MINISTRY OF EDUCATION AND TRAING

UNIVERSITY OF ECONOMICS HCMC

--------------

LE THI TUYET THANH

INFORMATIONAL ASYMMETRIES

AND THE DEMAND FOR

VEGETABLE IN HCMC

MAJOR: DEVELOPMENT ECONOMICS

CODE: 9310105

PH.D. DISSERTATION

EXECUTIVE SUMMARY

HO CHI MINH CITY 2019

INTRODUCTION

This study is motivated by the asymmetric information problem

on HCMC vegetable market, in which a large proportion of

vegetable does not comply with safety and hygiene standards. In

most cases, consumers can not recognize the safety attributes of

vegetable, and thus can not distinguish the conventional

vegetable and safe vegetable. This results in two problems. One

is that producers may apply unsafe production techniques to

maximize their own profits (moral hazards). The other is safe

vegetable can be driven out of the market being unable to

compete with conventional vegetable (adverse selection). These

two problems result in the failures of safe vegetable market in

HCMC. Both buyers and distributors have taken measures to

overcome these problems.

Lacking information, consumers try to seek information about

safety and hygiene standards of vegetable to be able to buy safe

vegetable. The prevalent information channels in HCMC are

internet, television, and newspaper, which can affect consumer

behavior and thus influence the development of safe vegetable

market. Understanding the effect of these channels on the

demand for safe vegetable, the preferences for the safety

attributes, and vegetable store choice are therefore important to

the safe vegetable distributors in expanding their market share.

The results of this study also help policymakers in providing

information and selecting the appropriate channels to develop the

market for safe vegetable.

The safe vegetable distributors, especially those of modern

distribution channels, confronted many difficulties in competing

with conventional vegetable in the context that buyers can not

distinguish the two types of vegetable. The first problem is how

to set vegetable prices in such a way that can compensate for the

high production and distribution costs and at the same time,

accepted by consumers. The second one is which safety

standards are important to consumers in order to have the

relevant development strategy. The last one is what store

attributes are critical to vegetable buyers in order to attract more

shoppers.

RESEARCH OBJECTIVES AND METHODOLOGY

Stemming from the above problems, this study proposes the

following research objectives.

Objective 1: Analyze the impact of information and informationseeking behavior of safety and hygiene standards on the demand

for safe vegetable. The research results help sellers realize the

response of consumer to the safe vegetable prices in order to develop

reasonable pricing strategy. The results also show the effect of

information on the demand for safe vegetable, thereby assisting

policymakers in providing information and selecting appropriate

information channels to promote the safe vegetable market. This

goal is solved by estimating the demand equation system for both

conventional and safe vegetables, with Linear Approximation

Almost Ideal Demand System (LA-AIDS), controlling for

endogeneity of prices and zero demand.

Objective 2: This ojective analyses the effect of information and

information-seeking behavior of safety and hygiene standards on

the willingness to pay for vegetable attributes. This is to have

the suitable policy in improving quality and safety attributes to

attract more buyers. The findings provide information for

vegetable distributors to develop price strategy in conjunction

with appropriate marketing procedures to maximize profits. This

objective applies Choice Experiment method to estimate the

willingness to pay for safe vegetable attributes, using

Conditional Logit (CL) and Mixed Logit (MX) models.

Objective 3: Investigate the influence of information and

information-seeking behavior of safety and hygiene standards on

the vegetable store choice. Other factors under consideration

include the distribution channel attributes and consumer

characteristics. The Multinomial Logit Model (MNL) is used to

analyze the impact of buyer characteristics on the store choice

decision of buying vegetable, while the Conditional Logit model

is used to explore the effect of distribution channel attributes on

the store choice decision. This study help sellers to perceive

factors that need to be improved in order to attract more

consumer. This study also helps to examine the target customer

group of each vegetable distribution channel, as well as find out

which information channel are mostly effective to appeal target

customer.

The conceptual framework of this study is summarized in Figure 1.

BUYER

CHARACTERISTICS

▪ Individual

characteristics:

age,

gender,

education,

occupation, bargaining

habit, number of vegan

days

▪ Household

characteristics: income,

number

of

meals,

household size, number

of childrens, elders.

▪ Market

price

(endogenous).

LA-AIDS

6 vegetable groups

▪ Frequency of accessing information

via TV, newspapers and internet

▪ Number of poisonings cases watched

▪ Number of poisoning cases

experienced by family members in the

past 12 months

MNL

models

CL/MX

models

Certification

Safety guarantee

Store format

Packaging & information

Price

Objective 3

Store

choice

Interaction

SAFE ATTRIBUTES

OF VEGETABLE

▪

▪

▪

▪

▪

Demand for

vegetable

INFORMATION

Interaction

DISTRIBUTION

CHANNEL

CHARACTERISTICS

Price

Distance

Freshness

Diversity

Input control

Preprocessing

Providing information

Safety level

Objective 1

Objective 2

CE with CL/MX

models

Figure 1: Conceptual framework.

WTP for

safety

attributes

The study uses survey data of 320 people who bought vegetables

in HCMC in 2018. The information factors and informationseeking behavior of buyers is the main explanatory variable in all

the three research objectives. Specifically, the information of

poisoning and safety and hygiene standard violation, as well as

the frequency of monitoring safety and hygiene standards

through communication channels, were included to analyze the

impact on safe vegetable demand, WTP for safe vegetable, and

store choice behavior.

OBJECTIVE 1: INFORMATION AND THE DEMAND

FOR VEGETABLE

Identification of vegetable groups

The study classifies vegetables on the market into 3 groups:

leafty vegetables (including flower and stem groups), root

vegetables and fruits (vegetable family). Thus, the estimated

equation system will comprise 6 equations: 3 equations for the

conventional vegetables of leaves, roots and fruits, and other 3

equations for the safe vegetables that eat leaves, roots and fruits.

Then measuring the average weekly quantity demanded of

vegetable groups in the past 3 months.

Prices and the issues of missing prices and endogeneity

Prices in this objective has an endogeneity problem caused by

the zero demand that would result in no information of prices,

and the heterogeneity in quality of vegetable, which results in the

simultaneity between expenditure share and prices.

To solve the endogeneity, prices are initially regressed on the

variables representing the survey location and household

characteristics. Households are surveyed in the same period

(several days) and location (these households supposedly under

the same cluster) have identical price level for each type of

vegetable, and the diffference of price among the households (if

any) because of variations in the quality and measurement errors.

The predicted price from sub-regression model would eliminate

these differences. The auxiliary regression is

ln 𝑝𝑟𝑖𝑐𝑒𝑖𝑘 = 𝛼0 + 𝛼1 𝐶𝑘 + 𝛼2 𝐷𝑘

(1)

where 𝑝𝑟𝑖𝑐𝑒𝑖𝑘 be the price of vegetables 𝑖 (there are 6 categories,

𝑖 = 1,2, … ,6, and 6 sub-regression equations respectively) for

household 𝑘, 𝐷𝑘 be the characteristics of buyers and households

𝑘, 𝐶𝑘 be the dummy variables describe the cluster. The cluster is

determined by ward/commune. The households surveyed in the

same ward/commune will be considered in the same cluster. The

variables of characteristics of buyers and households are shown

in Table 1.

Bảng 1: Characteristics of buyers and their families

Variable

Definition

Number of meals/week The average number of home-cooked

meals per week

Age

Age of vegetable buyer

Household

size Number of members who regularly eat at

(persons)

home in the household

Number of childrens

The number of childrens under 6 in the

household

Number of elders

The number of peoples over 60 in the

household

Household

income Total income of household members

(mil. VND/month)

Number of vegan The average number of vegan days per

days/month

month of buyer

Gender (1 = Male)

Dummy variable, the gender of vegetable

buyer, 1 = Male

Bargain (1 = Yes)

Dummy variable, 1 = Having bargain

habit while purchasing

Occupation (*)

Unskilled worker (baseline)

Office worker

Manager

Skilled worker

Homaker

Student

Others

Education (*)

Finished primary or less (baseline)

Finished secondary

Finished high school

Finished college

Finished university or higher

The adjusted price would be:

̂ 𝑖𝑘 = 𝛼

̅

ln 𝑝𝑟𝑖𝑐𝑒

̂0 + 𝛼

̂𝐶

̂𝐷

(2)

1 𝑘+𝛼

2 𝑘

̅𝑘 be the household characteristics, taking the average

where 𝐷

value of sample. The category variables are assigned by the most

frequent values in the sample, whereby the dummy variables of

education and occupation in the model are made for high school

and homemaker. The predicted price ln 𝑝𝑟𝑖𝑐𝑒

̂ 𝑖𝑘 would be used

into the model.

Zero demand and sample selection bias

The safe vegetable groups might have a significant number of

households with zero demand. As a result, the quantity demanded

or the expenditure ratio which is censored from 0 and the normal

estimation method could be bias. This goal employs two-step

estimation method of Heckman (1976) to deal with this problem.

According to this method, in step 1, the consumer is assumed to

decide whether consume each type of vegetable or not:

𝑑𝑖𝑘 = 𝛼𝐷𝑘 + 𝛾𝐼𝑘 + 𝑢𝑖𝑘

(3)

where 𝑑𝑖𝑘 be a dummy variable indicates whether household 𝑘

consumes vegetable 𝑖 or not, 𝐷𝑘 and 𝐼𝑘 be the exogenous

variable explains for the choice of consuming vegetable, in

which 𝐷𝑘 be the characteristics of individual and household, and

𝐼𝑘 be the information variables are shown in Table 2.

Bảng 2: The frequency of tracking safety and hygiene standards

information and poisonings

Variable

Tracking safety and

hygiene

standards

information via TV

Tracking safety and

hygiene

standards

Definition

The frequency of tracking safety and

hygiene standards information via TV (*):

Under once/month (baseline)

More than once/month

More than once/week

Per day

The frequency of tracking safety and

hygiene standards information

via

newspaper(*):

Under once/month (baseline)

information

newspaper

via

Tracking safety and

hygiene

standards

information

via

internet

Number of poisoning

cases watched per

month

Number of violations

of safety and hygiene

standards watched per

month

Number of poisonings

cases experienced in

12 months

More than once/month

More than once/week

Per day

The frequency of tracking safety and

hygiene standards information

via

internet(*):

Under once/month (baseline)

More than once/month

More than once/week

Per day

The number of food poisoning cases

watched per month, on average in the past 6

months

The number of violations of safety and

hygiene standards watched per month, on

average in the past 6 months

The number of poisoning symptoms of

family members experienced in the past 12

months

Then the demand (or expenditure) function would be defined as

𝑞𝑖𝑘 = 𝛽𝑋𝑖𝑘 + 𝜀𝑖𝑘

if 𝑑𝑖 > 0

𝑞𝑖𝑘 = 0

if 𝑑𝑖 ≤ 0

where 𝑋𝑖𝑘 is a vector of explanatory variables for demand

function would be described in the following section. The

expected quantity demanded of demand (or expenditure)

function with distributed assumption would be

𝐸(𝑞𝑖𝑘 |𝑋𝑖𝑘 , 𝑑𝑖 > 0) = 𝛽𝑋𝑖𝑘 + 𝐸(𝜀𝑖𝑘 |𝑢𝑖𝑘 > −𝛼𝐷𝑘 − 𝛾𝐼𝑘 )

(4)

and:

𝐸(𝜀𝑖𝑘 |𝑢𝑖𝑘 > −𝛼𝐷𝑘 − 𝛾𝐼𝑘 ) =

−𝛼𝐷𝑘 −𝛾𝐼𝑘⁄

𝜎2)

𝜎2 1−𝛷(−𝛼𝐷𝑘 −𝛾𝐼𝑘⁄𝜎 )

2

𝜎12 𝜙(

The demand function becomes

(5)

𝐸(𝑞𝑖𝑘 |𝑋𝑖𝑘 , 𝑑𝑖 > 0) = 𝛽𝑋𝑖𝑘 +

−𝛼𝐷𝑘 −𝛾𝐼𝑘⁄

𝜎2)

−𝛼𝐷

−𝛾𝐼

𝑘

𝑘⁄ )

𝜎2 1−𝛷(

𝜎2

𝜎12 𝜙(

(6)

where 𝜎2 is the standard deviation of 𝜀𝑖𝑘 and 𝜎12 is the

covariance between 𝜀𝑖𝑘 and 𝑢𝑖𝑘 .

Thus, this method ogriginally estimates the Probit model

explaining the choice of whether consume a type of vegetable or

not (i.e. estimating the coefficient 𝛼 in 𝑑𝑖 = 𝛼𝑍𝑖 + 𝑢𝑖 , then

calculating the Inverse Mills Ratio (IMR):

−𝛼𝐷 −𝛾𝐼

𝐼𝑀𝑅𝑖 = 𝜆 (

𝑘

𝑘⁄ )

𝜙(

𝜎2

−𝛼𝐷𝑘 − 𝛾𝐼𝑘⁄

𝜎2 ) = 1−𝛷(−𝛼𝐷𝑘−𝛾𝐼𝑘⁄ )

(7)

𝜎2

Then estimating the demand (or expenditure) function with IMR

𝑞𝑖𝑘 = 𝛽𝑋𝑖𝑘 + 𝜑𝐼𝑀𝑅𝑖 + 𝜀𝑖𝑘

(8)

where 𝜑 =

𝜎12

𝜎2

.

In this study, the equation system has 6 demand functions for

three types of vegetable that eat leaves, tubers and fruits, each of

these with conventional and safe type. Consequently, 6 Probit

models would be estimated correspondingly to 6 types of

vegetable. Following the calculation for 6 IMR variables for each

type of vegetable and using 6 estimated IMR variables for

measuring the LA-AIDS demand equation system.

Estimation methods

Demand function indicates quantity demanded that maximizes

utility and as a result of utility maximization problem. In addition

to estimating single demand equation, it is possible to estimate

the demand equation system. Estimating the single demand

equation has the critical limitation that is the separating goods

from related ones. This is an unreasonable assumption.

Moreover, the single demand equation will not help to show the

correlation among the goods, for example, cross-price elasticity

of demand can not be calculated. Thus, the demand equation

system with related goods is more preferred.

The dominant models are used to estimate demand equation

system including Linear expenditure system, Rotterdam System,

and Almost Ideal Demand System.

The AIDS model is the most regularly used, with two variants

LA-AIDS and QUAIDS. However, QUAIDS is more difficult to

estimate, especially the demand equation system has a censoring

problem with many cases of quantity demanded equals to 0.

While LA-AIDS is easier to estimate, despite the important

disadvantage is that the log of expenditure ratio is assumed to be

a linear function of the log of total expenditure. However, this is

not significant indeed, because of log-log formulation, the

expenditure ratio fulfills nonlinear function with respect to the

total expenditure. Solely the expenditure ratio is compelled to be

a square function of total expenditure – i.e. expenditure ratio

could be deviated when total expenditure reaches a certain level

- then LA-AIDS is virtually inappropriate. Consequently, the

study uses LA-AIDS to estimate the demand equation system of

vegetables.

LA-AIDSis developed by Deaton and Muellbauer (1980) from

the expenditure function:

1

ln 𝑒(𝑢, 𝑝) = 𝛼0 + ∑𝑖 𝛼𝑖 ln 𝑝𝑖 + 2 ∑𝑖 𝛾𝑖𝑗 ln 𝑝𝑖 ln 𝑝𝑗 + 𝑢𝛽0 ∏ 𝑝𝑖 𝛽𝑖

(9)

The estimated equation system would be:

𝑤𝑖 = 𝛼𝑖 + ∑𝑗 𝛾𝑖𝑗 ln 𝑝𝑗 + 𝛽𝑖 ln

𝑦

𝑃∗

+ 𝜑𝑖 𝐼𝑀𝑅𝑖 + 𝜖𝑖

(10)

where ln 𝑃 ∗ = ∑𝑗 𝑤𝑗 ln 𝑝𝑗

Note that the price variable used in this model is the predicted

price from the OLS models, 𝐼𝑀𝑅𝑖 is the Inverse Mills Ratio

estimated from the regression results of Probit models.

The demand function system in this objective has 6 equations,

however only 5 equations are estimated. The coefficient of the

6th equation would be computed by the theoretical constraints:

•

Additivity: ∑𝑖 𝛼𝑖 = 1, ∑𝑖 𝛽𝑖 = 0, ∑𝑖 𝛾𝑖𝑗 = 0

•

Homogeneity: ∑𝑗 𝛾𝑖𝑗 = 0

(with all i)

•

Symmetry: 𝛾𝑖𝑗 = 𝛾𝑗𝑖

Therefore, these constraints in the fact that are imposed on this

objective. Some studies also examined these constraints, but this

is not actually essential, because after testing if the constraints

are not satisfied, these constraints also are applied to this goal.

Elasticities

The price elasticity of demand or cross-price elasticity of demand

for commodity could be calculated by the estimated coefficients

from the demand equation system. There are 2 elasticities: noncompensated elasticity (referred to as “elasticity”, which is the

Marshallian demand) and compensated elasticity (Hicksian

elasticity). Marshallian elasticity is computed by the assumption

that total expenditure for vegatables in the demand equation

system be unchanged, while compensated elasticity is calculated

by the assumption that the utility is constant (and thus total

expenditure might be varied). Practically, the vegetable prices

change the consumer can adjust expenditure level for vegetables.

Therefore, compensated elasticity might have the greater

practical implications.

There are many procedures to calculate elasticity, but according

to Green and Alston (2008), the following calculation (applied to

LA-AIDS) is the most basically precise for Marshallian elasticity

𝜂𝑖𝑗 = −𝛿𝑖𝑗 +

𝛾𝑖𝑗

𝑤𝑖

− 𝛽𝑖

𝑤𝑗

𝑤𝑖

(11)

where 𝛿𝑖𝑗 is the Kronecker indicator, in which 𝛿𝑖𝑗 = 1 if 𝑖 = 𝑗

and 𝛿𝑖𝑗 = 0 if 𝑖 ≠ 𝑗, 𝛾𝑖𝑗 and 𝛽𝑖 is the estimated coefficients from

(10), 𝑤𝑖 is the expenditure ratio for vegetable 𝑖. The compensated

elasticity (Hicksian) be

∗

𝜂𝑖𝑗

= 𝜂𝑖𝑗 + 𝑤𝑗 (

1+𝛽𝑖

𝑤𝑖

)

(12)

OBJECTIVE 2: INFORMATION AND WTP FOR SAFE VEGETABLE

In this objective, the study evaluates the WTP for safety

attributes and investigating the effect of information on WTP

based on the Choice Experiment (CE). This research chooses two

vegetables, water spinach and carrots, to conduct the experiment

because the characteristics and storage time of two vegetables are

very different, thus the safety attributes might differ.

The experiment uses 5 attributes. The definition and attribute

levels are shown in Table 3.

Table 3: Attributes and their levels

Attribute

Definition

Price

% higher than

price in Choice 1

Vegetables are

sold in different

stores

Store format

Safety

certification

Vegetables are

granted by safety

standards

certificates

Safety

guagrantee

The

seller

commits

to

compensate

VND300 million

if the toxin

content exceeds

the prescribed

threshold

Packaging,

manufacturer

information and

traceability

stamps

Packaging

and

information

Level

of Level

of

Alternative 1 Alternative 2

0%

50%, 150%,

300%, 500%

Formal

Formal market

market

Supermarket

Safe vegetable

stores

No

No

certification

certification

VietGAP

certification

Organic

certification

No

safety No

safety

guagrantee

guagrantee

Safety

guagrantee

No

packaging, no

information,

no

traceability

stamps

Packaging,

manufacturer

information

Packaging,

manufacturer

information

and

traceability

stamps

With two choices (in which Choice 1 be fixed), 5 attributes and

levels in Table 3, it would be possible generate 4×3×3×2×2=144

alternatives. The study utilizes the interaction effect orthogonal

fractional factorial design to select 60 alternatives for

interviewing the buyers. 3 alternatives are grouped into 1 block,

there are 20 blocks respectively. Multiplying these 20 blocks

with 16 cases, the total is 320 blocks. Each respondent should be

answered 1 block for water spinach and 1 block for carrots, thus

each respondent answers a total of 6 choice tasks.

Model and estimation methods

With the database pointing out the preferences from the

hypothetical choice scenario mentioned before of two choices

𝑗 = 1,2, the model can estimate utility function:

𝑈𝑗 = 𝛽𝑋𝑗 + 𝜀𝑗

(13)

where 𝑋𝑗 is a set of attributes of choice 𝑗 described in Table 3 and

𝑈𝑗 is the utility of choice 𝑗 and 𝜀𝑗 is the residuals.

Phương trình ước lượng

This study initially estimates the basic utility function as shown

in (13)

𝑈𝑖𝑗 = 𝛽1 𝑝𝑟𝑖𝑐𝑒𝑖𝑗 + 𝛽2 𝑝𝑟𝑖𝑐𝑒𝑖𝑗 × 𝑖𝑛𝑐𝑜𝑚𝑒𝑖 + 𝛽2 𝑠𝑢𝑝𝑒𝑟𝑖𝑗 +

𝛽3 𝑠𝑝𝑒𝑐𝑖𝑎𝑙𝑡𝑦𝑖𝑗 + 𝛽4 𝑣𝑖𝑒𝑡𝑔𝑎𝑝𝑖𝑗 + 𝛽5 𝑜𝑟𝑔𝑎𝑛𝑖𝑐𝑖𝑗 +

𝛽6 𝑔𝑢𝑎𝑟𝑎𝑛𝑡𝑒𝑒𝑖𝑗 + 𝛽7 𝑖𝑛𝑓𝑜𝑖𝑗 + 𝛽8 𝑞𝑟𝑐𝑜𝑑𝑒𝑖𝑗 + 𝜀𝑖𝑗

(14)

where 𝑈𝑖𝑗 is the utility of buyer 𝑖 obtained from the choice 𝑗 (𝑗 =

1,2), 𝑝𝑟𝑖𝑐𝑒𝑖𝑗 is the prices of choice 𝑗 in the choice set of buyer 𝑖.

Note that 𝑝𝑟𝑖𝑐𝑒𝑖𝑗 is the specific amount with the unit of thousand

VND/kg, not the percent (%). 𝑠𝑢𝑝𝑒𝑟𝑖𝑗 and 𝑠𝑝𝑒𝑐𝑖𝑎𝑙𝑡𝑦𝑖𝑗 is dummy

variables indicating the store be the supermarket and safe

vegetable store (baseline: formal market). 𝑣𝑖𝑒𝑡𝑔𝑎𝑝𝑖𝑗 and

𝑜𝑟𝑔𝑎𝑛𝑖𝑐𝑖𝑗 is dummy variables denoting safety certificate of

vegetables of choice 𝑗 (baseline: no certification). 𝑔𝑢𝑎𝑟𝑎𝑛𝑡𝑒𝑒𝑖𝑗

is dummy variable equals to 1 if the seller commits to

compensate VND300 million in the case that the toxin content

exceeds the prescribed threshold. 𝑖𝑛𝑓𝑜𝑖𝑗 vand 𝑞𝑟𝑐𝑜𝑑𝑒𝑖𝑗 is

dummy variables presenting the having packaging and

manufacturer information 𝑖𝑛𝑓𝑜𝑖𝑗 , and having packaging,

manufacturer information and traceability stamps 𝑞𝑟𝑐𝑜𝑑𝑒𝑖𝑗 .

Assumme that these two attributes be prefferred to the consumer,

then 𝛽7 > 0, 𝛽8 > 0 and 𝛽8 > 𝛽7 , because the utility with

traceability stamps (𝛽8 ) must be higher than the utility without

traceability stamps.

With the expectation that those with higher income are less

sensitive to prices, the interaction between price and income is

introduced into the utility function. If the expectation is correct,

then 𝛽1 < 0 and 𝛽2 > 0.

Among the attributes, this study is particularly interested in the

safety attributes, including VietGAP certification and organic

certification. Additionally, the compensated guarantee of sellers

also is a kind of certificate (the self-certification of sellers),

because the sellers have to comply with safety standards of

vegetables even though hazards for compensation. To examine

the impact of information on WTP for safety attributes, the

information variables (Table 2) would be interacted with these

safety attributes.

If the information variables interact with 3 safety attributes are

simultaneously put into (14) which can cause the

multicollinearity. Therefore, each model can only include the

interaction variables of an attribute. As a result, 4 models are

estimated:

• The basic model: solely the attributes and the interaction

between price and income (14).

• VietGAP model: the basic model with the interaction

variables between VietGAP certification and information

variables.

• Organic model: the basic model with the interaction variables

between organic certification and information variables.

• Guarantee model: the basic model with the interaction

variables between safety guarantee and information variables.

These models are estimated by Maximum Likelihood. The

parameters 𝛽 are estimated by maximizing the log-likelihood

function:

𝐿𝐿 = ∑𝑖 ∑𝑗 𝑦𝑖𝑗 𝑝𝑖𝑗

(15)

𝑒

where 𝑝𝑖𝑗 = ∑

𝛽𝑋𝑗

𝑙𝑒

𝛽𝑋𝑙

is the probability of choosing 𝑗 by buyer 𝑖,

and 𝑦𝑖𝑗 is dummy variable indicating the observed choice from

the buyer, 𝑦𝑖𝑗 = 1 if the buyer 𝑖 selects choice 𝑗.

MX model is more complicated. In this model, log-likelihood

function is identical as CL, but the probability of choosing 𝑗 by

the buyer 𝑖 becomes:

𝑝𝑖𝑗 = ∫ (

∑

𝑒

𝑙

𝛽𝑋𝑖𝑗

𝑒 𝛽𝑍𝑖𝑙

) 𝑓(𝛽)𝑑(𝛽)

(16)

where 𝑓 (𝛽) be the probability distributed function of 𝛽, because

𝛽 is a random variable instead of fixed variable as in CL. To

estimate 𝛽, the model will choose randomly 𝑅 values from 𝑓 (𝛽)

and calculate the average probability from 𝑅. 𝑅 be the number of

draws, the larger 𝑅 the more accurate estimated parameters. The

research of Bierlaire (2003) shows that 𝑅 = 500 is the most

effective value, thus this study uses 𝑅 = 500.

Estimation of WTP

The parameter 𝛽 describes the marginal utility of attributes. For

example, the parameter 𝛽𝑣𝑖𝑒𝑡𝐺𝐴𝑃 of 𝑉𝑖𝑒𝑡𝐺𝐴𝑃 variable shows the

utility of VietGAP certificate. Comparing this utility to the

marginal utility of prices help to calculate the willingness to pay

for VietGAP certificate attribute:

𝛽

𝑊𝑇𝑃𝑉𝑖𝑒𝑡𝐺𝐴𝑃 = − 𝛽4

𝑝

(17)

where 𝛽𝑝 is the marginal utility of price (in thousand VND). In the

basic model, the marginal utility of 1 thousand VND to the buyer 𝑖

be:

𝛽𝑝 = 𝛽1 + 𝛽2 𝑖𝑛𝑐𝑜𝑚𝑒𝑖

(18)

𝑖𝑛𝑐𝑜𝑚𝑒𝑖 is the income of buyer 𝑖. If calculating at the mean value

of income, the marginal utility of price becomes

𝛽𝑝 = 𝛽1 + 𝛽2 𝑖𝑛𝑐𝑜𝑚𝑒

̅̅̅̅̅̅̅̅̅̅

(19)

where 𝑖𝑛𝑐𝑜𝑚𝑒

̅̅̅̅̅̅̅̅̅̅ is the average income of the sample.

OBJECTIVE 3: INFORMATION AND VEGETABLE STORE

CHOICE

This objective has two specific research objectives: (1) analyzing

the impact of buyer characteristics on the vegetable store choice

and (2) investigating the effect of distribution channel

characteristics on the vegetable store choice.

This study uses MNL and RUM models to identify the factors

influence store choice of buying vegetable. The collected data

shows each trip of buying vegetable in the last 7 days in detail

and thus each trip of buying vegetable is an observation – i.e. one

time of decision-making. Each trip of buying vegetable is

supposed to be exclusively travel to a unique store and therefore

the alternatives are mutually exclusive.

From the results of initial surveys and pilot surveys, the store of

buying vegetable is chosen to study as follows:

1. Hypermarket/Supermarket

2. Mini supermarket

3. Specialty store

4. Traditional market (legally established)

5. Street vendor (illegally established)

6. Local green grocery.

MNL Model

The assumption that the utility of consumer 𝑖 of choosing the

store 𝑚 from the shopping trip 𝑗 be a linear function as

𝑚

𝑈𝑖𝑗𝑚 = 𝛼𝑚 𝐷𝑖 + 𝛽𝑚 𝐼𝑖 + 𝜀𝑖𝑗

(20)

where 𝑚 = 1,2, … ,6 indicating the distribution channels

mentioned above, 𝐷𝑖 be a vector of the individual and household

characteristics, and household 𝑖 consists of variables shown in

Table 1 (The characteristics of individual and household), 𝐼𝑖 be

the information variables including the frequency of tracking

safety and hygiene, the number of safety and hygiene standards

violations and the number of poisoning cases experienced

depicted in Table 2.

The parameters 𝛼𝑚 and 𝛽𝑚 presenting the influence of

determinants on the utility of store choice 𝑚. Note that each of

store would have a vector of parameters. Each of 𝐷𝑖 variable

would have 𝛼1 defining the utility of extra supermarket to

individual with 𝐷𝑖 characteristics, 𝛼2 be the utility of minimart,

and similar to other stores.

Let the probability of consumer 𝑖 from the trip 𝑗 of choosing the

store 𝑚 be:

𝑝𝑖𝑗𝑚 = Pr(𝑦𝑖𝑗𝑚 = 1)

(21)

𝑚

where 𝑦𝑖𝑗

be the dummy variable defines the store choice 𝑚 of

the buyer 𝑖 from the trip 𝑗. Seting formal market is the base

category, the log-odds would be:

log

𝑝𝑖𝑗𝑚

𝑝𝑖𝑗4

= 𝛼𝑚 𝐷𝑖 + 𝛽𝑚 𝐼𝑖

∀𝑚 ≠ 1

(22)

𝑝1

𝑚

Note that log 𝑝𝑖𝑗1 = 0 and 𝜀𝑖𝑗

is assumed to be the logistic

𝑖𝑗

distribution. The probability of choosing the store 𝑘 would

become:

𝑘

𝑘

𝑝𝑖𝑗

=

𝑉

𝑒 𝑖𝑗

(23)

𝑚

𝑉

∑6𝑚=1 𝑒 𝑖𝑗

where 𝑉𝑖𝑗𝑚 = 𝛼𝑚 𝐷𝑖 + 𝛽𝑚 𝐼𝑖 . The MNL model will estimate the

parameters 𝛼𝑚 and 𝛽𝑚 for each of store (exclude 𝑚 = 4) by

Maximum Likelihood method:

log 𝐿 = ∑𝑖 ∑𝑗 ∑𝑚 𝑦𝑖𝑗𝑚 ln 𝑝𝑖𝑗𝑚

(24)

This model helps to analyze the influence of individual and

household characteristics, and the frequency of tracking safety

and hygiene information on the store choice of buying vegetable.

RUM

RUM model is applied to determine the impact of distribution

channel attributes on the store choice of buying vegetable.

Table 4: Attributes of vegetable stores

Variable

Definition

Price

The price indicator of the closest

store of each channel compared to

the closest formal market (the

price of the closest formal market

= 0%, the negative value in this

variable means that the price of

the closest store lower than the

formal market)

Measurement

units

%

Distance

Freshness

Diversity

Input control

Preprocessing

Providing

information

Safety level

Distance from the house to the

closest store (outlet) of each

distribution channel

Vegetables at each shopping

channel have different levels of

freshness

There are many types of

vegetables, the diversity of spicies

are sold at each distribution

channel

Whether vegetable stores have

strict input control or not

Vegetables at each channel before

being sold are preprocess:

cleaning, trimming, washing with

clean water, and packaging

The

level

of

providing

information on vegetable quality

(Expired date, the origin of

vegetables, processing, safety

standards, etc.) of each shopping

channel

Personal evaluation of buyers on

the probability of vegetables

fulfilling safety standards at each

channel

Minutes

1 = Freshness;

0

=

No

freshness

1 = Diversity;

0

=

No

diversity

1 = Strictly;

0

=

Not

strictly

1 = Carefully;

0

=

Not

carefully

1

=

Completely;

0

=

Not

completely

%

Model specification

This model will estimate the utility function

𝑈𝑖𝑗𝑚 = 𝐴𝑆𝐶𝑚 + 𝜌𝑋𝑚 + 𝜀𝑖𝑗𝑚

(25)

where 𝑋𝑚 is a vector of store attributes 𝑚. The attributes include

distance, diversity, preprocessing, freshness, providing

information and price. The definition of attribues are shown in

Table 4. These variables are the personal evaluation of buyers for

the store attributes of buying vegetable.

Interaction terms

Those with higher income might have different responses to the

price, thus income is interacted with price, is set into the model.

In some cases of buying vegetable, the buyer does not travel from

home to store but in such a convenient way with other works and

visit the store, therefore the parameter of distance in these cases

might have the difference compared to the cases of traveling to

stores from home. Hence, the interaction between distance and

dummy variable be convenient, is set into the model.

Consequently, the basic model will include the store attributes

and two interaction variables. Additionally, to investigate the

influence of information on the store choice of buying vegetable,

RUM model is set by the interaction variables between

information and “safety levels” attributes. This is to examine that

those with different frequency of tracking information might

have different responses to the safety levels of vegetables in the

stores. There would be 4 models are estimated:

• Model 1: Estimate equation (25) in which X includes the

attributes in Table 4 and two interaction variables

(income x price and distance x convenient)

• Model 2: This model comprises the variables in Table 1,

and the interaction variables between safety level and

frequency of tracking safety and hygiene standards

information via TV, newspaper and internet, is set into

the model.

• Model 3: The model consists of the variables in Model 1

together with the interaction variables between safety

level and the number of violations safety and hygiene

standards cases watched, the number of poisoning cases

watched, and the number of poisoning cases experienced

by family members.

• Model 4: This model contains the variables in Model 1

and all interaction variables used in Model 2 and 3.

The reason of developing Model 2 and 3 that is the interaction

variables might be correlated, resulting in the biasedness of

parameters.

MAIN FINDINGS

The demand for vegetable

This study categories vegetables into 3 groups: leafty vegetables,

root vegetable and fruits, each group have 2 kinds, the

conventional and safe vegetable. The safe vegetable is

considered by vegetables that are certificated or are sold in

supermarkets under strict input control procedures.

The results of survey show that the percentage of household

consumes safe vegetable is about 50%. The results of Probit model

explaining the choice of consuming 6 vegetable types indicate that

the households with small size, high income and having more

childrens under 6 tend to have higher likelihood of choosing safe

vegetable. Those who tend to buy higher safe vegetable including

male, elders, office workers, student, homemaker, vegetarian, and

those without bargain habit. The frequency of tracking safety and

hygiene standards information via TV, newspaper, and internet

generally has no impact on the choice of safe and conventional

vegetables. However, the information of poisoning and safety and

hygiene standards violations cases which induces the consumer

less purchase conventional vegetable and buying safe vegetable

instead. While the number of poisoning cases experienced by

family members in the past 12 months insignificantly effects the

choice of safe vegetable.

The results of demand equation system using LA-AIDS

describing the expenditure and quantity demanded of safe

vegetable, those who spend higher for safe vegetable consisting

the households with more children, male, office workers, student,

and homemaker. Meanwhile, the level of education insignificantly

affects the demand for safe vegetable. The frequency of tracking

safety and hygiene standards information in all stores has no

impact the demand for safe vegetable. The information of

violations cases watched has the influence in a manner of buying

less conventional leafty vegetables and buying more the safe

leafty and root vegetables, probably being the leafty vegetables

have hazards of safety violations higher than root vegetables and

fruits. While, the information of poisoning cases watched and the

number of poisoning cases experienced by family members does

not affect the demand for safe vegetable.

The estimated Hicksian and Marshallian elasticities of demand

are low, which shows that a weakness in the responses of

consumer to vegetable price. The conventional and safe

vegetables neither complements nor substitutes. In general, the

price of a conventional vegetable has the effect of quantity

demanded of other conventional vegetables, but has no impact

the quantity demanded of safe vegetable, and vice versa. This

implies that the price of conventional vegetable in the current

levels does not influence the demand for safe vegetable, the

consumer only responses to the safe vegetable as the price of

other safe vegetables be changed. Especially, the safe leafty

vegetable elastics with the price of conventional root vegetables

and fruits (substitution effects) and the price of conventional fruit

vegetables (income effects).

WTP for safe vegetable

This objective uses the choice experiments with hypothetical

scenario. The Conditional Logit and Mixed Logit model is

utilized to estimate the utility function for the attributes of two

vegetables. The regression results of two models are generally

identical. The result of Conditional Logit model is used to

calculate the willingness to pay for vegetable attributes.

The WTP estimation results show that consumers are not willing

to pay higher for VietGAP water spinach and VietGAP carrots

than conventional ones. The organic certification is more

preferred because buyers are willing to pay more than VND

12,000 /kg to buy organic water spinach and more than VND

22,000/kg to buy organic carrots. The organic certification is paid

higher price than VietGAP certification which is understandable,

but even the safety guarantee of sellers is more appreciated than

VietGAP. Water spinach which is guaranteed by sellers with

compensated amount of VND 300 million if the toxic contents

exceeding the safety threshold, is paid more than VND 11,000/kg

compared to the unguaranteed ones, and more than VND

22,000/kg of carrots.

The attribute of packaging and manufacturer information should

be used depending on vegetables. In this objective, the results

indicate that these attributes are important for water spinach,

while is not important for carrots. The buyers are willing to pay

more than VND 15,000/kg to buy water spinach with packaging

and manufacturer information, although they are not willing to

pay higher price for carrots. The buyers ask to have the origin

stamps for carrots instead, and are willing to pay more than VND

15,000/kg for carrots with traceability stamps.

While the Conditional Logit model shows that consumers are not

willing to pay for VietGAP, the result of Mixed Logit model

reveals more information. The standard deviation of VietGAP

coefficient is different from 0 and the value is high, which

implies that a half of buyers prefer to VietGAP water spinach and

are willing to pay higher for VietGAP ones, but the other half of

buyers does not prefer to VietGAP water spinach. Although the

average WTP equals to 0, there is a heterogeneity in preferences

for VietGAP water spinach. However, there is a homogeneity in

preferences for VietGAP carrots: most of the buyers are willing

to pay higher price for VietGAP carrots.

The MX model also shows that the willingness to pay higher for

water spinach and carrots with organic certification, the

preferences for this attribute is also heterogenous in both

vegetables. This implies that although most consumers prefer

organic certification and are willing to pay higher price, many

consumers are not considered in this certification.

The frequency of tracking safety and hygiene standards

information generally less affects the WTP for VietGAP

certification, organic certification and safety guarantee. The

trend of general influence that is those with moderate frequency

of tracking information are willing to pay higher than those with

irregular tracking information, while those with everyday

tracking information are not willing to pay higher than those with

irregular tracking information. This is probably because those

who follow the information at the frequent and moderate level,

are more sensitive about the safety and hygiene standards,

therefore are willing to pay higher price for safety attributes.

However, those who follow the information at very frequent

level might lose their trust in the certifications and guarantees

from the sellers, thus are not willing to pay for these attributes.

This indicates that, in a manner of speaking, the information of

safety and hygiene standards in the social media is failing to

encourage the development of safe vegetable.

Meanwhile, the information of safety and hygiene standards

violation cases, the number of poisoning cases, and the number

of poisoning cases experienced by family members does not

influence the preferences for VietGAP, organic certification, and

safety guarantee, therefore the willingness to pay for these

attributes is fixed.

Vegetable store choice

The third objective of study investigates the drivers affecting

vegetable store choice, including the factors of buyer

characteristics and store attributes, with the frequency of tracking

information which are the key variables. The study observes the

buyer choice of buying vegetable for all trip in the last 1 week,

using the MNL model to analyze the impact of buyer

characteristics, and the CL/MX model to analyze the effect of

store attributes.

The results of survey show that on average, each buyer purchases

vegetables 5 times per week, each time value approximately

VND 42,000.

Those who tend to purchase vegetables in the modern channels

consisting the households with less members, less elders, and

those who are male, high income, older people, without

bargaining habit, vegetarian, office worker, student and

homemaker. Education and the number of children in the

household has no effect.

The analysis of store attributes points out that the distance

variable is very important. The farther the store is, the less likely

store is chosen. This preference takes advantages for the

traditional channels that are closer to residential areas. Other

important attributes comprise the freshness, safety level and

providing information, while the diversity, input control and

preprocessing does not affect the vegetable store choice.

Regarding the impact of information, MNL model shows that the

effect of frequency of tracking safety and hygiene standards

information on the vegetable store choice is very limited in

general. The Internet has absolutely no effect on the store choice.

The frequency of tracking safety and hygiene information at the

moderate level is more likely to choose the traditional channel,

contrary to the frequency tracking information at everyday. The

information of safety and hygiene standards violation cases and

the number of poisoning cases show the general influence that a

decrease in the probability of choosing traditional channels and

an increase in the probability of choosing the modern channels.

Households have the higher number of poisoning cases

experienced by family members are more likely to choose the

minimart.

CL model with the interaction variables between information and

safety level shows that those with the higher frequency of

tracking safety and hygiene standards information via TV are less

sensitive to the safety level, while those with the higher

frequency of tracking information via internet and newspaper are

more sensitive to the safety level. The number of poisoning and

violation cases watched that induces the consumer’s respond

more strongly for the safety level and thus they tend to choose

the safer stores. However, the number of poisoning cases

experienced by family members in the past induces the

consumers to be familiar with and fewer responses to the safety

level.

POLICY IMPLICATIONS

The conventional vegetable does not substitute for safe

vegetable

Although the asymmetric information theory states that the

conventional vegetable has the lower production costs which

tend to push the safe vegetable out of the market, but the analysis

result of demand equation system in this study shows that the

conventional vegetable does not substitute for the safe vegetable.

In particular, the decrease in the price of conventional vegetable

will not significantly reduce the demand for safe vegetable. This

is a good signal, showing the asymmetric information will be

difficult to drive the safe vegetable out of the market. This is

probably being the difference in the product of the two groups.

Even though the traditional markets have the advantage of being

close to home and an ancient culture, but the modern markets

have its own advantages such as clean store’s space and modern

shopping way. These modern distribution channels therefore

should be maintained and promote its own advantages against the

traditional channels.

Low elasticities

The safe vegetable has the low price elasticity of demand that

also be a good signal. The safe vegetable buyers do not strongly

respond to the change in price. This implies that the safe

vegetable distributors, in the case of necessity, can raise the price

without considering about a significant decrease in quantity

demanded.

In fact, the price of vegetables in traditional markets are quite

fluctuating, depending on origin, quality, freshness and store

format. For example, the price of vegetables in the peak hour and

off-peak hour greatly differs. In many cases, the price of

vegetables in traditional markets is as high as or higher than the

price of the same vegetables in supermarket. This induces buyers

in the modern channels easily accept higher prices, resulting in

the demand for safe vegetable be inelastic. Moreover, higher

prices can be a signal of quality. Because vegetables sold at the

lower prices might be considered unsafe.

VietGAP is not really a signal of safety

While current retail channels, both traditional and modern

markets, use VietGAP term to send the signal of safe vegetable

to consumer, this research result shows that consumers are not

truely willing to pay higher for VietGAP vegetables, at least for

water spinach and carrots. There is high heterogeneity in the

preferences of buyers for VietGAP certification. Some buyers

appreciate VietGAP certification, the others do not. These buyers

might not believe in the compliance of manufacturer for

VietGAP production process. For the average calculated of the

sample, the buyers are not willing to pay any additional amount

for VietGAP vegetables. This implies that retailers need to have

another evidence to send the signal of quality and safety and

hygiene standards, at the same time VietGAP certification

organization need to reconsider the certification process and

supervise the compliance with safety regulatory standards.

Organic certification is a symbol of safety and hygiene

While VietGAP certification is not appreciated, this research

result reveals that the buyers are willing to pay more than VND

12,000/kg for organic water spinach against conventional water

spinach, and VND 22,000/kg for organic carrots against

conventional carrots. This is a high willingness to pay, close to

the market prices of these two vegetables (VND 16,000 for

conventional water spinach and VND 23,000 for conventional

carrots). This states that the buyers appreciate organic

UNIVERSITY OF ECONOMICS HCMC

--------------

LE THI TUYET THANH

INFORMATIONAL ASYMMETRIES

AND THE DEMAND FOR

VEGETABLE IN HCMC

MAJOR: DEVELOPMENT ECONOMICS

CODE: 9310105

PH.D. DISSERTATION

EXECUTIVE SUMMARY

HO CHI MINH CITY 2019

INTRODUCTION

This study is motivated by the asymmetric information problem

on HCMC vegetable market, in which a large proportion of

vegetable does not comply with safety and hygiene standards. In

most cases, consumers can not recognize the safety attributes of

vegetable, and thus can not distinguish the conventional

vegetable and safe vegetable. This results in two problems. One

is that producers may apply unsafe production techniques to

maximize their own profits (moral hazards). The other is safe

vegetable can be driven out of the market being unable to

compete with conventional vegetable (adverse selection). These

two problems result in the failures of safe vegetable market in

HCMC. Both buyers and distributors have taken measures to

overcome these problems.

Lacking information, consumers try to seek information about

safety and hygiene standards of vegetable to be able to buy safe

vegetable. The prevalent information channels in HCMC are

internet, television, and newspaper, which can affect consumer

behavior and thus influence the development of safe vegetable

market. Understanding the effect of these channels on the

demand for safe vegetable, the preferences for the safety

attributes, and vegetable store choice are therefore important to

the safe vegetable distributors in expanding their market share.

The results of this study also help policymakers in providing

information and selecting the appropriate channels to develop the

market for safe vegetable.

The safe vegetable distributors, especially those of modern

distribution channels, confronted many difficulties in competing

with conventional vegetable in the context that buyers can not

distinguish the two types of vegetable. The first problem is how

to set vegetable prices in such a way that can compensate for the

high production and distribution costs and at the same time,

accepted by consumers. The second one is which safety

standards are important to consumers in order to have the

relevant development strategy. The last one is what store

attributes are critical to vegetable buyers in order to attract more

shoppers.

RESEARCH OBJECTIVES AND METHODOLOGY

Stemming from the above problems, this study proposes the

following research objectives.

Objective 1: Analyze the impact of information and informationseeking behavior of safety and hygiene standards on the demand

for safe vegetable. The research results help sellers realize the

response of consumer to the safe vegetable prices in order to develop

reasonable pricing strategy. The results also show the effect of

information on the demand for safe vegetable, thereby assisting

policymakers in providing information and selecting appropriate

information channels to promote the safe vegetable market. This

goal is solved by estimating the demand equation system for both

conventional and safe vegetables, with Linear Approximation

Almost Ideal Demand System (LA-AIDS), controlling for

endogeneity of prices and zero demand.

Objective 2: This ojective analyses the effect of information and

information-seeking behavior of safety and hygiene standards on

the willingness to pay for vegetable attributes. This is to have

the suitable policy in improving quality and safety attributes to

attract more buyers. The findings provide information for

vegetable distributors to develop price strategy in conjunction

with appropriate marketing procedures to maximize profits. This

objective applies Choice Experiment method to estimate the

willingness to pay for safe vegetable attributes, using

Conditional Logit (CL) and Mixed Logit (MX) models.

Objective 3: Investigate the influence of information and

information-seeking behavior of safety and hygiene standards on

the vegetable store choice. Other factors under consideration

include the distribution channel attributes and consumer

characteristics. The Multinomial Logit Model (MNL) is used to

analyze the impact of buyer characteristics on the store choice

decision of buying vegetable, while the Conditional Logit model

is used to explore the effect of distribution channel attributes on

the store choice decision. This study help sellers to perceive

factors that need to be improved in order to attract more

consumer. This study also helps to examine the target customer

group of each vegetable distribution channel, as well as find out

which information channel are mostly effective to appeal target

customer.

The conceptual framework of this study is summarized in Figure 1.

BUYER

CHARACTERISTICS

▪ Individual

characteristics:

age,

gender,

education,

occupation, bargaining

habit, number of vegan

days

▪ Household

characteristics: income,

number

of

meals,

household size, number

of childrens, elders.

▪ Market

price

(endogenous).

LA-AIDS

6 vegetable groups

▪ Frequency of accessing information

via TV, newspapers and internet

▪ Number of poisonings cases watched

▪ Number of poisoning cases

experienced by family members in the

past 12 months

MNL

models

CL/MX

models

Certification

Safety guarantee

Store format

Packaging & information

Price

Objective 3

Store

choice

Interaction

SAFE ATTRIBUTES

OF VEGETABLE

▪

▪

▪

▪

▪

Demand for

vegetable

INFORMATION

Interaction

DISTRIBUTION

CHANNEL

CHARACTERISTICS

Price

Distance

Freshness

Diversity

Input control

Preprocessing

Providing information

Safety level

Objective 1

Objective 2

CE with CL/MX

models

Figure 1: Conceptual framework.

WTP for

safety

attributes

The study uses survey data of 320 people who bought vegetables

in HCMC in 2018. The information factors and informationseeking behavior of buyers is the main explanatory variable in all

the three research objectives. Specifically, the information of

poisoning and safety and hygiene standard violation, as well as

the frequency of monitoring safety and hygiene standards

through communication channels, were included to analyze the

impact on safe vegetable demand, WTP for safe vegetable, and

store choice behavior.

OBJECTIVE 1: INFORMATION AND THE DEMAND

FOR VEGETABLE

Identification of vegetable groups

The study classifies vegetables on the market into 3 groups:

leafty vegetables (including flower and stem groups), root

vegetables and fruits (vegetable family). Thus, the estimated

equation system will comprise 6 equations: 3 equations for the

conventional vegetables of leaves, roots and fruits, and other 3

equations for the safe vegetables that eat leaves, roots and fruits.

Then measuring the average weekly quantity demanded of

vegetable groups in the past 3 months.

Prices and the issues of missing prices and endogeneity

Prices in this objective has an endogeneity problem caused by

the zero demand that would result in no information of prices,

and the heterogeneity in quality of vegetable, which results in the

simultaneity between expenditure share and prices.

To solve the endogeneity, prices are initially regressed on the

variables representing the survey location and household

characteristics. Households are surveyed in the same period

(several days) and location (these households supposedly under

the same cluster) have identical price level for each type of

vegetable, and the diffference of price among the households (if

any) because of variations in the quality and measurement errors.

The predicted price from sub-regression model would eliminate

these differences. The auxiliary regression is

ln 𝑝𝑟𝑖𝑐𝑒𝑖𝑘 = 𝛼0 + 𝛼1 𝐶𝑘 + 𝛼2 𝐷𝑘

(1)

where 𝑝𝑟𝑖𝑐𝑒𝑖𝑘 be the price of vegetables 𝑖 (there are 6 categories,

𝑖 = 1,2, … ,6, and 6 sub-regression equations respectively) for

household 𝑘, 𝐷𝑘 be the characteristics of buyers and households

𝑘, 𝐶𝑘 be the dummy variables describe the cluster. The cluster is

determined by ward/commune. The households surveyed in the

same ward/commune will be considered in the same cluster. The

variables of characteristics of buyers and households are shown

in Table 1.

Bảng 1: Characteristics of buyers and their families

Variable

Definition

Number of meals/week The average number of home-cooked

meals per week

Age

Age of vegetable buyer

Household

size Number of members who regularly eat at

(persons)

home in the household

Number of childrens

The number of childrens under 6 in the

household

Number of elders

The number of peoples over 60 in the

household

Household

income Total income of household members

(mil. VND/month)

Number of vegan The average number of vegan days per

days/month

month of buyer

Gender (1 = Male)

Dummy variable, the gender of vegetable

buyer, 1 = Male

Bargain (1 = Yes)

Dummy variable, 1 = Having bargain

habit while purchasing

Occupation (*)

Unskilled worker (baseline)

Office worker

Manager

Skilled worker

Homaker

Student

Others

Education (*)

Finished primary or less (baseline)

Finished secondary

Finished high school

Finished college

Finished university or higher

The adjusted price would be:

̂ 𝑖𝑘 = 𝛼

̅

ln 𝑝𝑟𝑖𝑐𝑒

̂0 + 𝛼

̂𝐶

̂𝐷

(2)

1 𝑘+𝛼

2 𝑘

̅𝑘 be the household characteristics, taking the average

where 𝐷

value of sample. The category variables are assigned by the most

frequent values in the sample, whereby the dummy variables of

education and occupation in the model are made for high school

and homemaker. The predicted price ln 𝑝𝑟𝑖𝑐𝑒

̂ 𝑖𝑘 would be used

into the model.

Zero demand and sample selection bias

The safe vegetable groups might have a significant number of

households with zero demand. As a result, the quantity demanded

or the expenditure ratio which is censored from 0 and the normal

estimation method could be bias. This goal employs two-step

estimation method of Heckman (1976) to deal with this problem.

According to this method, in step 1, the consumer is assumed to

decide whether consume each type of vegetable or not:

𝑑𝑖𝑘 = 𝛼𝐷𝑘 + 𝛾𝐼𝑘 + 𝑢𝑖𝑘

(3)

where 𝑑𝑖𝑘 be a dummy variable indicates whether household 𝑘

consumes vegetable 𝑖 or not, 𝐷𝑘 and 𝐼𝑘 be the exogenous

variable explains for the choice of consuming vegetable, in

which 𝐷𝑘 be the characteristics of individual and household, and

𝐼𝑘 be the information variables are shown in Table 2.

Bảng 2: The frequency of tracking safety and hygiene standards

information and poisonings

Variable

Tracking safety and

hygiene

standards

information via TV

Tracking safety and

hygiene

standards

Definition

The frequency of tracking safety and

hygiene standards information via TV (*):

Under once/month (baseline)

More than once/month

More than once/week

Per day

The frequency of tracking safety and

hygiene standards information

via

newspaper(*):

Under once/month (baseline)

information

newspaper

via

Tracking safety and

hygiene

standards

information

via

internet

Number of poisoning

cases watched per

month

Number of violations

of safety and hygiene

standards watched per

month

Number of poisonings

cases experienced in

12 months

More than once/month

More than once/week

Per day

The frequency of tracking safety and

hygiene standards information

via

internet(*):

Under once/month (baseline)

More than once/month

More than once/week

Per day

The number of food poisoning cases

watched per month, on average in the past 6

months

The number of violations of safety and

hygiene standards watched per month, on

average in the past 6 months

The number of poisoning symptoms of

family members experienced in the past 12

months

Then the demand (or expenditure) function would be defined as

𝑞𝑖𝑘 = 𝛽𝑋𝑖𝑘 + 𝜀𝑖𝑘

if 𝑑𝑖 > 0

𝑞𝑖𝑘 = 0

if 𝑑𝑖 ≤ 0

where 𝑋𝑖𝑘 is a vector of explanatory variables for demand

function would be described in the following section. The

expected quantity demanded of demand (or expenditure)

function with distributed assumption would be

𝐸(𝑞𝑖𝑘 |𝑋𝑖𝑘 , 𝑑𝑖 > 0) = 𝛽𝑋𝑖𝑘 + 𝐸(𝜀𝑖𝑘 |𝑢𝑖𝑘 > −𝛼𝐷𝑘 − 𝛾𝐼𝑘 )

(4)

and:

𝐸(𝜀𝑖𝑘 |𝑢𝑖𝑘 > −𝛼𝐷𝑘 − 𝛾𝐼𝑘 ) =

−𝛼𝐷𝑘 −𝛾𝐼𝑘⁄

𝜎2)

𝜎2 1−𝛷(−𝛼𝐷𝑘 −𝛾𝐼𝑘⁄𝜎 )

2

𝜎12 𝜙(

The demand function becomes

(5)

𝐸(𝑞𝑖𝑘 |𝑋𝑖𝑘 , 𝑑𝑖 > 0) = 𝛽𝑋𝑖𝑘 +

−𝛼𝐷𝑘 −𝛾𝐼𝑘⁄

𝜎2)

−𝛼𝐷

−𝛾𝐼

𝑘

𝑘⁄ )

𝜎2 1−𝛷(

𝜎2

𝜎12 𝜙(

(6)

where 𝜎2 is the standard deviation of 𝜀𝑖𝑘 and 𝜎12 is the

covariance between 𝜀𝑖𝑘 and 𝑢𝑖𝑘 .

Thus, this method ogriginally estimates the Probit model

explaining the choice of whether consume a type of vegetable or

not (i.e. estimating the coefficient 𝛼 in 𝑑𝑖 = 𝛼𝑍𝑖 + 𝑢𝑖 , then

calculating the Inverse Mills Ratio (IMR):

−𝛼𝐷 −𝛾𝐼

𝐼𝑀𝑅𝑖 = 𝜆 (

𝑘

𝑘⁄ )

𝜙(

𝜎2

−𝛼𝐷𝑘 − 𝛾𝐼𝑘⁄

𝜎2 ) = 1−𝛷(−𝛼𝐷𝑘−𝛾𝐼𝑘⁄ )

(7)

𝜎2

Then estimating the demand (or expenditure) function with IMR

𝑞𝑖𝑘 = 𝛽𝑋𝑖𝑘 + 𝜑𝐼𝑀𝑅𝑖 + 𝜀𝑖𝑘

(8)

where 𝜑 =

𝜎12

𝜎2

.

In this study, the equation system has 6 demand functions for

three types of vegetable that eat leaves, tubers and fruits, each of

these with conventional and safe type. Consequently, 6 Probit

models would be estimated correspondingly to 6 types of

vegetable. Following the calculation for 6 IMR variables for each

type of vegetable and using 6 estimated IMR variables for

measuring the LA-AIDS demand equation system.

Estimation methods

Demand function indicates quantity demanded that maximizes

utility and as a result of utility maximization problem. In addition

to estimating single demand equation, it is possible to estimate

the demand equation system. Estimating the single demand

equation has the critical limitation that is the separating goods

from related ones. This is an unreasonable assumption.

Moreover, the single demand equation will not help to show the

correlation among the goods, for example, cross-price elasticity

of demand can not be calculated. Thus, the demand equation

system with related goods is more preferred.

The dominant models are used to estimate demand equation

system including Linear expenditure system, Rotterdam System,

and Almost Ideal Demand System.

The AIDS model is the most regularly used, with two variants

LA-AIDS and QUAIDS. However, QUAIDS is more difficult to

estimate, especially the demand equation system has a censoring

problem with many cases of quantity demanded equals to 0.

While LA-AIDS is easier to estimate, despite the important

disadvantage is that the log of expenditure ratio is assumed to be

a linear function of the log of total expenditure. However, this is

not significant indeed, because of log-log formulation, the

expenditure ratio fulfills nonlinear function with respect to the

total expenditure. Solely the expenditure ratio is compelled to be

a square function of total expenditure – i.e. expenditure ratio

could be deviated when total expenditure reaches a certain level

- then LA-AIDS is virtually inappropriate. Consequently, the

study uses LA-AIDS to estimate the demand equation system of

vegetables.

LA-AIDSis developed by Deaton and Muellbauer (1980) from

the expenditure function:

1

ln 𝑒(𝑢, 𝑝) = 𝛼0 + ∑𝑖 𝛼𝑖 ln 𝑝𝑖 + 2 ∑𝑖 𝛾𝑖𝑗 ln 𝑝𝑖 ln 𝑝𝑗 + 𝑢𝛽0 ∏ 𝑝𝑖 𝛽𝑖

(9)

The estimated equation system would be:

𝑤𝑖 = 𝛼𝑖 + ∑𝑗 𝛾𝑖𝑗 ln 𝑝𝑗 + 𝛽𝑖 ln

𝑦

𝑃∗

+ 𝜑𝑖 𝐼𝑀𝑅𝑖 + 𝜖𝑖

(10)

where ln 𝑃 ∗ = ∑𝑗 𝑤𝑗 ln 𝑝𝑗

Note that the price variable used in this model is the predicted

price from the OLS models, 𝐼𝑀𝑅𝑖 is the Inverse Mills Ratio

estimated from the regression results of Probit models.

The demand function system in this objective has 6 equations,

however only 5 equations are estimated. The coefficient of the

6th equation would be computed by the theoretical constraints:

•

Additivity: ∑𝑖 𝛼𝑖 = 1, ∑𝑖 𝛽𝑖 = 0, ∑𝑖 𝛾𝑖𝑗 = 0

•

Homogeneity: ∑𝑗 𝛾𝑖𝑗 = 0

(with all i)

•

Symmetry: 𝛾𝑖𝑗 = 𝛾𝑗𝑖

Therefore, these constraints in the fact that are imposed on this

objective. Some studies also examined these constraints, but this

is not actually essential, because after testing if the constraints

are not satisfied, these constraints also are applied to this goal.

Elasticities

The price elasticity of demand or cross-price elasticity of demand

for commodity could be calculated by the estimated coefficients

from the demand equation system. There are 2 elasticities: noncompensated elasticity (referred to as “elasticity”, which is the

Marshallian demand) and compensated elasticity (Hicksian

elasticity). Marshallian elasticity is computed by the assumption

that total expenditure for vegatables in the demand equation

system be unchanged, while compensated elasticity is calculated

by the assumption that the utility is constant (and thus total

expenditure might be varied). Practically, the vegetable prices

change the consumer can adjust expenditure level for vegetables.

Therefore, compensated elasticity might have the greater

practical implications.

There are many procedures to calculate elasticity, but according

to Green and Alston (2008), the following calculation (applied to

LA-AIDS) is the most basically precise for Marshallian elasticity

𝜂𝑖𝑗 = −𝛿𝑖𝑗 +

𝛾𝑖𝑗

𝑤𝑖

− 𝛽𝑖

𝑤𝑗

𝑤𝑖

(11)

where 𝛿𝑖𝑗 is the Kronecker indicator, in which 𝛿𝑖𝑗 = 1 if 𝑖 = 𝑗

and 𝛿𝑖𝑗 = 0 if 𝑖 ≠ 𝑗, 𝛾𝑖𝑗 and 𝛽𝑖 is the estimated coefficients from

(10), 𝑤𝑖 is the expenditure ratio for vegetable 𝑖. The compensated

elasticity (Hicksian) be

∗

𝜂𝑖𝑗

= 𝜂𝑖𝑗 + 𝑤𝑗 (

1+𝛽𝑖

𝑤𝑖

)

(12)

OBJECTIVE 2: INFORMATION AND WTP FOR SAFE VEGETABLE

In this objective, the study evaluates the WTP for safety

attributes and investigating the effect of information on WTP

based on the Choice Experiment (CE). This research chooses two

vegetables, water spinach and carrots, to conduct the experiment

because the characteristics and storage time of two vegetables are

very different, thus the safety attributes might differ.

The experiment uses 5 attributes. The definition and attribute

levels are shown in Table 3.

Table 3: Attributes and their levels

Attribute

Definition

Price

% higher than

price in Choice 1

Vegetables are

sold in different

stores

Store format

Safety

certification

Vegetables are

granted by safety

standards

certificates

Safety

guagrantee

The

seller

commits

to

compensate

VND300 million

if the toxin

content exceeds

the prescribed

threshold

Packaging,

manufacturer

information and

traceability

stamps

Packaging

and

information

Level

of Level

of

Alternative 1 Alternative 2

0%

50%, 150%,

300%, 500%

Formal

Formal market

market

Supermarket

Safe vegetable

stores

No

No

certification

certification

VietGAP

certification

Organic

certification

No

safety No

safety

guagrantee

guagrantee

Safety

guagrantee

No

packaging, no

information,

no

traceability

stamps

Packaging,

manufacturer

information

Packaging,

manufacturer

information

and

traceability

stamps

With two choices (in which Choice 1 be fixed), 5 attributes and

levels in Table 3, it would be possible generate 4×3×3×2×2=144

alternatives. The study utilizes the interaction effect orthogonal

fractional factorial design to select 60 alternatives for

interviewing the buyers. 3 alternatives are grouped into 1 block,

there are 20 blocks respectively. Multiplying these 20 blocks

with 16 cases, the total is 320 blocks. Each respondent should be

answered 1 block for water spinach and 1 block for carrots, thus

each respondent answers a total of 6 choice tasks.

Model and estimation methods

With the database pointing out the preferences from the

hypothetical choice scenario mentioned before of two choices

𝑗 = 1,2, the model can estimate utility function:

𝑈𝑗 = 𝛽𝑋𝑗 + 𝜀𝑗

(13)

where 𝑋𝑗 is a set of attributes of choice 𝑗 described in Table 3 and

𝑈𝑗 is the utility of choice 𝑗 and 𝜀𝑗 is the residuals.

Phương trình ước lượng

This study initially estimates the basic utility function as shown

in (13)

𝑈𝑖𝑗 = 𝛽1 𝑝𝑟𝑖𝑐𝑒𝑖𝑗 + 𝛽2 𝑝𝑟𝑖𝑐𝑒𝑖𝑗 × 𝑖𝑛𝑐𝑜𝑚𝑒𝑖 + 𝛽2 𝑠𝑢𝑝𝑒𝑟𝑖𝑗 +

𝛽3 𝑠𝑝𝑒𝑐𝑖𝑎𝑙𝑡𝑦𝑖𝑗 + 𝛽4 𝑣𝑖𝑒𝑡𝑔𝑎𝑝𝑖𝑗 + 𝛽5 𝑜𝑟𝑔𝑎𝑛𝑖𝑐𝑖𝑗 +

𝛽6 𝑔𝑢𝑎𝑟𝑎𝑛𝑡𝑒𝑒𝑖𝑗 + 𝛽7 𝑖𝑛𝑓𝑜𝑖𝑗 + 𝛽8 𝑞𝑟𝑐𝑜𝑑𝑒𝑖𝑗 + 𝜀𝑖𝑗

(14)

where 𝑈𝑖𝑗 is the utility of buyer 𝑖 obtained from the choice 𝑗 (𝑗 =

1,2), 𝑝𝑟𝑖𝑐𝑒𝑖𝑗 is the prices of choice 𝑗 in the choice set of buyer 𝑖.

Note that 𝑝𝑟𝑖𝑐𝑒𝑖𝑗 is the specific amount with the unit of thousand

VND/kg, not the percent (%). 𝑠𝑢𝑝𝑒𝑟𝑖𝑗 and 𝑠𝑝𝑒𝑐𝑖𝑎𝑙𝑡𝑦𝑖𝑗 is dummy

variables indicating the store be the supermarket and safe

vegetable store (baseline: formal market). 𝑣𝑖𝑒𝑡𝑔𝑎𝑝𝑖𝑗 and

𝑜𝑟𝑔𝑎𝑛𝑖𝑐𝑖𝑗 is dummy variables denoting safety certificate of

vegetables of choice 𝑗 (baseline: no certification). 𝑔𝑢𝑎𝑟𝑎𝑛𝑡𝑒𝑒𝑖𝑗

is dummy variable equals to 1 if the seller commits to

compensate VND300 million in the case that the toxin content

exceeds the prescribed threshold. 𝑖𝑛𝑓𝑜𝑖𝑗 vand 𝑞𝑟𝑐𝑜𝑑𝑒𝑖𝑗 is

dummy variables presenting the having packaging and

manufacturer information 𝑖𝑛𝑓𝑜𝑖𝑗 , and having packaging,

manufacturer information and traceability stamps 𝑞𝑟𝑐𝑜𝑑𝑒𝑖𝑗 .

Assumme that these two attributes be prefferred to the consumer,

then 𝛽7 > 0, 𝛽8 > 0 and 𝛽8 > 𝛽7 , because the utility with

traceability stamps (𝛽8 ) must be higher than the utility without

traceability stamps.

With the expectation that those with higher income are less

sensitive to prices, the interaction between price and income is

introduced into the utility function. If the expectation is correct,

then 𝛽1 < 0 and 𝛽2 > 0.

Among the attributes, this study is particularly interested in the

safety attributes, including VietGAP certification and organic

certification. Additionally, the compensated guarantee of sellers

also is a kind of certificate (the self-certification of sellers),

because the sellers have to comply with safety standards of

vegetables even though hazards for compensation. To examine

the impact of information on WTP for safety attributes, the

information variables (Table 2) would be interacted with these

safety attributes.

If the information variables interact with 3 safety attributes are

simultaneously put into (14) which can cause the

multicollinearity. Therefore, each model can only include the

interaction variables of an attribute. As a result, 4 models are

estimated:

• The basic model: solely the attributes and the interaction

between price and income (14).

• VietGAP model: the basic model with the interaction

variables between VietGAP certification and information

variables.

• Organic model: the basic model with the interaction variables

between organic certification and information variables.

• Guarantee model: the basic model with the interaction

variables between safety guarantee and information variables.

These models are estimated by Maximum Likelihood. The

parameters 𝛽 are estimated by maximizing the log-likelihood

function:

𝐿𝐿 = ∑𝑖 ∑𝑗 𝑦𝑖𝑗 𝑝𝑖𝑗

(15)

𝑒

where 𝑝𝑖𝑗 = ∑

𝛽𝑋𝑗

𝑙𝑒

𝛽𝑋𝑙

is the probability of choosing 𝑗 by buyer 𝑖,

and 𝑦𝑖𝑗 is dummy variable indicating the observed choice from

the buyer, 𝑦𝑖𝑗 = 1 if the buyer 𝑖 selects choice 𝑗.

MX model is more complicated. In this model, log-likelihood

function is identical as CL, but the probability of choosing 𝑗 by

the buyer 𝑖 becomes:

𝑝𝑖𝑗 = ∫ (

∑

𝑒

𝑙

𝛽𝑋𝑖𝑗

𝑒 𝛽𝑍𝑖𝑙

) 𝑓(𝛽)𝑑(𝛽)

(16)

where 𝑓 (𝛽) be the probability distributed function of 𝛽, because

𝛽 is a random variable instead of fixed variable as in CL. To

estimate 𝛽, the model will choose randomly 𝑅 values from 𝑓 (𝛽)

and calculate the average probability from 𝑅. 𝑅 be the number of

draws, the larger 𝑅 the more accurate estimated parameters. The

research of Bierlaire (2003) shows that 𝑅 = 500 is the most

effective value, thus this study uses 𝑅 = 500.

Estimation of WTP

The parameter 𝛽 describes the marginal utility of attributes. For

example, the parameter 𝛽𝑣𝑖𝑒𝑡𝐺𝐴𝑃 of 𝑉𝑖𝑒𝑡𝐺𝐴𝑃 variable shows the

utility of VietGAP certificate. Comparing this utility to the

marginal utility of prices help to calculate the willingness to pay

for VietGAP certificate attribute:

𝛽

𝑊𝑇𝑃𝑉𝑖𝑒𝑡𝐺𝐴𝑃 = − 𝛽4

𝑝

(17)

where 𝛽𝑝 is the marginal utility of price (in thousand VND). In the

basic model, the marginal utility of 1 thousand VND to the buyer 𝑖

be:

𝛽𝑝 = 𝛽1 + 𝛽2 𝑖𝑛𝑐𝑜𝑚𝑒𝑖

(18)

𝑖𝑛𝑐𝑜𝑚𝑒𝑖 is the income of buyer 𝑖. If calculating at the mean value

of income, the marginal utility of price becomes

𝛽𝑝 = 𝛽1 + 𝛽2 𝑖𝑛𝑐𝑜𝑚𝑒

̅̅̅̅̅̅̅̅̅̅

(19)

where 𝑖𝑛𝑐𝑜𝑚𝑒

̅̅̅̅̅̅̅̅̅̅ is the average income of the sample.

OBJECTIVE 3: INFORMATION AND VEGETABLE STORE

CHOICE

This objective has two specific research objectives: (1) analyzing

the impact of buyer characteristics on the vegetable store choice

and (2) investigating the effect of distribution channel

characteristics on the vegetable store choice.

This study uses MNL and RUM models to identify the factors

influence store choice of buying vegetable. The collected data

shows each trip of buying vegetable in the last 7 days in detail

and thus each trip of buying vegetable is an observation – i.e. one

time of decision-making. Each trip of buying vegetable is

supposed to be exclusively travel to a unique store and therefore

the alternatives are mutually exclusive.

From the results of initial surveys and pilot surveys, the store of

buying vegetable is chosen to study as follows:

1. Hypermarket/Supermarket

2. Mini supermarket

3. Specialty store

4. Traditional market (legally established)

5. Street vendor (illegally established)

6. Local green grocery.

MNL Model

The assumption that the utility of consumer 𝑖 of choosing the

store 𝑚 from the shopping trip 𝑗 be a linear function as

𝑚

𝑈𝑖𝑗𝑚 = 𝛼𝑚 𝐷𝑖 + 𝛽𝑚 𝐼𝑖 + 𝜀𝑖𝑗

(20)

where 𝑚 = 1,2, … ,6 indicating the distribution channels

mentioned above, 𝐷𝑖 be a vector of the individual and household

characteristics, and household 𝑖 consists of variables shown in

Table 1 (The characteristics of individual and household), 𝐼𝑖 be

the information variables including the frequency of tracking

safety and hygiene, the number of safety and hygiene standards

violations and the number of poisoning cases experienced

depicted in Table 2.

The parameters 𝛼𝑚 and 𝛽𝑚 presenting the influence of

determinants on the utility of store choice 𝑚. Note that each of

store would have a vector of parameters. Each of 𝐷𝑖 variable

would have 𝛼1 defining the utility of extra supermarket to

individual with 𝐷𝑖 characteristics, 𝛼2 be the utility of minimart,

and similar to other stores.

Let the probability of consumer 𝑖 from the trip 𝑗 of choosing the

store 𝑚 be:

𝑝𝑖𝑗𝑚 = Pr(𝑦𝑖𝑗𝑚 = 1)

(21)

𝑚

where 𝑦𝑖𝑗

be the dummy variable defines the store choice 𝑚 of

the buyer 𝑖 from the trip 𝑗. Seting formal market is the base

category, the log-odds would be:

log

𝑝𝑖𝑗𝑚

𝑝𝑖𝑗4

= 𝛼𝑚 𝐷𝑖 + 𝛽𝑚 𝐼𝑖

∀𝑚 ≠ 1

(22)

𝑝1

𝑚

Note that log 𝑝𝑖𝑗1 = 0 and 𝜀𝑖𝑗

is assumed to be the logistic

𝑖𝑗

distribution. The probability of choosing the store 𝑘 would

become:

𝑘

𝑘

𝑝𝑖𝑗

=

𝑉

𝑒 𝑖𝑗

(23)

𝑚

𝑉

∑6𝑚=1 𝑒 𝑖𝑗

where 𝑉𝑖𝑗𝑚 = 𝛼𝑚 𝐷𝑖 + 𝛽𝑚 𝐼𝑖 . The MNL model will estimate the

parameters 𝛼𝑚 and 𝛽𝑚 for each of store (exclude 𝑚 = 4) by

Maximum Likelihood method:

log 𝐿 = ∑𝑖 ∑𝑗 ∑𝑚 𝑦𝑖𝑗𝑚 ln 𝑝𝑖𝑗𝑚

(24)

This model helps to analyze the influence of individual and

household characteristics, and the frequency of tracking safety

and hygiene information on the store choice of buying vegetable.

RUM

RUM model is applied to determine the impact of distribution

channel attributes on the store choice of buying vegetable.

Table 4: Attributes of vegetable stores

Variable

Definition

Price

The price indicator of the closest

store of each channel compared to

the closest formal market (the

price of the closest formal market

= 0%, the negative value in this

variable means that the price of

the closest store lower than the

formal market)

Measurement

units

%

Distance

Freshness

Diversity

Input control

Preprocessing

Providing

information

Safety level

Distance from the house to the

closest store (outlet) of each

distribution channel

Vegetables at each shopping

channel have different levels of

freshness

There are many types of

vegetables, the diversity of spicies

are sold at each distribution

channel

Whether vegetable stores have

strict input control or not

Vegetables at each channel before

being sold are preprocess:

cleaning, trimming, washing with

clean water, and packaging

The

level

of

providing

information on vegetable quality

(Expired date, the origin of

vegetables, processing, safety

standards, etc.) of each shopping

channel

Personal evaluation of buyers on

the probability of vegetables

fulfilling safety standards at each

channel

Minutes

1 = Freshness;

0

=

No

freshness

1 = Diversity;

0

=

No

diversity

1 = Strictly;

0

=

Not

strictly

1 = Carefully;

0

=

Not

carefully

1

=

Completely;

0

=

Not

completely

%

Model specification

This model will estimate the utility function

𝑈𝑖𝑗𝑚 = 𝐴𝑆𝐶𝑚 + 𝜌𝑋𝑚 + 𝜀𝑖𝑗𝑚

(25)

where 𝑋𝑚 is a vector of store attributes 𝑚. The attributes include

distance, diversity, preprocessing, freshness, providing

information and price. The definition of attribues are shown in

Table 4. These variables are the personal evaluation of buyers for

the store attributes of buying vegetable.

Interaction terms

Those with higher income might have different responses to the

price, thus income is interacted with price, is set into the model.

In some cases of buying vegetable, the buyer does not travel from

home to store but in such a convenient way with other works and

visit the store, therefore the parameter of distance in these cases

might have the difference compared to the cases of traveling to

stores from home. Hence, the interaction between distance and

dummy variable be convenient, is set into the model.

Consequently, the basic model will include the store attributes

and two interaction variables. Additionally, to investigate the

influence of information on the store choice of buying vegetable,

RUM model is set by the interaction variables between

information and “safety levels” attributes. This is to examine that

those with different frequency of tracking information might

have different responses to the safety levels of vegetables in the

stores. There would be 4 models are estimated:

• Model 1: Estimate equation (25) in which X includes the

attributes in Table 4 and two interaction variables

(income x price and distance x convenient)

• Model 2: This model comprises the variables in Table 1,

and the interaction variables between safety level and

frequency of tracking safety and hygiene standards

information via TV, newspaper and internet, is set into

the model.

• Model 3: The model consists of the variables in Model 1

together with the interaction variables between safety

level and the number of violations safety and hygiene

standards cases watched, the number of poisoning cases

watched, and the number of poisoning cases experienced

by family members.

• Model 4: This model contains the variables in Model 1

and all interaction variables used in Model 2 and 3.

The reason of developing Model 2 and 3 that is the interaction

variables might be correlated, resulting in the biasedness of

parameters.

MAIN FINDINGS

The demand for vegetable

This study categories vegetables into 3 groups: leafty vegetables,

root vegetable and fruits, each group have 2 kinds, the

conventional and safe vegetable. The safe vegetable is

considered by vegetables that are certificated or are sold in

supermarkets under strict input control procedures.

The results of survey show that the percentage of household

consumes safe vegetable is about 50%. The results of Probit model

explaining the choice of consuming 6 vegetable types indicate that

the households with small size, high income and having more

childrens under 6 tend to have higher likelihood of choosing safe

vegetable. Those who tend to buy higher safe vegetable including

male, elders, office workers, student, homemaker, vegetarian, and

those without bargain habit. The frequency of tracking safety and

hygiene standards information via TV, newspaper, and internet

generally has no impact on the choice of safe and conventional

vegetables. However, the information of poisoning and safety and

hygiene standards violations cases which induces the consumer

less purchase conventional vegetable and buying safe vegetable

instead. While the number of poisoning cases experienced by

family members in the past 12 months insignificantly effects the

choice of safe vegetable.

The results of demand equation system using LA-AIDS

describing the expenditure and quantity demanded of safe

vegetable, those who spend higher for safe vegetable consisting

the households with more children, male, office workers, student,

and homemaker. Meanwhile, the level of education insignificantly

affects the demand for safe vegetable. The frequency of tracking

safety and hygiene standards information in all stores has no

impact the demand for safe vegetable. The information of

violations cases watched has the influence in a manner of buying

less conventional leafty vegetables and buying more the safe

leafty and root vegetables, probably being the leafty vegetables

have hazards of safety violations higher than root vegetables and

fruits. While, the information of poisoning cases watched and the

number of poisoning cases experienced by family members does

not affect the demand for safe vegetable.

The estimated Hicksian and Marshallian elasticities of demand

are low, which shows that a weakness in the responses of

consumer to vegetable price. The conventional and safe

vegetables neither complements nor substitutes. In general, the

price of a conventional vegetable has the effect of quantity

demanded of other conventional vegetables, but has no impact

the quantity demanded of safe vegetable, and vice versa. This

implies that the price of conventional vegetable in the current

levels does not influence the demand for safe vegetable, the

consumer only responses to the safe vegetable as the price of

other safe vegetables be changed. Especially, the safe leafty

vegetable elastics with the price of conventional root vegetables

and fruits (substitution effects) and the price of conventional fruit

vegetables (income effects).

WTP for safe vegetable

This objective uses the choice experiments with hypothetical

scenario. The Conditional Logit and Mixed Logit model is

utilized to estimate the utility function for the attributes of two

vegetables. The regression results of two models are generally

identical. The result of Conditional Logit model is used to

calculate the willingness to pay for vegetable attributes.

The WTP estimation results show that consumers are not willing

to pay higher for VietGAP water spinach and VietGAP carrots

than conventional ones. The organic certification is more

preferred because buyers are willing to pay more than VND

12,000 /kg to buy organic water spinach and more than VND

22,000/kg to buy organic carrots. The organic certification is paid

higher price than VietGAP certification which is understandable,

but even the safety guarantee of sellers is more appreciated than

VietGAP. Water spinach which is guaranteed by sellers with

compensated amount of VND 300 million if the toxic contents

exceeding the safety threshold, is paid more than VND 11,000/kg

compared to the unguaranteed ones, and more than VND

22,000/kg of carrots.

The attribute of packaging and manufacturer information should

be used depending on vegetables. In this objective, the results

indicate that these attributes are important for water spinach,

while is not important for carrots. The buyers are willing to pay

more than VND 15,000/kg to buy water spinach with packaging

and manufacturer information, although they are not willing to

pay higher price for carrots. The buyers ask to have the origin

stamps for carrots instead, and are willing to pay more than VND

15,000/kg for carrots with traceability stamps.

While the Conditional Logit model shows that consumers are not

willing to pay for VietGAP, the result of Mixed Logit model

reveals more information. The standard deviation of VietGAP

coefficient is different from 0 and the value is high, which

implies that a half of buyers prefer to VietGAP water spinach and

are willing to pay higher for VietGAP ones, but the other half of

buyers does not prefer to VietGAP water spinach. Although the

average WTP equals to 0, there is a heterogeneity in preferences

for VietGAP water spinach. However, there is a homogeneity in

preferences for VietGAP carrots: most of the buyers are willing

to pay higher price for VietGAP carrots.

The MX model also shows that the willingness to pay higher for

water spinach and carrots with organic certification, the

preferences for this attribute is also heterogenous in both

vegetables. This implies that although most consumers prefer

organic certification and are willing to pay higher price, many

consumers are not considered in this certification.

The frequency of tracking safety and hygiene standards

information generally less affects the WTP for VietGAP

certification, organic certification and safety guarantee. The

trend of general influence that is those with moderate frequency

of tracking information are willing to pay higher than those with

irregular tracking information, while those with everyday

tracking information are not willing to pay higher than those with

irregular tracking information. This is probably because those

who follow the information at the frequent and moderate level,

are more sensitive about the safety and hygiene standards,

therefore are willing to pay higher price for safety attributes.

However, those who follow the information at very frequent

level might lose their trust in the certifications and guarantees

from the sellers, thus are not willing to pay for these attributes.

This indicates that, in a manner of speaking, the information of

safety and hygiene standards in the social media is failing to

encourage the development of safe vegetable.

Meanwhile, the information of safety and hygiene standards

violation cases, the number of poisoning cases, and the number

of poisoning cases experienced by family members does not

influence the preferences for VietGAP, organic certification, and

safety guarantee, therefore the willingness to pay for these

attributes is fixed.

Vegetable store choice

The third objective of study investigates the drivers affecting

vegetable store choice, including the factors of buyer

characteristics and store attributes, with the frequency of tracking

information which are the key variables. The study observes the

buyer choice of buying vegetable for all trip in the last 1 week,

using the MNL model to analyze the impact of buyer

characteristics, and the CL/MX model to analyze the effect of

store attributes.

The results of survey show that on average, each buyer purchases

vegetables 5 times per week, each time value approximately

VND 42,000.

Those who tend to purchase vegetables in the modern channels

consisting the households with less members, less elders, and

those who are male, high income, older people, without

bargaining habit, vegetarian, office worker, student and

homemaker. Education and the number of children in the

household has no effect.

The analysis of store attributes points out that the distance

variable is very important. The farther the store is, the less likely

store is chosen. This preference takes advantages for the

traditional channels that are closer to residential areas. Other

important attributes comprise the freshness, safety level and

providing information, while the diversity, input control and

preprocessing does not affect the vegetable store choice.

Regarding the impact of information, MNL model shows that the

effect of frequency of tracking safety and hygiene standards

information on the vegetable store choice is very limited in

general. The Internet has absolutely no effect on the store choice.

The frequency of tracking safety and hygiene information at the

moderate level is more likely to choose the traditional channel,

contrary to the frequency tracking information at everyday. The

information of safety and hygiene standards violation cases and

the number of poisoning cases show the general influence that a

decrease in the probability of choosing traditional channels and

an increase in the probability of choosing the modern channels.

Households have the higher number of poisoning cases

experienced by family members are more likely to choose the

minimart.

CL model with the interaction variables between information and

safety level shows that those with the higher frequency of

tracking safety and hygiene standards information via TV are less

sensitive to the safety level, while those with the higher

frequency of tracking information via internet and newspaper are

more sensitive to the safety level. The number of poisoning and

violation cases watched that induces the consumer’s respond

more strongly for the safety level and thus they tend to choose

the safer stores. However, the number of poisoning cases

experienced by family members in the past induces the

consumers to be familiar with and fewer responses to the safety

level.

POLICY IMPLICATIONS

The conventional vegetable does not substitute for safe

vegetable

Although the asymmetric information theory states that the

conventional vegetable has the lower production costs which

tend to push the safe vegetable out of the market, but the analysis

result of demand equation system in this study shows that the

conventional vegetable does not substitute for the safe vegetable.

In particular, the decrease in the price of conventional vegetable

will not significantly reduce the demand for safe vegetable. This

is a good signal, showing the asymmetric information will be

difficult to drive the safe vegetable out of the market. This is

probably being the difference in the product of the two groups.

Even though the traditional markets have the advantage of being

close to home and an ancient culture, but the modern markets

have its own advantages such as clean store’s space and modern

shopping way. These modern distribution channels therefore

should be maintained and promote its own advantages against the

traditional channels.

Low elasticities

The safe vegetable has the low price elasticity of demand that

also be a good signal. The safe vegetable buyers do not strongly

respond to the change in price. This implies that the safe

vegetable distributors, in the case of necessity, can raise the price

without considering about a significant decrease in quantity

demanded.

In fact, the price of vegetables in traditional markets are quite

fluctuating, depending on origin, quality, freshness and store

format. For example, the price of vegetables in the peak hour and

off-peak hour greatly differs. In many cases, the price of

vegetables in traditional markets is as high as or higher than the

price of the same vegetables in supermarket. This induces buyers

in the modern channels easily accept higher prices, resulting in

the demand for safe vegetable be inelastic. Moreover, higher

prices can be a signal of quality. Because vegetables sold at the

lower prices might be considered unsafe.

VietGAP is not really a signal of safety

While current retail channels, both traditional and modern

markets, use VietGAP term to send the signal of safe vegetable

to consumer, this research result shows that consumers are not

truely willing to pay higher for VietGAP vegetables, at least for

water spinach and carrots. There is high heterogeneity in the

preferences of buyers for VietGAP certification. Some buyers

appreciate VietGAP certification, the others do not. These buyers

might not believe in the compliance of manufacturer for

VietGAP production process. For the average calculated of the

sample, the buyers are not willing to pay any additional amount

for VietGAP vegetables. This implies that retailers need to have

another evidence to send the signal of quality and safety and

hygiene standards, at the same time VietGAP certification

organization need to reconsider the certification process and

supervise the compliance with safety regulatory standards.

Organic certification is a symbol of safety and hygiene

While VietGAP certification is not appreciated, this research

result reveals that the buyers are willing to pay more than VND

12,000/kg for organic water spinach against conventional water

spinach, and VND 22,000/kg for organic carrots against

conventional carrots. This is a high willingness to pay, close to

the market prices of these two vegetables (VND 16,000 for

conventional water spinach and VND 23,000 for conventional

carrots). This states that the buyers appreciate organic

## Nghiên cứu sử dụng toàn bộ mạch ghép động mạch trong phẫu thuật bắc cầu động mạch vành

## BÀI THẢO LUẬN THIẾT kế kết cấu NHÀ dân DỤNG

## nâng cao năng lực cạnh tranh của các doanh nghiệp thủy sản xuất khẩu ( nghiên cứu tại tỉnh bà rịa - vũng tàu )

## KHAI THÁC kết cấu, TÍNH NĂNG KỸ THUẬT VÀ QUY TRÌNH KIỂM TRA, CHẨN ĐOÁN sửa CHỮA hệ THỐNG điều HÒA TRÊN XE HUYNDAI i30

## nghiên cứu thiết kế tăng cường khả năng chịu lực của kết cấu bê tông cốt thép bằng tấm dán cacbon

## KHAI THÁC kết cấu, TÍNH NĂNG KỸ THUẬT VÀ QUY TRÌNH KIỂM TRA, CHẨN ĐOÁN sửa CHỮA hệ THỐNG điều HÒA TRÊN XE HUYNDAI i30

## huy động vốn đầu tư phát triển kết cấu hạ tầng giao thông đường bộ trên địa bàn tỉnh bắc giang

## NÂNG CAO NĂNG LỰC CẠNH TRANH CỦA HỆ THỐNG NGÂN HÀNG THƯƠNG MẠI VIỆT NAM TRONG ĐIỀU KIỆN HỘI NHẬP KINH TẾ QUỐC TẾ

## Kết cấu nhà cao tầng BTCT Võ Bá Tầm

## Tiểu luận môn tài chính tiền tệ Thông tin bất đối xứng

Tài liệu liên quan