The Pennsylvania State University
The Graduate School
Department of Agricultural Economics and Rural Sociology
THREE ESSAYS ON ORGANIC MILK MARKETING AND
CONSUMER PURCHASE BEHAVIOR
A Dissertation in
Agricultural, Environmental, and Regional Economics
by
Yan Zhuang
© 2010 Yan Zhuang
Submitted in Partial Fulfillment of the Requirements
for the Degree of
Doctor of Philosophy
August 2010
27
ii
The dissertation of Yan Zhuang was reviewed and approved* by the following:
Edward Jaenicke Associate Professor of Agricultural Economics Graduate Program Chair Dissertation Advisor Chair of Committee
Spiro Stefanou Professor of Agricultural Economics
Alessandro Bonanno Assistant Professor of Agricultural Economics
Debashis Ghosh Associate Professor of Statistics
*Signatures are on file in the Graduate School
iii
ABSTRACT
Using fluid milk as a case study, this dissertation focuses on purchase behavior
associated with two related consumer choices: choosing organic or non-organic milk,
and choosing private label (store brand) or national brand milk. Milk serves as an
excellent sector for a case study because organic milk sales are growing at increasing
rates, non-organic private-label milk composes a large share of the market, and organic
private-label milk is becoming more widely offered by U.S. supermarkets.
Essay I constructs four price series for the organic and non-organic private label
and branded milk categories, and estimates a 3SLS system to see how these four prices
react to each other. For non-organic milk, we find that the PL price and the branded price
are positively related, a result that is generally consistent with other research. This type
of price reaction can be labeled as cooperative and symmetric. One the other hand, for
organic milk, the national brand price and PL price react to each other differently. The
price of organic PL milk increases with an increase of organic national brand milk price,
while instead the organic national brand milk decreases price with a price increase of
organic PL milk. This price reaction falls into the asymmetric dominant-fringe price
competition, and fits the reality that organic national brand’s market share is currently
dominant in organic milk.
Essay II estimates factors that influence consumers' hazard for first organic milk
purchase. A discrete time hazard model reveals that demographic variables, such as age,
education, and household size affect the time of first organic milk purchase. Another
finding is that the organic milk price affects an organic milk purchase significantly, while
the non-organic milk price does not. Comparing the models with and without frailty, we
conclude that neglecting unobserved heterogeneity underestimates the coefficients.
Essay III uses an estimation of a two-stage decision model where customers are
assumed to first decide whether to buy organic or non-organic milk, and then, conditional
on that decision, decide whether to buy a store's private label brand or a national brand.
iv
Results show that the effect of shopping patterns, coupon redemptions, and other
marketing factors affect households’ private label choice in a similar fashion, no matter
organic or non-organic milk is first selected. However, demographic factors depend on
1st stage selection. Age and education significantly influence the private label choice for
households who select organic but not for those who select non-organic, while income
and household size only significantly influence non-organic buyers.
v
TABLE OF CONTENTS
List of Figures .............................................................................................................. vi List of Tables ............................................................................................................... vii Acknowledgements ...................................................................................................... viii
Chapter 1 Introduction and Motivation....................................................................... 1
Objective ............................................................................................................... 2 Contribution .......................................................................................................... 5
Chapter 2 Strategic Milk Price Reactions with Organic and Private Label Expansion ............................................................................................................. 9
Literature Review ................................................................................................. 11 Model Specification .............................................................................................. 14 Data ...................................................................................................................... 22 Results ................................................................................................................... 23 Conclusion ............................................................................................................ 31
Chapter 3 Organic Milk Trial .................................................................................... 34
Literature Review ................................................................................................. 35 Model Specification .............................................................................................. 37 Data ...................................................................................................................... 44 Results ................................................................................................................... 46 Conclusion ............................................................................................................ 52
Chapter 4 Consumers’ Choice of Private Label and National Brand Ornganic and Non-organic Milk ................................................................................................. 54
Literature Review ................................................................................................. 57 Model Specification .............................................................................................. 60 Data ...................................................................................................................... 62 Results ................................................................................................................... 72 Conclusion and Discussion ................................................................................... 80
Chapter 5 Linkages and Implications ........................................................................ 83
Pricing Strategies .................................................................................................. 83 Targeting Customers ............................................................................................. 85
References .................................................................................................................... 90
vi
LIST OF FIGURES
Figure 2-1. Organic Price Premium ............................................................................ 23
Figure 3-1. Frequency of Organic Milk Trials by Week ............................................ 46
Figure 3-2. Organic Milk Trial Frequency for Income ............................................... 51
Figure 4-1. 2004 to 2006 Weekly Prices for Four Categories of Milk ....................... 69
Figure 4-2. Four Milk Category Annual Prices .......................................................... 70
vii
LIST OF TABLES
Table 2-1. Variables Descriptions for the Price Reaction Paper ................................. 19
Table 2-2. Types of Price Competition and Estimation Coefficients … ..................... 20
Table 2-3. 3SLS Results for Four Milk Categories .................................................... 26
Table 2-4. Four Category Competition Types ............................................................ 27
Table 2-5. Market Share Changes and Organic Price Premium ................................. 30
Table 3-1. Variable Descriptions for the Organic Milk Trial Paper ........................... 45
Table 3-2. Results from Discrete Time Hazard Model ............................................... 48
Table 3-3. Coupon Use and Organic Milk Trial ......................................................... 49
Table 3-4. Income and Organic Milk Trial ................................................................. 50
Table 4-1. Milk Purchase Frequency ........................................................................... 64
Table 4-2. Frequency of Multiple Milk Type Purchase .............................................. 65
Table 4-3. Variable Descriptions for the Private Label Choice Paper ........................ 67
Table 4-4. Household Size Distribution ...................................................................... 68
Table 4-5. Average Annual Prices for Four Milk Categories. ..................................... 70
Table 4-6. Choice between Organic and Non-Organic Milk ...................................... 73
Table 4-7. Choice between Private Label and National Brand Milk ......................... 74
viii
ACKNOWLEDGEMENTS
My sincere gratitude goes to my academic advisor, committee members, all my
friends, and family members. I cannot imagine myself finishing the dissertation without
your support.
I would like to express my deepest gratitude to Professor Edward Jaenicke, my
esteemed academic advisor, for his valuable directions. Professor Jaenicke was very
generous in time and advice. He always read and responded to my questions more
quickly than I could have expected. Although Professor Jaenicke’s schedule was tight,
he always carefully read and patiently edited each chapter of my dissertation. Some of
the chapters were even read during weekends. His spoken and written comments were
always extremely perceptive, insightful and helpful. I can never thank Professor Jaenicke
enough for his endless patience and inspiring guidance.
Enormous thanks go to my dissertation committee. My research for this
dissertation was made more efficient by help from dissertation committee members.
Professor Steafanou mentored me from the very first day of my PhD study. Professor
Bonanno was always willing to help me with his best suggestions. Professor Ghosh gave
specific helpful advices from statistics aspect. I could never have reached the heights or
explored the depths without the efforts from the whole dissertation committee.
Many thanks go to my friends who accompanied me during the past five happy
years. My dearest friends, Jie Zhang from the Department of Linguistics and Dr. Qin
Chen graduated from the Department of Agricultural and Biological Engineering, helped
me going through the toughest days in my life. Yanxiang Zhao and Shishuo Fu from the
ix
Department of Mathematics are not only my friends who shared my hope and pressure of
graduate study, but also excellent mathematicians who are always ready to answer my
questions in mathematics. Discussing questions with my friends from AEREC and the
Department of Economics was always a pleasant experience. PhD study would have
been a lonely journey without the fun with all my friends.
My very special thanks go to my parents, my son, and my husband, who share my
happiness in good times and cheer me up in bad times. Thank you for always believing
in me and encouraging me to pursue my dream.
1
Chapter 1
Introduction and Motivation
With the increasing awareness of foods’ effect on health and environment, the
global organic food market is growing rapidly since early 1990s, with a growth rate of
19% in 2007 (Research and Markets, 2008). The global organic food market is expected
to reach $70.2 billion by the end of 2010. Of all the countries and regions in the world,
Europe has the largest share in global organic food sales, and North America is the
second largest market, growing at a compound annual growth rate (CAGR) of around
21% during 2005-2007. These two regions contributed around 96% of global organic
food revenues in 2007. Some governments, like Spain, Singapore and India, promote
organic food market by policies. Organic food market in Asia-Pacific is forecasted to
grow at a CAGR of approximate 10% during 2008-2010 (Research and Markets, 2008).
With the rapid growth of organic food demand, sales through supermarkets are
replacing the original sales from farmers’ market. Supermarkets and mass merchandisers
account for 53 percent of U.S. organic sales in 2006, compared to 47 percent of organic
sales from natural food stores (Schultz 2008). However, as consumers increasingly
bought organic food from supermarkets rather than from natural food stores or directly
from farmers, and with the differences between organic products and conventional
products not easily visible by consumers, a certified organic label was deemed necessary.
These reasons led many countries to give specific criteria for organic food, and to have
organic certification indicating that certain criterions are satisfied. Implemented in 2002,
2
the “USDA Organic” seal or other organic claims on products mean that products are
made of organic ingredients. A product may carry the “USDA organic” seal if it contains
at least 95% organic ingredients, and the remaining ingredients are approved for use in
organic products. Products that contain at least 70% organic ingredients may label those
on the ingredient listing. The listing of organic claim and ingredients are voluntarily
instead of required (Organic Trade Association, 2008).
The USDA organic seal on dairy products certifies that minimal amounts of off-
farm inputs are used. More specifically, organic dairy animals must be fed organic
materials, live in healthy living conditions, not be overcrowded, and have regular access
to outdoor air, sunlight and pasture (Schultz 2008). All feeding and health care records
must be maintained. According to Schultz (2008), nearly 2,300 dairy cows were certified
organic in 1992, and organic milk appeared in markets in 1993. The number of certified
organic dairy cows increases 469 percent from 1992 to 1997, and 421 percent from 1997
and 2002. In 2005, 87,082 dairy cows were certified organic, accounting for about 1
percent of all dairy cows (Schultz 2008).
Objective
In this dissertation, I examine organic milk and private label purchasing behavior
in the U.S. market using A. C. Nielson Homescan data from 2004 to 2006. Former
research results (Guadagni and Little 1983, Gupta 1998) show that a consumer’s decision
making is based on demographic characteristics (income, household size, etc.), and
affected by marketing efforts (price, coupons, etc.). Based on these results, I study how
3
demographic variables and marketing strategies affect households’ choice of organic
milk. Specifically, I will study the following questions:
1) Controlling for demographic and marketing factors, how do organic and non-
organic milk prices, for both private label and branded products, react to each
other, and how do these reactions affect the organic price premium?
2) For consumers not regularly buying organic milk, what factors affect the
probability of an organic milk trial1 purchase?
3) Do regular buyers of organic milk and non-organic milk approach private label
milk differently?
These questions are examined in three separate essays. Each essay is described in
one chapter. The first essay describes the data set, and studies how the prices of organic
milk and non-organic milk, both private label and branded, react to each other using a
Three Stage Least Square (3SLS) model. This essay is focused on firms’ oligopoly
behavior with differentiated products, where firms play a pricing game. Price
instruments are recovered from econometric estimation in Essay 1 and later used in
Essays 2 and 3. Both the second and the third essays use models based on utility
maximizing behavior of consumers. In the second essay, a discrete time hazard model is
developed to study consumers’ organic milk trial decisions. Trial is observed when the
utility of purchasing organic milk exceeds an individual-specific threshold. In the third
essay, the choice between private label milk and national brand milk is studied by a two-
stage sample selection model. In the first stage, milk consumers decide whether or not to
1 The term “trial” is used for the initial purchase of organic milk.
4
buy organic milk. Relative prices, promotional variables, consumption patterns, and
demographic factors are assumed to influence this first-stage decision. In the second
stage, consumption patterns, promotions, demographics and a different set of relative
prices are assumed to influence the private-label or national brand choice, conditional on
the outcome of the first stage. The estimation in the first essay is related to firms’
oligopoly behavior with differentiated products, where firms play a pricing game. Both
the second and the third essays use models based on utility maximizing behavior of
consumers. All three models are applied to the same data set that contains households’
milk purchase data from 2004 to 2006. Looked at collectively, the three-essay
dissertation will provide a thorough investigation into the evolution of organic milk and
private label milk pricing and consumer behavior. More specifically, food researchers
and managers will have insight into at least three broad questions:
1) How the prices of organic and non-organic private label and national brand react to
each other, so that strategies focused on product positioning, promotional efforts, and
targeting to specific consumer demographics can be selected to maximize profit.
2) What kinds of households will try organic milk earlier than other households, and
what kinds of households prefer private label milk to national brand milk, so that the best
promotional strategies can be chosen and targeted at appropriate households.
3) How marketing strategies, like temporary price reductions and coupon offerings,
affect consumers’ decision making on what type of milk to buy, so that retailers can use
different marketing strategies for organic milk and non-organic milk, or private label
milk and national brand milk.
5
Contribution
Despite the two prominent market trends on the growth of organic and private label
foods, very little has been written about how milk prices are expected to react as both the
organic share and private label share continue to increase. Treating the organic private
label, organic national brand, non-organic private label and non-organic national brand
milk as potential rivals, essay I contributes to the initial investigations of 1) the
competitive interactions and games played among the four milk categories, and 2) how
private label and/or national brand milk market shares affect the organic price premium
for private labels and national brands separately. Estimation results from the Three Stage
Least Square (3SLS) model show that 1) non-organic private label and non-organic
national brands behave cooperatively, that is, one player increases the milk price when
the other increases the milk price. 2) Organic and non-organic private labels behave
independently. This is as expected because both categories are offered by retailers. 3)
The price competition between non-organic national brand and organic national brand,
and between organic national brand and organic private label is dominant-fringe. The
fringe category follows the price actions of a strong rival, while the dominant category
behaves non-cooperatively to keep its market share. This result is consistent with the fact
that non-organic national brand milk is dominant in market share compared to organic
national brand milk, and that organic national brand milk is dominant in market share
compared to organic private label milk. 4) The cross category competition (organic
private label vs. non-organic national brand and non-organic private label vs. organic
national brand) follows a leader-follower reaction. The followers react to the leader’s
6
changes in actions, while the leader does not. Organic and non-organic private labels are
both the followers in the cross category competition. 5) For private label milk, we find
that the organic premium decreases with an increase in the PL market share. One the
contrary, an increase in the organic market share is always related to a higher PL price
premium. 6) For national brand milk, organic price premium decreases as either the PL
share or organic share increase.
Essay II studies individual level first organic milk purchase. Three empirical
issues associated with this topic arise: 1) Organic milk trial may happen before the study
period, during the study period, after the study period, or never happen. As researchers,
we only have information during the study period. Therefore, random censoring is
essential. That is, we assume that the censoring time is independent of the failure time.
2) Prices of organic and non-organic milk may be endogenous because there are some
omitted variables (e.g. milk quality, milk taste, etc.) that are correlated with consumers’
choice. A Two Stage Residual Inclusion (2SRI) method is used to deal with the
endogeneity issue. Statistical significant of the included residuals also confirms the
endogeneity of prices. 3) Unobserved heterogeneity exists among households, so
households with the same demographics and facing the same marketing activities may try
organic milk at different time. We assume households’ unobserved heterogeneity follows
a Gamma distribution, and add gamma frailty to capture the unobserved heterogeneity.
LR test shows that unobserved heterogeneity is significant. Linking a random utility
model with a discrete time hazard model, this essay concludes that 1) marketing factors
such as coupons and prices affect the timing of organic milk trial significantly. Coupon
users are more likely to have organic trial purchase. Decreasing the price of organic milk
7
will make households more likely to have organic milk trial. 2) Income, college
education, being Asian or Hispanic, and out of category organic expenditure are
positively related to consumers’ organic milk trial hazard. 3) Household size and total
out of category expenditure are negatively related to consumers’ organic milk trial
hazard.
Essay III is among the first effort to investigate whether organic and non-organic
consumers approach private labels or national brands differently. We model the purchase
decision in two connected steps to correct for sample selection bias. Households first
decide whether to buy organic milk, and then whether to buy private label milk
conditional on the first step selection. The Two Stage Sample Selection model results
show that 1) when choosing between private label and national brand milk, marketing
factors do not depend on the first stage choice of organic and non-organic milk. For
example, shopping patterns, coupon redemptions, and prices affect organic and non-
organic buyers in a similar fashion. On the other hand, 2) demographic results depend on
first stage choice of organic and non-organic milk. For example, age and education
significantly influence organic buyers but not non-organic buyers, while income and
household size significantly affect non-organic buyers but not organic buyers. 3) From
the magnitude and significance level of prices, we can see that organic private label price
is the key price for organic private label choice, and non-organic national brand price is
the key price for non-organic private label choice. Along the way to these findings on the
private label choice, we uncover some interesting results from the first-stage organic
decision. 1) Income, college education, and appearance of children are positively related
to consumers’ organic choice, while household size and age are negatively related to
8
consumers’ organic choice. 2) One new result focuses on gender-based employment
levels in the household: For dual-headed households, we find that full-time employment
by a female head can, under some circumstances, decrease the likelihood of buying
organic milk.
9
Chapter 2
Strategic Milk Price Reactions with Organic and Private Label Expansion
In recent years, supermarket managers and food shoppers have witnessed the
intersection of two important food trends, namely the increasing prominence of private
label (PL) food products (also known as store brands), and the high-paced market growth
of organic foods. Dimitri and Oberholtzer (2009) focus attention on this intersection
when they report that the share of organic products sold under PLs has increased from 8
percent in 2003 to 17.4 percent in 2008. In the milk category, the focus of this research,
Dimitri and Oberholtzer (2009) report that the market share for organic PL more than
doubled recently.
Since the 1970s and 1980s, private label products have seen great improvements
in product quality and large gains in market share. Of all the food products, dairy is one
of the categories with highest private label expenditure and market share. Citing
Information Resources Inc., Smith (2005) claims in the professional journal Dairy Field
that private label milk sales rose 6.8% during the 52 weeks ending February 20, 2005,
with the dollar market share of private label milk being 58.7 percent. For the 52 weeks
ending May 19, 2007, the trade publication Progressive Grocer reports that total private
label sales across all grocery categories reached $46.5 billion. Among all categories,
private label milk leads the way with $6.5 billion (Progressive Grocer 2007).
The market for organic food is growing even faster than that for PLs. Dimitri and
Oberholtzer (2009) report that the retail sales of organic foods up to $21.1 billion in 2008
from $3.6 billion in 1997. Citing survey results from Hartman Group, Dimitri and
10
Oberholtzer (2009) report that 69 percent of adults bought organic food at least
occasionally in 2008, and 19 percent of consumers bought organic food weekly in 2008,
up from 3 percent in the late 1990s.
Despite the two prominent trends briefly documented above, very little has been
written about how prices are expected to react as both the organic share and PL share
continue to increase. Using milk as a case study, this paper analyzes purchase data to
investigate this issue. More specifically, this paper has two main objectives: (i) to model,
estimate, and empirically test for price reactions of PL and branded organic and non-
organic milk, and (ii) to explore, using the estimation results and counterfactuals , how
prices react to continued strong market growth. Fulfilling these two objectives will allow
us address the related question of how the observed organic price premium (for both PL
and branded milk products) may change if market growth continues.
Estimation results from the Three Stage Least Square (3SLS) model show that 1)
there is evidence of cooperative behavior non-organic private label milk prices and non-
organic national brand milk prices. A price increase in one category leads to a price
increase in the other. 2) The two milk categories offered by retailers, organic and non-
organic private label milk, are independent from each other, as expected. 3) The price
competition between non-organic national brand and organic national brand, and between
organic national brand and organic private label is dominant-fringe. The fringe one
follows the price actions of a strong rival, while the dominant one behaves non-
cooperatively to keep its market share. 4) The cross category competition (organic
private label vs. non-organic national brand and non-organic private label vs. organic
national brand) follows a leader-follower reaction. The followers react to the leader’s
11
changes in actions, while the leader does not. Organic and non-organic private labels are
both the followers in the cross category competition. Simulations of market share
changes show that 1) For private label milk, we find that the organic premium decreases
with the increase of PL market share. One the contrary, an increase in the organic market
share is always related to a higher PL price premium. 2) For national brand milk,
organic price premium decreases as either the PL share or organic share increase.
Literature Review
A pertinent line of research investigates strategic competition between private
label and national brand food products, as well as the relation between market share and
prices. This research focuses on price-setting competition, and how manufacturers in one
category may react to the pricing decisions by those in the other category. Examples of
this research include Putsis (1997and 1999); Putsis and Dhar (1998); Coterill and Putsis
(2000); Coterill, Dhar and Putsis (1999); Coterill, Putsis and Dhar (2000); Bonanno and
Lopez (2005); Ward et al. (2002); Bontemps et al. (2005); Bontemps et al. (2008).
Steiner (2004) reviews the history of national brand and private label competition.
Putsis (1997) provides a relevant example of this literature. He investigates price
interaction and market share effect using IRI scanner data. Results show that price of
private label are positively correlated with national brand products, and market share is
positively related to own price. Coterill, Dhar and Putsis (1999) develop a framework to
estimate market share and price reaction simultaneously, and find a positive relationship
between shares and prices on the supply side and a negative relationship on the demand
12
side. They also find that branded price is higher in the markets dominated by national
brands. Coterill and Putsis (2000) estimate a system of market share and price equations
simultaneously, and find that positive price reaction between private labels and national
brands is present, but not strong. They also find that markets with higher national brand
market share and supermarket concentration tend to have higher prices for both national
brands and private labels. Bontemps et al. (2008) study private labels, national brands
and food prices using data from a consumer survey for 218 food products. They find a
significant and positive relation between price of national brands and private label
development. After controlling of quality effect, the relation is still positive and
significant.
A wide range of studies investigate the price premium for organic products,
and/or why some consumers are willing to pay extra for it.
Some researches on willingness to pay for organic produce feature “hypothetical”
data about stated preference. Consumers are asked in a survey about their hypothetical
behavior that involves a willingness to pay for organic generally or the price premium
more specifically. Gil, Gracia and Sánchez (2000) use hypothetical data to study
willingness to pay for organic product. They examine survey data from two Spanish
regions with large organic product consumption. Results show that consumers that care
more about healthy diet and environmental damage are willing to pay a high premium for
organic food. Yiridoe et al. (2005) give a review of literatures that study the price and
willingness to pay for organic products. After reviewing 21 selected studies on price
premium for organic products, they conclude that 1) Consumers tend to pay a higher
13
price premium for organic products with a shorter shelf life; 2) The proportion of
respondents willing to pay a price premium decreases as the premium level increases.
Some researches use real purchase data to study willingness to pay for organic
products. Casadesus-Masanell et al (2009) use data from an outdoor wear maker, which
substituted organic cotton for conventionally grown cotton in all of its sportswear. They
find that customers were willing to pay significant premiums for organic garments.
Thompson and Kidwell (1998) study consumer choices of organic and conventional
produce using actual purchase data made in retail outlets. They find that organic food
prices are significantly higher than conventional food prices. The average actual price
premiums range from 40% to 175%.
Homescan data is a specific kind of real purchase data that are scanned by
participating households after each grocery purchase. There is a research line that
investigates how much more consumers would like to pay for organic product using
scanner data. Huang and Lin (2007) study the price premium of fresh tomatoes in
different regions using 2004 Nielsen Homescan panel data. Their estimation results show
that consumers pay $0.25 more per pound in the New York-Philadelphia market, $0.14
more in the Chicago-Baltimore/Washington and the Los Angeles-San Francisco markets,
and $0.29 more in the Atlanta-San Antonio market for organic tomatoes. Using Nielsen
Homescan panel data in 2005, Lin, Smith and Huang (2008) estimate price premiums and
discounts for five major fresh fruits and five major fresh vegetables in the United States.
The organic price premiums vary from 20% to 42% for fruits, and vary from 15% to 60%
for vegetables. Using retail purchases from the 2006 Nielsen Homescan panel data,
Smith, Huang, and Lin (2009) find that organic price premiums for half-gallon milk
14
range from $1.23 for whole private label organic milk (60%-68% above conventional
counterpart) to $1.86 for nonfat/skim-branded organic milk (89%-109% above
conventional counterpart).
In this essay, we investigate the milk category because it is a product that has both
a strong private label and organic presence. Milk is often found in the marketing and
agricultural economics literature. Citing Information Resources Inc., Barstow (2005)
claims that about 60 percent of milk is sold under a store brand. Bonanno and Lopez
(2005) examine how private label market share affects prices of reduced-fat and whole
milk using IRI data for 24 supermarket chains in 10 cities. The negative relationship
between milk prices and private label market share, and the positive relationship between
milk prices and the square of private label market share suggest a “U” shape relation
between milk prices and private label market share. Milk and organic milk have also
been studied as part of a demand analysis (Glaser and Thompson 2000, Tian and Cotterill
2005, and Chidmi, Lopez and Cotterill 2005).
Model Specification
In this paper, we will investigate the prices of four milk categories (i.e. i. organic
private label, ii. organic national brand, iii. non-organic private label, and iv. non-organic
national brand), and will explicitly assume that prices in each category can reflect price
setting behavior by dairy firms. In essence, the four categories of milk are treated as
potential rivals. One justification for this assumption is the presence of three large firms
in this sector, Horizon Organic, Organic Valley, and Aurora Dairy, a firm specializing in
15
private label organic milk (Dimitri and Venezia, 2007). Dimitri and Oberholtzer (2009)
report that Horizon Organic’s 2007 share of the organic milk market was 33 percent,
Organic Valley’s share was 19 percent, and the private label share was 27 percent. For
our investigation, we will estimate a simultaneous equation system where endogenous
prices of each of the four milk categories are modeled as being dependent on the other
category prices as well as other non-price factors. These other factors include market
shares, demographic variables, and product attributes. Our goals, therefore, are to see if
the four milk category prices do in fact react to each other, and, if they do, to use this
information to investigate outcomes reflecting hypothetical market conditions.
Factors that affect the price of the main milk category i (e.g., the price of private
label organic) can be divided into five groups: 1) a vector D of consumer demographic
information for specific market area , including the percentage of households with high
income, average household size, average age, and ethnicity (as measured by percentage
of Hispanic, African American, Asian, and White households); 2) a vector C of product
properties, including the percentage sold in large volume containers, and the percentage
having containers of certain materials (carton, glass, or plastics); 3) a vector X of market
structure variables, including market share of private labels and organic milk, and 4) a
vector P-i of prices of the other milk categories; 5) A time variable is also included to
capture the price change over time.
The four price reaction functions should include supply-side as well as demand-
side variables. Because prior research shows that the price of raw milk is found to have
no significant influence on milk prices (Bonanno and Lopez 2005; Chidmi, Lopez, and
Cotterill 2005), and because there is no raw milk price variable available in our data set,
16
we do not include price of raw milk in the model. However, the supply side is reflected
by the material of milk containers. We therefore estimate four equations as following:
for i = 1, 2, 3, 4
where θ α , β , γ , δ for 1,2,3,4 0,1,2,3,4,5 are coefficients, and
are error terms. We can also write the four equations in the following form using
variables in the estimation:
1
2
17
3
4
Definitions and statistics for the relevant variables are in Table 2-1.
Coefficients , , , and characterize the price interaction between private
labels and national brands within organic and non-organic milk separately. Coefficients
, , , and represent the price interaction between organic and non-organic for
private label and national brand milk separately. Coefficients , , , and represent
18
the crossing categories2 price interactions between organic private label and non-organic
national brand milk, and between organic national brand and non-organic private label
milk. Each of these twelve coefficients is the direct marginal effect of $1 increase in one
category on the price of another category. Using the mean values of prices, we can
calculate price elasticity.
Putsis and Dhar (1998) divides the competitive interaction into symmetric and
asymmetric patterns. Symmetric interaction implies a similar response to its rival, and
asymmetric interaction implies a different response to its rival. Independent, cooperative,
and non-cooperative interactions are three types of symmetric response. Independent
players do not respond to their rivals. Cooperative players increase prices with their
rivals, and non-cooperative players decrease prices when their rivals increase prices.
Leader-follower (Stackelberg) and dominant-fringe are two types of asymmetric price
reactions. Leader-follower behavior implies that the follower reacts to changes in the
leader’s actions, while the leader does not. The dominant-fringe form allows two rivals
to act oppositely: that is, one acts cooperatively while the other acts non-cooperatively.
For example, the fringe one may simply follow a strong rival, but the dominant one may
want to protect its share by taking non-cooperative actions.
2 By saying “crossing categories”, we mean organic private label vs. non‐organic national brand, organic national brand vs. non‐organic private label, non‐organic private label vs. organic national brand, and non‐organic national brand vs. organic private label.
19
Table 2-1. Variable Descriptions for the Price Reaction Paper
Mean Std. Err. Description orgplprice 0.0455267 0.0001432 Organic private label price orgnbprice 0.0526954 0.0001154 Organic national brand price norgplprice 0.0242183 0.0001514 Non-organic private label price norgnbprice 0.0385027 0.0002708 Non-organic national brand price plshare 0.6291639 0.0040392 Private label market share orgshare 0.3149908 0.0036999 Organic market share week 93.91596 0.8247687 Time index avgmaxage 6.889837 0.0093556 Average max household heads' age avghhsize 2.58179 0.0059671 Average household size highincper 0.3737727 0.0025533 Percentage of households with a high income Hispanicper 0.0832654 0.0016456 Percentage of hispanic AAper 0.0652076 0.0013117 Percentage of African American Asianper 0.0423366 0.0009992 Percentage of Asian Whiteper 0.8221613 0.0021526 Percentage of white bigvolorgplper 0.0020313 0.0001943 Percentage of big volume container for organic private label milk cartonorgplper 0.0713972 0.0013016 Percentage of carton container for organic private label milk glassorgplper 0 0 Percentage of glass container for organic private label milk bigvolorgnbper 0.021203 0.0007665 percentage of big volume container for organic national brand milk cartonorgnbper 0.1511884 0.0024642 Percentage of carton container for organic national brand milk glassorgnbper 0.0014234 0.0001219 Percentage of glass container for organic national brand milk bigvolnorgplper 0.3975402 0.004224 Percentage of big volume container for non-organic private label milk cartonnorgplper 0.0493456 0.0014533 Percentage of carton container for non-organic private label milk glassnorgplper 0.0000545 0.0000267 Percentage of glass container for non-organic private label milk bigvolnorgnbper 0.0383044 0.0015661 Percentage of big volume container for non-organic national brand milk cartonnorgnbper 0.0605139 0.001459 Percentage of carton container for non-organic national brand milk glassnorgnbper 0.0026618 0.0003436 Percentage of glass container for non-organic national brand milk orgplcerealper 0.8218611 0.0426685 Organic private label cereal expenditure orgnbcerealper 15.00919 0.2569846 Organic national brand cereal expenditure norgplcerealper 36.94441 0.4650527 Non-organic private label cereal expenditure norgnbcerealper 528.9248 3.937213 Non-organic national brand cereal expenditure
20
Table 2-2. Types of Price Competition and Estimation Coefficients
Symmetric Interaction Asymmetric Interaction
Price
Competition
Categories
Independent
(Nash)
Cooperative Non-
cooperative
Leader-
Follower
Dominant-
Fringe
Org PL vs. Org
NB
0,
0
0,
0
0, 0 0,
0
0, 0
0,
0
0, 0
Non-org PL vs.
Non-org NB
0,
0
0,
0
0, 0 0,
0
0, 0
0,
0
0, 0
PL Org vs.
PL Non-org
0,
0
0,
0
0, 0 0,
0
0, 0
0,
0
0, 0
NB Org vs.
NB Non-org
0,
0
0,
0
0, 0 0,
0
0, 0
0,
0
0, 0
PL Org vs.
NB Non-org
0,
0
0,
0
0, 0 0,
0
0, 0
0,
0
0, 0
NB Org vs. PL
Non-Org
PL Non-Org
0,
0
0,
0
0, 0 0,
0
0, 0
0,
0
0, 0
21
Following the competition types described by Putsis and Dhar (1998) and Putsis
(1999), we divide the price competition into several groups according to the signs and
significance of the price interaction coefficients (Table 2-2).
Two empirical issues arise when estimating the price reactions system above.
First, the error terms of the four equations may be correlated, because they are using the
same data set and similar variables. Secondly, some variables are endogenous. Price
variables are endogenous due to the structural system. Market shares may not be
orthogonal to the error terms (Bontemps et al. 2005; Ward et al. 2002). A Durbin-Wu-
Hausman test shows that private label milk market share (PLSHARE) and organic milk
market share (ORGSHARE) are endogenous for all the equations except for the one with
organic private label milk as dependent variable3. Therefore, we treat market shares as
endogenous variables in our model. Given these empirical issues, we choose for our
empirical model Three Stage Least Squares (3SLS), which provides consistent estimation
and allows correlation among error terms. 4 Cereal expenditures are included as
instrumental variables besides the exogenous variables in the four price reaction
equations. We choose cereal because it is usually purchased or consumed with milk, so
they are complementary goods to some extent. We calculated cereal expenditures on four
categories (organic private label, organic national brand, non-organic private label, and
non-organic national brand) for each market in each week, and use these expenditures as
additional instrumental variables. Therefore, in this paper, a four-equation 3SLS model
3 We tested the endogeneity of market shares by including the residuals of private label market share and organic market share in the regression system. The coefficients of market share residuals are all significant except for the equation with organic private label price as dependent variable. This non‐significant result may due to the low market share of organic private label milk. 4 For more details about 3SLS, please see Greene (2003).
22
is estimated to investigate milk prices, and reactions for four milk categories (organic and
non-organic private label, organic and non-organic national brand).
Data
The data used in this study are Nielsen Homescan data on food products across 52
geographic markets from 2004 to 2006. Homescan data are usually used to study
marketing activities and consumer purchase behavior in agribusiness and agricultural
economics literatures (Smith, Huang, and Lin 2009; Arnade, Gopinath, and Pick 2008).
The 52 markets are directly from the Nielsen standard coding, representing major and
mid-sized markets in the United States. When the original household-level data on milk
purchases are aggregated to the market level, we are left with 8104 observations in the
data set. Prices of four milk categories are the mean values of final milk purchase prices
in each market for each week. Figure 2-1 shows the weekly prices for organic private
label, organic national brand, non-organic private label, and non-organic national brand
milk in market 12 (San Francisco) from 2004 to 2006. Price trends of other markets are
similar.
Other variables in the model include market shares, demographics, and percentage
of various product properties. Figure 2-1 shows how private label and national brand
organic price premium change with time. Expenditures on cereal are included as
instrumental variables. We choose the outside category of fresh and frozen meat and
produces because milk is usually purchased on a weekly basis with fresh and frozen meat
and produce. We assume that the shopping behavior in the category of fresh and frozen
23
meat and produces reflect a households’ grocery shopping patterns. What is more, there
are no products that are obviously correlated with the choice of milk.
Figure 2-1. Organic Price Premium
Results
Table 2-3 presents the results of the four price equations (1) – (4) estimated via
3SLS. The results and interpretation of specific factors are discussed in turn.
0
0.5
1
1.5
2
2.5
3
3.5
0 20 40 60 80 100 120 140 160 180
Ratio of Organ
ice an
d Non
‐organ
ic M
ilk Price
Week
Organic Price Premium
PL Organic Price Premium NB Organic Price Premium
24
Price Reactions
For non-organic milk, we find that the private label price and the branded price
are positively related. This result is generally consistent with other research (Putsis 1997;
Bontemps et al. 2008).5 According to Table 2-3, non-organic private label and non-
organic national brand have the symmetric cooperative price reaction. Non-organic
private label price will increase by $0.167 if non-organic national brand price increases
by $1. In terms of elasticity, that is a 0.27 percent increase in organic private label price
with 1 percent increase in the organic national brand price6. On the other hand, non-
organic national brand price will increase by $0.203 if non-organic private label price
increases by $1, which is a 0.13 percent increase in non-organic national brand price
increase with 1percent increase in non-organic private label price.
One the contrary, for organic milk, the national brand price and private label price
react to each other differently. The price of organic private label milk increases by
$0.294 with $1 increase of organic national brand milk price, while organic national
brand milk decreases its price by $0.271 with $1 increase of organic private label milk.
These reactions in absolute dollar values correspond to a 0.34 percent increase in organic
private label milk price with 1 percent increase in organic national brand price, and a 0.23
percent decrease in organic national brand price with 1 percent increase in organic private
label milk price. This price reaction fall into the asymmetric dominant-fringe price
competition, and fits the reality that organic national brand is dominant in organic milk.
5 These researches study the price reaction between private labels and national brands. Since non‐organic products dominate organic products for most product categories, we believe the results are similar to those of non‐organic products. 6 All the price elasticity are calculated by STATA command “eyex”.
25
For private label milk, organic private label price decreases with the increase of
non-organic private label price, and vice versa. However, neither of the reactions is
significant at 5% level, and the magnitude is relatively low (both below $0.007 in
absolute dollar value and 0.09 percent in elasticity). This result suggests that private label
organic milk and private label non-organic milk set their price independently.
For national brand milk, a $1 increase in non-organic national brand price will
bring a $0.131 increase in organic national brand price, corresponding a 0.08 percent
increase in organic national brand price with 1 percent increase in non-organic national
brand milk price. On the other hand, a $1 organic national brand price will bring $0.606
decrease in non-organic national brand price, corresponding a 0.829 percent decrease in
non-organic national brand milk price with 1 percent increase in organic national brand
milk price. These responses suggest an asymmetric dominant-fringe price competition.
Comparing to organic national brand using the Homescan data, non-organic national
brand is dominant in market share.7
For crossing categories, the organic and non-organic private label milk prices
react to cross categories significantly, while organic and non-organic national brand milk
reacts to cross categories non-significantly. To be specific, the price of organic private
label milk will increase by $0.102 if the non-organic national brand price increase by 1$,
corresponding a 0.08 percent in terms of price elasticity, while the non-organic national
brand does not react to organic private label significantly. The non-organic private label
7 Using Nielsen Homescan data from 2004 to 2006, we find that non‐organic national brand milk doubles the market share of organic national brand milk (21.4% vs. 11.2%).
27
26
Table 2-3. 3SLS Results for Four Milk Categories
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
orgplprice orgnbprice norgplprice norgnbprice orgnbprice 0.2944592*** 0.0919 orgplprice -0.271264*** 0.0586 orgplprice -0.0482048 0.0551 orgplprice 0.2092923 0.1282
norgplprice -0.06326* 0.0373 norgplprice -0.040852 0.0313 orgnbprice -0.359517*** 0.0772 orgnbprice -0.6057241*** 0.1740
norgnbprice 0.1018668** 0.0411 norgnbprice 0.1305724*** 0.0331 norgnbprice 0.1669019*** 0.0307 norgplprice 0.2029762*** 0.0670 plshare -0.0009981 0.0020 plshare -0.0034466** 0.0015 plshare 0.0095621*** 0.0014 plshare 0.0117516*** 0.0038
orgshare -0.0012834 0.0028 orgshare -0.0036946** 0.0016 orgshare -0.0103321*** 0.0016 orgshare 0.0104255*** 0.0039
week 0.0000184*** 0.0000 week 0.0000674*** 0.0000 week 0.0000412*** 0.0000 week -0.000021* 0.0000
avgmaxage 0.00000243 0.0004 avgmaxage -0.0000509 0.0003 avgmaxage -0.0018586*** 0.0003 avgmaxage 0.0019735*** 0.0007
avghhsize 0.0017082*** 0.0006 avghhsize -0.0002389 0.0005 avghhsize -0.0019358*** 0.0004 avghhsize 0.0009949 0.0011
highincper 0.001719 0.0014 highincper 0.0067523*** 0.0010 highincper 0.0021708** 0.0011 highincper 0.0067808*** 0.0026 Hispanicper 0.0030892 0.0030 Hispanicper -0.0017267 0.0023 Hispanicper 0.0054972** 0.0024 Hispanicper -0.0125496** 0.0056 AAper 0.0134241*** 0.0041 AAper 0.0086965*** 0.0032 AAper 0.0236963*** 0.0032 AAper -0.006297 0.0078
Asianper 0.0304655*** 0.0051 Asianper 0.0161547*** 0.0042 Asianper -0.0133759*** 0.0042 Asianper
-0.0369346*** 0.0099
Whiteper 0.009274*** 0.0035 Whiteper 0.0000891 0.0027 Whiteper 0.0039648 0.0028 Whiteper-0.0182915*** 0.0065
bigvolorgplper -0.1271176*** 0.0150 bigvolorgnbper
-0.0454457*** 0.0042 bigvolnorgplper
-0.0298981*** 0.0013 bigvolnorgnbper -0.042445*** 0.0052
cartonorgplper 0.0043783 0.0037 cartonorgnbper -0.0065754*** 0.0021 cartonnorgplper 0.0148174*** 0.0020 cartonnorgnbper 0.0150419*** 0.0052
glassorgplper (dropped) glassorgnbper 0.1341249*** 0.0202 glassnorgplper 0.020627 0.0717 glassnorgnbper 0.0315934** 0.0143 _cons 0.0117572* 0.0069 _cons 0.0570387*** 0.0043 _cons 0.0557475*** 0.0054 _cons 0.0472648*** 0.0138 Note: *** means significant at 1%, ** means significant at 5%, and * means significant at 10%. Coefficient for “glassorgplper” is dropped in the 2nd column because it equals 0 for each market in each week.
27
price will decrease by $0.36 if the organic national brand increase price by$1,
corresponding a 0.78 percent decrease in terms of price elasticity, while the organic
national brand price does not react to non-organic private label price significantly. These
responses suggest a leader-follower type of asymmetric price competition. The non-
organic national brand is the leader in the competition with organic private label milk,
and organic national brand is the leader in the competition with non-organic private label
milk.
Table 2-4 summarized the price competition types among the four milk
categories.
Table 2-4. Four Category Competition Types
Price Comepition Categories Price Competition Types Organic PL vs. Organic NB
Asymmetric Dominant (Organic NB)-Fringe (Organic PL)
Non-org PL vs. Non-org NB Symmetric Cooperative PL Organic vs. PL Non-org Symmetric Independent NB Organic vs. NB Non-org
Asymmetric Dominant (NB Non-org)-Fringe (NB organic)
Organic PL vs. Non-org NB
Asymmetric Leader (Non-org NB)-follower (Organic PL)
Non-org PL vs. Organic NB
Asymmetric Leader (Organic NB)-follower (Non-org PL)
Market Shares
The market shares of private label milk and organic milk affect milk prices
significantly except for organic private label milk. A higher average private label market
28
share is related to lower organic national brand price, but higher non-organic private label
and national brand prices. A higher organic market share is related to higher non-organic
national brand price, but lower organic national brand and non-organic private label
price. We can see from Table 3 that the magnitude of organic share coefficient is always
similar to or higher than private label share coefficient.
Demographics and Products Attributes
Table 2-3 shows that market-level demographic variables affect the milk
categories’ prices differently. We summarize seven demographic results: 1) Higher
average age of household heads is related to lower price of non-organic private label milk
price and higher non-organic national brand milk price. It does not affect two organic
milk categories significantly. 2) Larger average household size in a market is related to
higher organic private label price, and lower non-organic private label price. It does not
affect two national brand milk categories. 3) A higher percentage of high income8
households in a market relates to higher prices for all the four milk categories. 4) A
higher percentage of Hispanic households relates to a higher non-organic private label
price and a lower non-organic national brand price. It does not affect two organic milk
categories significantly. 5) The percentage of African Americans does not affect the non-
organic national brand milk price significantly, but it affects all the other three milk
categories positively and significantly. 6) A higher percentage of Asian households in a
8 The Homescan data contains 16 category codes for particular income ranges. Households in the second highest income category ranges from $70,000 to $99,999, and the top income category containing households with more than $100,000 a year in income. We define high income as the households with annual income in the highest two categories.
29
market is related to lower prices of non-organic milk categories and higher prices of
organic milk categories. 7) The percentage of white households is related to higher price
of organic private label milk and a lower price of non-organic national brand milk, but it
does not affect the other two categories significantly.
In each of the four milk categories, product attributes such as big volume
containers, carton packaging and glass packaging are calculated as percentage of products
in the market. Here, three results emerge: 1) A higher percentage of own category big
volume packaging is always related to lower milk prices. 2) Own category carton
container percentages is related to higher milk prices except for organic national brand
milk. 3) Glass container percentage has positive relation with category prices otherwise,
with the effect on non-organic private label milk prices non-significant. They are zero
for organic private label milk in all the weeks, so the coefficient is dropped for this
category.
Organic Price Premiums
Using the Table 2-3 results, we construct counterfactuals based on new scenarios
for market shares of private label milk and organic milk. These scenarios include 10, 20,
and 30 percent absolute value increases in market shares. Table 2-5 reports these
counterfactual results and shows how an organic price premium for private label and
national brand milk separately will change with the changes in the market share of private
label milk and/or organic milk. We can see that private label organic premium decreases
with the increase of private label market share. A 30 percent increase in private label
30
market share will result in a 21 percent decrease in private label organic price premium.
One the contrary, the increase of organic market share is always related to a higher
private label price premium. A 30 percent increase in the organic share will result in a 26
percent increase in private label organic market share. Because of the opposing effects
on the private label organic price premium, the simultaneous increase of private label
share and organic share can even out the effect. For example, the private label price
premium are virtually unchanged if both private label share and organic share increase by
10 percent.
Table 2-5. Market Share Changes and Organic Price Premium
Market Share PL-O/PL-Non NB-O/NB-Non Current Shares 1.879847058 1.368615707PLShare + 10% 1.804480725 1.319391903OrgShare+10% 1.958089282 1.323188849PL&OrgShareEach + 10% 1.876390252 1.276541979PLShare + 20% 1.734626321 1.273002538OrgShare+20% 2.043626252 1.28010141PL&OrgShareEach + 20% 1.872915481 1.19397413PLShare + 30% 1.669707497 1.229208008OrgShare+30% 2.137528056 1.239166571PL&OrgShareEach + 30% 1.869414232 1.11952148
As opposed to the private label organic price premium, the national brand organic
price premium decreases with the increase of the private label share and organic share. A
30 percent increase in private label share will result in a 14 percent decrease in national
brand organic price premium, and a 30 percent of increase in organic share will result a
31
13 percent decrease. If private label share and organic share increase by 30 percent each,
national brand organic price premium will decrease by 25 percent.
These counterfactuals demonstrate the important role that the private label and
national brand segments may have over time on the price premium. While the national
brand price premium may shrink as organic proponents fear, the private label premium is
likely to remain. This result would generally be received as good news for organic dairy
farmers or milk producers.
Conclusion
Using Nielsen Homescan data set from 52 markets in the United States, this paper
assesses the price interactions among the four fluid milk categories (organic private label,
organic national brand, non-organic private label and non-organic national brand), how
demographic variables and product properties in a market affect milk prices, and the
impacts of private label and organic milk market shares on milk prices. Results from
empirical analysis show that 1) private label organic premium decreases with private
label market share but increase with organic market share, while national brand organic
price premium decreases with both private label share and organic share; 2) Types of
price competition among the four milk categories include symmetric cooperative (non-
organic PL vs. non-organic NB), symmetric independent (organic PL vs. non-organic
PL), asymmetric dominant-fringe (organic NB vs. organic PL and non-organic NB vs.
organic NB), and asymmetric leader-follower (non-organic NB vs. organic PL and
organic NB vs. non-organic PL).
32
Although this paper focuses on the price competition among the four milk
categories, the methodology can be easily used on price competitions among different
brands. Results from this paper will give food manufacturers, retailers, and food
researchers insights on how to make effective price strategies, how to react to expansion
of private label and organic products, and how to target a new market. To be specific,
this paper has the following implications:
Firstly, managers can use the results about organic price premium to run market
campaign. For example, when a manager plans to run a campaign to increase the market
share of private label milk, he or she should be aware that higher private label market
share is related to a lower organic price premium for both private labels and national
brands in a market. Changes in organic price premium may result in the changes of
profit. Therefore, when evaluating the return of investment for a campaign, managers
may want to consider market share and organic price premium collectively.
Secondly, private label and organic market shares have opposite effect on private
label organic price premium, while have similar effect on national brand organic price
premium. Therefore, product managers should be aware of that the simultaneous
expansion of private label and organic market share may not affect private label organic
price premium a lot, because the effects may be evened out, however, it will affect
national brand organic price premium in a large magnitude, because both market shares
are related with lower national brand organic price premiums.
Thirdly, managers of organic national brand milk may find it useful to know that
their price changes are followed by organic private label and non-organic private label
33
milk, and non-organic national brand milk will act non-cooperatively to them keep the
dominant market share.
Fourthly, managers of private label milk may find the information in this paper
helpful for them to predict the reactions from national brand milk. Results of this paper
show that private labels, both organic and non-organic milk, follow the price changes of
national brands. Private label milk managers will expect cooperative price reactions from
non-organic national brand to non-organic private label milk, no responses from non-
organic national brand to organic private label and from organic national brand to non-
organic private label milk, and non-cooperative price reaction from organic national
brand to organic private label milk.
Finally, the relation between market structure and prices may help managers to
target a new market. For example, organic national brand milk managers may want to
locate a new market with lower private label market share, lower organic milk market
share, and more high income households than the other markets. On the other hand, non-
organic national brand milk managers may want to locate a new market with higher
private label market share, higher organic market share, higher average age of
households’ heads, more high income households than the other markets.
34
Chapter 3
Organic Milk Trial
Organic milk sales are growing at increasing rates since the mid-1990s.
According to a USDA report, U.S. retail sales of organic milk and cream edging over $1
billion in 2005, up 25 percent from 2004 (Dimitri and Venezia, 2007). A survey by the
Food Marketing Institute and Prevention magazine shows that 30 percent of respondents
purchased organic milk or other dairy products in the first six months of 2006 (Food
Marketing Institute and Prevention, 2006). According to the Organic Trade Association
(2008), sales of organic milk in 2007 were over $1.3 billion, accounting for 2.7 percent of
the nation’s total milk sales.
The fast growth of organic milk sales is based on repeated purchase as well as
trial (first time) purchase. Trial purchase of organic milk is important because consumers
usually test the actual quality of milk by experiencing it. Hein (2008) states on website
Brandweek that fifty-eight percent of customers will buy a product again after trying it.
So, organic milk trial is critical for increasing organic milk sales. Despite the importance
of organic milk trial, little research is done about factors affect consumers’ organic milk
trial purchase timing. In this paper, we will estimate purchase timing decisions for
households’ fist time organic milk purchase.
35
Literature Review
Trial decisions can be considered as two separate decisions. On the one hand,
consumers need to decide whether to switch from the current category to a new category.
On the other hand, consumers implicitly need to decide when to switch. Therefore,
research results from brand switching and purchase timing will give some insights of trial
analysis.
The behavior of brand switching has been the research of interest for marketing
researchers for a long time (Lehmann 1972; Bass 1974; Kalwani and Morrison 1977;
Jain, Bass, and Chen 1990). Vilcassiam and Jain (1991) use IRI data to study frequently
purchased products, and find that marketing mix and demographic variables can explain a
large part of variation in brand switching rates. Switching rate induced by promotion is
negatively related to the share of purchase. Wedel et al. (1995) study brand switching
using an exponential mixture hazard model. They accommodate heterogeneity by
allowing segments of consumers to switch and repeat at different rates. Grover and
Srinivasan (1992) define a loyal segment and a switching segment. In the loyal segment,
consumers are loyal to the individual brands, while in switching segment, consumers
switch among preferred set of brands. The two segments have the same mathematical
models and variables, but the parameters are estimated separately for each segment.
Bonfrer and Chintagunta (2004) found that the probability of purchasing a control brand,
i.e. a private label brand, tends to be higher for store loyal customers than for store
switchers. Rhee and Bell (2002) study shoppers’ inter-store selection behavior. Since
shoppers often shop at different stores, they define “main store” as the store that a
36
shopper spent most money in, and examine the tendencies of shoppers to transition away
from the current main store. The model is established in a discrete time hazard
framework and estimated as random effect probit model using Bayesian methods. Strong
state dependence shows that shoppers are unwilling to give up the benefits of knowledge
about information of main store. The mean of transition rate is 0.183, which means that
most shoppers do not switch or switch very few times.
Purchase timing is one important aspect of purchase decision. Many marketing
papers focus on households’ purchase time decision (Jeuland, Bass, and Wright 1980;
Jain and Vilcassim 1991, 1994; Gupta 1991; Helsen and Schmittlein 1993; Gonul and
Srinivasan 1993; Wedel, Kamakura, Desarbo, and Hofstede 1995). Guo and Villas-Boas
(2007) investigate how consumers adjust purchase timing of storable-goods in response
to expected price change. Vilcassim and Jain (1991) study households’ purchase timing
and brand switching together using a continuous-time semi-Markov approach. They find
that unobserved heterogeneity plays an important role in purchase timing decision.
Seetharaman and Chintagunta (2003) use the proportional hazard model to study
purchasing timing of households using scanner data of laundry detergents and paper
towels. These authors study purchase timing by investigating the probability of
purchasing a product after last purchase. A baseline model and a covariate model were
examined in this paper. The baseline model captures a household’s intrinsic purchase
pattern over time, and the covariate model captures the influence of marketing variables.
They compare the continuous time and discrete time proportional hazard model, and find
that discrete time hazard model outperforms the continuous time hazard model in terms
of explaining the observed purchase outcomes.
37
Song (2005) studies trial decision of online grocery shopping. A discrete time
hazard model is used to examine how neighborhood effects influence the trial decisions
of individuals who have not used online shopping before. This research models the
conditional probability of first trial in a region given trials in neighboring regions. He
shows that speed to trial is strongly influenced by education levels and household
composition of a region.
Based on former research on brand switching and purchase timing, this paper will
model household level trial decisions about organic milk. Household level variables,
such as education, age, income, as well as marketing variables will be analyzed. Results
will show what factors speed up the trial process, and what factors slows down the trial
process. These results will give food industry researchers and managers information on
how to identify customers that are more likely to try organic milk, and what marketing
strategies should be used to increase organic milk users.
Model Specification
In this paper, we propose to estimate a discrete-time hazard model where the
dependent variable is time-dependent binary choice. Before the econometric model is
specified, however, I will talk about assumptions related to discrete time hazard model.
38
Censoring
When analyzing data for a time to event study, one usually focuses on data within
a specific period of time. Each individual in the study has his or her own starting point
and event time. Some individuals may have experienced the event at the beginning of
study period, some individuals may experience the event during the study period, and
some individuals may not experience the event before the end of the study period. The
researcher has no information about whether these individuals will experience the event
after the study period. These kinds of incomplete observations are usually described as
censored observation in survival analysis. Aalen, Borgan and Gjessing (2008),
Kalbfleisch and Prentice (2002), and Yamaguchi (1991) provide information about
difference types of censoring.
One critical assumption about our organic milk trial model is random censoring.
Random censoring means that the censoring time is independent of the failure time. This
assumption is valid for most end-of-study censoring. For our study, the study period is
determined by the researcher, and purchasers do not known the censoring time before
they make milk purchase decisions, therefore, this assumption is likely to hold for this
study.
Censoring time of each household is assumed to be independent of each other in
our paper. That is, a household’s organic milk trial decision is not affected by other
households. This assumption may not be always true in reality, but for parsimony of the
paper, we assume this assumption holds9.
9 Song (2005) studies neighborhood effect on first time online purchase.
39
Frailty
The term frailty was introduced by Vaupel et al. (1979) and used in proportional
hazard model to capture unobserved heterogeneity. A latent variable for frailty is
introduced in the model. This latent frailty variable is a random variable varying over the
households. Gamma frailty is widely used in proportional hazard modeling. One reason
for using gamma distribution is feasibility and simplicity in integration (Duchateau and
Janssen 2008). Gamma distribution can take the appropriate range of zero to infinity and
is mathematically tractable (Rodríguez 2005). Another reason is that the gamma
approximation can reduce the error with truncated data (Abbring and Van den Berg
2007). Therefore, we will follow many authors (e.g., Jenkins 1995) to assume
unobserved (or omitted) heterogeneity between individuals is Gamma distributed.
In the content of organic milk purchase decision, the purpose of including frailty
is to describe unobserved heterogeneity among households’ organic milk trial purchase
decisions. The same marketing activities may have different effect on different
households, even when the households have the same demographic variables. Frailty is
introduced to account for such differences. Households possess different frailties in
frailty model, and households that are most frail will try organic milk earlier than other
households.
In this paper, frailty is assumed to be constant for a particular household, but
different among households. That is, it does not change over time for a household.
Therefore, we will have a single value of (latent frailty variable) that is common to a
40
group of time series purchases for the same household. This latent frailty variable is also
assumed to be independent of observed factors (right-hand-side variables).
Endogeneity
Endogeneity arises in the organic milk trial decision analysis because there are
likely to be omitted variables in the error terms that are correlated with prices. For
example, quality of milk is usually believed to affect households’ purchase decision, and
it is positively correlated with milk prices. If we do not have a variable for milk quality,
the effect will be thrown into the error term, and the error term will be correlated with
price variables. Neglecting endogeneity in a model leads to inconsistency in estimation
results.
A control function method includes extra variables in the empirical model to
condition out the unobserved factor that is not independent of the endogenous variables10
(Petrin and Train 2003 and 2010). We will use the control function method in this paper
to deal with endogeneity in price variables. To be specific, we will estimate the model in
two steps. In the first step, price variables are regressed on observed characteristics and
instruments. The residuals of this regression are saved as new variables. In the second
step, the organic milk trial decision model is estimated with all the right-hand-side
variables, plus the price residual variables from the first step.
10 This control function method is called “Two Stage Residual Inclusion” in some literatures (e.g. Terza, Basu, and Rathouz 2007)
41
Empirical Model
To study households’ organic milk trial decision making, we link individual
utility maximization model with a discrete-time hazard model. The utility maximization
model is closely related to consumers transition decision from non-organic milk to
organic milk, and the discrete-time hazard model estimates the timing of trial purchase.
Let denote the time that an organic milk trial occurs for individual . Hazard
models focus on survival probability and hazard probability . For my study,
the survival probability is the probability that individual does not purchase organic milk
at least to time t, and the hazard probability is the probability that individual buys
organic milk at time t. If time is treated as continuous variable, the survival function is
The hazard rate of individual at time is
lim∆
t ∆| t∆
If time is treated as a discrete variable, i.e., if we treat trial event as occurring
within meaningful discrete time intervals, then the hazard function becomes
, .
The hazard rate is the conditional probability that an organic milk trial happens
within the given time interval given that it has not happened before. Most hazard models
are formulated with a particular form. The most widely used hazard model takes the
42
functional form of Cox’s (1972) proportional hazard model (Allison 1982, Seetharaman
and Chintagunta 2003, Nam et al 2008):
It is called proportional hazard model because “the ratio of the hazard rates for
any two individuals at any point in time is a constant over time” (Allison 1982). Taking
exponential on both sides, we get
exp exp
Denote exp , we get
where is an unspecified baseline hazard function; is a vector of observed
factors that potentially influence the time of organic trials; and is a vector of
coefficients to be estimated. Following Kalbfleisch and Prentice (2002), we assume the
time of organic milk trials follows a Weibull distribution.
The hazard probability is based on the underlying random utility maximization
model. Suppose individual is a non-organic milk buyer at period 0, and has the
possibility of buying organic milk beginning at period 1. Whenever the underlying utility
of buying organic milk exceeds the utility of buying non-organic milk, an organic milk
trial happens. The observed trial variable is
1 0
Therefore, if an organic milk trial never occurs for individual , is a
sequence of zeros; and if organic milk trial occurs, is a sequence of zeros with the
last being one.
43
We propose to estimate two discrete time hazard regression models using
maximum likelihood estimation. The first model (Model I) includes observed
heterogeneity of households, prices, coupons, promotions, and a time variable. The
results from this model will suggest how demographic variables and marketing activities
affect the timing of organic milk trial.
The second model (Model II) considers frailty. Trial decisions are affected by
household members’ unobserved factors such tastes and preferences. Households with
same demographic characteristics that face same marketing activities may have organic
milk trial at different time because of unobserved heterogeneity. Therefore, frailty is
added in Model II. The hazard rate is now
. . . log
Where is Gamma distributed frailty with unit mean and variance . The unit mean is
not restrictive though, as long as it is finite, because any non-unit mean can be
normalized to a unit one (Lancaster 1992, Chapter 4).
The coefficients in Model II are expected to be larger in absolute value than in
Model I, because not including unobserved heterogeneity results in an under estimate of
hazard rate, and attenuates the magnitude of the impact of coefficients on hazard rate
(Lancaster 1992, chapter 4).
Stata command “pgmhaz8” written by Professor Stephen P. Jenkins is used to
estimate models in this paper.
44
Data
Nielsen Homescan milk purchase data from 2004 to 2006 are used in this paper.
Homescan panel provides food purchase information scanned by households after each
grocery shopping. Each purchase record includes purchase date, quantity, volume,
expenditure, promotion type, and households’ demographic information.
Since the target event of interest is the organic milk trial purchase, subsequent
purchases after organic milk trial are not included in this study. This leaves the data set
with 236227 observations from 22258 households. Figure 3-1 shows the frequency of
organic milk trial by week.
Based on the random censoring assumption, we assume a household’s first
organic milk purchase observed in the data set is the organic milk trial purchase. Left
censoring is an issue here because all the households that always buy organic milk are
treated as repeated organic purchaser on their first milk purchase trip. 2236 households
purchased organic milk during their first milk trip in our data set, accounting for 9.13%
of all the 24494 households. To check how this left censoring issue affects the robustness
of estimation results, we did the sensitivity analysis by using the subset data of 2005 and
2006. Results are pretty consistent between the two models. Only one variable "African
American" has opposite sign, but it is non-significant in both models. This suggests that
left censoring does not affect our estimation significantly.
45
Table 3-1. Variable Descriptions for the Organic Milk Trial Paper Variable Name Mean Description Organic 0.0038 Dummy variable for organic. Equals 1 if the product is organic milk, 0 otherwise. Household Demographics
income 63.98543
Household income (in $1,000s). Converted from 16 income ranges by using the midpoint. The top-most category ($100k+) is arbitrarily assigned the value of twice the mean ($170k) of the second highest income range ($70k to $100k).
Income2 6275.931 Income squared.
maxage 7.209521 The maximum age category of male and/or female household heads. Categories are linear, and category 7 represents ages from 50 to 54.
dumedu 0.479072 Equals 1 if at least one household head has at least college education. hhsize 2.533991 Household size African American 0.057171
Equals 1 if the household head is African American. (White represents the reference case.)
Asian 0.029744 Equals 1 if the household head is Asian.Hispanic 0.07913 Equals 1 if the household head is Hispanic. children 0.271679 Equals 1 if the household has children under 18.
h2_1inc 0.215496 Equals 1 if the household has both a male and female head, but only one income. (Households with only one head represents the reference case.)
h2_mf_fp 0.100063 Equals 1 if male head works full time (35 hours/week), and female head works part time (34 hours/week).
h2_mp_ff 0.014373 Equals 1 if male head works part-time (34 hours), and female head works full time (35 hours/week).
h2_mf_ff 0.202804 Equals 1 if both heads work full time (35 hours/week). married 0.662059 Equals 1 if household heads are married, 0 otherwise. Shopping Patterns and Marketing Variables dumcpn 0.021256 Equals 1 if coupon is used, 0 otherwise. otherpro 0.176385 Equals 1 if other promotion is used, 0 otherwise. expfpm 16.62183 Total weekly expenditure on fresh or frozen produce and meat. exporgfpm 0.166702 Total weekly expenditure on organic fresh or frozen produce and meat. Prices and Estimated Price residuals logorgprice -2.97848 Logarithm value of organic milk price. lognorgprice -3.65304 Logarithm value of non-organic milk price. rlogorgprice -0.00069 Residual for organic milk price (from an instrument equation). rlognorgprice 0.000804 Residual for non-organic milk price (from an instrument equation).
46
Figure 3-1. Frequency of Organic Milk Trials by Week
We assume weak separablility for fluid milk product, because milk purchase
decisions are usually made independent of other categories. That is, there is no obvious
substitution for fluid milk. Table 3-1 provides statistical summary and description of all
variables.
Results
Results of estimation are listed in Table 3-2. We will discuss the results of
marketing variables, demographic, and employment and shopping patterns in turn.
0.00%
20.00%
40.00%
60.00%
80.00%
100.00%
120.00%
0
500
1000
1500
2000
2500
30001 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106
113
120
127
134
141
148
155
Cumulative Pe
rcen
tage
Freq
uency
Week
Frequency of Organic Milk Trials by Week
47
Marketing Variables
Marketing variables include coupons, other promotions, and prices. Coupon
variable (dumcpn) indicates whether a coupon is used in a milk purchase trip. This
variable has positive and significant coefficients in both models, which suggests that the
time to the organic milk trial purchase decreases with the use of coupon. Table 3-3
shows the frequency of coupon use for organic milk trial purchases and non-organic milk
purchases. Nevo and Wolfram (2002) note that this rationale of couponing behavior is
just one of several possible rationales for coupon use. On the contrary, other promotion
(otherpro) has negative insignificant coefficients in both models, which suggests that
other promotion methods such as displays do not have much impact on organic milk trial
purchase.
While the model includes both the organic and non-organic price, only organic
price is significant in both models. This suggests that organic price instead of non-
organic price affects the time to organic milk trial. Organic milk price has the expected
negative signs for both models, meaning that higher organic milk price will prolong the
time to organic milk trial. Specifically, a 1 percent increase in the organic milk price
will bring a 2.52 percent decrease in purchasing hazard based on Model I and 4.59
percent based on Model II. This result is consistent with Seetharaman (2003), suggesting
that failing to account for frailty underestimates the effectiveness of price.
48
Table 3-2. Results from Discrete Time Hazard Model Org Trial Hazard Model I (Without Frailty) Model II (With Frailty) Coef. Std. Err. Coef. Std. Err. income 0.0058733* 0.0034914 0.0091773* 0.0048646 income2 -0.0000135 0.0000165 -0.0000181 0.0000232 maxage -0.0879533*** 0.022738 -0.0327421 0.033377 dumedu 0.4477526*** 0.0822892 0.5398363*** 0.1154625 hhsize -0.1523438*** 0.045965 -0.1577004** 0.0636544 African American 0.2381753* 0.1431589 0.077525 0.1965266 Asian 0.6075112*** 0.1456066 0.9670452*** 0.2531649 hispanic 0.4949494*** 0.1106832 0.7665323*** 0.1754092 children 0.0044932 0.1208946 0.0164722 0.1768066 h2_1inc 0.1988711 0.1210963 0.3643086** 0.177148 h2_mf_fp 0.2110992 0.1503561 0.3342379 0.2222541 h2_mp_ff -0.2215379 0.3315795 -0.0732834 0.4610722 h2_mf_ff -0.0195027 0.128983 -0.0389908 0.1891237 married -0.1197024 0.1117265 -0.2308095 0.1572482 dumcpn 1.128068*** 0.1502392 1.294362* 0.1887423 otherpro -0.0882738 0.1030442 -0.1041621 0.1209466 logorgprice -2.51586*** 0.8325367 -4.591357*** 1.041327 lognorgprice 0.0268568 0.1993534 -0.2604156 0.2540521 expfpm -0.0030765 0.0020981 -0.0042114* 0.0023839 exporgfpm 0.1195192*** 0.0075778 0.2234635*** 0.0194406 rlogorgprice 2.63276*** 0.8996115 5.330475*** 1.101634 rlognorgprice 0.5564161 0.3388474 0.9353773** 0.3853948 _cons -12.62808*** 2.856788 -19.80108*** 3.530103 LR test of Gamma var. =0 chibar2(01) =339.766 Prob.>=chibar2 = 0 Notes: *** means significant at 1%, ** means significant at 5%, and * means significant at 10%.
The price residuals have expected signs and significance for organic milk. The
positive and significant coefficient for organic milk price residual means the price of
organic milk is higher than what has been explained by other variables. This is because
49
organic milk possesses desirable attributes that is not included in the analysis, such as
high quality, good taste, and environmental friendly. This finding is consistent with
Petrin and Train (2010).
Table 3-3. Coupon Use and Organic Milk Trial
Coupon No coupon Percentage of Coupon Use Organic 57 822 6.93%
Non-organic 4948 228214 2.17%
Demographic Variables
Income is significant at 10 percent level in both models. The positive signs show
that higher income households are more likely to have organic milk trial purchase,
holding other factors constant. Table 3-4 and Figure 3-2 show the frequency of organic
milk trial for households with different income levels. Income square is not significant in
both models. Consistent with other organic purchase studies (not based on first trials),
the education variable, which indicates that a household head holds a college degree, is
positive and significant in both models. Higher education households are more likely to
try organic milk sooner. Household size is significant and negative in both models,
meaning that larger households are less likely to try organic milk. Age (maxage) is only
significant in the Model I. Marriage and appearance of children are not significant in
both models.
50
Race affects first time organic milk purchase. Results show that Asian and
Hispanic are more likely to make organic milk trial purchase. The coefficient of African
American is significant in Model I, but not significant in Model II.
Table 3-4. Income and Organic Milk Trial
Household Income
Organic Trial Frequency
Cumulative Organic Trial Frequency
Percentage of Cumulative Frequency
Under $5000 4 4 0.46%
$5000-$7999 8 12 1.37%
$8000-$9999 4 16 1.82%
$10,000-$11,999 5 21 2.39%
$12,000-$14,999 12 33 3.75%
$15,000-$19,999 30 63 7.17%
$20,000-$24,999 42 105 11.95%
$25,000-$29,999 35 140 15.93%
$30,000-$34,999 49 189 21.50%
$35,000-$39,999 52 241 27.42%
$40,000-$44,999 60 301 34.24%
$45,000-$49,999 56 357 40.61%
$50,000-$59,999 97 454 51.65%
$60,000-$69,999 92 546 62.12%
$70,000-$99,999 166 712 81.00%
$100,000 & Over 167 879 100.00%
51
Figure 3-2. Organic Milk Trial Frequency for Income
Employment Variables and Shopping Patterns
Employment variables show interesting impacts on organic milk trial purchase.
Gender differences in employment may affect the hazard of organic milk trial purchase.
Comparing to households with only one head, results from Model II show households
with two heads but only one income (h2_1inc) has a significant higher hazard of having
organic milk trial purchase. However, the insignificant estimate for households with one
part time income from female head and one full time income from male head (h2_mf_fp)
imply that households with an additional part time income from female head are not more
likely to have organic milk trial. Similarly, we do not see this significant increase of
hazard for dual-headed households with a fully employed female head and part-time
0.00%
20.00%
40.00%
60.00%
80.00%
100.00%
120.00%
020406080
100120140160180
Cumulative Freq
uency in Percentage
Freq
uency
Income
Organic Milk Trial Frequency for Income
52
employed male head (h2_mp_ff), nor the households with two full time employed heads
(h2_mf_ff). Most coefficients of employments variables in Model II are larger than those
in Model II in magnitude.
All the shopping pattern variables are significant in Model II. Households
spending more on out of category organic products are more likely to try organic milk,
while households have higher total expenditure on that category are less likely to try
organic milk.
Comparing the coefficients of Model I and Model II, we find that the magnitude
of coefficients in Model II is larger than in Model I for most variables. This is consistent
with Lancaster (1990, chapter 4) that accounting for unobserved heterogeneity increases
the magnitude of the impact of covariates on hazard rate. LR test of Gamma variable
shows that unobserved heterogeneity (frailty) is significant.
Conclusion
This study is among the first effort to study purchase timing of organic milk trial.
We use Nielson Homescan data from 2004 to 2006 to study factors that affect the timing
of non-organic buyer’s first organic purchase. Random utility maximization model gives
the rational of consumers’ trial behavior. Based on this economic model, the discrete
time hazard model gives the empirical estimates.
Censoring, endogeneity, and frailty are discussed in this study. Sensitivity test
shows that left censoring does not affect the empirical estimation significantly. A
control function method (also called two stage residual inclusion) is used to correct
53
endogeneity. Heterogeneity is captured by Gamma frailty. Accounting for unobserved
heterogeneity increases the magnitude of the impact of covariates on hazard rate.
Results show that coupons and organic milk price affect the timing of organic
milk trial purchase significantly. Income, education level, and out of category organic
expenditure are positively related to the hazard. Household size and out of category total
expenditure are negatively related to the hazard. Consumers’ race also affect their
organic milk trial hazard. Results from this paper may help managers to expand market
share, and target the customers effectively.
54
Chapter 4
Consumers’ Choice of Private Label and National Brand Organic and Non-organic Milk
Since the 1970s and 1980s, private label products (also known as store brands)
have seen great improvements in product quality and large gains in market share. Once
considered a low-quality, low-price alternative, some private label products evolved to
compete with high-quality, market-leading brands, including organic brands (Burt, 2000).
Citing a study in 2006 produced by the Private Label Manufacturers Association,
Haberkorn (2006) notes that 41 percent of shoppers buy private label goods frequently,
up from 36 percent in 2001 and 12 percent in 1991. In 2000, the market share of private
label brands exceeded most national brands in about 50 percent of product categories,
ranking first or second in 131 out of 266 categories (German, 2001).
Dairy is one grocery category where the private label expenditure level and
market share are among the highest. For the 52 weeks ending May 19, 2007, the trade
publication Progressive Grocer reports that total private label sales across all grocery
categories reached $46.5 billion. Among all categories, private label milk leads the way
with $6.5 billion, followed by bread and baked goods with $3.4 billion, and cheese with
$2.9 billion (Progressive Grocer 2007). Citing Information Resources Inc., Barstow
(2005) claims that about 60 percent of milk is sold under a store brand.
These private label trends extend to the organic food market, which itself has
rapidly grown annually since early 1990s and now has a 2007 growth rate of 19 percent.
The global organic food market is expected to reach $70.2 billion by the end of 2010
55
(Research and Markets, 2008). According to the trade journal Gourmet Retailer (2008),
private labels are responsible for 17.4 percent of all organic sales, with dairy and produce
items having the highest shares. In the market for organic milk, two national brand milk
companies, Organic Valley and Horizon Organic, have led the market from the late
1980s. As of May 2007, these two producers provide 75 percent of U.S. organic milk
supply (Dimitri and Venezia, 2007). After these two brands, private label organic milk
occupies third position nationally, comprising just under 10 percent of the market share
(Ihde, 2002). Dimitri and Oberholtzer (2009) find that private label’s share of organic
milk sold at retail stores has more than doubled between 2004 and 2007, increasing from
12 percent to 27 percent.
With a growing market share as background, both for organic food generally and
organic private label milk specifically, this paper investigates consumers’ linked choices
between organic and non-organic milk, and between private label and branded milk.
Using household-level demographic information, market-level prices, and other
shopping-trip variables, we attempt to identify factors that (a) separate milk-buying
households into organic or non-organic milk consumers, and (b) separate these same
households once more into private label or branded milk consumers. Our main objective
is to understand whether organic milk buyers approach the private label decision in the
same way as non-organic milk buyers.
While the consumer’s choice can be partially decomposed into two related
decisions, one on organic and another on private label, an accurate investigation of this
question must account for potential selection effects that link the two decisions.
Therefore, this paper relies on a sample selection model, where in the first stage milk
56
consumers decide whether or not to buy organic. In the second stage, consumers decide
whether or not to buy private label milk, conditional on the outcome of the first stage.
This two-stage decision model is meant to correspond closely to shoppers’ actions within
supermarkets, where one typically finds private label products placed as close as possible
to their national brand competitors (Sayman, Hoch, and Raju, 2002). In our case, we
expect that supermarket shoppers will find organic private label milk close to its national
brand counterpart, and non-organic private label milk close to its counterpart.11
Using a definition described below for a weekly purchase of a main milk
category, our results show that a large number of demographic, promotional, and
shopping-pattern factors do affect consumers’ decisions to buy private label rather than
national brand milk. More specifically, we find that the impacts of marketing and
shopping trip-related variables on the private label decision do not depend on the organic
choice. However, the influence of many demographic factors on the private label choice
does depend on the organic selection. While our main focus is on the second-stage
private label decision, a number of important results emerge from the first-stage decision
that separates organic from non-organic buyers. One novel result involves how the
number of hours worked by female household heads affects the likelihood of purchasing
organic milk. We find that dual-headed households with a fully employed woman are
less likely to buy organic milk than households where only one head works or where two
11 Other discrete choice models, such as a nested logit or a mixed logit, might also be useful for investigating the private label or national brand decisions. However, the two‐stage sample selection model closely mimics a fairly common supermarket setting where organic milk is found in a separate section from non‐organic milk. The sample selection model links the two decisions by including information from the first stage (e.g., the estimated inverse Mills ratio) in the second stage or by estimating the two stages simultaneously. Alternatively, a nested logit could links the two decisions via the nest structure, and a mixed logit could link the two decisions via some unobservable components of the error term.
57
heads work but a woman works part time. No other research documents this decreased
likelihood of buying organic resulting from full-time employment by female heads.
These and other results are discussed in more detail below. First, however, we frame our
work in context to existing research, and then we present our econometric model and
discuss the data used in the estimation.
Literature Review
Current trends suggest that the private label choice is increasingly important for
retailers. In general, most of the marketing research that investigates the private label
versus national brand choice (for example, Richardson, Jain and Dick, 1996; and Batra
and Sinha, 2000) focuses on consumers’ demographics and perceptions. Richardson,
Jain, and Dick (1996) provide an extensive list of factors affecting consumers’ private
label choice: 1) Demographic variables, such as income and family size, where lower
income and larger size households are more likely to buy private label brands; 2)
Extrinsic cues, such as name, price and packaging, where better extrinsic cues increase
the likelihood of purchase; 3) Perceived factors, such as the perceived value for money,
risk, and quality variation, where perceived values are positively linked to private label
preference; 4) Former experience, such as familiarity with store brand, where more
familiarity means lower perceived risk and quality variation associated with private label,
58
which makes consumers less dependent on extrinsic cues, hence higher private label
preference.12
Several research papers that focus on some combination of private label and
organic are particularly relevant for our study. Hassan and Monier-Dilhan (2006)
investigate private label offerings when there is also a public-based quality label (such as
the European Union’s Protected Designation of Origin or Protected Geographical
Indication, or France’s Label Rouge) and show that these quality labels offer good
opportunities for retailers’ private label offerings. Hammarlund’s (2002) investigation of
organic milk purchases and Thompson and Kidwell’s (1998) research on organic fruit
and vegetable purchases also help frame our study. Jonas and Roosen (2008) and Zhang
et al. (2008) investigate organic milk demand in Germany and fresh organic produce
purchase decisions in the U.S., and each, respectively, provides some demographic
results relevant for our study. Jonas and Roosen (2008) paper divided all the milk
purchases into three categories: organic milk, conventional national brand milk, and
conventional private label milk. In the first stage, a Probit model is used to estimate the
probability of choosing each category. In the second stage, a simultaneous equation
system of LA/AIDS model is used to estimate market shares. Two stage sample selection
model is similar to a censored demand model used in Jonas and Roosen (2008) in the first
step, while different in the second step. In the first step, both models account for the
probability of being censored, and both use the probability calculated in the first step in
the second step of estimation. In the second step, two stage sample selection model
12 Another line of research focuses on strategic competition between branded products and private labels. Empirical examples in this line include Ward et al. (2002), Bonanno and Lopez (2005), Bontemps et al. (2005) and Bontemps et al. (2008).
59
focuses on a consumer’s individual choice (could be a binary choice or quantity
demanded), while a censored demand model usually focuses on a demand
system(LA/AIDS is widely used).
Based on Zhang et al. (2008), Dettmann and Dimitri (forthcoming), Jonas and
Roosen (2008), Hammarlund (2002), Glaser and Thompson (2000), and Thompson and
Kidwell (1998), among others, we expect to find a number of demographic factors
positively or negatively affecting a household’s decision to buy an organic product such
as milk, which corresponds to the first-stage decision in our model. While some previous
demographic results are mixed, we expect income, education, Hispanic ethnicity to
increase a household’s likelihood of buying organic. Previous results for age and the
presence of children are mixed, with some researchers (Zhang et al. 2008, and Jonas and
Roosen 2008) finding that households with heads in upper age groups more likely to buy
organic, and households with children less likely to buy organic. Hammarlund (2002)
finds larger-sized households as well as households with higher numbers of children are
less likely to buy organic milk. Thompson and Kidwell (1998) conversely find
households with higher numbers of children are more likely to buy organic produce.
No study we know of directly examines the potential differences between a
private label and branded organic product purchase event. Therefore, we have no prior
expectations on how household demographic variables might impact households’ second-
stage decision to buy private label organic or non-organic milk. General findings by
Richardson, Jain, and Dick (1996) do, however, suggest that prior experience with private
label or organic products should positively affect a household’s decision. Their findings
60
also suggest that higher-income households would be more likely to buy national brand
milk.
Model Specification
Consistent with the random utility model framework, our model of private label
choice of organic or non-organic milk follows a sample selection model that involves
incidental truncation.13 For example, the choice of private label or branded organic milk
is observed only if a consumer has first chosen organic milk over non-organic milk. We
first assume that UO equals the unobserved utility from choosing organic milk, UNO
equals the unobserved utility from choosing non-organic, and that an organic milk
purchase is observed when UO > UNO ; otherwise we observe a non-organic milk
purchase. Given an organic milk purchase, in similar fashion we let UO,PL equals the
unobserved utility from choosing organic private label milk and UO,NB equal the
unobserved utility from choosing organic national brand milk. As before, organic
private-label milk is purchased when UO,PL > UO,NB; otherwise, organic national brand
milk is purchased. Our model is therefore similar to other double hurdle models in the
agribusiness and marketing literature where the second-stage choice variable is
continuous rather than binary (for examples, see Zhang et al. 2008; Yen and Line 2006).
In the following discussion, we specify the two cases where consumers choose a private
label or national brand conditional on first choosing either the organic or non-organic
milk.
13 For more on sample selection models generally, and binary response models with sample selection more particularly, see Wooldridge (2002)
61
Starting first with the selection mechanism, we posit that , the latent variable
underlying the observed milk choice for organic (O) or non-organic (NO) milk such
that j = {O, NO} by household i, depends on a vector Zj of household demographic
attributes, marketing variables reflecting both price and promotion, and household
shopping patterns. This underlying latent variable takes the following form:
(1) , for j = O, NO,
1 if 0, and 0 if 0,
where γj is a vector of parameters to be estimated and μji is the error term.
Conditional on observing , the observed second-stage choice of private label or
national brand milk, , and the latent variable is then specified to depend on a
similar but not identical vector Xj of demographic, marketing, and shopping-pattern
attributes:
(2) , , , for j = O, NO,
1 if 0 and 1, and 0 if 0 and
1,
where βj is a vector of parameters and εji is the error term. In (2), therefore, a
private label milk is purchased when 1, while a branded milk is chosen when
0.
We assume that (μi, ε i) are distributed bivariate standard normal, such that (μi, ε
i) ∼ BVN(0, 0, 1, 1, ρ). Given this distributional assumption and (1) and (2), the log-
likelihood function for our bivariate probit model with sample selection is:
62
(3) ∑ Φ, , ,
∑ Φ, , ,
∑ Φ
where Φ is the univariate standard normal cumulative density function, Φ2 is the
bivariate standard normal cumulative density function distribution, and ρ is the
correlation between μj and εj. The full model in (1), (2), and (3) is estimated via
maximum likelihood estimation after first addressing probable endogeneity of milk
prices, which are elements of both X and Z.14
Data
This study uses shopping trip and household-level Nielsen Homescan data that
provide market-related information such as purchase date, dollars paid, promotion type,
brand information, and other information related to the UPC product code. 15 The
Homescan data are unique in that they provide demographic information, including
household size, education, age, race, and other demographic information on the more
than 40,000 households in the Nielsen panel. We specifically use data on milk for all
14 More specifically, the Heckprob procedure in Stata is used for maximum likelihood (ML) estimation of (1), (2), and (3) and corresponding robust standard errors. The ML estimation procedure is sensitive to identification and requires the X and Z vectors to contain some different elements. In our case, differences are achieved through price variables of different milk categories, and through previous purchase variables of different milk categories. Furthermore, we have found that when the previous purchase variables are omitted, the X and Z vectors become too similar and ML convergence fails. In this case, a two‐step estimation leads to similar results. Puhani (2000) discusses the pros and cons of the two‐step estimator. 15 Zhang et al. (2008) use Homescan data for 2003 in their analysis of organic product purchase decisions, though they did not attempt to recover price information for non‐purchasing households. Einav, Leibtag, and Nevo (2008) discuss the overall accuracy of the Homescan data.
63
U.S. markets in 2004, 2005, and 2006.16 To make the vast amount of Homescan data
manageable, we filter and manipulate it in several ways. First, we use a week as discrete
time interval. Second, because a household may buy different types of milk in one week,
we define a “main milk category” that captures the highest expenditure in one of four
milk categories: organic private label, organic national brand, non-organic private label,
and non-organic national brand.17 Third, we limit the data to households within Nielsen’s
52 market areas, which represent major and mid-sized markets. Our method for
recovering price information, discussed shortly, is responsible for this restriction. After
aggregating the data set on a weekly basis, we have 283,728 weekly shopping trips with
milk purchases. Among these trips, 67.35 percent (191,103 weekly trips) represent a
main purchase of private label milk purchase, and 32.65 percent (92,625 weekly trips)
represent a national brand milk purchase. Within the four milk categories, non-organic
private label milk has the highest share (64.42 percent), and organic private label milk
has the lowest share (2.94 percent). Table 4-1 shows the frequencies of main milk
purchases among the four categories. Table 4-2 gives the frequency for multiple milk
type purchase in a week. The reason for the low organic private label market share is that
organic private label milk is a new developing category instead of a well established
category. 1) From supply side, only some of the retailers provide organic private label
milk, while most of the retailers provide non-organic private label milk. 2) Consumer
perception: Consumers need time to perceive the quality of organic private label milk. 3)
16 These years represent a time frame after the USDA implemented national organic standards and labeling requirements and before recent major macro events (such as escalating food prices followed by a major recession). 17 Rhee and Bell (2002) identify a household’s “main” store based on a weekly allocation of expenditure at a number of stores. We loosely use the same technique to identity a main category of milk based on a weekly allocation of expenditure across four milk categories.
64
Price: The price difference between organic private label and organic national brand is
relatively low. 4) Customer segmentation: Retailers are still learning about how to target
the right customers for organic private labels.
Table 4-3 presents the definitions of all the variables used in the analysis. The
variables roughly fall into three categories: household demographics, shopping patterns,
and prices and instruments.
Table 4-1. Milk Purchase Frequency
Non-Organic Organic Total
National Brand60,723 (21.4%) 31,902 (11.2%)
92,625
Private Label 182,775 (64.4%) 8,328 (2.9%)
191,103
Total 243,498 40,230 283,728
Household Demographics
For the most part, Homescan’s demographic data are used with only minimal
transformations. In some cases, categorical variables are converted to binary dummy
variables; in other cases, some of the categorical variables are combined. Table 4-2
describes these transformations and lists the variable means.18 Table 4-4 provides a brief
summary of how private label or national brand purchases vary with household size. It
18 Nielsen Homescan data use the terms Black and Oriental, which we have re‐interpreted as African American and Asian.
65
shows that households with four or fewer members compose 91.7 percent of the weekly
milk purchase trips. Small households with one or two members occupy over 60 percent
of the weekly milk purchase trips. Measured by milk purchase trips, private label
purchases exceed national brand purchases in any household size.
In our estimated model, we include a large number of demographic variables,
some of which have not often been investigated. For example, we include a second order
term for income to investigate nonlinearities. We also focus on household heads’
employment status to investigate whether the likelihood of purchasing organic milk is
different for households within the following categories: households with a single head,
two heads but one income, one full-time employed male head and one part-time
employed female head, one part-time employed male head and one full-time employed
female head, and two full-time employed heads. Finally, all the demographic variables
are included in both the Z and X vectors from (1) and (2).
Table 4-2. Frequency of Multiple Milk Type Purchases
Numbers of Milk Types Frequency Percentage 1 280920 99% 2 2794 1% 3 14 0% 4 0 0%
Total 283728 100%
66
Shopping Patterns and Product Promotion Variables
In addition to demographic information on the purchasing household, the Homescan data
provide some detailed information on the actual purchase event. (i) To reflect prior
experience with a product, we construct a prior-purchase binary variable that equals one
if a household had bought a product in each of the previous four weeks. (ii) To reflect a
household’s overall level of weekly expenditures that is independent of milk purchases,
we construct an out-of category variable based on an aggregation of all weekly
expenditures by each household in the broad category defined by Nielsen as fresh and
frozen packaged meat and produce (abbreviated by Nielsen as FPM). (iii) To reflect a
household’s commitment to organic, we aggregate weekly household expenditures for all
organic purchases in the FPM category. And (iv) to reflect a household’s loyalty or trust
in private label products, we aggregate weekly expenditures of private label purchases in
the FPM category. Finally, we construct two product promotion variables: one if a
coupon was used with the purchase, and another if the product purchased was part of a
store promotion. 19 These variables are described in Table 4-2. Keep in mind, however,
that the coupon variable is only observable on the coupon’s redemption, not its issuance.
In our data, there are 6,463 coupon redemption observations (which represent 2.29
percent of the total observations) and 46,451 other observations of promotion use (which
represents 16.43 percent of the total observations).
19 Some states have regulations and laws that prohibit pricing milk below cost. In some cases, these regulations and laws may prohibit the use of coupons on milk products.
67
Table 4-3. Variable Descriptions for the Private Label Choice Paper Variable Name Mean Description Private Label 0.70626 Dummy variable for private label. Equals 1 if the product is private label milk, 0 Organic 0.16329 Dummy variable for organic. Equals 1 if the product is organic milk, 0 Household Demographics
income 70.2053 Household income (in $1,000s). Converted from 16 income ranges by using the midpoint. The top-most category ($100k+) is arbitrarily assigned the value of twice the mean ($170k) of the second highest income range ($70k to $100k).
Income2 7383.06 Income squared.maxage 7.07672 The maximum age category of male and/or female household heads. Categories
are linear, and category 7 represents ages from 50 to 54. dumedu 0.53760 Equals 1 if at least one household head has at least college education.hhsize 2.56075 Household sizeAfrican American 0.05276 Equals 1 if the household head is African American. (White represents the Asian 0.04138 Equals 1 if the household head is Asian.Hispanic 0.08933 Equals 1 if the household head is Hispanic.children 0.29196 Equals 1 if the household has children under 18. h2_1inc 0.22639 Equals 1 if the household has both a male and female head, but only one income.
(Households with only one head represents the reference case.)h2 mf fp 0.11066 Equals 1 if male head works full time (35 hours/week), and female head works h2 mp ff 0.01386 Equals 1 if male head works part-time ( 34 hours), and female head works full h2 mf ff 0.21048 Equals 1 if both heads work full time (35 hours). married 0.67969 Equals 1 if household heads are married, 0 otherwise. Shopping Patterns and Marketing Variables prev org 0.10486 Equals 1 if organic milk was a main purchase in the previous four weeks.prev orgPL 0.01435 Equals 1 if organic private label milk was a main purchase in the previous four prev orgNB 0.09051 Equals 1 if organic branded milk was a main purchase in the previous four prev norgPL 0.61501 Equals 1 if non-organic private label milk was a main purchase in the previous prev norgNB 0.13622 Equals 1 if non-organic branded milk was a main purchase in the previous fourdumcpn 0.01994 Equals 1 if coupon is used, 0 otherwise.otherpro 0.16023 Equals 1 if other promotion is used, 0 otherwise. expfpm 17.0246 Total weekly expenditure on fresh or frozen produce and meat. exporgfpm 2.57095 Total weekly expenditure on organic fresh or frozen produce and meat.expPLfpm 0.41082 Total weekly expenditure on private label fresh or frozen produce and meat.Prices and Estimated Price residuals logorgprice -2.97772 Logarithm value of organic milk price.lognorgprice -3.68313 Logarithm value of non-organic milk price.logorgplprice -3.08614 Logarithm value of organic private label milk price. logorgnbprice -2.9313 Logarithm value of organic national brand milk price. lognorgplprice -3.79344 Logarithm value of non-organic private label milk price. lognorgnbprice -3.31944 Logarithm value of non-organic national brand milk price. rlogorgprice 0.00376 Residual for organic milk price (from an instrument equation). rlognorgprice 0.01554 Residual for non-organic milk price (from an instrument equation).rlogorgplprice 0.00777 Residual for organic private label milk price (from an instrument equation).rlogorgnbprice -0.01183 Residual for organic national brand milk price (from an instrument equation).rlognorgplprice -0.01915 Residual for non-organic private label milk price (from an instrument equation).rlognorgnbprice 0.00475 Residual for non-organic national brand milk price (from an instrument
68
Table 4-4. Household Size Distribution
HHSize National Brand Private Label Freq. Freq. Percent 1 22,676 35,700 58,376 20.65 2 38,524 78,576 117,100 41.43 3 14,403 30,511 44,914 15.89 4 11,159 27,737 38,896 13.76 5 3,749 11,918 15,667 5.54 6 1,189 3,655 4,844 1.71 7 365 1,507 1,872 0.66 8 227 391 618 0.22 9 82 292 374 0.13
Total 92,374 190,287 282,661 100
Prices and Instruments
Because the Homescan data contain only the prices paid on products purchased by
individual households, a major task associated with these transaction-based data is the
construction of weekly price vectors facing each household for products not purchased.
In other words, prices of non-purchased items must be inferred. Based on the assumption
that households usually shop within a home-based market area, we calculate average
weekly category prices by market. There are 52 markets identified in the Homescan data
set. The algorithm is as follows: 1) Calculate the realized unit price of each milk
purchase, 2) Based on these transaction-level prices for each milk category, calculate the
mean price by market for each week, and 3) Assign these mean weekly prices to
individual households in the data set. In this way, we obtain the average market-based
organic private label price, organic national brand price, non-organic private label price
and non-organic national brand price for each week. Using a similar algorithm, we
obtain market-based prices for the more aggregated categories of organic and non-
69
organic milk. Our hope is that accuracy lost through averaging based on market-level
geographic boundaries is offset by the usefulness of including a full set of price vectors.
Source: Nielsen Homescan Data. Used according to confidential agreement with
USDA/ERS.
Figure 4-1. 2004 to 2006 Weekly Prices for Four Categories of Milk
For all 52 Homescan markets as a whole, Figure 4-1 shows weekly prices for the
four milk categories, and Table 4-5 summarizes price information by presenting the
average annual prices. Both the figure and the table show that organic milk prices
generally increase and non-organic milk prices generally decrease over the years. Figure
.02
.03
.04
.05
.06
Pric
es (D
olla
rs)
0 50 100 150WEEK
PL Org NB OrgPL Non-org NB Non-org
70
4-2 shows that, in each year, the private label categories have lower average prices than
national brand for both organic and non-organic milk, as expected.
Figure 4-2. Four Milk Category Annual Prices
Table 4-5. Average Annual Prices for Four Milk Categories
2004 2005 2006Frequency 63,608 90,923 129,197Organic Private Label Price ($/gallon) 6.22 6.90 7.29Organic National Brand Price ($/gallon) 6.64 6.93 7.47Non‐Organic Private Label Price ($/gallon) 3.70 3.51 3.31Non‐Organic National Brand Price ($/gallon) 4.59 4.53 4.43
At least three empirical issues further complicate the construction and use of the
price vectors. First, because organic private label milk only has a small market share,
some markets have no purchase records in certain weeks; hence prices are missing for
6.226.64
3.7
4.59
6.9 6.93
3.51
4.53
7.29 7.47
3.31
4.43
0
1
2
3
4
5
6
7
8
Org PL Org NB Non‐Org PL Non‐Org NB
Prices in
Dollars/G
allon
Four Milk Categories
2004 2005 2006
71
those weeks. Previous studies (such as Keane 1997 and Gupta 1988) have used prices in
adjacent weeks to approximate the missing prices. Because of the large number of
observations in the Nielsen data, we choose instead to eliminate the observations with
missing prices.
A second issue concerning prices is the choice between the shelf price and the
realized price. The shelf price is the listed price in a store and includes most price
deductions except for coupons or other promotions that are deducted at the register. On
the other hand, the realized price is the final price households paid for the purchase and
this price accounts for coupon use or any other register-based promotions. Coupon
availability, which we do not observe, affects this decision. Because we assume that
households live in the same market have similar access to coupons, and because we
construct market-based average prices, we believe that a realized price (rather than a shelf
price) provides a more accurate representation of the actual price.
The third issue involves the potential endogeneity of prices. To account for
endogeneity, we estimate instruments for prices using two types of exogenous variables
from the Homescan data as regressors. For the first set of regressors, we use market-level
demographic variables that are similar to the household-level variables used at the in the
sample selection model. Second, to reflect relative costs across the four milk categories,
we calculate the market level-based percentage of products sold with paper carton,
plastic, and glass containers, and include these variables in the price instrument
equations.
After estimating price instrument equations for the four milk categories and the
two aggregate categories, one last estimation issue remains to be addressed. Several
72
researchers, including Smith and Blundell (1986) and Rivers and Vuong (1988), have
noted that the traditional two-stage instrumental variable method of replacing the
endogenous variable with its predicted value recovered from a separate instrument
regression equation leads to inconsistent estimators when the model is a probit or another
nonlinear, limited dependent variable model. These authors propose an alternative
method, explained succinctly by Terza, Basu, and Rathouz (2008) and Wooldridge (2002,
p. 474), which involves recovering the estimated residuals from the instrumental variable
regression and adding them to the underlying model that contains the endogenous
variable. One of the first uses of this residual-inclusion technique in the agricultural
economics literature is Thompson and Kidwell (1998); Petrin and Train (2010) employ it
toward a marketing application on cable television purchase options.
Results
After estimating price instrument equations for all of our price series and
collecting the estimated residuals, we use a maximum likelihood procedure to estimate
the sample selection bivariate probits specified in (1), (2), and (3). As noted earlier, two
sample selection models are estimated: one where organic milk is selected in the first
stage ( = 1) and the organic private label decision is estimated in the second stage, and
another where non-organic milk is selected in the first stage ( = 1) and the non-
organic private label decision is estimated in the second stage. Table 4-6 presents the
first-stage coefficient estimates for both the organic and non-organic selection decisions;
whereas Table 4-7 presents the second stage coefficient estimates for both the organic
73
Table 4-6. Choice Between Organic and Non-Organic Milk Variable Organic Choice Non-organic Choice
Organic Coef. Robust Std. Err. Coef. Robust Std. Err.
income 0.0015275*** 0.0005611 -0.0008399* 0.0005026 income2 7.43e-06*** 2.62e-06 -0.0000113*** 0.00000236 maxage -0.1063468*** 0.0036104 0.1133024*** 0.0032924 dumedu 0.2095197*** 0.0127753 -0.1961619*** 0.0114149 hhsize -0.1133284*** 0.0071197 0.1036622*** 0.0062452 African American 0.0706791*** 0.0258537 -0.0183092 0.0243977 Asian 0.258573*** 0.0253442 -0.1482747*** 0.0245746 Hispanic 0.1733171*** 0.0188917 -0.0889717*** 0.0175745 children 0.061304*** 0.0188312 -0.0423302** 0.0167391 h2_1inc 0.0423945** 0.0190119 -0.0310881* 0.0172693 h2_mf_fp 0.0396856* 0.0231829 -0.0850872*** 0.0208463 h2_mp_ff -0.0141359 0.0442415 0.0384415 0.0399466 h2_mf_ff -0.1882561*** 0.0202597 0.1572035*** 0.0184243 married 0.1049857*** 0.0183875 -0.0777667*** 0.016499 prev_org 3.447071*** 0.0214401 -3.537215*** 0.0242137 dumcpn 0.141517*** 0.0348415 -0.1744714*** 0.0311307 otherpro -0.018219 0.0156678 0.0063959 0.014085 logorgprice -04.90311*** 0.1043794 4.775937*** 0.1286989 lognorgprice -0.4031711*** 0.0302995 0.2195187*** 0.0288745 week -0.0001468 0.0001425 0.0013449*** 0.0001293 expfpm -0.0022442*** 0.0003621 0.0027969*** 0.0003247 exporgfpm 0.1153371*** 0.0034634 -0.1250868*** 0.0038929 rlogorgprice 4.992942*** 0.1183293 -4.882907*** 0.1441386 rlognorgprice 0.1374301*** 0.0516398 -0.0122305 0.0483407 _constant -17.1971*** 0.3695816 15.73769*** 0.4375948
ρ -0.1024042*** 0.0455156 -0.1049881*** 0.0290288
Notes: *** means significant at 1%, ** means significant at 5%, and * means significant at 10%. Results in Tables 5 and 6 stem from simultaneous estimation.
74
Table 4-7. Choice Between Private Label and National Brand Milk Variable Organic Private Label Choice Non-organic Private Label Choice Private Label Coef. Robust Std. Err. Coef. Robust Std. Err. income 0.000293 0.0011986 0.0017161*** 0.0006228 income2 -2.11e-07 5.53e-06 -0.00000575** 0.00000301 maxage -.026884*** 0.0073981 0.0000931 0.0046098 dumedu -.0496965* 0.0291259 -0.000348 0.0146698 hhsize 0.0258312 0.016378 0.0248122*** 0.0081674 African American -0.193269*** 0.0583295 0.0066959 0.0303868 Asian -0.0821475 0.0503081 -0.0445566 0.0356111 Hispanic 0.0731773* 0.0376525 -0.0313818 0.0236598 children -0.0753281** 0.0369271 0.0223182 0.0227671 h2_1inc 0.0330445 0.0421079 -0.0007144 0.0209825 h2_mf_fp -0.0633123 0.0478642 -0.0581816** 0.0269928 h2_mp_ff -0.172328* 0.0974012 -0.0292437 0.0589527 h2_mf_ff -0.081692* 0.0463812 -0.0301226 0.0234118 married 0.0435488 0.0398381 0.0130047 0.0190873 prev_orgPL 1.068917*** 0.0919303 prev_orgBR -1.621175*** 0.0864612 prev_norgPL 1.919725*** 0.0152622 prev_norgBR -2.030665*** 0.0177649 dumcpn -0.310787*** 0.0863774 -0.6808372*** 0.0374002 otherpro 0.4440535*** 0.0358085 0.0674837*** 0.0192887 logorgPLprice -1.658044*** 0.2435658 logorgNBprice 0.4979778 0.3493393 lognorgPLprice -0.0043353 0.0361416 lognorgNBprice 0.9657441*** 0.0607786 week 0.0007298** 0.0003336 0.0021721*** 0.0001653 expfpm -0.003811*** 0.0007884 -0.0003646 0.0004333 expPLpm 0.0225614*** 0.0026386 0.0119132*** 0.0016552 rlogorgPLprice 1.708004*** 0.2541381 rlogorgNBprice -0.7061858* 0.3673382 rlognorgPLprice -0.0693239 0.0538169 rlognorgNBprice -0.9211407*** 0.0671356 _constant -3.67684*** 1.249312 3.139806*** 0.2462953 Note: *** means significant at 1%, ** means significant at 5%, and * means significant at 10%.
75
private label and the non-organic private label decision. Comparing the two sets of first-
stage estimates of Table 4-6 is almost trivial: the results are similar except reversed in
sign, with differences stemming from the joint estimation of (2). On the other hand,
comparing the second-stage results Table 4-7 is nontrivial and shows how identical or
nearly identical factors may influence the private label decision in different ways. Our
discussion, therefore, focuses on Table 4-7 first.
Results from the second-stage private label decision
Examining Table 4-7’s second-stage results in detail allows us to compare the
ways organic and non-organic milk buyers approach the choice between private label and
national brand milk. An important general finding is that all the shopping pattern and
marketing variables affect the private-label decision in similar fashion, no matter whether
organic or non-organic milk is selected. On the other hand, many of the demographic
factors affect the private-label decision in different ways depending on the organic or
non-organic selection. This interesting finding suggests that while various household
types (based on demographics) approach private label milk differently depending on
whether it’s organic or non-organic, various shopping patterns and marketing tactics
aimed at branded or private label products have similar effects for both organic and non-
organic milk. At least six results, found in Table 4-7 and marked (i) through (vi) below,
support the finding that shopping patterns and marketing tactics have similar effects;
alternatively, six other results, marked (vii) through (xii), support the finding that
demographic factors have different effects depending on the first-stage selection.
76
Similar effects from shopping and marketing variables: In the following six
cases, estimated coefficients in both Table 6 columns, the organic and non-organic
private label choice, have like signs and levels of statistical significance.
(i) Table 6 shows that estimated coeffecients for prev_org_PL and prev_norg_PL
are positive and significant, whereas coefficients for prev_org_NB and prev_norg_NB are
negative and significant. In words, no matter whether organic or non-organic is selected,
households who previously purchased private label milk more likely to buy the private
label product again.
(ii) The coefficient for dumcpn is negative, while the coefficient for otherpro is
positive. This result means households using coupons for the milk purchase are less
likely to buy private label organic or non-organic milk. However, households taking
advantage of in-store promotions are more likely to buy private label milk. These results
suggest that coupon-based promotions are an effective marketing tactic of branded milk
products, and the effectiveness holds for both organic and non-organic milk. On the
other hand, in-store promotions are shown to be effective with private label milk, and
again the effectiveness holds for both organic and non-organic milk.
(iii) Estimated coefficients corresponding to product prices have the expected
signs, so higher own-product prices decrease the likelihood of purchase, and higher
alternative prices (i.e., cross prices) increase the likelihood.
(iv) A positive estimate for expfpm implies that households spending more in the
outside category of fresh and frozen produce and meats are less likely to buy private label
(organic or non-organic) milk.
77
(v) Similarly, a positive estimate for expPLfpm implies that households spending
more on outside-category private label products are more likely to buy private label
(organic or non-organic) milk.
(vi) Finally, a positive sign of time trend variable week for both organic and non-
organic milk buyers shows that households are more likely to buy private label (organic
or non-organic milk) as time progresses.
Selection-dependent demographic factors: In the following six cases, estimated
coefficients in Table 4-7’s two main columns have differences in sign and/or levels of
statistical significance.
(vii) Based on the estimated coefficients on income in Table 4-7, one sees that
household income positively affects the decision to buy non-organic private label. On the
other hand, income has no statistical effect on the decision to buy organic private label.
(viii) Turning to the estimates for maxage and dumedu, one sees that households
with older heads, and households with a college education, are each less likely to buy
organic private label milk, but neither age nor college education has a significant effect
on the decision to buy non-organic private label milk.
(ix) Households with more members (hhsize) are more likely to buy non-organic
private label milk, but they are statistically no more or less likely to buy organic private
label milk.
(x) African American households are less likely than white households to buy
organic private label milk, and Hispanic households are more likely. On the other hand,
ethnic or racial background has no statistically significant effect on non-organic milk
purchases.
78
(xi) Households with children are less likely to buy organic private label milk, but
they are statistically no more or less likely to buy non-organic private label milk.
(xii) For dual-income households, full-time employment by a female head
increases the likelihood of buying branded milk conditional on selecting organic, whereas
a female's full-time employment has no statistical impact on the private label or branded
milk decision conditional on selecting non-organic.
From the First-Step Organic Decision
Even though the private label decision is our main focus, estimation results presented for
the organic selection decision (Table 4-6’s left column) add to existing research on
organic purchasing behavior. These results show that demographic variables play an
important role in a household’s choice between organic and non-organic milk. By virtue
of positive and significant estimated coefficients, the following household demographic
factors are shown to significantly increase the probability of a household choosing
organic milk as their main weekly purchase: (a) higher income, where the effect even
grows stronger increases as income increases (i.e., the second order term is positive), (b)
college education, (c) having children under 18, (d) being Asian, African American, or
Hispanic (as opposed to white). Conversely, households that have (e) older heads or (f)
large sizes have a decreased probability of choosing organic milk. These two results are
supported by negative and significant estimated coefficients in Table 4-6. In many cases,
our results are consistent with previous findings (e.g., Zhang et al. 2008, Dettmann and
79
Dimitri forthcoming, and Jonas and Roosen 2008). However, the positive effects for
income’s square and for African American households have not previously been shown.
Perhaps the most novel aspect of our results is how household heads’ employment
status affects the organic choice. Compared to the base case, composed of households
with only one head, Table 4-6’s positive coefficient estimate for the dummy variable that
represents households with two heads but only one income (h2_1inc) implies that having
a second head significantly increases the household’s likelihood of buying organic milk.
Likewise, the positive estimate for households with a fully employed male head and a
part-time employed female head (h2_mf_fp) also implies that, compared to the base case,
the addition of a second part-time income by a female head also increases the
household’s odds of buying organic milk. However, we do not see this increased
likelihood for dual-headed households with a fully employed female head and part-time
employed male head (i.e., h2_mp_ff). In this case, the effect is not statistically different
than the base case. Finally, note that coefficient estimate in Table 4-6 for h2_mf_ff is
negative and statistically different from zero. This result implies that households with
fully employed male and female heads actually have a decreased likelihood of buying
organic milk relative to the base case. Taken collectively, these results have strong
implications on how gender differences in employment may affect organic purchase
decision. In general, we find that households with a woman head working full-time are
less likely to buy organic milk than other household types. Marketers of organic products
may use this result to investigate its underlying reasons, which might relate to
convenience or time spend preparing meals at home. An important caveat is that our data
are not detailed enough to show who, a male or female head, has done the shopping.
80
Almost all of the marketing and shopping pattern variables in Table 4-6
significantly affect the organic/non-organic milk choice. Our results show that
households redeeming a coupon are more likely to buy organic milk. Households
spending more on out-of-category organic products (exporgfpm) are more likely to buy
organic milk; however, households that spend more across all products in that category
(expfpm) are less likely to buy organic milk.
Finally, results in Table 4-6 show an unexpected result for at least one price. As
expected, we find that a higher organic milk price makes a household less likely to buy
organic milk. In fact, the price effect is quite strong. Based on estimated marginal
effects, a 1 percent increase in the organic milk price decreases the odds of buying
organic milk by over 0.57 percent. However, we surprisingly find that a higher non-
organic milk price also makes a household less likely to buy organic milk. This
unexpected price effect, however, is quite small. Based on estimated marginal effects, a
1 percent increase in the non-organic milk price decreases the odds of buying organic
milk by 0.05 percent.
Conclusion and Discussion
This research is among the first effort to investigate whether organic and non-
organic consumers approach private labels or national brands differently. Using milk as a
case study, we model the purchase decision in two connected steps to correct for sample
selection bias. Households first decide whether to buy organic milk, and then decide
whether to buy private label milk. We manipulate household-level purchase data from
81
2004 to 2006 to recover market-level prices, account for endogeneity using appropriate
methods, and include prices along with household demographic and marketing variables
in both an organic selection equation and a private label choice equation in the
simultaneously estimated model. The results of the model indicate how these factors
affect how households approach the private label choice depending on whether they
selected organic or non-organic milk. The model also informs us about the organic
selection decision.
Regarding households’ choice for private label or branded milk, we find the
striking result that most of the demographic results depend on households’ selection in
the first stage, while on the other hand, shopping and marketing factors are invariant to
the first stage selection. For example, the effect of shopping patterns, coupon
redemptions, and other marketing factors affect households’ private label choice in a
similar fashion, no matter whether organic or non-organic milk is first selected.
However, demographic factors including age and education significantly influence the
private label choice for households who select organic but not for those who select non-
organic. Alternatively, other demographic factors such as income and household size
significantly affect the private label choice for households who select non-organic milk
but not those selecting organic milk.
Along the way to these findings on the private label choice, we uncover some
interesting results from the first-stage organic decision. Our results showing a positive
influence towards organic milk purchases of income, education, marriage, and some
racial or ethnic backgrounds. These results are strengthened by specifically controlling
for role that (endogenous) prices play in the organic decision. One new result focuses on
82
gender-based employment levels in the household: For dual-headed households, we find
that full-time employment by a female head can, under some circumstances, decrease the
likelihood of buying organic milk.
From the point of view of food manufacturers and retailers, the results of this
paper will help managers understand who is buying organic private label milk, and how
marketing actions (such as prices and promotions) affect consumers’ decision making.
Some of our results help identifying issues in need of further investigation. For example,
private label manufacturers and retailers might explore further the unintuitive result that
college-educated households are negatively inclined to buy a private label if they are in
the market for organic but not so disinclined if they are in the market for non-organic
milk. Future research might also focus on several issues omitted from this paper,
including the potential influence of advertising, and the role played by loyalty to
individual brands or stores.
83
Chapter 5
Linkages and Implications
This dissertation investigates organic milk marketing and consumer purchase
behavior from three aspects. Each of these three topics is discussed in a single chapter.
That is, Chapter 4 studies how organic private label, non-organic private label, organic
national brand, and non-organic national brand milk prices react to each other; Chapter 5
examines what factors affect consumers’ time to their first organic milk purchase;
Chapter 6 compares the decision making process of organic and non-organic buyers’
private label choice. Although each chapter explores a relatively independent topic, there
are strong linkages among these three chapters. This chapter draws from the conclusions
of Chapters 2, 3, and 4 to examine linkages and implications of the results.
Pricing Strategies
Pricing strategies play an important role in business world because they can affect
profit and sales volume dramatically. Managers want to price their products in a way that
their products are competitive against their rivals and return a relatively high profit.
Given the supply prices and substitution products in a market, competitors’ prices and
consumers’ responses are two major factors managers consider.
Chapter 2 gives predictions and suggestions about competition based pricing.
Treating organic private label, organic national brand, non-organic private label, and non-
organic national brand milk as potential rivals, we provided the following insights to
84
managers: 1) After a price increase, the manager of organic private label milk should
expect a price decrease in organic national brand milk, and no price reaction from non-
organic private label and non-organic national brand milk. 2) After a price increase, the
manager of organic national brand milk should expect a price increase in organic private
label and non-organic private label milk, and a price decrease in non-organic national
brand milk. 3) After a price increase, the manager of non-organic private label milk will
see a price increase in non-organic national brand milk, and no responses from organic
private label and non-organic national brand milk. 4) After a price increase, the manager
of non-organic national brand milk will see a price increases in organic private label,
non-organic private label and organic national brand milk.
Chapter 3 and Chapter 4 predict consumers’ responses to price changes.
Theoretically, consumers’ probability of choosing a product decreases with own price
increase, and increases with substitute’s price increase. However, our research results
show that only some prices affect consumers’ decision making significantly. For
example, the following information may help milk manages’ strategic pricing: 1)
Regarding consumers’ first organic milk purchase, only the organic milk price
significantly affects consumers’ organic milk trial. The coefficient of the non-organic
milk price is not significant, and the magnitude is much smaller than that of organic milk
price. Therefore, managers may want to focus on organic milk price to promote organic
milk trial purchases. 2) When it comes to consumers’ main purchases or organic or non-
organic milk, however, both organic milk price and non-organic milk price significantly
affect consumers’ organic milk choice. However, the effect of organic milk price is
much larger than that of non-organic milk price on main purchases. Therefore, managers
85
may want to focus on organic milk price to increase organic milk sales. 3) When it
comes to the private label decision, prices affect organic and non-organic buyers
differently. Consumers’ organic private label choice is only affected by the organic
private label price, not by the organic national brand price. On the other hand,
consumers’ non-organic private label choice is only affected by non-organic national
brand price, and not non-organic private label price. Therefore, private label milk
managers may want to focus on organic private label price to increase organic private
label milk sales, and focus on non-organic national brand milk price to increase non-
organic private label milk sales.
Putting all three findings together, lowering the organic private label price can
increase private label sales relative to branded organic, and it might have a positive effect
of new trials. However, because this price cut could be accompanied by a price increase
from branded organic milk, the positive impact on collective organic trials will be
lessened while private label’s position relative to branded organic milk will be
strengthened. On the other hand, lowering the organic national brand price can decrease
private label sales relative to branded organic, and it has a positive effect of new trials.
What’s more, because this price cut could be accompanied by a price decrease from
private label organic milk, the positive impact on organic trials will be strengthened.
Targeting Customers
Researching customers in depth is critical for a successful marketing strategy.
Customers are usually divided into two or more segments based on distinguishable
86
aspects. Managers would like to tailor the marketing and sales efforts to specifically
reach the segment that is more likely to buy their product. Consumer markets can be
segmented on consumers’ geographic, demographic, psychographic, or behavioral
characteristics. The results from this dissertation gave insights about consumers’
demographic and behavioral characteristics, and should allow managers to better target
their marketing efforts.
Demographic segmentation is to divide consumers into several groups based on
consumers’ social-economic status. Potential customers are identified by demographic
variables such as income, education level, age, household size, occupation, marital status,
children, and races. Results from Chapter 3 and Chapter 4 can help managers find out
demographic segmentation strategies for the four milk categories.
Managers are able to use the information in the dissertation, and make consumer
segmentation based on income, college education levels, the presence of children, and
age. Each of these demographic factors are discussed in turn: 1) Income: Managers may
want to focus on household groups with high income to promote organic milk trial
purchase as well as organic milk repeated purchase. 2) College education: By dividing
households into two groups, households with heads having and not having a college
degree, managers can use different marketing strategies on organic, non-organic, private
label and national brand milk. For example, managers may want to focus on households
with college education for organic milk marketing, because higher education level is
positively correlated with probability of organic milk purchase. 3) Children: By
targeting households with children, managers can focus on organic milk marketing in
general and organic national brand milk marketing in particular as children’s presence is
87
positively linked to both types of purchases. 4) Age: Households can be segmented
based on age groups, as consumers’ needs and preferences change with age for a given
type of product. Managers may want to focus on households with older age heads to
increase non-organic milk sales and organic national brand milk sales. This is because
households with older heads are more likely to purchase non-organic milk when choosing
between organic and non-organic milk. For the organic buyers, households with older
heads are more likely to buy organic national brand milk.
Behavioral segmentation divides customers into segments based on actual
behavior toward products. Customers are divided by benefits sought, usage rate, brand
loyalty, etc. Using results from Chapter 2 and Chapter 3, managers can 1) make
consumer segmentation based on the usage rate of private labels. This dissertation shows
that coupon users are more likely to buy national brand milk, and households using other
promotion methods such as display and featuring are more likely to purchase private
labels. Therefore, when making promotion strategies, managers may want to focus on
coupons to attract national brand buyers, and focus on other promotion methods to attract
private label buyers. 2) Milk managers can get information about consumers’ shopping
patterns in other categories, and use this information to help customer segmentation. For
example, this research shows that households that spend more on fresh and frozen
produce and meat (FPM) organic products are more likely to purchase organic milk, and
households that spend more on FPM private label products are more likely to purchase
private label milk. Therefore, managers may want to divide consumers into groups based
on their FPM organic and private label purchase, consider the group with higher organic
expenditure to be the segments with higher probability of purchasing organic milk, and
88
consider the group with higher private label expenditure to be the segments with higher
probability of purchasing private label milk. 3) Brand loyalty is a consumer’s
commitment to repurchase a product. The loyalty variables measured by previous
purchases are very strong, showing that previous purchase is a strong predictor for future
purchase. Managers may want to segment customers based on previous purchase, and
focus marketing effort on loyal customers. Finally, some of these behavior segmentation
can be put together. For example, putting (2) and (3) together, managers might find that
a check-out coupon is particularly effective for branded organic milk, but it plays to both
the loyalty effect and the coupon effect associated with this particular product. Targeting
a new market is closely related to targeting customers. One difference is that targeting a
new market usually focuses on market level characteristics instead of individual level
characteristics. To get a high product price, milk managers may use the information in
the dissertation and make the following decisions: 1) Organic private label managers may
want to focus on new markets with older average household head age and high market
share of big volume organic private label milk package. 2) Organic national brand milk
managers may want to expand the market to locations with low private label market
share, low organic market share, low market share of big volume and carton package
milk, and high percentage of high income household. 3) Non-organic private label
managers may want to focus on a potential market with high private label market share,
low organic market share, young average age of household heads, smaller average
household size, high market share of small volume milk, and high percentage of high
income households. 4) Non-organic national brand managers may want to focus on
markets with high private label market share, high organic market share, older average
89
age of household heads, low market share of big volume milk package, high market share
of carton package, and high percentage of high income households.
90
References
Aalen, O.O., Ø. Borgan and H.K. Gjessing Survival and Event History Analysis, 2008).
Abbring, J.H. and G.J.V.D. Berg "The Unobserved Heterogeneity Distribution in
Duration Analysis." Biometrika, Vol. 94, (2007) pp. 87-99.
Arnade, C., M. Gopinath and D. Pick "Brand Inertia In U.S. Household Cheese
Consumption." American Journal of Agricultural Economics, Vol. 90, (2008).
Barstow, C. "Natural and organic private labels become recognizable brand names."
Natural Grocery Buyer, (2005).
Bass, F.M. "The Theory of Stochastic Preference and Brand Switchin." Journal of
Marketing Research, Vol. 11, (1974) pp. 1-20.
Bonanno, A. and R.A. Lopez "Private Label Expansion and Supermarket Milk Prices."
Journal of Agricultural & Food Industrial Organization, Vol. 3, (2005).
Bonfrer, A. and P.K. Chintagunta "Store Brands: Who Buys Them and What Happens to
Retail Prices WHen They Are Introduced?" Review of Industrial Organization,
Vol. 24, (2004) pp. 195-218.
Bontemps, C., V. Orozco and V. Réquillart "Private Labels, National Brands and Food
Prices." Review of Industrial Organization, Vol. 33, (2008) pp. 1-22.
Bontemps, C., V. Orozco, V. Réquillart and A. Trevisiol "Price Effects of Private Label
Development." Journal of Agricultural & Food Industrial Organization, Vol. 3,
(2005).
Casadesus-Masanell, R., M. Crooke, F. Reinhardt and V. Vasishth "Households’
Willingness to Pay for “Green” Goods: Evidence from Patagonia’s Introduction
91
of Organic Cotton Sportswear." Journal of Economics & Management Strategy,
Vol. 18, (2009) pp. 203-233.
Chidmi, B., R.A. Lopez and R.W. Cotterill "Retail oligopoly power, dairy compact, and
Boston milk prices." Agribusiness, Vol. 21, (2005) pp. 477-491.
Cotterill, R.W., R. Dhar and J. William P. Putsis "On the Competitive Interaction
Between Private Label and Branded Grocery Products": University of
Connecticut, 1999).
Cotterill, R.W. and J.W.P. Putsis "Market Share and Price Setting Behavior for Private
Labels and National Brands." Review of Industrial Organization, Vol. 17, (2000)
pp. 17-39.
Cotterill, R.W., W.P. Putsis, Jr. and R. Dhar "Assessing the Competitive Interaction
between Private Labels and National Brands." The Journal of Business, Vol. 73,
(2000) pp. 109-137.
Dimitri, C. and L. Oberholtzer "Marketing U.S. Organic Foods: Recent Trends From
Farms to Consumers", in USDA (ed.), (Washington, D.C.: Economic Research
Service, 2009).
Dimitri, C. and K.M. Venezia "Retail and Consumer Aspects of the Organic Milk
Market": Economic Research Service/USDA, 2007, pp. 1-18).
Duchateau, L. and P. Janssen The Frailty Model, Springer New York, 2008).
Food Marketing Institute and Prevention, "The Shopping for Health 2006: Making
Healthy Eating Easier ", (2006).
92
Gil, J.M., A. Gracia and M. Sa´nchez "Market segmentation and willingness to pay for
organic products in Spain." International Food and Agribusiness Management
Review, Vol. 3, (2000) pp. 207-226.
Glaser, L.K. and G.D. Thompson "Demand for Organic and Conventional Beverage
Milk", Western Agricultural Economics Association Annual Meeting (Vancouver,
British Columbia, 2000).
Gonul, F. and K. Srinivasan "Consumer Purchase Behavior in a Frequently Bought
Product Category: Estimation Issues and Managerial Insights From a Hazard
Function Model With Heterogeneity." Journal of the American Statistical
Association, Vol. 88, (1993) pp. 1219-1227.
Greene, W.H. Econometric Analysis (Upper Saddle River, NJ, 2003).
Grover, R. and V. Srinivasan "Evaluating the Multiple Effects of Retail Promotions on
Brand Loyal and Brand Switching Segments." Journal of Marketing Research,
Vol. 29, (1992) pp. 76-89.
Guadagni, P.M. and J.D.C. Little "A Logit Model of Brand Choice Calibrated on Scanner
Data." Marketing Science, Vol. 2, (1983) pp. 203-238.
Guadagni, P.M. and J.D.C. Little "When and what to buy: A nested logit model of coffee
purchase." Journal of Forecasting, Vol. 27, (1998) pp. 303-326.
Guo, L. and J.M. Villas-Boas "Consumer Stockpiling and Price Competition in
Differentiated Markets." Journal of Economics & Management Strategy, Vol. 16,
(2007).
Gupta, S. "Stochastic Models of Interpurchase Time with Time-Dependent Covariates."
Journal of Marketing Research, Vol. 28, (1991) pp. 1-15.
93
Hein, K. "Study: Sampling Works": Brand Week, Sept 29, (2008).
Helsen, K. and D.C. Schmittlein "Analyzing Duration Times in Marketing: Evidence for
the Effectiveness of Hazard Rate Models." Marketing Science, Vol. 11, (1993) pp.
395-414.
Huang, C.L. and B.-H. Lin "A Hedonic Analysis of Fresh Tomato Prices among Regional
Markets." Review of Agricultural Economics, Vol. 29, (2007) pp. 783-800.
Jain, D.C., F.M. Bass and Y.-M. Chen "Estimation of Latent Class Models With
Heterogeneous Choice Probabilities: An Application to Market Structuring."
Journal of Marketing Research, Vol. 27, (1990) pp. 94-101.
Jain, D.C. and N.J. Vilcassim "Investigating Household Purchase Timing Decisions: A
Conditional Hazard Function Approach." Marketing Science, Vol. 10, (1991) pp.
1-23.
Jain, D.C. and N.J. Vilcassim "Estimating Household Purchase Rates for Consumer Non-
durable Goods." Applied Stochastic Choice Models and Data Analysis, Vol. 10,
(1994) pp. 15-26.
Jenkins, S.P. "Easy estimation methods for discrete-time duration models." Oxford
Bulletin of Economics and Statistics, Vol. 57, (1995) pp. 129-138.
Jeuland, A.P., F.M. Bass and G.P. Wright "A Multibrand Stochastic Model
Compounding Heterogeneous Erlang Timing and Multinomial Choice Processes."
Operations Research, Vol. 28, (1980) pp. 255-277.
Kalbfleisch, J.D. and R.L. Prentice The Statistical Analysis of Failure Time Data
(Hoboken, New Jersey, John Wiley & Sons, 2002).
94
Kalwani, M.U. and D.G. Morrison "A Parsimonious Description of the Hendry System."
Management Science, Vol. 23, (1977) pp. 467-477.
Lancaster, T. The Econometric Analysis of Transition Data, Cambridge University Press,
(1992).
Lehmann, D.R. "Judged Similarity and Brand Switching Data as Similarity Measures."
Journal of Marketing Research, Vol. 9, (1972) pp. 331-334.
Lin, B.-H., T.A. Smith and C.L. Huang "Organic premiums of US fresh produce."
Renewable Agriculture and Food Systems, Vol. 23, (2008).
Organic Trade Association, "Organic Agriculture and Production", (2008).
Progressive Grocer, "Store brands by the numbers", (2007).
Putsis, J.W.P. and R.W. Cotterill "Share, Price and Category Expenditure-Geographic
Market Effects and Private Labels." Managerial and Decision Economics, Vol.
20, (1999) pp. 175-187.
Putsis, J.W.P. and R. Dhar "The Many Faces of Competition." Marketing Letters, Vol. 9,
(1998) pp. 269–284.
Putsis, W.P., Jr. "An Empirical Study of the Effect of Brand Proliferation and Private
Label-National Brand Pricing Behavior." Review of Industrial Organization, Vol.
12, (1997) pp. 355-371.
Putsis, W.P.J. "Empirical analysis of competitive interaction in food product categories."
Agribusiness, Vol. 15, (1999) pp. 295-311.
Research and Markets, "Emerging Organic Food Markets", (2008).
Rhee, H. and D.R. Bell "The Inter-store Mobility of Supermarket Shoppers." Journal of
Retailing, Vol. 78, (2002) pp. 225-237.
95
Rodríguez, G. "Unobserved Heterogeneity", 2005).
Schultz, M. "Organic Dairy Profile", 2008).
Seetharaman, P.B. and P.K. Chintagunta "The proportional hazard model for purchase
timing: A comparison of alternative specifications." Journal of Business &
Economic Statistics, Vol. 21, (2003).
Smith, P.A. "A promising sector: private label milk and dairy products are making the
most of recent success." Dairy Field, (2005).
Smith, T., C. Huang and B. Lin "Estimating organic premiums in the US fluid milk
market." Renewable Agriculture and Food Systems, Vol. 24, (2009) pp. 197-204.
Song, S. "Essays on neighborhood effects and spatial diffusion: Evidence from online
grocery retailing", (Phyladelphia: University of Pennsylvania, 2005).
Steiner, R.L. "The Nature and Benefits of National Brand/ Private Label Competition."
Review of Industrial Organization, Vol. 24, (2004).
Thompson, G.D. and J. Kidwell "Explaining the Choice of Organic Produce: Cosmetic
Defects, Prices, and Consumer Preferences." American Journal of Agricultural
Economics, Vol. 80, (1998) pp. 277-287.
Tian, L. and R.W. Cotterill "Constrained Price, Address, or Logit Brand Demand
Models: An Econometric Comparison in the Boston Fluid Milk Market."
Agribusiness, Vol. 21, (2005) pp. 149-166.
Vaupel, J.W., K.G. Manton and E. Stallard "The Impact of Heterogeneity in Individual
Frailty on the Dynamics of Mortality." Demography, Vol. 16, (1979) pp. 439-454.
96
Vilcassim, N.J. and D.C. Jain "Modeling Purchase-Timing and Brand-Switching
Behavior Incorporating Explanatory Variables and Unobserved Heterogeneity."
Journal of Marketing Research, Vol. 28, (1991) pp. 29-42.
Ward, M.B., J.P. Shimshack, J.M. Perloff and M. Harris "Effects of the Private-Label
Invasion in Food Industries." American Journal of Agricultural Economics, Vol.
84, (2002) pp. 961-973.
Wedel, M., W.A. Kamakura, W.S. Desarbo and F.T. Hofstede "Implications for
Asymmetry, Nonproportionality, and Heterogeneity in Brand Switching from
Piece-Wise Exponential Mixture Hazard Models." Journal of Marketing
Research, Vol. 32, (1995) pp. 457-462.
Yamaguchi, K. Event History Analysis, Sage Publications, 1991).
Yiridoe, E.K., S. Bonti-Ankomah and R.C. Martin "Comparison of consumer perceptions
and preference toward organic versus conventionally produced foods: A review
and update of the literature." Renewable Agriculture and Food Systems, Vol. 20,
(2005) pp. 193-205.
27
YAN ZHUANG
VITA
CONTACT INFORMATION Cell phone: (814)321-5834 Email: [email protected] RESEARCH INTERESTS Food Marketing, Consumer Economics, Industrial Organization
EDUCATION PhD in Agricultural, Environmental and Regional Economics and Operations Research (double major); Minor in Statistics Penn State University, University Park, PA August 2010 Master in Institutional Economics Tsinghua University, Beijing, China Bachelor in Economics and Political Science and Public Administration (double major) Beijing/Peking University, Beijing, China
PAPERS UNDER REVIEW Zhuang, Yan, Carolyn Dimitri, and Edward Jaenicke. "Consumer Choice of Private Label or National Brand: The Case of Organic and Non-Organic Milk." Agribusiness: an International Journal: Revise and resubmit.
PAPERS PRESENTED AT PROFESSIONAL MEETINGS Zhuang, Yan, Carolyn Dimitri, and Edward Jaenicke. "Consumer Choice of Private Label or National Brand: The Case of Organic and Non-Organic Milk." Presented on AAEA (Agricultural & Applied Economics Association) & ACCI (The American Council on Consumer Interests) 2009 Joint Annual Meeting. Milwaukee, WI, 2009. Zhuang, Yan, Carolyn Dimitri, and Edward Jaenicke. "Price Reactions and Organic Price Premiums for Private Label and Branded Milk." Accepted by the Joint European Association of Agricultural Economics (EAAE)/ Agricultural & Applied Economics Association (AAEA) Seminar. Freising, Germany, 2010 AWARDS AND HONORS Sigma Gamma Delta, Agricultural Honor Society, Penn State University 2009
