Price
Butter
1 Olive Oil Dolble Log .827 -.158 -1.078 .559
‘ AIDS .685 -.642 -.643 .395
Vegetable Oil Dolbla Log .475 -.935
AIDS .303 -.900
Butter Double Leg .659 V -.813
AIDS

 

 

 

 

 

 

 

Tole 5- 14: Neat Elasticities fro 6 da Trevei l (1990)

 

 

 

 

 

 

 

Double Leg .793 -1 4.10
AIDS .863 -1.310
Mutton Double Log .947 -.986 .312 n
1 nos .957 -1.006 .310
3 Chicken Double Log .641 -.436 ..532
___g __ A ,1 A g __ y E345 -.643

 

120

  

   

 

 

 

 

 

 

 

 

 

 

1.151. 5-15: V - - Elasticities fro-f . - Trevail (1) 7
Total Price Tea Price Price m
E . . . Coffee Sugar
Double Log -.006
AIDS .581 -.737 -.089 -.060
Coffee Double Log .832 .227 -.665 -.778
AIDS .732 .140 -.781 -.816
Sugar Doible Log .514 -.034 -.519
AIDS .581 - .036 - .634
Hill: Double Log .952 -.752
‘ AIDS .849 -1.231

 

The expenditure elasticities estimated in this thesis
tend to be slightly higher than those estimated by the AIDS
and double log specifications of the Groupe de Travail.

This is probably due to the inclusion here of the inverse
Mill's ratio. The estimates of the expenditure elasticities
for barley in the Groupe de Travail models, however, seem to
be an average of the estimates here, illustrating the
importance of estimating elasticities by area: the Groupe de
Travail estimates mask inter-regional differences in
elasticities.

For chicken, mutton, and beef, the price elasticities
estimated here are close to those of the Groupe de Travail.
This is especially true for beef, where the price
elasticities do not vary much by study area. The own-price
elasticity for chicken in the Groupe de Travail models is

roughly the average of those estimated here, again

121
suggesting the importance of estimating elasticities by

area. This is also true of the sugar own-price elasticity.

CHAPTER 6: CONCLUSIONS AND POLICY IMPLICATIONS

Policy Implications

The income and price elasticities presented in chapter
5 indicate that there are regional differences in how
households respond to income and price changes. This
implies that the effect of a uniform national food policy
will vary by region.

Since barley is an inferior good in urban Sidi Kacem
and Casablanca, it is suitable for self-targeting in these
areas. It is also a Giffen good in these areas so the
poorest of the poor should benefit if it is subsidized. A
price subsidy will result in decreased barley consumption,
even for the poor. But the consumption of other foods would
presumably increase. Barley is a necessity in Marrakech and
rural Sidi Kacem, although households in Marrakech are more
income and price sensitive than those in rural Sidi Kacem.
Since Marrakech and rural Sidi Kacem are poor areas, a
subsidy on barley in these areas would benefit most
households (especially in Marrakech where barley is heavily
consumed) with negligible leakage to non-needy households.

Three caveats regarding a subsidy on barley are
necessary. If a household is a net buyer of barley, that
household will benefit from a subsidy on barley to the
extent that the household has access to the subsidized

barley. A net seller of barley, however, may be hurt from a

122

123
subsidy on barley if the producer price is suppressed to
keep the consumption price of barley low. Furthermore, many
households acquire their barley from home production and do
not currently participate in the barley market. A barley
subsidy may then have little effect if the consumption price
is not low enough or the producer price high enough to lure
them into the market. In that case, barley marketing reform
and/or a complimentary producer price policy may be
necessary to bolster market activity.

Second, a subsidy on barley may induce livestock
producers to use barley as animal feed, thus funneling the
benefit of the subsidy to a non-target population.

Third, since barley is mostly nontradable,
implementation of the subsidy is difficult.

A subsidy on hard wheat would also have distributional
consequences by region. Households in Marrakech are the
most sensitive to the price of hard wheat. Presumably, as
the price declined with a subsidy, households in Marrakech
would increase their consumption of hard wheat the most.
Furthermore, even if hard wheat is subsidized, the subsidy
will benefit-wealthier households since they consume more
hard wheat than do the poor, and the percentage of wealthy
households that consume hard wheat is higher than the
percentage of poor households that consume hard wheat. A
subsidy on hard wheat will also leak more than does the

subsidy on soft wheat because there is a larger increase in

124
the consumption of hard wheat as incomes increase than the
increase in soft wheat.

Soft wheat is a necessity in all four study areas, for
all income quartiles. Furthermore, households in all four
study areas and in all income quartiles are highly sensitive
to the price of soft wheat. Although some benefits probably
accrue to the higher-income households, these results,
coupled with those from chapter 4 that show the importance
of soft wheat to low-income households, suggest that the
poor are probably benefiting from the subsidy on soft wheat.
Although the cost of the subsidy program is high, it may be
effective in reaching the poor. A further advantage of
subsidizing soft wheat is that it is fairly easy to
subsidize imports and thus lower domestic prices to
consumers.

If the subsidy on soft wheat is removed, however, it
should be removed gradually and replaced with a highly
visible, better-targeted program. This could include a
subsidy on barley, or a non-price food program such as
supplemental feeding program or food-for-work. Morocco
probably does not have the infrastructure to support a food
stamp program, but the program type could vary by region
depending on the infrastructure available.

Since income and price response does change for some
commodities by study area and by income quartile, a regional

analysis of the these responses using the whole sample

125
should be undertaken. Target populations can then be easily
identified and their responses to price and income changes
anticipated.

Although the subsidies on vegetable oil and sugar are
less costly and less contentious than the subsidy on soft
wheat, some discussion of the effects of removing these
subsidies is warranted. A large percentage of the
population, even among the lower-income households, consume
both sugar and vegetable oil. Sugar is an especially
important budgetary item in Marrakech, where 10 percent of
the household food budget is spent on sugar. Marrakech also
has the highest own-price elasticity of sugar, so a change
in the price of sugar will lead to the largest change in the
consumption of sugar. The poor in Marrakech will probably
be hurt the most if the sugar subsidy is removed.

The expenditure elasticity for olive oil is not
significantly different from zero in urban Sidi Kacem and
Casablanca. The own-price elasticity of olive oil is large
in absolute value, so a subsidy on olive oil would lead to
large increases in consumption of olive oil in all four
areas. The results from the probit analysis in chapter 5
indicate that households may prefer olive oil to vegetable
oil, but given the substantially higher price of olive oil,

households are forced to consume vegetable oil.

126
Methodological Conclusions
Unit Values
As described in chapter 3, using unit values as proxies
for prices in cross-sectional data analysis leads to several
problems. The problems outlined by Deaton (1988b) are:

1) Consumers choose the quality of their purchases,
and the unit values reflect this choice:

2) Quality choice may itself reflect the influence of
prices, as consumers respond to price changes by
altering both quantity and quality: and ‘

3) Measured unit values are contaminated by errors of
measurement in expenditures and in quantities and
are likely to be spuriously negatively correlated
with measured quantities.

The Groupe de Travail (1990) tested the importance of
the "quality effect" by regressing unit values on per capita
income, adult equivalent per person, and study area dummy
variables. The results indicated the quality elasticity is
not significantly different from zero in most cases.

To test whether there were any within-cluster effects
of income on unit values, the same regression as above was
run but with deviations from cluster means for the
independent and dependent variables. This is a more
detailed analysis than that of the Groupe de Travail. The
results indicate no significant within-cluster effects of
income on unit values.

From both of these tests, one can conclude that the

quality effect described by Deaton is not significant and

that unit values are safe to use as proxies for prices.

127
This analysis is fairly crude, however. The quality effect
may be masked by the fact that all of the variables for all
of the commodities are the result of aggregation. Soft
wheat, for example, comprises soft wheat flour, bread made
of soft wheat, and other products made of soft wheat, all
converted to a "grain equivalent." The unit value is
therefore an average unit value for grain products. But
this unit value does not account for the cost of converting
grain into the various products. The unit value is a
function of not only the quality, but also of the
aggregation process. The income effect on unit values may

therefore be obscured.

Zero Consumption

The inverse Mill's ratio was included to account for
the otherwise nonzero mean of the disturbance term that
results when the sample is censored. Despite the large
percentage of households reporting zero consumption of most
of the goods (see table 3-2), the inverse Mill's ratio was
significant in only two demand equations (hard wheat and
butter). Although these are products for which less that 50
percent of households sampled reported positive consumption,
the Mill's ratio was insignificant for other products with
similar percentages of consuming households (mutton and
olive oil, for example). The Mill's ratio was included,

however, even when its coefficient was insignificant.

128
Dropping the Mill's ratio substantially altered the
estimates of the other coefficients in the equations, and in
some cases made the coefficients unreasonable. The
insignificance of the inverse Hill's ratio is probably due
to high correlation between the Mill's ratio and the other
variables in the equations, since the Mill's ratio is itself
a function of those variables. Since a large proportion of
households reported zero consumption of most of the goods,
the censoring problem is probably more important than the
multicollinearity that arises by including the inverse
Mill's ratio.

Model Specification

A functional form similar to that of the AIDS was used
here to estimate demand parameters. It has the advantabe
over the double log that interaction terms are not necessary
to do a regional analysis. This can reduce significantly
the number of coefficients to be estimated. The
modifications were the following: 1) the expenditure term
was not deflated by a price index, 2) the expenditure term
itself represented expenditure on all goods rather than
expenditure on the goods in the system, 3) unit values were
used as proxies for price, and 4) the system was estimated
equation by equation using OLS rather than as a complete
system of demand equations. Several additional variables

other than those specified by the AIDS were included in the

129
model. These included a squared income term (when
significant), regional dummy variables, interaction terms on
the price and expenditure variables (when significant), the
inverse Mill's ratio, and household size. The diversions of
the model here from the AIDS make it difficult to justify
the model based on the theoretical soundness of the AIDS.
The AIDS was, however, the theoretical starting point for
estimating the model in this way. The modifications made
were ones to make the model more realistic and feasible.
The addition of the squared income term, for example,
allowed goods to be necessities for one group and luxuries
or inferior for another, a distinction not possible with the
AIDS. Barley was found to be a necessity for some groups
and inferior for others. Although the theoretical soundness
of the AIDS is not preserved, the modified linear AIDS can
be considered an alternative functional form whose
coefficients may be compared to those from other functional
forms.

There are several advantages and disadvantages to the
modified linear AIDS. One of the disadvantages is that
since the elasticities are a function of the coefficients of
the model, the estimation of the elasticities requires
additional calculations. If the theoretical soundness of
the AIDS is not preserved, it is not clear that the modified
AIDS yields any benefits that warrant the cost of the extra

calculations required.

130

The modified AIDS is, however, another functional form
that can be used to estimate elasticities. These can then
be compared to estimates from other specifications such as
the double-log to test the robustness of the elasticities.
This may be the major benefit of using the AIDS or a
functional form similar to the AIDS, since it is difficutl
to estimate slope dummies very well with the double log

because there are so many variables.

GAUSS Program

Although the model can be estimated using any standard
statistical package such as SAS or SPSS, it is more
expedient to use a flexible program such as GAUSS. The
additional calculations can be coded into the program to be
updated automatically. Calculation of the elasticities by
study area, by income quartile, by both, or by any other
criterion is simple once the program has been written.
Furthermore, the variance-covariance matrix of the
coefficients needs to be adjusted when the Mill's ratio is
included to account for the fact that the Mill's ratio is a
predicted variable, not an observed one. This adjustment is
not possible using SAS or SPSS but is easily coded into the
GAUSS program. Further modifications or respecifications
can be easily accomplished using the GAUSS program. GAUSS
can also be used to estimate a nonlinear system in the case

of imposing symmetry and homogeneity. It can be used to do

131
maximum likelihood estimation instead of Heckman's two stage
procedure. GAUSS is time consuming to learn, but powerful

once mastered.

Regional Analysis

Accounting for regional differences in consumption
patterns was accomplished by three means: 1) including
regional dummy variables in the regression equations, 2)
including regional interaction terms on price and income to
allow the slope coefficients to vary by study area, and 3)
estimating elasticities with study area budget shares.

The interaction terms were insignificant for all
commodities except chicken and hard wheat. Regional dummy
variables were significant in all equations. The major
source of difference in elasticities between the study areas
was the different budget shares in each study area.

The greatest difference in price and expenditure
elasticities by study area was for barley, hard wheat, and
sugar. If further analysis is done using the whole national

sample, detailed regional analysis would be advised.

Further Research
The National Household Budget Survey of 1984/85 is a
rich data set that provides many opportunities for further

research. The following is a non-exhaustive list of

132
exercises that could be done using either the subsample used
here or the national sample.

The data could be used to estimate the tobit model
using maximum likelihood estimation. The results could be
compared to the those from the Heckman two-stage procedure
to see how divergent the results are.

One could also estimate a true AIDS and compare the
results with the modified version. Homogeneity and symmetry
could be imposed and compared with a system without
homogeneity and symmetry imposed. Estimating the AIDS would
require calculating the price index and using non-linear
estimation techniques on the system of demand equations. It
would also require including the prices of all goods in the
system in each equation, and recalculation of the budget
shares and expenditure term. This exercise would best be
done on the entire data set.

This analysis assumes that the coefficients of items
such as shelter, education, and clothing that are not in the
system are not significantly different from zero. It would
be interesting to see how the results change if these
consumption goods are included in the model.

This analysis has ignored the production or supply side
of the economy. Since many households are producers of the
products they consume, it would be more thorough to have an

analysis of how producers would be affected by changes in

133
food prices. The net effect on households could then be
determined.

Deaton (1988a, 1988b) discusses the problems associated
with using unit values as proxies for prices. He argues
that the measurement problem is the most important problem
resulting from using unit values as proxies for prices.
Correction for the measurement error could be done following
the methods employed by Laraki (1988).,

The National Household Budget Survey of 1984/85 was
sampled with a complex sampling design including both
stratification and clustering. Yet the formulas used here
to calculate the standard errors of the coefficients assumes
simple random sampling. One could use a program such as PC
CARP that accounts for the sampling design to see if it made
any difference in the standard errors.

Finally, the elasticity estimates calculated here could
be used to do a more formal welfare analysis of the subsidy
programs. Compensating and equivalent variation could be
calculated to see how much households would be willing to

pay, or be compensated, for alternative programs.

134

APPENDIX 2'1: "ADDING-UP" RESTRICTION IRITT- IN TERMS
OP BUDGET SNARE AND ELASTICITY

The "adding-up" restriction implies that

2:331: '3 x

Taking the derivative of this budget constraint with respect

to x: 8g
2317‘”

Multiplying the left-hand side by
k

IQ

xlxlgql

yields:

 

 

since

we can see that:

23wggk= 1

 

135

APPENDIX 3-1: OTHER DEMAND MODELS

The plethora of demand models available makes the
choice of functional form an extremely difficult one for the
researcher. Each functional form offers certain advantages
over others, but all entail some restrictions. The
discretion of the practitioner must be used in deciding
which characteristics are most important. This section
briefly outlines some commonly used models and highlights
their advantages and disadvantages. While the list is in no
way exhaustive, it may be useful as a sampling with which
to compare the AIDS. The double logarithmic approach
represents a single equation approach while the LES, the
indirect translog and the Rotterdam models all generate

complete systems of demand equations.

Double Logarithn
The double logarithm is a member of the single equation

category and is usually expressed as:

n

logqiarcx1 +eilogx+g eulogn (5'13)

where e1 is the total expenditure elasticity and en is the

cross-price elasticity of the kth price on the ith demand.

136

137
Although the double logarithmic equation is easy to
estimate and yields coefficients that are easy to interpret,
it violates "adding-up," and testing symmetry is not
possible. Furthermore, the justification for the choice of
functional form is ad hoc. The double log specification
can, however, be made more flexible by adding a quadratic

income term and income price interaction terms.

Linear Expenditure System (LBS)

The LES is a complete system of demand equations and is
derived by maximizing a particular utility function subject
to the budget constraint. The specific utility function is
the Klein-Rubin utility function given by:

U';; 911°9(91'71) (5'14)

'where 01 and y; are parameters.

Maximization of this utility function subject to a budget

constraint yields the following demand equation:

15.

q1371+(— i(173:?ij (5‘15)

Written in expenditure form, this becomes:

138

p1q1=P171+pi (Y’; ijJ) (5'16)

where:

p1 =- price of commodity i.

q, =- quantity of commodity i,

ya = subsistence or committed quantities of commodity i,

y = total expenditure,
B,:- marginal share of the budget spent on commodity i.

The LES may be interpreted as follows: the y; are the
minimum quantities of good i. Expenditure on the ith
commodity thus consists of some minimum expenditure (p171)
after which the minimum expenditure on all other j goods (2
IHYJ) is subtracted from total income. The remainder is
spent on the ith good according to the marginal share of the
budget spent on commodity i (the 01's).

Because the model is derived from a direct utility
function, all the classical restrictions on the demands will
hold. Namely, they will satisfy "adding-up," "homogeneity,"
and "symmetry." Yet it must be noted that the choice of
this particular utility function has several implications.
No two goods may be complements, every good must be a

substitute for every other good, and inferior goods cannot

be modeled.

139
Indirect Translog

The indirect translog is obtained by specifying the
indirect utility function and deriving the demand equations
by applying Roy's Identity21 The indirect translog is the
second-order approximation to an arbitrary, twice
differentiable indirect utility function.

This model is flexible and is consistent with demand
theory. However, the derived demand equations are
non-linear in their parameters and estimation requires
sophisticated software. Furthermore, the number of
unrestricted and unknown structural parameters is high and

may pose estimation problems (Johnson, et al., 1984).

Rotterdam
The Rotterdam model is a flexible specification of
demand equations based on the direct differentiation of the
general form of Marshallian demand equations to which the

results of utility theory are then applied.

 

21'Roy's Identity allows one to obtain the Marshallian
demand functions from the indirect utility function by the
following:

6v

391-
av_-q3(xnp)

ax

 

where:
v is the indirect utility function

140
One of the restrictive features of the Rotterdam model
is that it implies that the income elasticities for

necessities increase as the expenditure share decreases.

141

APPENDIX 3-28 TRANSPORMING THE AIDS COST FUNCTION INTO
BUDGET SHARES

Shephard's Lemma states that, where they exist, the
partial derivatives of the cost function with respect to
prices are the Hicksian demand functions, that is:

6c(U.p) =1h(u

I ) =
P1 p q:

multiplying both sides by

 

 

P: .
c(u,pi'
yields:
5c(u.p)
W . pg.
_g C(11)?)
P1
This may also be written as:
blogc(u,p) :w
lilogp1 *

Taking the derivative of the AIDS cost function in the text
with respect to log p1 will yield wi.

 

142

APPENDIX 3-38 CALCULATION OP EXPENDITURE AND PRICE
ELASTICITIES

Expenditure Elasticities

The estimated equation for the AIDS without squared income

term is:

I
"1"“; +23: yulogpfflllog (x) 10% +1tilogCU+an1+8izDz+513D3

or

P1 91
X

 

I
=01”; yulogpfflilog (x) +15% -1~1t,log(.'U+8nD1+812D2+813D3

taking the derivative with respect to x:

 

b(fzgi) 5X'
T’Bt—x'

P: x 591 ‘P1 91 0x. 51!
1—

x2 x

p1(xaqi -qi 5X) 2 B 5x
1

143

8x
pibqi-EquL_=Bibx

5x
p,6q,=.€i.qx1__+310x

32*. P35 + Bi 6"
g: 1! Pi qi

 

bq1
<3: B.
1.? “v;
7

The standard error of the expenditure elasticity without

squared term:

 

S” I‘Jvar({;+ l)

i

 

144

 

= \Jvarbg—i) + var(1) + 2cov(%:)(1)
i

 

The estimated equation for the AIDS with squared income term

is:

0W0)
9(x’0)

3 3 3
+ nilogCU-k; 51.17;“; 2; CuD‘logpJ +2; 1: 1,Dglogx

W. = auZJjndloqpfifluloqu)+B.ztloq(X)12w

piqi
x

 

I
.-.- 111+; 711109P1+Bnl°9(x)+Biz[1°g(x) 124,0 24:23;

a a
+ nilogCU+2 8111);: 2; (1,0210ng +2; tizDzlogx
z- J z-

3
z-

145

taking the derivative with respect to x:

P1 qi
M x ) bx

T'gflu ? I 2312(109 If)???

 

x6 - q 6x 5 5
pi qixzpi i _.._ fli 7x + 2612 (109 x)?!

P1 (X bqi - qi OX)

x = [31 8x + 2812 (log x)8x

 

p. x 69. _ p. 9. 5x

x x I B, bx + 2012 (log x)8x

Pi qi 5"
x

p1 8g1 = + 81 8x + 2812 (log x)6x

q1 8x + 8, 5x + 2012 (log x)8x

b =
q1 x P1 P1

 

 

 

192 = _6_x + [31 6x + 21312 (log x)6x
g; X P1 q1 p1 qi

 

 

146

 

 

 

a
(11 X X
=1+ +20 (loqu )
Ti 31 (P1 qt) 12 p1 qi
X
32
q: = Bi 2512
_J_f 1 + _w_1 + w, log 2:
x

The standard error of the expenditure elasticity with

squared term:

 

2912 51 2512
“’1 log x) +2cov(.;.1.)( w, log 11:)

 

 

S.‘ = Jvu(%) + var(
i

 

= \J_1_'2var(fli) + %(log X)z V8113“ )+4—]-'-(%-g—J-‘COV(511 )(Biz)
wi W1 wi

 

= _17(/var(Bi) + 4(log it)2 warm“) + 4 log x cov([311 )(Biz)
1

147
Own- and Cross-Price Elasticities
The estimated equation for the modified AIDS when the
income term is not deflatd by the price index is:

w1= 411+: yulogpf-Bilog (1:) +0 00: __.g_)_::) +1tilogCU1-6nDl+512D2+613D3

P191
X

 

= «1+;yulogpd1-Bilog(x) +o%(__(xx,g) 1-1tilogCU1-6HDI+612Dz-1-61303

P1q1

5( )

 

5P1

—;(P1591 1’ 915M) 3 711—
P1

a
(0.60. + 0.00.) -- (ya—51)::

i

9 fig
T-i F: I q1P1 = 711"

P1

148
P15q1

131— : 713x - q1pi
P1

dividing both sides by qdn:

55E

q, -=13£-—1
‘51- W1

77

By similar calcuation, the cross-price elasticity is e1d '

7.,
V1

The standard error of the own-price elasticity is:

l

"1

where on is the standard error of the own-price coefficient

from the original equation.

The standard error of the cross-price elasticity is:

1

W1

149
where 013 is the standard error of the cross-price

coefficient from the original equation.

APPENDIX 3-4 3

Table A3-4:

150

GROUPE DE TRAVAIL RESULTS

Regressions of Unit Values

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

W
Log of
q Household
Log of Size in Dummy
Unit Per Cap Adult Var. Dummy Dummy
Value Expend. EqLuiv. CASA Var SKR Var. SKU
Hard 0.028 0.027 0.138* -0.127* -0.172*
Wheat 0.028 0.034 0.053 0.060 0.045
Soft 0.006 -0.112*’ 0.301* -0.152 -0.365*
Wheat 0.023 0.027 0.037 0.050 0.034
Barley' 0.039 -0.034 0.140* 0.201* -0.059
0.025 0.029 0.035 0.067 0.066
Mutton 0.023* 0.008 ”-0.0831' -0.029 0.000
0.012 0.014 0.018 0.023 0.017
Beef 0.023 -0.008 1~0.088* -0.032 -0.014
3.580 0.011 0.015 0.019 0.015
Chicke 0.041* 0.047*1-0.248* -0.059 0.224*
n 0.017 0.020 0.028 0.036 0.025
Veg 0.007 -0.010 1-0.l48* -0.033 -0.026*
011 0.008 0.009 0.013 0.018 0.012
Olive. 0.012 -0.024 1~0.073* -0.099* -0.102*
011 0.012 0.013 0.018 0.027 0.018
Butter’ 0.036* 0.003 1~0.277* -0.172* 0.021
0.018 0.022 0.028 0.037 0.029
Tea 0.048* 0.005 1-0.045* 0.127* 0.020
0.013 0.0151 0.020 0.028 0.018
Coffee 0.052* 0.046* <-0.187* 0.033 0.005
0.016 0.019 0.026 0.035 0.024
Sugar' 0.021* -0.011. F0.179* -0.228* -0.117*
0.008 0.010 0.013 0.019 0.012
Fresh -0.017 -0.010 1-0.092* 0.063* -0.l72*
Milk 0.011 0.012 0.029 0.028 0.013
Note: Standarc error is be ow coeIEIcient.

 

 

CASA = Casablanca

SKR = rural Sidi Kacem

SKU = urban Sidi Kacem

* indicates significant at least at the 5% level.

151

APPENDIX 3-5: CLUSTERING AND STRATIPICATION

The simplest type of sampling is simple random sampling
(SRS) in which every element in the population has an equal
chance of being selected for inclusion in the sample. The
sampling error of survey estimators reported in textbooks
and calculated by computer programs are based on the
assumption of simple random sampling (with replacement) of
the population. Simple random sampling is rarely used in
practice, however, usually due to its high costs of
implementation. Sampling errors based on SRS may thus over-
or understate the true sampling error, depending on the type
of sampling used.

Two other types of commonly used sampling procedures

are stratification and clustering.

Stratification

Stratification involves the classification of the
pepulation into subpopulations, or strata, based on
supplementary information about the population. A sample is
then drawn from each stratum. It is desirable that each
stratum be internally homogeneous since only a portion of
each stratum will be included in the sample.

The sample may be drawn by either proportionate
sampling, in which the sample drawn is proportional to the

size of the stratum, or disproportionate sampling, in which

152
the sampling fraction varies across strata. Proportionate
stratification generally yields estimators which are more
precise than those from SRS because stratification ensures
that the different strata in the population are accurately
represented in the sample. Disproportionate sampling is
used when some strata are much more variable than others.
Higher-variability strata would be sampled at a higher rate
to capture the diversity, while a smaller fraction of more
homogeneous strata would suffice. The cost of sampling the
different strata also influences the sampling rate: higher
cost strata are sampled at a lower rate, lower cost strata
at a higher rate. While proportionate sampling must lead to
estimators which are no less precise than those from SRS,
disproportionate sampling may or may not lead to more
precise estimators. Standard errors based on proportionate
sampling will tend to overestimate the standard error,

while those based on disproportionate sampling may be over-

or underestimated.

Clustering
Clustering involves the division of the population into
groups from which a sample of groups will be drawn. All
elements of a selected group will then be included in the
sample. If the groups are further divided into smaller

groups from which a selection is made, the process is called

153
two-stage cluster sampling. Multi-stage cluster sampling
involves the division into several layers.

Clustering is usually undertaken as a means of
economizing on resources. If clusters are delineated
geographically, for example, choosing a sample of clusters
will require less geographic coverage by enumerators than
would a SRS of the same size. However, clustering leads to
a loss in precision compared to SRS. The loss of precision
arises because clusters tend to miss whole groups of
elements. If the population is the population of a state,
for example, and the clusters are counties from which a
selection of counties will be drawn, people living in the
same county are more likely to be like each other than they
are to be like people in the rest of the state. The more
heterogeneous the population within the cluster, the less
will be the loss of precision: the less heterogeneous the
greater will be the loss of precision. In general, however,
clustering tends to underestimate the standard error ofthe
estimators.

When clusters are different sizes, if each cluster has
the same probability of being drawn, the size of the final
sample is largely dependent upon which clusters are drawn.
If the cluster sizes are substantially different, this could
result in a wide range of possible sample sizes. In order
to reduce the variability in sample size, probability

proportional to size (PPS) is often used in selecting

154
clusters. The probability of choosing any one cluster is
proportional to its size. Thus if one cluster is twice the
size of another, the probability of choosing the former is

twice that of choosing the latter.

Design Effect

The design effect is a measure of the divergence
between the variance of an estimator based on a particular
survey design to the variance of the estimator based on a
SRS of the same size. The design effect of any estimator z
is designated 02V(z)/V(z°) . A sample estimate of the design
effect for z is designated d2(z) . A design effect greater
than one implies that the variance of the estimator is
greater with the particular survey design than it would have
been under SRS.

Most surveys employ a combination of both
stratification and clustering, in which case the effect on
the precision of the estimators is ambiguous since
stratification leads to equal or more precision than SRS
while clustering leads to less. However, Kalton (1983, p.
77) advises that empirical findings from sampling error
computations that have been performed for estimates based on
complex survey designs demonstrate that the loss of
precision from clustering tends to outweigh the gains from
proportionate stratification, so that complex sample design

provides less precise estimators than an unrestricted sample

155
of the same size: that is, the design effects are greater
than 1. The use of sampling error formulae thus tend to
overstate the precision of survey results based on complex

survey design.

156

APPENDIX 3-6: SAMPLING METHODS USED POR THE MOROCCAN DATA
SET

Urban Areas

The primary sampling units for the urban areas (PSU-U)
were defined as geographical zones delimited by natural
borders, each containing 3-4 districts from the 1982
population census. Each PSU-U has an average of 600
households.

After stratification by type of dwelling (luxury,
modern, new medina, medina, shantytown, douars, small
center), a sample of 536 primary units was chosen by a PPS
selection procedure.

Each urban primary unit chosen was divided into
secondary sampling units (SSU-Us) with an average size of 50
households and having well defined boundaries. Each PSU-U
had an average of 12 SSU-Us. Three SSU-Us were chosen from
each PSU-U using PPS selection process.

A list of households in each secondary unit was
established with the address of the household, first and
last names of head of household, size of household,
professional situation of head (working salaried, working
non-salaried, unemployed having never worked, unemployed).
The households lists were then stratified by profession of
the head of household and size of household. Five
households were chosen from each SSU-U following the

principles of proportionate sampling.

157
Rural Areas

The primary sampling unit in the rural areas (PSU-R) is
a zone formed by a number of douarsu'and has an average
size of 1000 households. These units are based on the 1982
census. Stratification of the PSU-Rs was done using
geographic and agro-economic criteria. 432 PSU-Rs were then
selected using PPS.

Douars were chosen as secondary sampling units for the
rural areas (SSU-R).

Three douars were chosen in each PSU with PPS. As in
the urban areas, households in the chosen douars were listed
with information on the first and last name of the head of
household, the size of household, the first and last name of
the eldest son or daughter, and profession of the head of
household. After stratification of households of each douar
of the sample according to the profession of the head of
household and the size of household, five of households were
chosen using a system of proportionate sampling.

In order to capture the effects of seasonal
fluctuations on the structure of consumption, the data
collected covered an entire year (each household was
surveyed over a one week period) with the survey equally

spread over the course of the year.

 

.n A douar is defined as a group of households

connected by social factors (tribe, extended family, etc.)
or economic factors (agriculture, sharing land worked on)
and are under the authority of one person, the moqadem.

158
The entire survey consisted of 14,520 households in
seven different economic regions. The analysis in this
thesis uses a subsample of the original survey. The
subsample contains approximately 1000 households in 3

provinces.

159

APPENDIX 4-18 UNIT VALUES

7 Table A4-1: Unit Value by Sty

 

 

 

 

 

 

  

 

  

 

   

 

   

 

  

 

 

  

 

    

 

 

 

 

Urban Rural 1.
Product Sidi Sidi Marrakech Casablanca
Racem Races k
Mutton 31.3 32.4 32.6 31.2 i
(DH/kg) (1.4) (1.8) (3.6) (7.1) I
Beef 41.8 41.8 42.5 40.7 1
(DH/kg) (3.6) (2.8) (3.7) (9.1) 1
Chicken 23.6 31.3 24.3 19.5 '
(DH/kg) (7.8) (10.0) (6.4) (4.6)
Hard Wheat 2.1 1.9 2.3 256 ;
(DH/kg) (0.9) (0.9) (0.9) (0.9) 1
Soft Wheat 1.9 1.4 2.1 2.7 E
(DH/kg) (0.9) (0.7) (0.9) (0.9) |
Barley 2.1 1.6 2.0 2.2 1
(DH/kg) (0.3) (0.5) (0.9) (0.7) 3
Vegetable 6.3 6.3 6.5 5.8 i
011 (DH/kg) (0.4) (0.4) (0.5) (1.3) ;
Olive 011 15.7 15.7 17.2 16.6 :
(DH/kg) (1.8) (3.2) (1.6) (3.9) :
Butter 23.9 28.8 28.1 22.1
(DH/kg) (4.8) (5.1) (5.1) (6.2)
Coffee 41.4 39.2 38.7 35.7
(DH/kg) (8.9) (7.3) (6.8) (12.1)
Tea 56.5 49.8 48.2 49.4
(DH/kg) (9.8) (9.9) (7.3) (13.6) 5
Fresh Milk 3.3 2.6 3.1 2.8 9
(DH/litre) (0.3) (0.5) (0.2) (0.3)
Sugar 3.5 3.8 4.2 3.7 5
Wk?) <0-5> <-> -4 «W J
F gures n parentheses are standar- errors.‘

 

 

” Unit values are calculated by dividing expenditure
on a commodity by the quantity consumed of that commodity.

 

 

 

160

APPENDIX 4-2: noussuonn size a! swan! AREA AND INCOME
generics

Table A4-2: Households Size by Study Area
andIncome Quartile

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

[ __ _. _ 7
i Income Mean Number of 1
1 Study Area Quartile Persons in Household |;
‘Urban 1 5.9 f
‘ Sidi Kacem (2.2) fl
1 2 5.0 3
i (3.2) ‘
; 3 5.1 f
i ' (2.3) I
1 4 4.0
J (3.0) g
éRural 1 8.1
1 Sidi Kacem (3.3)
J 2 7.4
i (3.6)
i 3 6.8
1 (3.1)
: 4 5.9
J (4.3)
lMarrakech 1 6.8 j
i (2.7) 1
2 6.9 i
(3.7) x
3 6.0 ,
(3.0) 1
4 4.6 ;
(2.8) i
Casablanca 1 7.1 i
(3-2) i
2 6.7 4]
(3.3)
3 5.6

 

 

 

 

 

Noto?‘* 7

 

Cereals
Soft Wheat:

Hard Wheat:

Barley:

Beef:

Chicken:

APPENDIX 4-3:

grain
bread
flour
semoule
couscous
pasta

grain
bread
flour
semoule
couscous

grain
bread
flour
semoule
couscous

boned
with bones
minced

live
not live
roasted

161

CONVERSION FACTORS

1.00
1.00
0.76
0.68
0.68
1.00

1.00
1.00
0.79
0.68
0.68

1.00
1.00
0.79
0.68
0.68

1.00
0.75
0.75

0.50
1.00
1.00

162

APPENDIX 5-1: VARIANCE-COVARIANCE MATRIX WHEN MILL'S RATIO

INCLUDED
Tobit model:
. if Y. > 0 I l
= 1 1 Y ==X +-u 11~N’ 2
Y1 {Y0 otherwise 1 15 i ' 1 (0'0)

MXIY)

E(Y|Y>O) =x’ +0
‘ ‘ ”3 9114(1)

==XfB +<uq

e {6
where
= B
7 a
r 4(va
1- _—_7_-
¢Wx17)
1x1
91:
F1

 

 

asyVar (02515) = 0211

where

A = (6:61)" - (6(611‘101111. - A.X.(x'm)"x{A.J¢.(G{G.)"

163

 

 

1 - Al 0 0 1
o 1 - 12 0
A1 = .
1 0 0 1 ' 4'11,
717; 0 °
0= 0 7275 0
L o 0 . 7.711.

 

 

 

 

9!
G1 = 5
91¢,
Xi
X 2
X a 1 8 XI:
x,J ‘
.XN’J

 

 

I
y; . iN187)

 

1408’?)
N1=# (Y1>0)
No=#(Y1=0)

N=N1+No

164

APPENDIX 5'2: PROEIT RESULTS PROM PIRST STAGE OP HECXMAN

PROCEDURE
5.12m
PARAMETER ESTIMATE STANDARD ERROR
CONSTANT 0.5480 0.0589
LZABT -0.0286 0.0010
LZABD 0.0182 0.0012
LZAOR 0.0079 0.0017
LDPTP -0.0905 0.0128
LDTPER2 0.0039 0.0007
LUNICON -0.0082 0.0012
SIGMA 0.0088 0.0056
emu
PARAMETER ESTIMATE STANDARD ERROR
CONSTANT 0.3058 0.0253
LZORGE 0.0134 0.0023
LDPTP -0.0283 0.0015
LUNICON -0.0131 0.0042
DUMCASA -0.0264 0.0065
DUMSKR -0.0332 0.0136
DUMSKU -0.0303 0.0081
DUMXOR 0.0246 0.0119
SIGMA -0.0257 0.0177
MEAT
PARAMETER ESTIMATE STANDARD ERROR
CONSTANT 0.6368 0.0200
LZABD -0.0311 0.0053
LDPTP -0.0625 0.0020
LUNICON -0.0280 0.0027
DUMSKU -0.2531 0.0227
DUMSKR -0.0926 0.0210
DUMCASA -0.3600 0.0198
DUMXBD 0.0267 0.0046
OWNPSKU 0.0447 0.0052
OWNPSKR 0.0579 0.0032
OWNPCASA 0.0479 0.0045
LDTPSKU 0.0269 0.0025
LDPTPSKR 0.0091 0.0026
LDPTPCASA 0.0399 0.0021
SIGMA -0.0137 0.0065

 

165

W

PARAMETER ESTIMATE STANDARD ERROR
CONSTANT 0.1116 0.0195

LZHLT 0.0057 0.0067

LDPTP -0.0115 0.0009

LUNICON -0.0108 0.0025

DUMCASA 0.0080 0.0051

DUMSKR 0.0146 0.0035

DUMSKU 0.0110 0.0037

SIGMA 0.0182 0.0181
m1

PARAMETER ESTIMATE STANDARD ERROR
CONSTANT 0.2878 0.1681

LZHLO -0.0001 0.0056

LDPTP -0.0220 0.0145

LUNICON -0.0317 0.0158

DUMSKU 0.0210 0.0439

DUMSKR 0.0144 0.0218

DUMCASA 0.0206 0.0436

SIGMA -0.0546 0.0560

cm

PARAMETER ESTIMATE STANDARD ERROR
CONSTANT -0.0606 0.0791

LZBR 0.0015 0.0024

LDPTP 0.0057 0.0070
LUNICON 0.0062 0.0071

DUMCASA -0.0001 0.0058

DUMSKR 0.0040 0.0015

DUMSKU -0.0009 0.0043

SIGMA 0.0306 0.0161

m

PARAMETER

CONSTANT

LZPOUL
LZVNDB
LDPTP
LUNICON
DUMSKU
DUMSKR
DUMCASA
CPISKU
CP1CASA
SIGMA

PARAMETER
CONSTANT

LZVNDB
LZVNDM
LDPTP
LUNICON
DUMCASA
DUMSKR
DUMSKU
SIGMA

m

PARAMETER
CONSTANT

LZVNDM
LZPOUL
LDPTP
LDTPER2
LUNICON
DUMCASA
DUMSKR
DUMSKU
SIGMA

ESTIMATE

0.0109
0.0175
0.0185
-0.0101
-0.0097
0.2099
0.0209
0.1364
-0.0582
-0.0372
-0.0010

ESTIMATE

0.1280
-0.0183
0.0133
-0.0069
0.0022
0.0139
0.0011
0.0121
-0.0144

ESTIMATE

-0.4059
-0.0025

0.0183

0.0964
-0.0055
-0.0049
-0.0089
-0.0014
-0.0145

0.0031

STANDARD ERROR

0.0049
0.0003
0.0022
0.0003
0.0004
0.0069
0.0005
0.0088
0.0020
0.0024
0.0013

STANDARD ERROR

0.0561
0.0073
0.0070
0.0044
0.0056
0.0067
0.0017
0.0052
0.0227

STANDARD ERROR

0.0478
0.0004
0.0009
0.0088
0.0004
0.0008
0.0017
0.0013
0.0020
0.0032

 

PARAMETER

CONSTANT

LZSUCRE
LDPTP
LUNICON
DUMCASA
DUMSKR
DUMSKU
SIGMA

ESTIMATE

0.1773
0.0152
-0.0167
-0.0115
-0.0137
-0.0111
-0.0249
0.0319

167

STANDARD ERROR

0.0185
0.0083
0.0017
0.0026
0.0027
0.0029
0.0035
0.0295

168

LI§I_QI_!ABIAELE§

CP1CASA Interaction between LZPOUL and DUMCASA
CPlsKU Interaction between LZPOUL and DUMSKU
DUMCASA Dummy variable for Casablanca

DUMSKR Dummy variable for rural Sidi Kacem

DUMSKU Dummy variable for urban Sidi Kacem

DUMXBD Dummy variable for own production of hard wheat
DUMXOR Dummy variable for own production of barley
LDPTP Per capita total expenditure

LDPTPCASA Interaction between LDPTP and DUMCASA
LDPTPSKR Interaction between LDPTP and DUMSKR
LDPTPSKU Interaction between LDPTP and DUMSKU
LDTPER2 Per capita total expenditure squared
LUNICON Log of number or adult equivalents

LZABT Log of unit value for soft wheat
LZABD Log of unit value for hard wheat
LZAOR Log of unit value for barley

LZBR Log of unit value for butter

LZHLO Log of unit value for olive oil
LZHLT Log of unit value for vegetable oil

LZPOUL Log of unit value for chicken

LZSUCRE Log of unit value for sugar

LZVNDB Log of unit value for beef

LZVNDM Log of unit value for mutton

OWNPCASA Interaction between LZABD and DUMCASA
OWNPSKR Interaction between LZABD and DUMSKR

OWNPSKU Interaction between LZABD and DUMSKU

SIGMA Coefficient on inverse Mill's ratio

169

BIBLIOGRAPHY

Alderman, H. "The Effect Food Price and Income Changes on
the Acquisition of Food by Low-Income Households."
IFPRI report. Washington, D.C.: World Bank. May, 1986.

Alderman, H., and J. van Braun. "The Effects of the Egyptian
Food Ration adn Subsidy System on Income Distribution
and Consumption." IFPRI report #45. Washington, D.C.:
World Bank. July, 1984.

Blanciforti, L., and R. Green. "An Almost ideal Demand
System Incorporating Habits: An Analysis of
Expenditures on Food and Aggregated Commodity Groups."

65(1983):511-515.

Buse, R.C., and J.P. Chavas. "The Effects of Food Stamps on
Food Expenditures: Comment." Ale- 72(1990)4:1081-1083.

Capps, O., and J. Havlicek. "Concepts of Consumer Demand
Theory." in “ ‘

 

. R. Raunikar, and C. L. Huang,
(eds.). Ames, Iowa: Iowa State University Press, 1987.

Capps, O., J.R. Tedford, and J. Havlicek. "Household Demand
for Convenience and Nonconvenience Foods."

American
19urnal_9f_A9risultural_Econemisa- (1985)=862-869-

Cox, T. L., and M. K. Wohlgenant. "Prices and Quality Effects
in Cross-Sectional Demand Analysis."

of_Asriculfural_Ecen9mics November. 1986. PP 908-919-

Deaton, A. "Estimation of Own- and Cross-Price Elasticities

From Household Survey Data." lggrnal_gf_£ggngme§;ig_.
36(1987):7-30.

Deaton, A. "Quality, Quantity, and Spatial Variation of
Price: Estimating Price Elasticities from Cross-
Sectional Data." LSMS Working Paper. World Bank. 1988a.

Deaton, A. "Quality, Quantity, and Spatial Variation of
Price." AER. 78(1988b)3:418-430.

Deaton, A., and A. Case. "Analyis of Household
Expenditures." The World Bank Living Standards
Measurement Study, Working Paper No. 28. Washington,
D.C. 1988.

170

Deaton, A., and J. Muellbauer. "An Almost Ideal Demand
System." AER. 70(1980a)3:312-326.

. Cambridge:

 

Cambridge University Press. 1980b.
Direction de la Statistique, Ministere du Plan. Consommation

 

Roxanne du Maroo-

El-Eraky, M. B. "Analysis of Food Consumption in Egypt: A
Demand Systems Approach." Ph. D Dissertation. Michigan
State University. 1987.

Green, R., and J.M. Alston. "Elasticities in AIDS Models."
5153. 72(1990)2:442-445.

Groupe de Travail Pour l'Estimation des Elasticities de
Demande des Produits Agro-Alimentaires. "Analyse de la
Consommation Alimentaire au Maroc." Working Paper No.
1. September, 1990.

Heckman, J.A. "The Common Structure of Statistical Models of
Truncations, Sample Seletion and Limited Dependent
Variables and a Simple Estimator for Such Model."

WW- 5(1976)475-
492.

Heckman, J.A. "Sample Selection Bias as a Specification
Error." Eggngmgtzigg. 47(1979)1:153-161.

Johnson, S. R., A. A. Hassan, and R. D. Green.
. Ames, Iowa: The
Iowa State University Press. 1984.

Johnson, S. R., R. D. Green, z.A. Hassan, and A. N. Safyurtlu.

"Market Demand Functions." in EQQQ.D§DAD§.ADAI¥§1§1

MW. 0- Capps and B
Senauer (eds. ). Sponsored by the S-165 Southern

Regional Research Committee and the Farm Foundation,
1986.

Kalton, G. "Introduction to Survey Sampling." Sage
University Paper series on Quantitative Applications in
the Social Sciences. Series no. 07-035. Beverly Hills
and London: Sage Publications. 1983.

Kennedy, E.T. "Alternatives to Consumer-Oriented Food
Subsidies for Achieving Nutritional Objectives." flood
si 'es ' ev ntries' C s ' 5 nd
£911§y_92tign§. ed. P. Pinstrup-Anderson. Baltimore:
The Johns Hopkins University Press. 1988.

171

Kmenta, J. Elemen;§_gf_figgngmetrigs. New York: Macmillan
Publishing Company. 1986.

Kumar, S.K., and H. Alderman. "Food Consumption and
Nutritional Effects of Consumer-Oriented Food
Subsidies." E99d_Suhsidies_in_nexelening_99untriesi

. ed. P. Pinstrup-
Anderson. Baltimore: The Johns Hopkins University
Press. 1988.

   
 
  

    

Laraki, K. . T,. < . ” _ we: I ‘
. h.D. Dissertation.

  

Cornell University. 1988.

Moschini, G., and K.D. Meilke. "Modeling the Pattern of
Structural Change in US Meat Demand." Ameriggn_gggzna;

9f_Agrisultural_Esoncmics- 71(1989)=253-261.

Mudbhary, P.K. "A Demand System Analysis of Food Consumption
in Nepal." Ph.D Dissertation. Michigan State
University. 1988.

Pakpahan, A. "Food Demand Analysis in Urban West Java." Ph.D
Dissertation. Michigan State University. 1988.

Payne, R. "Food Deficits and Political Legitimacy: The Case
of Morocco." in Africals_Agrarian.£risisl_The_Eeets_9f
Famine. eds. S.K. Commins, M.F. Lofchie, and R. Payne.
Boulder, Colorado: Lynne Rienner Publishers, Inc. 1986.

Pinstrup-Andersen, P. "The Social and Economic Effects of
Consumer-Oriented Food Subsidies: A Summary of Current
Evidence." E99d_Subsidies_in_DexelQPing_Ceuntriesl
C9sts1_Benefitsi_and_zelisx_QPtiens. ed- P- Pinstrup-
Anderson. Baltimore: The Johns Hopkins University
Press. 1988.

Pinstrup-Andersen, P., and H. Alderman. "The Effectiveness
of Consumer-Oriented Food Subsidies in REaching
Rationing and Income Transfer Goals." Food Subsidies in

v s: sts '
thigns. ed. P. Pinstrup-Anderson. Baltimore: The Johns
Hopkins University Press. 1988.

Pitt, M.M. "Food Preferences and Nutrition in Rural

Bangladesh-" Esxieu_9f_Economics_and_§tatistis§. 65(Feb
1983):105-ll4.

Pollak, R.A., and T.J. Wales. "Estimation of Complete Demand
Systems from Household Budget Data: The Linear and
Quadratic Expenditure Systems." e e ' m’c
Bgyigg. 68(1978):348-359.

172
Raunikar, R., and Chung-Liang Huang. Eggd_pgmdnd_Ang1y§1§;
i2r9blemsl_Issues1_and_EmPirical_Exidense

. Iowa: Iowa
State University Press. 1987.

Ray, Ranjan. "Analysis of A Time Series of Household
Expenditure Surveys for India.” Thg_fiey1gy_df
. 62(1980):595-602.

Rogers, B.L. "Design and Implementation Considerations for
Consumer-Oriented Food Subsidies." Egdd_§nb§1die§_in
. ed. P. Pinstrup-Anderson. Baltimore: The Johns
Hopkins University Press. 1988.

Rogers, B.L., and A.J. Swindale. "Determinants of Food
Consumption in the Dominican Republic." Report prepared
cooperatively by Tufts University School of Nutirtion
and USAID Office of Nutrition. Medford. 1988.

Royaume du Maroc, Ministers du Plan, Direction de la
Statistique. "Enquete Nationals sur la Consommation et
les Depenses des Manages 1984-1985 - Rapport
Methodologique." Rabat. February, 1988.

Sawadogo, K. "Food Consumption Analysis and Related
Parameters of Ouagadougou, Burkina." Apprdpxidtg
- . H.W.
Ohm and J.G. Nagy (eds.). West Lafayette: Purdue
University Press. (1985):338-350.

Savadogo, R., and J. Brandt. "Household Food Demand in
Burkina Faso: Implications for Food Policy. "

Agricultural_Eceaniss 2(1988)= 345- -364-

Swamy, G., and H. P. Binswanger. "Flexible Consumer Demand
Systems and Linear Estimation: Food in India." Amezidgn

l9urnal_9f_Asrisultural_Esonomics 65(1983)= 675- -684-
Theil, H., C. F. Chung, and J. L. Seale. Adydnde§_in

01-117 1- 09' 3\ ._-, ‘ 0! 09:--119 I!
Patterns. Supplement 1. Connecticut: JAI Press Inc..
1989.

Timmer, C. P., and H. Alderman. "Estimating Consumption
Parameters for Food Policy Analysis."

of_Agrisultural_Econemics 61(1979)= 982-987

Timmer, C.P., W.P. Falcon, and S.R. Pearson. Edod Poligy

Addlygis. Baltimore: The Johns Hopkins University
Press. 1983.

173

Tobin, J. "Estimation of Relationships for Limited Dependent
Variables." EEQDQEQIIiEd- 26(1958)24-36.

Tuluy, H.A., and B.L. Salinger. "A Comparative Study of the
Political Economy of Agricultural Pricing Policies."
World Bank Working Paper. June. 1988.

Valdes, A. "Explicit versus Implicit Food Subsidies:
Distribution of Costs." MW

WW ed. P-
Pinstrup-Anderson. Baltimore: The Johns Hopkins
University Press. 1988.

Wohlgenant, M.K. "Conceptual and Functional Form Issues in
Estimating Demand Elasticities for Food."

American
WW- 66(1984)=2'11-215-

"‘111111111111