Less is More Expensive: Income Differences in Bulk

Buying

By Mallick Hossain∗

Quantity discounts are an effective way for households to save

money. I explore how large these quantity discounts are, how
bulk buying differs by income, and how other factors affect the bulk
buying decision. Using Nielsen’s granular store- and household-level
data, I establish two empirical facts. First, quantity discounts are
large for a wide range of products. Second, low-income households
are less likely to buy in bulk than high-income households. I estimate
that low-income households could reduce their grocery expenditures
by 5%, saving an aggregate of $5.4 billion annually, if they bought
in bulk to the same extent as high-income households. I augment
Nielsen data with new data that I collect on state-level unit-price
regulations and on warehouse club entry. I find that a combination
of cognitive costs, store preferences, budget constraints, and storage
costs discourage low-income households from realizing these savings.
I then estimate a discrete choice model of household purchasing
behavior to quantify how bulk buying changes when cognitive costs
and storage costs are reduced. Counterfactual simulations show that
mandating the display of unit prices, which has only been adopted
by nine states, would reduce the difference in how often the highest-
and lowest-income households buy in bulk by 26%.

∗ Federal Reserve Bank of Philadelphia, Ten Independence Mall, Philadelphia, PA, 19106,
[email protected]. The views expressed here are those of the author and do not necessarily
represent the views of the Federal Reserve Bank of Philadelphia or the Federal Reserve System. This
article is based on a chapter of my dissertation. I thank my advisers, Katja Seim, Holger Sieg, Aviv
Nevo, and Sarah Moshary for their advice and constructive criticism. Thanks to Mike Abito, Francesco
Agostinelli, Emek Basker, Minsu Chang, Chris Cronin, Frank DiTraglia, Eileen Divringi, Hanming Fang,
Jim Ferry, Ishan Ghosh, Joao Granja, Kathleen Hui, Ben Hyman, Jeff Lin, Paolo Martellini, Davin Reed,
Claudia Sahm, Paul Sangrey, Andrew Shephard, Petra Todd, Anna Tranfaglia, and Keith Wardrip for
their comments. Thanks to Dmitri Koustas for providing data on warehouse clubs. The conclusions
drawn from the Nielsen data are those of the researcher and do not reflect the views of Nielsen. Nielsen is
not responsible for, had no role in, and was not involved in analyzing and preparing the results reported
herein.
1
2 INCOME DIFFERENCES IN BULK BUYING

I. Introduction

Grocery purchases account for a sizable share of a household’s discretionary

spending, especially for the lowest-income households (BLS, 2019). To save money,
households often wait for sales, redeem coupons, or purchase generic brands
(Griffith et al., 2009). Low-income households also increase their home production
to reduce their spending (Aguiar and Hurst, 2005). Quantity discounts are another
way for households to save money. Even though households buy more on a given
shopping trip, they pay lower unit prices, and reduce their overall spending.
I examine how large these quantity discounts are, how bulk buying varies by
income groups, and which factors influence the decision to buy in bulk. Most
explanations for why households pay different prices for the same product relate
to differences in search behavior. Prices vary based on where households shop or
whether they use coupons (Talukdar, 2008; Griffith et al., 2009). However, even in
the absence of sales and coupons, prices for the same product often differ within a
store due to quantity discounts.
This paper contributes new findings that, despite the substantial savings available
from quantity discounts, low-income households are less likely to buy in bulk than
high-income households.1 Kunreuther (1973) provides the first evidence of this
“bulk buying gap” for a few specific products and Orhun and Palazzolo (2019)
expands this finding to a whole product category. Since households purchase a
variety of products when shopping, I show that the bulk buying gap exists across
the full range of product categories that households purchase.
I find that cognitive costs, store preferences, budget constraints, and storage
costs all contribute to this bulk buying gap. First, the cognitive costs of assessing
price differences across products can prevent households from making economical
decisions (Mitchell, Lennard and McGoldrick, 2003). Providing price information
reduces the effort needed to compare prices, and households change their purchase
decisions when such information is prominently displayed (Chetty, Looney and
Kroft, 2009; Bogomolova and Jarratt, 2016). Posting unit prices reduces the
cognitive costs of comparing unit prices across different products. Sixteen states
have regulations governing the display of unit prices, but no study has evaluated the
impact of these regulations on consumer behavior. I provide the first nationwide
study of the impact of displaying unit prices on bulk purchasing and find that
households are significantly more likely to buy in bulk when retailers are mandated
to display unit prices.
Second, households may make different purchase decisions based on where they
live or where they shop (Chung and Myers Jr, 1999; Talukdar, 2008; Allcott et al.,
2019). I show that even within neighborhoods, there are large differences in bulk
buying between high- and low-income households. Even after controlling for where
households shop, income differences in bulk buying persist.

1 Throughout this paper, “high-income” refers to households making over $100,000 and “low-income”
refers to households making under $25,000.
 INCOME DIFFERENCES IN BULK BUYING 3

Third, budget constraints affect bulk buying because low-income households

may not have enough cash on hand to purchase a bulk package. Leveraging
within-month variation in budgets, I show that the lowest-income households
slightly decrease their bulk purchasing towards the end of month, presumably
when their budget constraint is binding. In contrast, middle- and high-income
households either do not change their bulk buying or slightly increase their bulk
buying towards the end of the month.
Fourth, storage costs also affect the bulk buying decision because even though
large packages provide lower unit prices, they are more cumbersome to store. I
show that households are more likely to buy in bulk when they live in larger
homes and when products are smaller. I provide a new approach to estimate
storage costs cross-sectionally using differences in product “concentration.” One
way of identifying household-level storage costs compares purchase frequencies
of households with the same demand for a product (Hendel and Nevo, 2006). A
household with high storage costs will purchase small packages more frequently.
I propose a complementary approach: I compare purchases of products that
are otherwise identical, but differ in their level of concentration. All else equal,
households who buy smaller, more concentrated packages have higher storage costs
than those that buy larger, less concentrated packages. Based on this approach,
I find that low-income households have higher storage costs than high-income
households. On balance, this means that firms extract higher rents from low-income
households because they have less ability to store for future consumption.
For my analysis, I combine household- and store-level datasets to study income
heterogeneity in bulk buying. Nielsen’s Consumer Panel data are a nationally
representative panel survey of household grocery purchases, and Nielsen’s Retail
Scanner data are a national panel of weekly UPC-level sales data from over
30,000 stores. I construct a new dataset of state-level per-unit pricing regulations,
including a measure of regulatory stringency. I also use data on entry dates and
locations of over 1,400 warehouse clubs in the United States. As a result, I have
a comprehensive view of a household’s possible product choices, available price
information, retail environment, and resulting expenditures.
I find that low-income households could realize substantial savings from buying
in bulk at the same rate as high-income households. To do this, I estimate the
average bulk discount for each product category based on Nielsen’s weekly store-
level price and product data. The average discount across all product categories
is such that a 10% larger package has a 5% lower unit price. Then, I estimate
how much each household buys in bulk using Nielsen’s household-level purchase
data. Given each product category-specific bulk discount and household-level bulk
buying, I predict how much low-income households could save if they increased
their bulk buying intensity to match that of high-income households. I find that
low-income households would reduce their annual grocery expenditures by 5%
if they bought in bulk like high-income households, saving an aggregate of $5.4
billion annually.
4 INCOME DIFFERENCES IN BULK BUYING

I then employ three differences-in-differences models to determine how much

cognitive costs, store preferences, and budget constraints affect bulk buying. The
first model uses a novel dataset that I compiled of state regulations regarding
the display of per-unit prices and exploits the fact that a significant share of
households in the Nielsen Consumer Panel moves between regulatory regimes
when they move from one state to another. Before households move to a state
without unit price posting requirements, their bulk buying behavior is similar to
that of households that remain in the same regulatory regime. After households
move, however, I find that their bulk buying is 4–5% lower than that of households
who did not experience a regulatory change.
The second differences-in-differences model shows how much store preferences,
particularly for warehouse clubs, affect bulk buying. The bulk buying gap narrows
substantially after controlling for the types of stores households shop at; high-
income households spend a larger portion of their budget at warehouse clubs. To
estimate the effect of warehouse clubs on bulk buying, I examine how bulk buying
changes within households before and after a warehouse club enters nearby. Before
a warehouse club enters, bulk buying is similar for households that will and will not
experience a warehouse club entry. After a warehouse club enters, however, I find
bulk buying increases by 5–10% compared to households that did not experience
an entry within a 15-mile radius. This increase is only limited to middle- and
high-income households and is due to increased shopping at warehouse clubs.
The third differences-in-differences model analyzes how budget constraints affect
bulk buying. To do this, I analyze within-household changes in bulk buying over
the course of a month because the liquidity of poorer households decreases towards
the end of the month (Orhun and Palazzolo, 2019). Using information on weekly
bulk buying from Nielsen, I find that low-income households slightly decrease
their bulk buying by 0.5–1% between the first half and second half of the month
while middle- and high-income households slightly increase their bulk buying by
0.5–1.5% over the same period.
I also assess the importance of storage costs using differences in bulk buying
relative to the size of a household’s home and relative to the physical size of the
product being purchased. I examine how a household’s bulk buying changes when
it moves to a different type of housing, after controlling for other within-household
changes. Bulk buying is 3–4% higher for the same household when it lives in
a single-family home relative to when it lives in an apartment, controlling for
demographic differences between households. I also find that the bulk buying gap
is smaller for products with smaller physical footprints.
Finally, I construct a discrete-choice model of consumer purchasing behavior
to quantify consumer preferences and disentangle the contributions of cognitive
and storage costs to the bulk buying decision. I estimate this model using data
on toilet paper purchases. Households choose a product based on price, quantity,
quality, and package size, which serves as a proxy for storage costs. I can separate
preferences for quantity from size preferences because I demonstrate that toilet
 INCOME DIFFERENCES IN BULK BUYING 5

paper comes in varying “concentrations.” I allow state-level unit pricing mandates

to affect a household’s unit price sensitivity. From this demand model, I simulate
household responses to two counterfactuals: 1) universally posting unit prices and
2) reducing storage costs.
My model predicts that requiring stores to post unit prices would reduce
the bulk buying gap in package size purchased between high- and low-income
households by 26%. Reducing storage costs would close the remainder of the gap
and low-income households would actually buy more in bulk than high-income
households. As a result of these policies, households would buy larger quantities
of toilet paper and pay lower unit prices. Universally displaying unit prices would
encourage households to better utilize quantity discounts by reducing cognitive
costs, increasing bulk buying, and helping consumers save money.
The rest of the paper is structured as follows. Section II describes the data.
Section III documents new facts of quantity discounting. Section IV presents
evidence of contributing factors to the bulk-buying gap. Section V introduces the
model. Section VI presents estimation and counterfactual results. Section VII
concludes.

II. Data

In this section, I describe the datasets used for my analysis and give a brief
overview of their respective features.2 Nielsen’s Consumer Panel data provide
information on households’ shopping and purchasing decisions. Nielsen’s Retail
Scanner data provide information on weekly product assortments and prices. A
new regulatory dataset I construct contains information on state-level regulations
regarding the display of per-unit pricing. Finally, warehouse club data provide
information on the location and entry dates of warehouse clubs across the United
States. By combining these data, I have a comprehensive view into a household’s
possible product choices, available price information, retail environment, and their
resulting purchase decision.

A. Nielsen Consumer Panel Data

I use the Nielsen Consumer Panel dataset from 2004–2017. This dataset is a
panel of about 178,000 unique households. I observe about 40,000 households
each year from 2004–2006 and about 60,000 households each year from 2007–2017.
Households scan all items that they purchase and then input information about
quantities, prices, date of purchase, and store. Nielsen retains about 80% of its
panel from year to year with the mean and median tenure of a household being
four and three years, respectively.

2 Researcher’s own analyses derived based in part on data from The Nielsen Company (US), LLC
and marketing databases provided through the Nielsen Datasets at the Kilts Center for Marketing Data
Center at The University of Chicago Booth School of Business.
6 INCOME DIFFERENCES IN BULK BUYING

I consider food, drink, and non-food grocery (e.g., paper towels, toilet paper,
detergent) purchases made at grocery stores, discount stores, dollar stores, ware-
house clubs, and drug stores. These outlets account for over 90% of household
expenditures in these categories. I exclude alcohol, tobacco, health, and general
merchandise products from my analysis since these products (e.g., cigarettes,
painkillers) may have different consumption patterns than grocery products or
are not suited for bulk purchases (e.g., printers, cookware, linens). I also exclude
households with a student or military head of household as well as those with an
annual income of less than $5,000 and those living in mobile homes. Only about
7% of households are excluded and I use the remaining 166,000 households for my
analysis. See Appendix A.A1 for further details of sample construction.
Nielsen computes projection weights to ensure their sample is nationally represen-
tative. Weights are calculated to match population moments based on household
size, income, age, race, ethnicity, education, occupation, and presence of children.
All analyses use these projection weights unless otherwise stated. Nielsen groups
household income into 16 different income bins. Due to the large number of
bins, in tables and parts of the text, I will report differences by income quartiles.
However, where possible (especially in graphs), I will report estimates for each
income bin. Table 1 presents descriptive statistics for households in the sample.

Table 1—: Nielsen Consumer Panel Summary Statistics

Variable Mean SD 25th Pctile 75th Pctile

Household income ($000s) 57.13 31.17 27.5 85
Household size 2.56 1.45 1 3
Age 52.59 14.39 41.5 63
College educated 0.39 0.49 0 1
Child present 0.33 0.47 0 1
Married 0.50 0.50 0 1
N (Household-Years) 734,724
N (Households) 166,163

Note: Household income is grouped into bins. Midpoints of each bin are used in order to calculated
sample moments. Data are weighted for national representativeness.
Source: Nielsen Consumer Panel (2004–2017)

B. Nielsen Scanner Data

The Nielsen Scanner data contain average weekly prices and volume sold of
individual products at about 35,000 stores from about 90 retail chains between
2006–2016. Average prices are weighted by the volume sold. Only products with
positive sales in a given week are recorded. I match the Retail Scanner data with
 INCOME DIFFERENCES IN BULK BUYING 7

the Consumer Panel data based on store identification numbers and purchase
dates. By matching the two datasets, I recover the set of products available to a
household and the product it chose to purchase.

C. Unit Pricing Regulations

I compile a novel dataset on state-level regulations regarding the display of unit

prices. The data are based on annual regulatory updates aggregated in Handbook
130 published by the National Institute of Standards and Technology (NIST, 2019).
I cross-check this information with state regulatory codes and state officials to
ensure accuracy. This data include information on which states have regulations,
when they were adopted, and how stringent these regulations are. More details
are discussed in Section IV.A.

D. Warehouse Club Data

I also use data on all warehouse clubs in the United States between 2004–
2015 gathered by Coibion, Gorodnichenko and Koustas (2017). This data record
information on the opening dates, locations, and identity of all warehouse clubs
in the United States. It was gathered by combining information available on
company websites, annual reports, and by contacting firms.

III. Stylized Facts

In this section, I document two new facts about quantity discounts. First, I show
that quantity discounts apply to 91% of grocery categories. Second, I document
that households making over $100,000 are 26% more likely to buy non-food items
in bulk than households making $5,000–$8,000 annually, compared to only 3%
for food items. Combining these findings, I estimate that low-income households
could reduce their grocery expenditures by 5%, saving an aggregate of $5.4 billion
annually, simply by buying in bulk at the same rate as high-income households.

A. Quantity Discount Prevalence

Quantity discounts are a specific form of non-linear pricing in which unit prices
decrease as package size increases. To establish the prevalence and magnitude
of quantity discounts, I use Nielsen’s Retail Scanner data from 2016. I estimate
quantity discounts using the following regression for each of the 693 product
categories:3

(1) ln(P )ibm = α + β ln(Size)ibm + λbm + ibm ,

where Pibm is the unit price (package price divided by package size) of product
i from brand b purchased in market m (defined as a store-week). Sizeibm is the
3 48 categories could not be estimated typically because the data did not have sufficient variation.
These were generally uncommon categories like mushroom sauce, canned grapes, and canned chow mein.
8 INCOME DIFFERENCES IN BULK BUYING

item’s package size, which is the number of units included in a UPC (e.g., quart,
square feet, count, pound). λbm is a brand-store-week fixed effect. Variation in
unit prices across package sizes within the same brand-store-week identify β.4 If
retailers offer quantity discounts, then β will be negative.
Figure 1 plots the distribution of β across product categories (statistically
insignificant betas are zero). 91% of all product categories have a statistically
significant and negative β and non-food items generally have larger discounts
than food items.5 The median β is -0.51 for non-food products, which means that
a 10% increase in package size is associated with a 5.1% decrease in unit price.
This discount is larger than the median β for food items (-0.43).6 The size and
near-universality of quantity discounts suggest they offer substantial savings to
households without sacrificing consumption.7

B. Bulk Purchasing

Given how common and how large quantity discounts are, households can use
quantity discounts to save money on a wide range of items. However, since food
products deteriorate while non-food products do not, bulk buying will likely differ
between food and non-food items. Because of these differences, I analyze food and
non-food products separately. Following the literature, I classify a product as “bulk”
if it is in the top two quintiles of the size distribution for that product category
(Griffith et al., 2009).8 Then, for each household, I compute the expenditure
share of bulk purchases of food and non-food items. I then regress this “bulk
share” on household income and other household characteristics that could affect
consumption patterns and may be correlated with income, and plot the income
coefficients. The equation below is estimated on food and non-food purchases
separately:
X
(2) BulkShareimt = α + β q Incomeimt + γXimt + λm + λt + imt ,
q

where BulkShareimt is household i’s share of bulk purchases in market m in year

t (a market is a county). Incomeimt consists of dummies for each income bin q.

4 Some readers may be concerned that the positive sales threshold limits the number of weeks products
are observed. I find that a large majority of products (at the UPC level) are observed for over half of the
year. The unobserved weeks can be attributed to a variety of reasons including zero sales, discontinued
products, or missing reports from retailers. Observing products for most weeks of the year limits the
possibility that quantity discounts are estimated on a limited subset of weeks.
5 Some products do have a significantly positive coefficient, indicating that unit prices increase with
package size. These quantity “surcharges” are less common, but have been highlighted before (Sprott,
Manning and Miyazaki, 2003).
6 These findings are robust to outliers. Winsorizing unit prices at the 98th and 90th percentile
produces almost identical estimates.
7 For a comparison of quantity discounts with coupons, see Appendix A.A2.
8 This definition avoids the risk of too narrowly defining bulk and only capturing purchases that occur
solely at warehouse clubs. This broader definition helps capture large sizes that are available at grocery
stores and mass merchandisers.
 INCOME DIFFERENCES IN BULK BUYING 9

Figure 1. : Distribution of Bulk Discounts by Product Type

1.5

Median: −0.51

1.0
Density

Median: −0.43

0.5

0.0

−2 −1 0 1
Elasticity of Unit Price w.r.t. Package Size

Product Type Food Non−Food

Note: Figure plots the distribution of coefficients from a regression of log unit price on log package size
(Equation (1)) for individual product categories. Regression controls for store-brand-week fixed effects.
Histogram plots 645 product categories.
Source: Nielsen Retail Scanner (2016)

Ximt consists of household characteristics (age, household composition, marital

status, education, housing type, tract-level vehicle access).9 Year and market fixed
effects are captured by λt and λm .
Figure 2 illustrates that bulk purchases compose a 10 percentage point larger
share of non-food expenditures for households making over $100,000 compared
to those making $5,000–$8,000. As income increases, bulk purchases make up an
increasing share of expenditures. For food items, there is a more muted increase
of one percentage point across income groups.
The 10 percentage point gap is quite large. For the average household making
between $5,000–$8,000, 39.6% of their non-food grocery spending is on bulk
packages. Hence, households making over $100,000 are 26% more likely to buy in
bulk relative to the lowest-income group.
These patterns are consistent with high-income households buying in bulk,
obtaining low unit prices, and consuming out of storage. Given the existence of
quantity discounts, larger packages generally correspond to lower unit prices. The
fact that low-income households are less likely to buy these storable items in bulk

9 These characteristics are used consistently throughout the paper. See Appendix A.A1 for details of
demographic variables and how they are collected.
10 INCOME DIFFERENCES IN BULK BUYING

Figure 2. : Bulk Purchasing by Household Income and Product Type

10.0
Difference in Bulk Purchasing
(Percentage Points)

7.5

5.0

2.5

● ●
●
● ● ● ●
0.0 ● ● ● ● ● ●
●

0 25 50 75 100
Annual Household Income ($000s)

Product Type ● Food Non−Food

Note: Figure plots the income bin coefficients from Equation (2), which regresses the share of annual
purchases that were bulk packages on household characteristics as well as market and year fixed effects.
Nielsen projection weights are used to ensure national representativeness. Households making $5–$8k
are the reference group. Coefficient values are reported in Appendix Tables A7 and A8. 95% confidence
intervals are shown.
Source: Nielsen Consumer Panel (2004–2017)

suggests that some obstacles may prevent them from buying and storing large
packages.10
Because the bulk buying gap is largest for non-food products, the rest of this
paper focuses on non-food products. These products are ideal for analyzing bulk
purchasing because they isolate the key features that make bulk buying and
quantity discounts attractive for households. Primarily, households can store
items for future consumption. Additionally, these products generally do not have
substitutes and they cannot be produced at home (e.g., toilet paper, diapers).
My findings carry over to food products, but one must be careful to account
for perishability, which counteracts product storability. Additionally, many food
products have close substitutes (e.g., soda, juice, water) and home production
(e.g., cooking meals) can substitute for many products (Aguiar and Hurst, 2005,
2007).

10 This relationship persists across most categories. Appendix A.A3 shows the same pattern for a few
popular categories and Figure 8 illustrates the difference for all non-food categories.
 INCOME DIFFERENCES IN BULK BUYING 11

C. Savings from Bulk Buying

In this subsection, I calculate the savings that low-income households could

achieve from buying in bulk like high-income households. For each product category,
I compute the average difference in package sizes purchased by estimating the
following regression:
X
(3) ln(AvgSize)imt = α + β q Incomeimt + γXimt + λm + λt + imt ,
q

where AvgSizeimt is the quantity-weighted average package size purchased by

household i in market m in year t, where a market is a DMA.11 Incomeimt
is an indicator for a household’s income quartile. X controls for household
characteristics. Market and year fixed effects are included through λm and λt .
In this regression, β q gives the average log-difference between the package size
purchased by a household in income quartile q and the lowest-income quartile
(households making less than $25,000).12 To compute savings, I multiply this
average difference in package size purchased by the category-specific quantity
discount estimated in Section III.A. For example, high-income households buy 30%
larger packages of toilet paper which has a quantity discount of 0.216. Therefore,
low-income households could save 0.3 × 0.216 = 0.0648 or 6.5% from buying big
packages like high-income households do. Aggregating across all categories where
high-income households buy larger packages gives an estimated savings of 5%, or
$215, per year.13,14
Saving 5% on these common household purchases is substantial for low-income
households. For the bottom quintile of the income distribution, these items account
for 30% of their discretionary spending compared to 19% for the top quintile of
the distribution.15 If the 24.4 million households making under $25,000 were to
obtain these savings, that would be an overall savings of $5.4 billion annually,
assuming no supply-side response.16 For context, this is equal to 8% of the $68
billion federal Supplemental Nutrition Assistance Program budget in 2017 (USDA,

11 Average package size is weighted by quantity to account for the fact that an unweighted average
would favor small packages.
12 I use quartiles to reduce the number of income bins from 15 to 4, but results hold at more granular
levels. Disaggregated results are available upon request.
13 This averages only across categories where high-income households buy larger packages. There are
some categories, such as septic tank cleaners, in which high-income households buy in smaller packages.
Imposing that low-income households buy the same average size across all categories reduces projected
savings to 2.3%.
14 The first-best calculations of savings would identify the product with the lowest unit price given
a household’s brand and store choice and compute savings based on that product. This estimate will
likely be substantially higher than what I computed, so I view the estimated 5% savings as a conservative
estimate of potential savings. See Appendix A.A4 for calculations of savings on popular product categories.
15 Discretionary spending is defined as total expenditures minus expenditures on shelter, utilities,
transportation, healthcare, cash contributions, personal insurance, and pensions. Calculation is based on
expenditure data on food at home and housekeeping supplies from Table 1 of the 2017 Consumer Expen-
diture Survey available at https://www.bls.gov/opub/reports/consumer-expenditures/2017/home.htm
16 Household count comes from Table B19001 of the 2017 1-year American Community Survey.
12 INCOME DIFFERENCES IN BULK BUYING

2019). These potential savings do not require low-income households to buy more
over the course of the year because buying in bulk does not necessarily change
how much households consume. It just changes how much they buy at one time.

IV. Factors Affecting Bulk Buying

In this section, I show that cognitive costs, store preferences, budget constraints,
and storage costs affect the bulk buying decision. To do this, I use plausibly
exogenous variation and natural experiments to estimate the causal impact of unit
pricing regulation and warehouse club entry on bulk purchasing. Since the biggest
differences in bulk buying are for non-food grocery items, all analysis is restricted
to non-food products.

A. Cognitive Costs

Cognitive costs are the first possible contributor to the bulk buying decision.
Consumers may not be aware of the quantity discount (or how valuable it is)
because they do not compute unit prices when making purchases. To test this
hypothesis, I utilize a novel hand-collected dataset of state-level unit-price regu-
lations requiring retailers to display per-unit prices. Displaying per-unit prices
reduces cognitive costs and households can more easily compare products and pick
the one with the best value.
Unit price labeling dates back to the late 1960s and early 1970s. During this
period, a large consumer protection movement pushed for unit prices to be posted
so consumers could compare different brands and sizes of products (Miyazaki,
Sprott and Manning, 2000). As a result, some states passed laws requiring
retailers to display unit prices. These laws varied widely with some giving retailers
discretion over how to display unit prices and other states specifying formatting
requirements, such as minimum font sizes and background colors to aid readability
and clarity (Rose, 2000).
Using annual regulatory updates published by the National Institute of Standards
and Technology (NIST), I compile state-level regulations on unit pricing (NIST,
2019). For states with regulations, I cross-check NIST’s designation with state
regulatory codes and consult with state officials to ensure accuracy. Figure 3
shows that, as of 2017, 16 states have regulations on the display of unit prices and
34 have no regulations.17
If these regulations affect household decisions, then bulk buying should differ
between states with and without these regulations. I first document how aggregate
patterns in bulk buying differ between states with different regulations and then I
will provide causal evidence for the impact of these regulations. I estimate the
following regression:

(4) BulkShareit = α + β1 Regit + γXit + λt + it ,

17 Summary statistics of these groups are reported in Appendix Table A11.
 INCOME DIFFERENCES IN BULK BUYING 13

Figure 3. : Unit Price Regulations by State (2017)

Regulation Status Mand. Disp, Strict Mand. Disp Vol. Disp No Reg

Note: Figure plots whether or not a state has regulations in place governing the display of unit prices
as of August 1, 2017. “No Reg” denotes that no regulations are in effect. “Vol. Disp” denotes states
where regulations apply if retailers choose to display unit prices. “Mand. Disp” denotes states where all
retailers must display unit prices. “Mand. Disp, Strict” denotes states where strict display formatting
requirements are in effect.
Source: NIST Handbook 130

where BulkShareit is the annual share of expenditures that were bulk purchases
for household i in year t. Regit is an indicator for whether or not unit-price
regulations are in effect. Xit controls for household characteristics. I control for
year fixed effects through λt . Standard errors are clustered by state because these
regulations are at the state level.
Since 2004, no state has modified its regulations on unit prices, so the coefficient
on unit pricing regulation is identified from cross-sectional variation between states
that have regulations and those that do not.18,19 Columns (1) and (2) of Table
2 reveal that bulk purchasing is 3.6 percentage points higher in states with unit
price regulations compared to states without unit price regulation, even after
controlling for household characteristics and year fixed effects.
I then analyze these unit pricing regulations at a higher level of detail. State
regulations vary across two dimensions: Posting and Formatting. Table 3 shows
the breakdown of states along these dimensions. First, states can opt to have unit

18 Becausethere is no time variation in regulations, I cannot include state fixed effects in the estimation.
19 In
2013, the District of Columbia passed a law requiring retailers to display unit prices, but no
households in my sample live in DC.
14 INCOME DIFFERENCES IN BULK BUYING

Table 2—: Unit Price Regulations and Bulk Buying

(1) (2) (3) (4)

Regulation 0.035∗∗ 0.036∗∗
(0.017) (0.016)
Vol. Disp 0.050∗ 0.011
(0.027) (0.010)
Mand. Disp 0.038∗∗∗ 0.038∗∗∗
(0.009) (0.009)
Mand. Disp, Strict 0.028∗∗∗ 0.028∗∗∗
(0.006) (0.006)
Avg Bulk 0.5 0.5 0.5 0.49
Demographics N Y Y Y
Omit California N N N Y
Observations 732,512 732,512 732,512 668,791
Adjusted R2 0.006 0.057 0.045 0.037
Note: Table shows the results of estimating Equation 4. The dependent variable is the annual share of
bulk purchases made by households and the independent variables are either a binary indicator of the
presence or absence of regulation (Columns (1) and (2)) or an ordered measure of regulatory stringency
(Columns (3) and (4)). Column (4) omits California because it is the only state that has voluntary unit
price, but strict requirements on how unit prices are displayed. Standard errors are clustered by state.
∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01.
Source: Nielsen Consumer Panel (2004–2017)
 INCOME DIFFERENCES IN BULK BUYING 15

price posting be voluntary (seven states) or mandatory (nine states). Second, states
can specify how unit prices are formatted when they are displayed.20 Formatting
regulations specify features including minimum font sizes, background colors, and
label positioning. With the exception of California, only states that mandate unit
price posting have formatting requirements. Excluding California, regulations
are naturally ordered: no regulation, voluntary posting, mandatory posting (no
formatting requirements), and mandatory posting (with formatting requirements).

Table 3—: Unit Price Regulations by State

No Formatting Rules Strict Formatting Rules
Voluntary Arkansas Montana California
Posting Florida Nevada
Hawai’i West Virginia
Mandatory Maryland Vermont Connecticut New York
Posting New Hampshire Massachusetts Rhode Island
Oregon New Jersey
Note: Table reports whether unit price posting is mandatory or voluntary for retailers and whether or
not there are strict formatting requirements on how unit prices should be displayed (minimum font size,
color, etc.).

Columns (3) and (4) continue the earlier analysis, but leverage the stringency
of the regulations. Column (3) shows that mandatory posting is associated with
significantly higher bulk buying, but states with voluntary requirements may have
higher rates of bulk buying. However, as Table 3 shows, California is an outlier in
this regulatory environment because is the only state with the unique combination
of voluntary posting and strict formatting requirements. Because of this, I exclude
California and re-estimate the regression. Column (4) reveals that California is
the primary driver of this effect and states with voluntary posting do not have
significantly higher bulk purchasing. On the other hand, mandatory unit price
posting is associated with a 2.8–3.8 percentage point increase in bulk buying. The
point estimates for bulk buying in states with strict formatting requirements are
lower than those in states without formatting requirements, but these estimates
are not significantly different from each other. This pattern supports the intuition
that standardized unit price presentation reduces cognitive costs, increases the
salience of unit prices, and facilitates comparisons for consumers.
This estimation provides strong evidence of a relationship between unit pric-
ing regulations and bulk purchasing. However, there is a risk of selection bias
since these regulations were primarily adopted in the Northeast and West Coast
regions of the United States. To provide causal evidence, I examine about 13,000
households that move once during their tenure in the data. About 11% of these
20 All states with these regulations standardize how unit prices are to be calculated, which is what
makes the voluntary states different from states without regulations.
16 INCOME DIFFERENCES IN BULK BUYING

households move between regulatory regimes while the remainder are either local
moves or moves that maintain their current regulatory regime. To estimate the
effect of unit-price regulations on these movers, I use a differences-in-differences
specification:

(5) BulkShareit = α + β1 Regit + γXit + λi + λt + it ,

where the variables are the same as in Equation (4), but I control for household
fixed effects and standard errors are clustered at the household level.21 With
this specification, β1 is identified by changes in bulk purchases for households
that move from a state with unit-price regulations to a state without unit-price
regulations (or vice versa).22 Since the “direction” of a household’s move may
matter (i.e., whether they start in a state without regulations and move to a state
with regulations or vice versa), in subsequent specifications, I will account for the
direction of the move.23
This specification relies on the assumption that households would have continued
buying in bulk like other households that moved, but did not change their regula-
tory regime. To provide evidence supporting this “common trends” assumption, I
plot an event study by estimating a modified version of Equation (5):
X
(6) BulkShareit = α + β1τ Y rit + γXit + λi + λt + it ,
τ 6=−1

where Y r is a dummy for each year before or after a household moves to a state
with a different unit pricing regime. The reference group is t = −1 so all effects
are relative to the year before the household moves. Figure 4 plots the annual
coefficients. Figure 4 shows that there are no significant pre-trends. Furthermore,
households decrease their bulk buying when they move from a state with unit-
price regulations to a state without unit-price regulations. On the other hand,
households that move from states without unit-price regulations to states with
unit pricing regulations do not significantly change their bulk buying.
Table 4 reports the results of estimating Equation (5). Columns (1) and (2) show
that a household’s bulk buying is about one percentage point higher when they are
in a state with unit price regulations, but this effect is only marginally significant.
This specification implicitly assumes that the effect of moving to a state with
unit price regulations will be the same as moving to a state without regulations
(i.e., the effect is symmetric). Column (3) treats the different directions of moving
differently and shows that moving to a state without unit price regulations
significantly decreases bulk buying by 1.4 percentage points while moving to a
state with regulations does not significantly change bulk buying.

21 Clustering
at the state level does not affect the estimates.
22 Projection
weights are not used because the weights are not designed for this subsample of movers.
23 Appendix Table A12 reports summary statistics for various groups of movers. Groups are relatively
similar, but movers are slightly richer, older, more educated, and have fewer children.
 INCOME DIFFERENCES IN BULK BUYING 17

Table 4—: Event Study of Movers to Different State Regulatory Regimes

All Movers No Law To Law Law To No Law

(1) (2) (3) (4)
Regulation 0.008 0.009∗ 0.003 0.014∗∗
(0.005) (0.005) (0.007) (0.006)
Avg Bulk 0.5 0.5 0.5 0.5
Household FE Y Y Y Y
Year FE Y Y Y Y
Demographics N Y Y Y
Observations 92,739 92,739 86,712 88,479
Adjusted R2 0.625 0.627 0.627 0.628
Note: Table shows estimates of Equation 5 which regresses household bulk buying on unit price regulation
after controlling for household fixed effects. “Regulation” denotes the estimated effect of moving from a
state without regulation to a state with regulation. Columns (3) and (4) only include one set of movers
that switch regimes in each specification with the remaining households that move, but do not switch
regimes. Column (3) excludes households that move to states without unit price regulations to restrict
identification of the regulatory effect to households moving to states with regulations. Column (4) excludes
households that move to states with unit price regulations to restrict identification of the regulatory effect
to households moving to states without regulations. Standard errors are clustered at the household level.
∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Source: Nielsen Consumer Panel (2004–2017)
18 INCOME DIFFERENCES IN BULK BUYING

Figure 4. : Event Study of Movers and Unit Price Regulations

Difference in Bulk Purchasing

0.02
(Percentage Points)

●
●

0.00 ● ●

● ●

−0.02

−0.04
−2 0 2
Years After Moving

Law Status After Household Move ● To Law To No Law

Note: Figure plots the β1t coefficients and 95% confidence intervals from Equation (6), which regresses
household bulk purchasing on dummies for years before and after a household moves to a state with
a different unit pricing regime than the state it moves from. The regression controls for household
characteristics as well as household and year fixed effects. Standard errors are clustered at the household
level. “To” reports estimates for households that move from a state without unit price regulations to a
state with unit price regulations. “Away” reports estimates for households that move from a state with
unit price regulations to a state without regulations.
Source: Nielsen Consumer Panel (2004–2017)

The asymmetric effect of unit pricing indicates the importance of both cognitive
costs and consumer education. For households that move to states without
regulation, the negative coefficient suggests that cognitive costs are discouraging
households from buying in bulk. For households that move to states with regulation,
they may not know how to best use the information provided and therefore
consumer education may help them recognize the value of quantity discounts and
buy in bulk more.
Unit pricing regulations are relatively simple to implement for both policymakers
and retailers. Retailers will bear some initial setup costs of redesigning their price
labels, but ongoing costs will likely be similar to current menu costs that firms
bear.24 Adopting unit pricing policies (like those recommended by the National
Conference on Weights and Measures) would encourage bulk buying while imposing

24 In 1975, the Government Accountability Office (then the General Accounting Office) estimated that
implementation and maintenance would cost about 0.1% of sales (General Accounting Office, 1975). This
was estimated before the adoption of bar codes and other efficiency-improving practices of the retail
sector. Implementing unit pricing now is likely to cost substantially less than those early estimates.
 INCOME DIFFERENCES IN BULK BUYING 19

few costs. These findings support the broader assertion that increasing price
transparency allows households to choose products that deliver more value.

B. Store Preferences

The second potential contributor to the bulk buying gap is store preferences;
low-income households may not live in areas where bulk sizes are available or may
not shop at stores that offer bulk sizes. In this subsection, I provide evidence that
the bulk buying gap persists within neighborhoods and within store types. Then,
I show that warehouse club entry increases bulk buying by 4.0–7.3%, but these
increases hold only for middle- and high-income households.

Inequality Within Markets and Retail Chains

If supply factors are the primary driver of the bulk buying gap, then the gap
should disappear when comparing households in the same neighborhood since
they have the same set of stores to choose from. I show that the bulk buying gap
still persists within zip codes. This remaining gap corresponds to the amount
that cannot be explained by differences in access, at least as approximated by
geography.
Even within zip codes, there may be other factors affecting where households
shop, such as whether or not a household has a vehicle, access to public transit,
or a warehouse club membership. To account for possible differences, I examine
how much of the bulk buying gap persists within chains. This exercise assumes
that within a chain, households have access to the same assortment of goods
(DellaVigna and Gentzkow, 2019). I also examine the bulk buying gap within
store types (i.e., “channel”) to account for the fact that bulk buying differences
may primarily be between channels (discount versus dollar) instead of between
retailers within a channel (Walmart versus Target).25
I estimate within-zip and within-chain bulk buying gaps using a modified form
of Equation 2:
X
(7) BulkShareimt = α + β q Incomeimt + γXimt + λmt + imt ,
q

where Incomeimt is an indicator for the income bin of household i in market m in

year t. Ximt consists of household characteristics. For the analysis of bulk buying
within zip code, BulkShareimt is the share of bulk purchases made by household
i in zip code m in year t and λmt is a zip-year fixed effect. For the analysis of bulk
buying within retail chains, BulkShareimt is the share of bulk purchases made by
household i in retail chain m in year t and λmt is a retail chain-year fixed effect
and/or a channel-year fixed effect.

25 Retailer names are only for expository purposes. Retailer identities are anonymized in the Nielsen
data.
20 INCOME DIFFERENCES IN BULK BUYING

Figure 5 plots the income coefficients with and without fixed effects for each
regression. Adding zip-year fixed effects reduces the gap between the highest
and lowest income groups by 9% (from 10.5 percentage points to 9.6 percentage
points). Results are virtually unchanged if I use county-year fixed effects instead
of zip-year fixed effects. Using channel-year fixed effects reduces the bulk buying
gap by a more substantial 66% (from 7.4 percentage points to 2.5 percentage
points). Adding retail chain-year fixed effects on top of channel-year fixed effects
does not significantly affect the bulk buying gap. This implies that a large share
of the bulk buying gap is related to the types of stores households shop at, but
not the specific chain they choose within a particular store type.

Figure 5. : Bulk Buying Within Zip Codes or Stores

8
Difference in Bulk Purchasing

Difference in Bulk Purchasing

●
9
6
(Percentage Points)

(Percentage Points)
●

●
6
● 4
●
●
●
●
● ● ●
3 ● ●
2 ● ● ● ● ●
●
●
● ● ●
●● ● ●
●
0 0

0 25 50 75 100 0 25 50 75 100
Household Income ($000s) Household Income ($000s)

Type ● With FE Without FE Type ● Channel + Retailer FE Channel FE Without FE

(a) Zip-Year Fixed Effects (b) Retailer/Channel-Year Fixed Effects

Note: Figure plots the income bin coefficients from Equation (7), which regresses the share of annual
purchases that were bulk packages on household characteristics as well as either zip code-year, retailer-year,
or channel-year fixed effects (a “channel” is a type of store). Nielsen projection weights are used to ensure
national representativeness. Households making $5–$8k are the reference group. Coefficient values are
reported in Table A9.
Source: Nielsen Consumer Panel (2004–2017)

Overall, within zip codes, the bulk buying gap between high- and low-income
households persists. However, within store type (or retail chain), the bulk buying
gap is substantially reduced. Two important conclusions can be drawn from these
patterns. First, in an accounting sense, the type of store a household shops at
accounts for two-thirds of the bulk buying gap. This is likely an overestimate of
the contribution of store preferences because those preferences may be driven by
more fundamental factors (e.g., high storage costs or budget constraints could
prevent households from shopping at warehouse clubs, as opposed to households
having a low preference for warehouse clubs). Second, the bulk buying gap still
persists within channels and retail chains. These patterns suggest that where a
household shops and what they choose within a store are much more important
than where a household is located. The next section explores how store preferences
are related to income and how warehouse clubs affect bulk buying.
 INCOME DIFFERENCES IN BULK BUYING 21

Store Preferences by Income

The previous section shows that while the bulk buying gap persists within
zip codes, it is narrower within store types and retail chains. In this section, I
show that the biggest shopping differences between income groups are related
to warehouse clubs. I then estimate the effect of warehouse club entry on bulk
buying.
To demonstrate differences in store preference by household income, I examine
the relationship between where households shop and their income using the
following regression:
X
(8) ChannelShareimt = α + β q Incomeimt + γXimt + λm + λt + imt ,
q

where ChannelShareimt is the share of annual spending that household i in market

m in year t made in a particular channel (grocery store, discount store, dollar
store, drug store, or warehouse club). Incomeimt is an indicator for a household’s
income bin. Ximt captures other household characteristics. Finally, market and
year fixed effects capture differences in spending shares across markets and over
time.
Figure 6 reveals that while there are small differences in the share of annual
expenditures at grocery, drug, and discount stores, there are dramatic differences
in whether households shop at warehouse clubs or dollar stores: households
making over $100,000 spend about 13 percentage points more of their non-food
expenditures at warehouse clubs than households making under $25,000.
Because the biggest differences are in warehouse clubs and these stores almost
solely stock bulk sizes, I focus on how warehouse clubs affect bulk buying. The
following analysis of warehouse clubs uses data on over 1,400 warehouse club
locations between 2004–2015.26 The first possibility is that high-income households
shop at warehouse clubs because they are closer. Table 5 shows that low-income
households are about 15 miles away from the nearest warehouse club compared to
only 8 miles away for high-income households.
The ideal experiment would randomly assign warehouse clubs to neighborhoods
and then their effect on bulk buying could easily be calculated. Even though store
locations are not randomly assigned, within a household, it is exceedingly unlikely
that a warehouse club opening could be co-incident with a shift in bulk buying,
so any observed changes are likely causal. Leveraging the panel structure of the
Nielsen Consumer Panel, I estimate how a household’s bulk purchasing changes
after a warehouse club opens using the following equation:

(9) BulkShareimt = α + βEntryimt + γXimt + λim + λt + imt ,

26 Data provided by the authors of Coibion, Gorodnichenko and Koustas (2017) and covers BJ’s,
Costco, and Sam’s Club.
22 INCOME DIFFERENCES IN BULK BUYING

Figure 6. : Annual Spending Shares By Store Type, Relative to Low-Income

Households
Diff. in Annual Budget Share

10
(Percentage Points)

0
●
●
●
●
●
●
−10 ●
● ● ●
● ●
● ●

0 25 50 75 100
Annual Household Income ($000s)

Type ● Dollar Store Grocery/Discount/Drug Store Warehouse Club

Note: Figure plots the income bin coefficients from Equation (8), which regresses the share of annual
purchases at each store type on household characteristics as well as year and market fixed effects. Nielsen
projection weights are used to ensure national representativeness. Households making $5–8k are the
reference group.
Source: Nielsen Consumer Panel (2004–2017)

Table 5—: Average Distance to Warehouse Club by Income (Miles)

Household Income Mean SD 25th Pctile 75th Pctile
<25k 14.79 18.98 3.26 20.53
25-50k 12.79 17.51 3.03 15.94
50-100k 10.61 15.16 2.85 12.01
>100k 7.92 12.08 2.51 8.45
Note: Table reports the distance between zip code centroids of warehouse locations and household
locations. Nielsen projection weights are used to ensure national representativeness.
Source: Nielsen Consumer Panel (2004–2015)

where BulkShareimt is the share of bulk purchases made by household i in market

m in quarter t. Entryimt is an indicator for whether or not a warehouse club
entered within 15 miles of household i in quarter t.27,28 I include a household-
27 In cases where a household is located near multiple warehouse clubs, I use the earliest entry date
since the first warehouse club would generate the largest supply shock.
28 According to the 2017 National Household Travel Survey, the average household traveled about seven
 INCOME DIFFERENCES IN BULK BUYING 23

market and year-quarter fixed effects λ to ensure that β is identified by within-

household changes in bulk buying before and after a warehouse club opens instead
of households that may move to areas closer to warehouse clubs. X controls for
possible demographic changes within the household.
This specification relies on the assumption that households would have continued
buying in bulk like other households that did not experience a warehouse club
entry. To provide evidence supporting this “common trends” assumption, I plot
an event study by estimating a modified version of Equation 9, but replace the
entry indicator with dummies for each quarter pre- and post-entry:
X
(10) BulkShareimt = α + β q Qtrimt + γXimt + λim + λt + imt ,
q

where Qtrimt is a dummy for each quarter prior to entry and after entry, with
the quarter immediately before entry (q = −1) as the reference group. Figure 7
plots the quarterly coefficients and shows that for most income groups there are
no significant pre-trends. For households in the lowest income quartile, there is
some evidence that those that experienced a warehouse club entry buy in bulk
more often than other low-income households that do not experience an entry.
After a warehouse club enters, there are significant increases in bulk buying for
middle- and high-income households and these effects are persistent up to eight
quarters after a warehouse club has opened.
Table 6 shows the regression results. Overall, households that experienced a
warehouse club entry increased their bulk purchasing by two percentage points.
However, when I interact household income with warehouse club entry, the increase
in bulk buying is due to changes for households making over $25,000 and is
increasing in income, with households making over $100,000 increasing their bulk
buying by 3.5 percentage points. Households in the lowest quartile do not have
any significant change in their bulk buying. One likely reason that low-income
households do not change their bulk buying is that even after a warehouse club
enters, households do not purchase a membership (fees range from $45–$120
depending on the chain and membership level). Other possible reasons are that
low-income households do not have access to transportation that can carry items
home, do not have the space to store the items, or even if they had a membership,
they still would not purchase extremely large sizes available at warehouse clubs
due to budget constraints.29
This analysis estimates the intent to treat effect since not all households shop
at the entrant warehouse club after it opens. As a result, this is a conservative

miles to buy goods, with low-income households traveling about one or two miles less than higher-income
households (Federal Highway Administration, 2017). Allowing for the possibility that households might
travel farther to shop at a warehouse club, I use a cutoff of 15 miles. Appendix Table A13 shows that
this pattern is robust to other cutoffs.
29 As an example, Philadelphia provides public transit access to a warehouse club. However, carrying
club-sized items on a bus is infeasible for more than two or three items. A personal vehicle would be
necessary.
24 INCOME DIFFERENCES IN BULK BUYING

Table 6—: Effect of Warehouse Club Entry on Bulk Buying

(1) (2) (3)

Post-Entry 0.020∗∗∗ 0.020∗∗∗
(0.004) (0.004)
Post-Entry : <25k 0.001
(0.008)
Post-Entry : 25-50k 0.019∗∗∗
(0.006)
Post-Entry : 50-100k 0.025∗∗∗
(0.005)
Post-Entry : >100k 0.035∗∗∗
(0.009)
Avg Bulk 0.48 0.48 0.48
Household-Zip FE’s Y Y Y
Year-Quarter FE’s Y Y Y
Demographic Controls N Y Y
Observations 2,401,038 2,401,038 2,401,038
Adjusted R2 0.428 0.428 0.428
Note: Table reports coefficient estimates for Equation (9) which regresses households’ quarterly bulk
purchase shares on an indicator for warehouse club entry, household characteristics as well as household-zip
code and year-quarter fixed effects. Projection weights are not used. ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Source: Nielsen Consumer Panel (2004–2015)
 INCOME DIFFERENCES IN BULK BUYING 25

Figure 7. : Event Study of Warehouse Club Entry

<25k 25−50k

●
0.05 ● ● ● ●
● ● ●
● ● ● ● ●
●
Change in Bulk Purchasing

● ● ● ●
● ● ● ● ●
● ● ●
● ●
0.00 ● ●
●
●
●
●
(Percentage Points)

−0.05

50−100k >100k

0.05 ● ●
● ● ● ● ●
● ● ● ●
● ● ●
● ● ●
●
● ● ●
0.00 ●
● ● ●
●
● ●
● ●
●
●
●
●

−0.05

−5 0 5 −5 0 5
Quarters After Entry
Note: Figure plots the quarterly coefficients from Equation (10)—the effects of warehouse club entry on
bulk purchasing of households before and after warehouse club entry—using 2004–2015 household-by-
quarter Nielsen Consumer Panel data. The regression controls for household characteristics as well as
household-zip code fixed effects. All coefficients are relative to bulk purchasing in the quarter before entry
(q = −1). Error bars denote 95% confidence intervals.

lower bound on the actual treatment effect on households that shop at warehouse
clubs.30 The effect is quite substantial even given how conservative it is.
To examine whether this change in bulk buying comes from households shopping
more at warehouse clubs, I estimate Equation 9 on different margins of bulk
buying. The increase in bulk buying after warehouse club entry could be coming
from three possible margins. First, households could increase their bulk buying
at non-warehouse club stores (intensive non-warehouse club margin). Second,
households could increase their bulk buying at warehouse club stores (intensive
warehouse club margin). Third, households could increase their bulk buying by
switching shopping to warehouse club stores (extensive margin). The observations
used in estimation are household-quarter bulk shares at warehouse club and
non-warehouse club stores.
Table 7 shows the regression results. Column (1) is estimated using quarterly
bulk buying shares at non-warehouse club stores as the dependent variable. After
warehouse club entry, there is no significant change in bulk buying at non-warehouse
club stores for any income group. Furthermore, the standard errors are quite
30 Even though low-income households do not change their bulk buying, other research suggests that
they may be worse off because existing retailers are more likely to increase prices for storable products as
a competitive response (Bauner and Wang, 2019).
26 INCOME DIFFERENCES IN BULK BUYING

Table 7—: Effect of Warehouse Club Entry on Bulk Buying Along Different
Margins

Non-Club Club Club Intensive

(1) (2) (3)
Post-Entry 0.0001 0.075∗∗∗ −0.005
(0.004) (0.007) (0.004)
Post-Entry : 25-50k −0.003 0.018∗∗∗ −0.003
(0.004) (0.006) (0.004)
Post-Entry : 50-100k −0.001 0.009∗∗∗ 0.0005
(0.002) (0.003) (0.003)
Post-Entry : >100k −0.001 0.012∗∗∗ −0.001
(0.003) (0.004) (0.003)
Avg Bulk 0.4 0.27 0.97
Household-ZIP FE’s Y Y Y
Year-Quarter FE’s Y Y Y
Demographic Controls Y Y Y
Observations 2,401,038 2,401,038 661,240
Adjusted R2 0.254 0.567 0.193
Note: Table reports coefficient estimates for Equation (9) which regresses households’ quarterly bulk
purchase shares on an indicator for warehouse club entry, household characteristics as well as household-zip
code and year-quarter fixed effects. Column (1) uses bulk buying shares only at non-warehouse club
stores as the dependent variable. Column (2) uses bulk buying shares at warehouse club stores (inclusive
of zero spending quarters) as the dependent variable. Column (3) uses bulk buying shares at warehouse
club stores (excluding zero spending quarters) as the dependent variable. Projection weights are not used.
∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Source: Nielsen Consumer Panel (2004–2015)
 INCOME DIFFERENCES IN BULK BUYING 27

small, so if there were changes, they are minimal. Column (2) is estimated using
quarterly bulk buying shares at warehouse club stores as the dependent variable,
which includes many zeros because there are quarters where households do not
shop at warehouse club stores. For households that never shop at warehouse club
stores, they have a string of zeroes. After warehouse club entry, quarterly bulk
buying at warehouse clubs increases by a statistically significant 7.5 percentage
points and is higher for higher income groups. This increase could be generated by
households shopping more often at warehouse clubs, and therefore increasing the
number of quarters with non-zero bulk spending at warehouse clubs. It could also
be generated by households increasing the share of bulk purchases they already
make at warehouse clubs. Column (3) focuses on only quarters with positive
bulk shares at warehouse clubs and shows that there are no significant changes
in bulk buying, conditional on shopping at a warehouse club, after a warehouse
club enters. Therefore, the significant increase in Column (2) is generated by
households switching to shopping more at warehouse clubs after a warehouse
club enters. Overall, warehouse club entry increases bulk buying by encouraging
households to make more shopping trips to warehouse clubs.

C. Budget Constraints

Budget constraints may be another contributing factor to the bulk buying gap.
Importantly, the necessary budget constraints must bind over short time periods
(e.g., months) because over the course of a year, households spend more than they
would have for the same amount of goods if they had taken advantage of bulk
discounts. Such short-horizon budget constraints are most binding for the lowest
income groups, but are unlikely to bind for middle- and high-income groups.31
To test this explanation, I examine over-the-month changes in liquidity. Low-
income households are more likely to have higher liquidity at the start of the
month compared to the end of the month (Stephens Jr, 2003; Orhun and Palazzolo,
2019). Consistent with this fact, the spending of low-income households tends to
decline over the course of the month while the spending of higher income groups
is relatively flat (Orhun and Palazzolo, 2019).
I use within-household variation in the timing of purchases to estimate a
differences-in-differences model. This model tests the coincidence of changes
to bulk buying with times of the month when households may be liquidity con-
strained. This approach is similar to Orhun and Palazzolo (2019). I use weekly
bulk buying information to estimate the following regression:
(11)
4
β1qw 1{week = w} ∗ Inciw + β2 Inciw + γXiw + λi + iw ,
XX
BulkShareiw = α +
q w=2

31 Middle- and high-income households may still face monthly budget constraints, but grocery spending
is unlikely to be a major factor for all but the lowest-income households.
28 INCOME DIFFERENCES IN BULK BUYING

where BulkShareiw is the share of bulk purchases made by household i in week

w. 1{week = w} is an indicator for the second, third, or last week of the month.
Inciw indicates the household’s income bin q. Xiw is the usual set of household
characteristics and λ is a household fixed effect. The results are reported in Table
8.
Overall, households making less than $25,000 decrease their bulk buying by
a slight 0.2 to 0.4 percentage points during the third and fourth weeks of the
month while households making over $25,000 have no decline (and possibly a slight
increase) in their bulk buying during the third and fourth weeks of the month.
This analysis does not rule out the possibility of liquidity constraints contributing
to differences in bulk buying, but it does rule out that local changes in liquidity
significantly affect bulk buying. Liquidity shocks larger than intra-month paycheck
variation may be necessary to increase bulk buying, but more work would be
necessary to determine whether that is the case.

D. Storage Costs

Storage costs are the fourth contributing factor that I examine. Intuitively,
households that buy in bulk need a place to store large packages, which could be
in a basement, pantry, or cabinets. Households without available storage space
may want to save money through quantity discounts, but choose not to because
they have limited storage space.
The ideal experiment would randomly assign households to various home sizes
and then observe their bulk purchasing behavior to identify storage costs. However,
exogenously changing a household’s living situation is infeasible. The next best
option is to test some intuitive implications of storage costs. First, while I cannot
randomly assign households to different home sizes, there are many households
that move while they are in the Nielsen panel. I observe whether households
live in single-family homes or apartments, which generates variation in available
storage space. According to the American Housing Survey, the median single-
family home is about twice as large as the median apartment. Since at least
1999, new single-family homes have had a median size of 2,000–2,400 square feet
while the median apartment is only 1,000–1,100 square feet and this holds true
within Census regions as well. Therefore, households that move into single-family
homes are likely to have more available storage space and this will increase their
willingness to buy in bulk.
To test this hypothesis, I estimate how bulk buying changes when households
change their housing size, by estimating Equation 12:

(12) BulkShareit = α + βHouseit + γXit + λi + λt + it ,

where Houseit is a dummy for whether a household i lives in a single-family house

in year t (apartments are the reference group). Xit controls for changes in other
household characteristics. Household and year fixed effects, λ, ensure that β is
 INCOME DIFFERENCES IN BULK BUYING 29

Table 8—: Over-the-Month Changes in Bulk Buying

(1) (2) (3)

Week 2 −0.002∗∗∗ −0.001 −0.001
(0.0004) (0.001) (0.001)
Week 3 −0.002∗∗∗ −0.004∗∗∗ −0.004∗∗∗
(0.0004) (0.001) (0.001)
Week 4 0.002∗∗∗ −0.002∗∗ −0.002∗∗
(0.0004) (0.001) (0.001)
25-50k 0.003∗∗∗ 0.002∗∗
(0.001) (0.001)
Week 2 : 25-50k −0.002∗∗ −0.002∗∗
(0.001) (0.001)
Week 3 : 25-50k 0.001 0.001
(0.001) (0.001)
Week 4 : 25-50k 0.003∗∗∗ 0.003∗∗∗
(0.001) (0.001)
50-100k 0.009∗∗∗ 0.004∗∗∗
(0.001) (0.001)
Week 2 : 50-100k −0.001 −0.001
(0.001) (0.001)
Week 3 : 50-100k 0.002∗ 0.002∗
(0.001) (0.001)
Week 4 : 50-100k 0.005∗∗∗ 0.005∗∗∗
(0.001) (0.001)
>100k 0.025∗∗∗ 0.012∗∗∗
(0.001) (0.001)
Week 2 : >100k 0.001 0.001
(0.001) (0.001)
Week 3 : >100k 0.006∗∗∗ 0.006∗∗∗
(0.001) (0.001)
Week 4 : >100k 0.007∗∗∗ 0.007∗∗∗
(0.001) (0.001)
Mean Bulk 0.46 0.46 0.46
Household FE’s Y Y Y
Demographics N N Y
Observations 2,854,905 2,854,905 2,854,905
Adjusted R2 0.402 0.402 0.404
Note: Table displays the regression coefficients from estimating Equation 11 which regresses a household’s
weekly share of bulk purchases of non-food products on the week of the month, income, and other
household characteristics and includes a households fixed effect. ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Source: Nielsen Consumer Panel (2004–2017)
30 INCOME DIFFERENCES IN BULK BUYING

identified off of within-household changes in housing. Standard errors are clustered

at the household level. Table 9 shows that bulk buying is one percentage point
higher when households are in single-family homes compared to when they are in
apartments.

Table 9—: Relationship Between Bulk Buying and Housing Changes

(1) (2) (3)

Single-Family Home 0.012∗∗∗ 0.009∗∗∗ 0.009∗∗∗
(0.002) (0.002) (0.002)
Avg Bulk 0.5 0.5 0.5
Market FE’s Y Y Y
Year FE’s Y Y Y
Demographics N Y Y
Future Income N N Y
Observations 731,762 731,762 566,535
Adjusted R2 0.688 0.688 0.691
Note: Table shows the results of estimating Equation 12. The dependent variable is the annual share
of bulk purchases made by households and the independent variables are housing and other household
characteristics. Estimation includes household, market, and year fixed effects. “Future Income” denotes a
household’s income one year in the future. Standard errors are clustered by household. ∗ p<0.1; ∗∗ p<0.05;
∗∗∗ p<0.01
Source: Nielsen Consumer Panel (2004–2017)

Column (1) shows that bulk buying is 1.2 percentage points higher when a
household lives in a single-family home. However, since housing changes can be
due to other within-household shifts, such as marriage or having children, column
(2) also controls for other within-household demographic changes. The increase
in bulk buying is slightly reduced, but there is still an increase when households
move into larger spaces. Finally, households may move into larger housing if they
expect to earn more and this expectation of future income may also increase their
bulk buying. Column (3) also includes a household’s one-year-ahead income and
there is no change to the bulk buying increase after a household moves into a
single-family home. This result is not causal, but it supports the intuition that
when households have more storage space, they are more able to buy in bulk.
Another implication of storage costs is that products with a smaller “footprint”
(physical volume) have lower storage costs. Therefore, if storage costs influence
bulk buying, there should be a smaller gap in bulk buying for smaller products
(like plastic wrap) relative to large, cumbersome products (like paper towels and
toilet paper). To test this implication, I estimate a modified form of Equation 3
 INCOME DIFFERENCES IN BULK BUYING 31

that relates average package sizes with household income:

X
ln(AvgSize)imt = α + β q Incomeimt + γXimt + λm + λt + imt ,
q

where AvgSizeimt is the average package size purchased by household i in market

m in year t. Incomeimt consists of dummies for each income quantile q. Ximt
consists of household characteristics. Year and market fixed effects are captured
by λ.
Figure 8 plots the income coefficients from the regression for all non-food grocery
categories. I have highlighted some popular product categories. The bulk buying
gap is largest for the physically biggest products such as paper towels and toilet
paper while the gap is smaller for less bulky items such as liquid detergent and
diapers. Overall, this pattern supports the hypothesis that storage costs contribute
to the bulk buying gap, but the persistence of the gap even for smaller products
suggests that other factors are at play.

Figure 8. : Bulk Buying Gap For Non-Food Grocery Products

Paper Towels
Log−Point Increase Over

0.50 Toilet Paper

Poorest Households

Detergent

Diapers
0.25

0.00

Product Category

Household Income 25−50k 50−100k >100k

Note: Figure plots the β coefficients from Equation (3), which regresses average package size purchased
on household income. The regression controls for household characteristics as well as market and year
fixed effects.
Source: Nielsen Consumer Panel (2004–2017)

Overall, these results provide evidence that storage costs and bulk buying are
related. When households move to larger homes (relative to apartments), they buy
32 INCOME DIFFERENCES IN BULK BUYING

more in bulk. Similarly, product categories with larger physical footprints exhibit
larger bulk buying gaps relative to product categories with smaller footprints. To
more precisely quantify storage costs, I estimate a simple model of the consumer
purchase decision.

V. Model

The previous analyses show that cognitive costs and storage costs affect the
bulk buying decision. To decompose the contribution of each factor, I embed
them into a discrete choice model of the household’s purchase decision. The ideal
setting would include a homogeneous good where demand is uncorrelated with
income. Given substantial price, package size, and regulatory variation, differences
in large and small purchases between households would identify storage costs and
differences in buying between regulatory regimes would identify cognitive costs.
This setting is approximated by one where products have limited dimensions of
differentiation and storage costs can be separately identified from demand.
A discrete choice model of toilet paper purchases closely approximates this
ideal setting. Toilet paper is an excellent product for this analysis because it is a
necessity item with easily observable dimensions of differentiation, namely price,
quality, quantity, and package size. It is offered in a wide range of package sizes
and stores stock numerous brands and sizes (grocery and mass merchandise stores
usually stock 35–40 unique brand-sizes). The top five brands and private-label
store brands account for 86% of sales. I focus on the most common package sizes,
which range from 4- to 24-roll packages. I define a product as a unique brand-size
combination.32 Additionally, underlying toilet paper consumption is primarily a
function of household composition and age, not income.33 High-income households
consume a similar amount as low-income households but make fewer purchases
(Orhun and Palazzolo, 2019). Finally, toilet paper cannot be easily substituted for
another product nor can it be obtained through home production.34
The biggest identification challenge is separately identifying storage costs from
underlying demand (i.e., households may buy large quantities because they have
high consumption or because they have low storage costs). To separate storage costs
from demand, I use variation induced by differences in product “concentration,”
which I define as the yield of the product per unit volume. Product concentration
breaks the direct link between volume and consumption. In the detergent category,
a product’s yield is the number of washes it will supply. A concentrated detergent
can wash the same number of loads but requires a smaller fluid volume than

32 Specifically, this is a unique brand-roll count-sheet count because packages can differ in their
“concentration” due to “double,” “mega,” and “super mega” rolls.
33 A 100-fold cross-validated elastic net regression of annual purchases on household characteristics
rules out income as significantly predictive. See Appendix A.A6 for details.
34 While a bidet is a possible alternative, this is more likely a lifestyle choice instead of a situation
where households switch between toilet paper and bidets. Furthermore, in the United States, 98% of
households report that they use toilet paper (the remainder either said no or did not respond) (Statista,
2019).
 INCOME DIFFERENCES IN BULK BUYING 33

diluted detergent. Therefore, given the same number of washes, households that
choose concentrated detergent must have higher storage costs than those choosing
diluted detergent, assuming quality does not differ based on concentration.
The same reasoning holds true for toilet paper. Households do not demand a
particular number of rolls (the primary determinant of package size), but choose
how long they want their supply to last (i.e., purchase enough to last for two weeks,
a month, two months, etc.).35 Toilet paper comes in a variety of concentrations with
“mega” rolls being four times more concentrated than “regular” rolls. Therefore, a
household that purchases 24 “regular” rolls has the same demand for toilet paper
as a household that purchases six “mega” rolls, but the former household has
lower storage costs since they can store the bigger package.
To illustrate the varying concentrations of toilet paper, Figure 9 plots the
distribution of quantity (measured in number of days the supply will last for a
single person) against package sizes (measured in rolls) for toilet paper products in
the Nielsen data. As expected, there is an increasing relationship between how long
the package will last and the number of rolls in a package, but there is substantial
variation within packages containing the same number of rolls. The dashed lines
denote the 25th and 75th percentiles of the average days’ supply purchased by
households. A wide range of package sizes fall within this range for each brand.36
For example, a household demanding a 60-day supply of Charmin could purchase
a package containing anywhere from 8 to 24 rolls. This overlap generates the
necessary variation to separate storage costs from underlying demand.

A. Model Setup

I model a household’s purchase decision using a discrete choice framework with

random coefficients. When making a purchase, households consider the price, unit
price, quality, quantity, and size of each package and choose the package that
maximizes their utility. These features are captured in the household i’s indirect
utility function:

(13) Uijt =β1 P ricejt + β2i U nitP ricejt + β3 U nitP ricejt × Regi +
β4i log(Daysj ) + β5i BigP ackj + β6 BigP ackj × Housei +
β7i SmallP ackj + β8 SmallP ackj × Housei + θb(j) + ijt ,

where P ricejt is the total price of product j at time t. Regi is an indicator for
whether unit price regulations are in effect for household i. Daysj is the number
of days the package will last (a function of the number of total sheets in the
package and the number of people in the household). U nitP ricejt is the per-day,
per-person price of the package, since the yield of a package is how many days
35 According to a 2007 Charmin survey, the average person uses 57 sheets per day. I assume this
consumption rate when computing how long a product will last (Jaffe, 2007).
36 Scott toilet paper is an exception because it does not offer different roll types. All rolls have 1000
sheets.
34 INCOME DIFFERENCES IN BULK BUYING

Figure 9. : Scatterplot of Toilet Paper Package Size and Quantity

ANGEL SOFT CHARMIN CTL BR

25 ●●
●●
● ●●
● ●● ●● ●
●
● ● ●
● ●● ●●
● ●
● ●●● ● ● ● ● ● ●●
●
●
●
●●
●
●●
●●●●
●●● ●●
●●
●●●● ●●
●●
●
●●
●●●
●●● ●●●
●●●●
● ● ● ●●
●
● ●●
●
●●● ●
● ● ● ● ● ● ●● ● ● ●
●
● ●●
● ●
●
●●
●●
●
●
●●
●
●
●
●●●
●
●●
●●
●
●●●
●
●
●
● ●●
● ●●
●
● ●
●
●
●●●●
●●
● ● ●● ● ●● ●●●●●●●
●
●●●● ●●
●
●● ●●
● ●●
●●● ●●
●●●
●
● ●●
●● ●
●●●● ● ●● ●● ● ● ● ● ● ●● ● ● ●
●
●
●
●

● ● ●●● ●●
20 ●● ● ● ● ●
● ● ● ● ● ● ●● ●
● ●●●
●●●
●
●
● ●
●●●●
● ●
●● ●●
● ●●
●●● ● ● ●● ● ● ●
● ● ●●●
●●
●●
●
● ● ●
●●● ●●
● ●
●
● ● ●
●
●
●
●
● ● ● ● ● ●● ● ● ●
●

● ● ●
●● ● ● ● ● ● ●
● ● ● ● ● ●●●
●●
●●
●●
● ● ●●●
●●●
●● ●
●● ● ●●
●●● ● ● ● ● ●
●
● ● ● ●
●● ●
●
● ●●●
●
●
● ●
●●
●●
●● ● ●
●●●
● ● ● ●●
●● ● ●● ● ● ●● ● ●● ●● ● ● ● ●

●●
●●
● ●●●
● ● ●●
●●
●●●
●
●●
●
●●●●●
●● ●
●●●●●
● ●
● ●
● ●●
● ●●
●●● ● ●● ● ● ● ●
●● ● ●
●
●
●
●
●●●
●●●● ●● ●●
●● ● ●
● ●●
● ●
● ● ●●
● ●
● ● ● ● ● ● ● ● ●

15 ●
●●● ● ● ●

● ● ●● ●●
● ●● ●●●
●● ● ● ● ●● ● ●
●●●●
●●
●●
●●
●
●●
●
●●●
● ●●
● ● ● ●●
●
●
●●
●●
●
●
●●
●
●
●●
●●●
● ●
●●●
●● ●●
● ●
●
●
●●
●
●●
●
●
●
●●
●
●
●
●
● ●
●●
●
●●
●
●●●
●
●
●●
●
●
●
●
●●
●
●
●●
●
●
●●
●
●
●
●●
●●●
●● ●●
●●● ●●
●●● ● ●
●● ● ●
● ●●●● ● ●● ●●
●●
●● ●● ●
● ●● ● ● ●●●
●●●●
●●●●
●●●●● ●● ●
● ●●
●
●●● ●● ● ● ●
●●
●
● ●●
●●●
●●●
●
●●●
●●
●●
●
● ●●●●●●
●●●● ●●●●
●
●●●
●●
●
●● ● ●
● ● ●● ● ● ●
● ●● ● ●

10 ●●●
●●●●
●●●●●
● ●
●●
●●
● ● ● ●
● ● ●● ●● ● ●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●●
●
● ●
●
●
●
●●
●●
●●
●
●
●●
●●
● ●●●
●●●
●●
●●

●● ●● ●●● ●
●
●●
●●

● ●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●●
●
●
●●●
●
●
●●
●●
●●●
●
●
●●●●
●●●
●
●●●●
●●●
●
●●● ●● ●●
●
● ●●●
● ● ●● ● ●

● ● ● ● ● ●
●●●
●●
●●
●●
●●
●
●●
●
●●●
● ●
●●●●●● ● ● ● ●●
● ●●
● ●
●●●●
● ●
●
●●
●●●● ●●●
●●
● ●
● ● ●● ● ●●● ●●
● ● ● ●
● ● ● ●
●
●

●●●
●
● ●●● ●
●
●
●●
●●
●
●
●●
●
●
●●
●
●
●
●●
●
●●
●
● ●●●
●
●
●
●●
●
●
●
●●
●
●
●
●●
●●
●
●
●
● ●
●
●●●●
●●
●
●●● ● ● ● ●
●
●●●
● ●
●●
●
●●
●●
●●●
●●●●●●●
●●●
●●●
●●●
● ●●
●●●
●●●
●
● ●
●●
●●●
●
●
● ●●
●●●●
●●●
●●●
●
● ●
●●● ● ● ●

5 ●
●
●
●
●
●●
●●
●
●●
● ●●
●
●
●●●● ●●●● ● ● ● ●
●
●
●
●
●
●●
●
●●
●
●●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●●
●●●●●
●●●
●●●
●●
●●● ●
●
●
●
●
●
●●
●
●
●
●●
●●
●
●
●
●●
●
●
●●
●●●
●
●
●
●
●●
●●
●
●●
●
●
●
●
●
●
●●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●●
●●
●●
●
●
●
●
●
●
●●
●●
●●
●●●
●●
● ●
●●
●●
●●
●●●
●
●●
●
●
●●
●●
●●●
●●●● ● ● ●● ●

●
●●
●
●●

● ● ●
●
●
●
●●
●
●● ●
●
●
●
●
●
●●
●●
●
●
●
●
●
●●
●●
●
●●
●●●● ●

0
Rolls

KLEENEX COTTONELLE QUILTED NORTHERN SCOTT 1000

25 ●●
●●●●
●
● ●
●● ●
●●●
●●
● ●●●
●●
●
●●
● ●●●
●
● ● ●● ● ●●
● ●●● ●●
● ● ● ●
●●
●●
● ●●●
●
●
●●●●●
●●●
●●
● ●● ●
●● ●●
●
●
● ● ●●
●● ● ● ●● ● ●●
● ● ● ● ●● ●
●
●

●● ●● ●● ●
20 ●●●●
●●
●●●●
●
● ●
●●●
●
●● ● ●
●●●
● ●●
●●● ●
● ● ●
●
●●●●●
●
● ● ● ●
●
●
● ●

●
●
●●
●●
●
●●
●●
●
●
●●
●●●●●●
●
●●
●●
●
●●●●
●
●
●●●●●
●●
●●● ● ● ● ●
●●● ● ●
●●
●●●●
●● ●●●
●● ●
● ● ●●
●
●
● ●
● ●●● ● ● ● ● ● ●●
●●
● ●
●
● ●

●
●● ● ●● ●● ●
●● ● ●● ● ● ● ●
●● ● ●
●● ●
●● ●
●●● ● ● ● ●

15 ● ●
● ●

●
●
●
●
●
●●
●
●
●●
●
●
●●
●
●●
●
●●
●
●●●
●
●●●
●●●
●
●● ●
●
●●●
●
●
●●●
●●
●
●●
●
●
●●
● ●●
●●● ●●
● ● ●●
●
● ●● ●
●●
●●●●●
●
●●
●●
●●●●
●
●●●
●●●● ●●● ● ● ● ● ● ●● ●
●●●● ●
●
●●●
●
●●● ●●●● ●●●●●
● ● ● ●● ● ● ● ●

10 ●●●● ●
●●
●
●● ●●
● ●●● ●● ●●
●●
●●● ●●
●
●
● ●
●

●
●●
●
●
●
●
●●●
●
●●
●
●
●
●●●
● ●●
●●● ●●
●
●●●
●●●
●● ●
●●
●●●
●●
●●●●
●● ●
● ● ●●
● ● ●
●●
● ● ●

● ●●●●
● ●
● ● ●
● ● ● ●
●●●● ●● ● ●● ● ●● ● ● ● ●●
●
● ●

●
●
●●
●●
●
●
●●
●●
●
● ●
●
●● ●
●
●●
●
●●
●
●●
●●
●●
●
●●●● ●
● ●
●
●
●●
●●●●
●●●●●
●
●● ●●●
●●
●●● ●●●● ●●
●●●●
●● ●● ●
●●
●●
● ●● ●

5 ●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●●
●●
●●
●
●
●●
●
●●
●●
●●
●●●
●
●● ●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●●●
●
●
●
●
●
●●
●●
●
●
●●
●●●●●
●●●● ● ●● ●
●
● ●

●
●
●●

●
● ● ●
●
●
●
●●●
●
●●
●
●
●
●

0
0 50 100 150 200 0 50 100 150 200 0 50 100 150 200
Days' Supply
Note: Figure plots the package sizes and quantities of the top five toilet paper brands and private-label
products. The y-axis represents the number of toilet paper rolls contained in a package while the x-axis
represents the number of days a product will last a single person household assuming a consumption rate
of 57 two-ply sheets per day (Jaffe 2007). Noise is added vertically to better illustrate the number of
products available within package sizes since roll counts are discrete. Dashed lines indicate the 25th and
75th percentiles of the average days’ supply purchased by households.
Source: Nielsen Consumer Panel (2004–2017)

it will last. BigP ackj is a dummy for the package having more than 12 rolls
and SmallP ackj is a dummy for less than 12 rolls.37 Housei is an indicator for
whether the household lives in a single-family home, with the alternative being
an apartment. Finally, θb(j) is a brand fixed effect. Brand fixed effects capture
quality differences between products. Because households may weigh unit prices
or package sizes differently based on unobserved factors, I allow β2 , β4 , β5 , β7 to
vary. In particular, I assume they are normally distributed and allow for them to
be correlated. I assume ijt is iid Type 1 extreme value.
This simple model incorporates the key features necessary to quantify the
contribution of cognitive and storage costs to the bulk-buying gap. Preferences
for package size (a measure of storage costs) are captured by β5 , β6 , β7 , and β8 ,
while the effect of displaying per-unit prices is captured by β3 .
The price coefficient is identified using price variation across shopping trips
due to shopping at different stores or sales. The size coefficient is identified by

37 Households bunch at 12-roll packages, so this allows for different package preferences around this
bunching point.
 INCOME DIFFERENCES IN BULK BUYING 35

variation in the product “concentration” as illustrated in Figure 9. That is, given

their preferred days’ supply (x-value), some households choose large packages
and some choose small packages (y-value). I estimate the model using simulated
maximum likelihood.38 To increase accuracy and reduce computational burden,
I use pseudo-random Halton draws in the estimation procedure (Hensher and
Greene, 2003).

VI. Estimation and Counterfactual Results

I estimate this model separately for each income quartile using household
purchases from 2016. I observe about 45,500 toilet paper purchases across about
14,800 households at grocery stores and mass merchandisers.
Table 10 reports model estimates for the random coefficients specification. The
estimation results show that both the price and unit price coefficients are negative,
implying that all else equal, households prefer lower prices. The interaction terms
reveal that when unit prices are posted, all households are more sensitive to
unit prices. This pattern supports the assertion that households respond to the
provision of new price information. All households prefer to have more days’
supply of toilet paper compared to less. In terms of storage costs, all households
select against large sizes and, with the exception of the highest-income households,
this preference is not significantly different based on their housing type. On the
other hand, all households also dislike small packages. Under a pure storage costs
story, the small packages would have been expected to have a positive sign for
low-income households. However, as mentioned in the model specification section,
there is bunching at 12-roll packages across households of all types, so this negative
sign on the small size is likely a result of that bunching.
Each of the random coefficients displays substantial heterogeneity.39 Overall,
lower income households exhibit more heterogeneity in their sensitivity to unit
prices, as implied by the standard deviations of the unit price coefficient.
Figure 10 plots the distribution of own-price elasticities for each product using
the random coefficients estimates. The majority of elasticities fall between -1.7
and -5.1 with poorer households having larger elasticities (in magnitude).40

A. Model Fit

I examine model fit by comparing how well the model predicts the average
size purchased for each income group. Since coefficients are random, the choice

38 I use the ‘mlogit‘ package which implements Ken Train’s Matlab code in R (Revelt and Train, 1998;
Croissant, 2020).
39 Since each random coefficient was assumed to be normally distributed, some households may have a
non-intuitive valuation for product attributes, such as a positive valuation for unit price. Sign restrictions
can be imposed by assuming alternative distributions, such as a log-normal distribution.
40 Table 4 of Cohen (2008) reports elasticities ranging from -1.94 to -2.54 for paper towels. My estimates
cover this range, but have larger tails.
36 INCOME DIFFERENCES IN BULK BUYING

Table 10—: Random Coefficient Estimation Results (2016)

<25k 25-50k 50-100k >100k

(1) (2) (3) (4)
Total Price −0.300∗∗∗ −0.287∗∗∗ −0.284∗∗∗ −0.196∗∗∗
(0.011) (0.007) (0.006) (0.009)
Unit Price −4.114∗∗∗ −2.642∗∗∗ −2.969∗∗∗ −3.243∗∗∗
(0.290) (0.124) (0.105) (0.155)
. : Reg −1.914∗∗∗ −1.759∗∗∗ −0.819∗∗∗ −0.503∗∗∗
(0.330) (0.161) (0.123) (0.149)
Log(Days) 1.146∗∗∗ 1.381∗∗∗ 1.452∗∗∗ 0.843∗∗∗
(0.078) (0.050) (0.047) (0.071)
Large Size −1.654∗∗∗ −1.194∗∗∗ −1.288∗∗∗ −1.104∗∗∗
(0.174) (0.096) (0.098) (0.181)
. : Home 0.273∗ 0.101 −0.016 0.422∗∗
(0.147) (0.097) (0.093) (0.179)
Small Size −0.437∗∗∗ −0.345∗∗∗ −0.504∗∗∗ −0.807∗∗∗
(0.083) (0.062) (0.066) (0.139)
. : Home −0.173∗ −0.241∗∗∗ −0.221∗∗∗ −0.202
(0.095) (0.067) (0.069) (0.140)
σunitP rice 6.009∗∗∗ 4.973∗∗∗ 4.791∗∗∗ 3.849∗∗∗
(0.241) (0.116) (0.092) (0.113)
σlog(Days) 1.620∗∗∗ 1.499∗∗∗ 1.812∗∗∗ 1.651∗∗∗
(0.062) (0.039) (0.035) (0.056)
σLarge 1.847∗∗∗ 1.327∗∗∗ 1.647∗∗∗ 1.846∗∗∗
(0.142) (0.073) (0.065) (0.085)
σSmall 2.111∗∗∗ 2.181∗∗∗ 2.011∗∗∗ 2.353∗∗∗
(0.083) (0.048) (0.042) (0.067)
Brand FE’s Y Y Y Y
Observations 4,968 12,950 17,875 7,942
Log Likelihood −14,502.240 −37,679.860 −51,245.220 −22,705.370
Note: Table presents MLE estimates from Equation 13. “Total Price” denotes the total price of the
package while “unit price” is the price per day that the package will last (assuming constant consumption
of 57 sheets per day (Jaffe 2007)). “Reg” indicates whether unit price regulations are in effect. “Large”
indicates packages that are larger than 12 rolls and “small” indicates packages that are smaller than 12
rolls. A 12-roll package is the reference group. “House” indicates if the household lives in a single-family
home (reference group is an apartment). ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Source: Nielsen Consumer Panel (2016) and Nielsen Retail Scanner (2016)
 INCOME DIFFERENCES IN BULK BUYING 37

Figure 10. : Distribution of Price Elasticity by Household Income

0.6
0.4

<25k
0.2
0.0
0.6

25−50k
0.4
0.2
Density

0.0
0.6

50−100k
0.4
0.2
0.0
0.6

>100k
0.4
0.2
0.0
−6 −4 −2 0
Elasticity

Note: Figure plots the distribution of price elasticities resulting from the estimation of Equation 13, using
random coefficients.
Source: Nielsen Consumer Panel (2016) and Nielsen Retail Scanner (2016)

probabilities take the following form:

0
eβ xijt
Z
(14) Pijt = P β 0 x f (β)dβ,
je
ijt

I use simulation to approximate the integral by taking 1,000 draws from the
joint distribution of β. Table 11 compares the overall model predictions to the
actual data. Overall, the fit is close, but the model over-predicts the amount
purchased across all households, primarily because it over-predicts the purchases
of particularly large generic packages. For example, a particular generic 12-pack
has a 1–2% share for each income group, but based on its characteristics, the
model predicts a 3–5% share. The model assumes that all generic brands are equal,
but in reality, it may be the case that generic brands differ based on the retailer
that sells them. This additional dimension of heterogeneity could be captured by
more granularly defining brands by the retailer that sells them.

B. Counterfactuals

Using the parameter estimates from the previous section, I predict how house-
holds respond to lower storage costs and universal unit price regulation. For these
38 INCOME DIFFERENCES IN BULK BUYING

Table 11—: Random Coefficient Model Fit (Days’ Supply Purchased)

Income Data Model

<25k 48.64 50.67
25-50k 49.07 52.45
50-100k 51.23 54.48
>100k 53.85 54.93
Note: Table compares the average days’ supply of toilet paper purchased in the data with the predicted
purchase from the model. I assume an average daily consumption rate of 57 two-ply sheets per day (Jaffe
2007).
Source: Nielsen Consumer Panel (2016) and Nielsen Retail Scanner (2016)

counterfactual exercises, I compare all counterfactual results to a “base case” of

predicted purchases given their current shopping environment. I consider two
counterfactual scenarios:

1) Unit-Price Regulation: Unit-price regulations are adopted everywhere.

2) Reduced Storage Costs: All households have the same storage costs (i.e.,
size preferences) as high-income households.

For the unit-price regulation scenario, I set each household’s unit price coefficients
equal to the sum of its coefficient and the regulation interaction term. For
households making under $25,000, their unit price coefficient becomes −4.114 −
1.914 = −6.028. For the reduced storage cost scenario, I set all size coefficients
equal to the coefficients for households making over $100,000.
Table 12 reports the counterfactual predictions for the random coefficients
model.

Table 12—: Bulk Purchasing Counterfactual Simulation Results

Income Base + Unit Price Regs + Rich Storage

<25k 50.67 52.67 56.86
25-50k 52.45 55.23 58.11
50-100k 54.48 55.91 58.47
>100k 54.93 55.83 55.83

Note: Table reports predicted package quantities purchased by households using model estimates of
Equation (13). Units are number of days the chosen package will last assuming average daily consumption
rate of 57 two-ply sheets (Jaffe 2007). The “Unit Price Regs” scenario imposes unit price regulations
everywhere. The “Rich Storage” scenario imposes that all households have the same preferences for “large”
and “small” packages as households making over $100k. Scenarios are cumulative.
 INCOME DIFFERENCES IN BULK BUYING 39

The random coefficient counterfactuals, while overpredicting the average days’

supply purchased, predicts a gap of 4.26 days’ supply between high- and low-
income households. After universally adopting unit price regulations, all households
increase their purchasing, but the gap between high- and low-income households
shrinks to 3.16 days’ supply. Equalizing storage costs actually reverses the gap
with the lowest income households buying 1.03 days’ supply more than high-
income households. This dramatic increase is driven by the fact that middle- and
low-income households continue to be highly price-sensitive and have a stronger
preference for large quantities.
These counterfactuals support the main findings from Section IV which showed
that unit price regulations increase bulk buying and that storage costs are a
substantial factor preventing households from buying in bulk. Since the earlier
sections examined bulk purchasing across all non-food products, I repeat the
earlier analysis on mover households specifically for toilet paper purchases. I
estimate a modified version of Equation 5 which replaces share of bulk purchases
with log days’ supply of toilet paper:

(15) Log(DaysSupply)it = α+β1 Regit +β2 SingleF amilyit +γXit +λi +λt +it ,

Table 13 shows that households increase the days’ supply purchased by 3.5% when
unit prices are posted and by 2.6% when they move into a single-family home. The
model predictions are in line with these changes. The random coefficients model
predicts that purchasing increases by 1.6%–5.3% when unit prices are posted
(compared to 3.5% above) and by 4.6%–8.0% when storage costs are reduced
(compared to 2.6% above). The reduced-form estimates and model predictions
line up quite well with regards to posting unit prices. However, there are some
differences between the two types of estimates with respect to storage costs. Part
of this difference may be because home type (apartment or single-family home)
does not capture true storage costs while the random coefficients model offers
more flexibility to capture heterogeneity between households even with the same
type of housing. For example, the presence of a basement, garage, or even the
number of bathrooms may all influence the storage costs for toilet paper, but
those are all differences that can exist within single-family homes.
Overall, reducing cognitive costs and increasing the salience of unit prices helps
households make better value decisions, and generate a strong boost to bulk buying.
Adopting unit price regulations are a relatively straightforward policy approach to
encourage bulk buying, especially compared to the challenge of feasibly reducing
storage costs for low-income households.

VII. Conclusion

This paper documents the new fact that low-income households are less likely to
take advantage of quantity discounts relative to high-income households. This gap
is especially large for storable, necessity items like toilet paper and paper towels.
40 INCOME DIFFERENCES IN BULK BUYING

Table 13—: Effect of Unit Price Regulations and Housing Changes on Toilet Paper
Purchases

(1) (2)
Regulation 0.029∗ 0.035∗∗
(0.015) (0.015)
Single-Family Home 0.026∗∗∗
(0.006)
Household FE Y Y
Year FE Y Y
Demographics N Y
Observations 4,553,957 4,553,957
Adjusted R2 0.507 0.508
Note: Table shows estimates of Equation 5 which regresses household bulk buying on unit price regulation
after controlling for household fixed effects and changes in household characteristics. “Regulation” denotes
the estimated effect of moving from a state without regulation to a state with regulation. “Single-Family
Home” indicates that household lives in a single-family home with the reference category being an
apartment. Standard errors are clustered at the household level. ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Source: Nielsen Consumer Panel (2004–2017)

If low-income households bought in bulk like high-income households, they could

save 5% on grocery items, saving an aggregate of $5.4 billion annually. I provide
evidence that cognitive costs, store preferences, budget constraints, and storage
costs contribute to this gap.
By using state-level variation in whether or not retailers have to display unit
prices, I find that displaying unit prices reduces cognitive costs and increases bulk
buying. Then, I show that where a household shops accounts for a large portion
of this disparity and that warehouse clubs increase bulk buying, but only for
middle- and high-income households. Low-income households are unlikely to shop
at warehouse clubs, even if they are nearby. Next, I demonstrate that low-income
households slightly decrease bulk buying towards the end of the month, when
budgets are tighter. Finally, I show that households increase bulk buying when
they move to larger housing, supporting the fact that storage costs also influence
the bulk buying decision.
Combining these features into a discrete choice model of toilet paper purchases,
I predict how households’ bulk purchasing changes if unit-price regulations are
adopted universally and if storage costs are reduced. I find that posting unit prices
closes the bulk buying gap by 26% and reducing storage costs completely closes
the gap with middle- and low-income households buying larger quantities than
high-income households.
This paper is one of the first to focus on consumer’s take-up of quantity discounts
and explore the factors that contribute to this decision. It provides evidence that
 INCOME DIFFERENCES IN BULK BUYING 41

cognitive costs, store preferences, budget constraints, and storage costs affect a
household’s bulk buying decision. These differences have substantial financial
consequences for the poorest households and are likely to generate systematic
underestimates of consumption inequality if quantity discounts offset quality
differences between products. Additionally, if the prices of large and small packages
evolve differently, then households may experience substantial changes in their
buying power. Future work will determine the extent to which inequality and
inflation measures are underestimated because of quantity discounts.

REFERENCES

Aguiar, Mark, and Erik Hurst. 2005. “Consumption versus expenditure.”

Journal of Political Economy, 113(5): 919–948.

Aguiar, Mark, and Erik Hurst. 2007. “Life-cycle prices and production.”
American Economic Review, 97(5): 1533–1559.

Allcott, Hunt, Rebecca Diamond, Jean-Pierre Dubé, Jessie Handbury,

Ilya Rahkovsky, and Molly Schnell. 2019. “Food deserts and the causes of
nutritional inequality.” The Quarterly Journal of Economics, 134(4): 1793–1844.

Bauner, Christoph, and Emily Wang. 2019. “The Effect of Competition

on Pricing and Product Positioning: Evidence from Wholesale Club Entry.”
International Journal of Industrial Organization.

BLS. 2019. “Consumer Expenditures in 2017.” Bureau of Labor Statistics 1080.

Bogomolova, Svetlana, and Ian Jarratt. 2016. “Unit pricing in supermarkets:

Review of past evidence from academic and industry studies.” Available at SSRN
2853977.

Chetty, Raj, Adam Looney, and Kory Kroft. 2009. “Salience and taxation:
Theory and evidence.” American Economic Review, 99(4): 1145–77.

Chung, Chanjin, and Samuel L Myers Jr. 1999. “Do the poor pay more
for food? An analysis of grocery store availability and food price disparities.”
Journal of Consumer Affairs, 33(2): 276–296.

Cohen, Andrew. 2008. “Package size and price discrimination in the paper towel
market.” International Journal of Industrial Organization, 26(2): 502–516.

Coibion, Olivier, Yuriy Gorodnichenko, and Dmitri Koustas. 2017. “Con-

sumption inequality and the frequency of purchases.” National Bureau of Eco-
nomic Research Working Paper No. 23357.

Croissant, Yves. 2020. “mlogit: Multinomial Logit Models.” R package version

1.1-0.
42 INCOME DIFFERENCES IN BULK BUYING

DellaVigna, Stefano, and Matthew Gentzkow. 2019. “Uniform pricing in

us retail chains.” The Quarterly Journal of Economics, 134(4): 2011–2084.

Federal Highway Administration. 2017. “2017 National Household Travel

Survey.” U.S. Department of Transportation.

General Accounting Office. 1975. “Food Labeling: Goals, Shortcomings, and

Proposed Changes.” General Accounting Office.

Griffith, Rachel, Ephraim Leibtag, Andrew Leicester, and Aviv Nevo.

2009. “Consumer shopping behavior: how much do consumers save?” Journal
of Economic Perspectives, 23(2): 99–120.

Hendel, Igal, and Aviv Nevo. 2006. “Measuring the implications of sales and
consumer inventory behavior.” Econometrica, 74(6): 1637–1673.

Hensher, David A, and William H Greene. 2003. “The mixed logit model:
the state of practice.” Transportation, 30(2): 133–176.

Jaffe, Chuck. 2007. “On a roll.” MarketWatch.

Jarmin, Ronald S, Shawn D Klimek, and Javier Miranda. 2009. “The role
of retail chains: National, regional and industry results.” In Producer dynamics:
New evidence from micro data. 237–262. University of Chicago Press.

Kunreuther, Howard. 1973. “Why the poor may pay more for food: theoretical
and empirical evidence.” The Journal of Business, 46(3): 368–383.

Mitchell, Vincent-Wayne, David Lennard, and Peter McGoldrick. 2003.

“Consumer awareness, understanding and usage of unit pricing.” British Journal
of Management, 14(2): 173–187.

Miyazaki, Anthony D, David E Sprott, and Kenneth C Manning. 2000.

“Unit prices on retail shelf labels: an assessment of information prominence.”
Journal of Retailing, 76(1): 93–112.

NIST. 2019. “Uniform Laws and Regulations in the Areas of Legal Metrology
and Fuel Quality (Handbook 130).”

Orhun, A Yeşim, and Mike Palazzolo. 2019. “Frugality Is Hard to Afford.”

Journal of Marketing Research, 56(1): 1–17.

Revelt, David, and Kenneth Train. 1998. “Mixed logit with repeated choices:
households’ choices of appliance efficiency level.” Review of Economics and
Statistics, 80(4): 647–657.

Rose, Veronica. 2000. “Unit Pricing and Electronic Shelf Labels.” Connecticut
Office of Legislative Research 2000-R-1044.
 INCOME DIFFERENCES IN BULK BUYING 43

Sprott, David E, Kenneth C Manning, and Anthony D Miyazaki. 2003.

“Grocery price setting and quantity surcharges.” Journal of Marketing, 67(3): 34–
46.
Statista. 2019. “U.S. Population: Do you use toilet paper?”
Stephens Jr, Melvin. 2003. “” 3rd of tha Month”: Do Social Security Recip-
ients Smooth Consumption Between Checks?” American Economic Review,
93(1): 406–422.
Talukdar, Debabrata. 2008. “Cost of being poor: retail price and consumer
price search differences across inner-city and suburban neighborhoods.” Journal
of Consumer Research, 35(3): 457–471.
USDA. 2019. “Supplemental Nutrition Assistance Program Participation and
Costs.”
Zou, Hui, and Trevor Hastie. 2005. “Regularization and variable selection via
the elastic net.” Journal of the Royal Statistical Society: Series B (Statistical
Methodology), 67(2): 301–320.
44 INCOME DIFFERENCES IN BULK BUYING

Appendix

A1. Data Appendix

The Nielsen Consumer Panel consists of about 40,000–60,000 US households

that provide information on their shopping purchases using in-home scanners or
Nielsen’s mobile app. Panelists are geographically dispersed and demographically
balanced. Households are recruited based on key demographic characteristics,
primarily household size, income, age, education, presence of children, race,
ethnicity, and occupation. To generate national averages, Nielsen assigns each
household a projection factor.
Households are recruited through direct mail and online invitations. To incen-
tivize households to remain in the panel, Nielsen provides monthly prize drawings,
sweepstakes, points, and regular communication and support to panelists. Nielsen
tries to ensure that incentive methods are non-biasing and regularly tests for its
correlation with retention rates. To ensure data quality, Nielsen filters out any
households that are poor reporters and do not meet minimum spending thresholds
based on their household size. All households in the sample meet this threshold
for the full year.
Demographic variables are recorded and updated annually. For my analysis,
I collapse some of the demographic variables into more aggregate categories.
Household composition measures the number adults and children residing in the
home. Marital status is an indicator for whether the head of household is married
or not (I do not distinguish between single, divorced, or widowed). Education
is an indicator for whether at least one head of household completed college.
Housing variables indicate whether a household lives in a single-family home, and
apartment, or a mobile home. Finally, age is the age of the head of household. In
the case of two heads, I average the two ages.
To construct my analysis sample, I remove any households where the head of
household is a student or a member of the military because these households
likely have different living arrangements that are not representative of a typical
household’s decision (i.e., on campus housing or barracks are different than
traditional homes and apartments). I drop any households living in mobile homes
as well because this type of housing could include a wide range of house types
including RVs and manufactured homes. I also remove any households making
less that $5,000 and those that could not be geocoded based on their zip code.41 .
Finally, some households were dropped because they could not be matched to
tract-level vehicle access data.42 Table A1 reports how many households were
removed based on this cleaning procedure.

41 I use the 2017 Census Gazetteer to assign zip codes to the latitude and longitude of their population-
weighted centroid
42 Vehicle access data comes from the 2009–2013 American Community Survey, which asks how
individuals get to work. There is limited variation in this measure since most respondents have vehicle
access. For context, only 4% of Nielsen households live in Census tracts less than 90% access to cars.
 INCOME DIFFERENCES IN BULK BUYING 45

Table A1—: Homescan Sample Construction

Step HH
Starting Households: 178, 232
Exclude Military and Students: 175, 102
Exclude Households under 5k: 174, 106
Exclude Mobile Homes: 167, 065
Drop Zips Not Geocoded: 166, 366
Cannot Be Matched to Car Access: 166, 164
Note: Table reports the number of unique households in the sample after each step of data refinement.
Source: Nielsen Consumer Panel (2004–2017)

In the purchase data, I exclude alcohol, tobacco, pet items, health and beauty
items, general merchandise, “magnet,” and “deferred” product categories from my
analysis. Alcohol and tobacco are excluded because of their addictive qualities,
which may induce peculiar purchase patterns. For example, a smoker may choose
to only buy one pack of cigarettes with the intention of quitting even though a full
carton may deliver a better value. Pet items are excluded to focus on products
intended for human consumption. I exclude health and beauty items and general
merchandise because these products such as trash cans, printers, eye shadow,
and antacids are unlikely to be bought in bulk or have irregular consumption
patterns. “Deferred” categories are categories that Nielsen has stopped tracking,
so to maintain a consistent sample of products, these are excluded from my
analysis. Finally, “magnet” purchases are items which do not have a UPC codes
such as fresh fruits and vegetables, deli counter items, or bakery items. Because
these items are only recorded for a subset of Nielsen households and are not
standardized, I also exclude them from my analysis. This process leaves me with
721 unique product categories. Because this paper focuses on bulk purchases, I also
exclude 28 categories that have five or fewer sizes across all possible products.43
Overall, the products analyzed are common household staples including almost
all food categories, basic toiletry items, and non-food essentials like toilet paper,
soaps/detergents, and diapers. See Table A2 for summary statistics of the top 20
product categories by annual spending.
To compare sizes across different product categories, I assign each product to
its quintile in the size distribution for that product category. I assign quintiles
based upon the sample quintiles of product sizes to ensure that each quintile has

43 These excluded categories are: jelled aspic salad, sour cream sauce mix, canned roast beef, canned
roast beef hash, retort pouch bags, prepared sandwiches, canned rice, canned dumplings, canned bread,
frozen vegetables in pastry, frozen grapefruit juice, frozen grape juice, frozen orange juice, frozen cream
substitutes, canned ham patties, bathroom accessory, packaged soap, borateem, dry starch, grease relief,
bathroom brushes, miscellaneous brushes, thermometers, dustpans, feather dusters, laundry baskets,
sanitary belts, gift package with candy or gum.
46 INCOME DIFFERENCES IN BULK BUYING

Table A2—: Summary Statistics of Top 20 Product Categories in Nielsen Homescan

Data (2017)

Product Annual SD Avg. Price SD Avg. Size SD

Spending
Soft Drinks 79.38 139.02 4.75 4.17 85.87 53.81
Diet Soft Drinks 74.82 132.79 4.73 5.07 84.18 65.06
Milk 65.65 77.02 3.11 1.79 97.79 35.00
Cereal 57.97 68.37 4.06 2.10 18.05 8.17
Toilet Paper 56.15 49.47 11.44 7.09 17.09 10.51
Yogurt 55.00 75.68 3.28 2.17 17.25 15.22
Coffee 53.97 61.69 8.60 5.74 21.84 11.05
Bread 50.03 47.09 2.88 1.52 20.54 4.64
Cookies 46.97 57.60 3.59 3.44 13.02 6.39
Fresh Meat 46.96 62.86 7.75 5.03 30.48 24.97
Frozen Pizza 44.48 60.64 5.99 3.67 20.69 12.48
Bottled Water 44.06 73.46 4.21 3.75 261.91 181.39
Fresh Fruit 42.68 64.91 4.28 2.06 1.93 1.31
Choc. Candy 41.05 53.83 3.91 3.67 8.64 9.15
Detergent 40.17 45.29 10.05 7.85 99.52 61.23
Shred. Cheese 39.16 42.80 4.21 2.45 13.37 10.98
Bacon 37.63 45.44 6.87 4.67 17.42 11.88
Ice Cream 37.36 50.34 4.43 2.03 46.80 24.47
Potato Chips 35.99 41.71 3.04 1.89 8.87 3.81
Canned Soup 32.39 38.36 3.21 2.22 22.07 17.33

Note: Table reports summary statistics for the top 20 product categories by total spending. Annual
spending is the average spending in that product category among households that purchased in that
product category over the course of the year. Average price and average size are the average prices and
sizes of products purchased in their corresponding category. All estimates are weighted using Nielsen’s
projection weights. Prices are in nominal 2017 dollars. Sizes are reported in common units for for that
category (e.g. ounces for milk).
Source: Nielsen Consumer Panel (2004–2017)
 INCOME DIFFERENCES IN BULK BUYING 47

20% of available products in its support. An alternative strategy would assign

quintiles based on cutting the range of product sizes into equal intervals. However,
in some product categories, this risks generating quintiles with sparse support
when there is an especially large package available. As an example, consider eggs.
Most packages contain 6, 12, or 18 eggs, but there are some products that offer
up to 15-dozen eggs (180 eggs). Generating quintiles by cutting the available
range into equal intervals would generate quintiles of 1-36, 37-72, 73-108, 109-144,
145-180 which would assign almost all packages to the first quintile and the fifth
quintile. Using the sample quintiles generates a more even distribution ensuring
better support of each quintile. For products with a narrow range of sizes that fall
in multiple quintiles, I assign the product to the minimum quintile. For example,
over 60% of egg products are dozens, which covers three quintiles. I assign all
products with 12 or fewer eggs to the first quintile.

A2. Quantity Discounts and Coupon Savings

This section compares savings from quantity discounts to savings from coupons.
To be conservative, I compare the savings from redeemed coupons (likely higher
than the average savings of all coupons offered) to savings offered by quantity
discounts (likely lower than quantity discounts actually redeemed). For each
product purchased in the Consumer Panel data, households can input the value
saved if they used a coupon. For each product category, I compute the average
discount across all products in that category.
I then estimate quantity discount savings based on moving from a product in
the second quintile to the fourth quintile of the size distribution. This leaves out
small product sizes that may have high unit prices due to convenience (e.g., a
20-oz soda bottle at the checkout counter) and especially large sizes that may not
be widely available at all stores. This range covers sizes that households are likely
to consider when making their purchase decision.
Figure A1 plots the distribution of coupon savings and estimated bulk savings
for food and non-food products. Coupon savings are narrowly clustered with a
median savings of 31% for non-food products and 33% for food products. Bulk
discounts have lower median savings for non-food and food products of 27% and
23%, respectively, but are more widely dispersed, even exceeding 50% savings for
some non-food products.44 Coupon savings are similar across product categories
while there is substantial variation in quantity discounts with non-food products
offering higher savings.

A3. Bulk Buying Across Popular Categories

Across popular spending categories, these gaps are particularly large in storable,
non-food categories like paper towels and toilet paper, where households making
44 Smaller shifts, such as from the second to third quintile or third to fourth quintile generate smaller
savings, but still preserve the long right tail primarily for non-food products.
48 INCOME DIFFERENCES IN BULK BUYING

Figure A1. : Estimated Savings from Coupons and Bulk Discounts

0.08

0.06

Bulk
0.04

0.02
Density

0.00
0.08

0.06

Coupon
0.04

0.02

0.00
0 25 50 75 100
Savings (%)

Product Type Food Non−Food

Note: Figure plots the distribution of savings from coupons and quantity discounts. For each coupon
redemption, the percent savings are the ratio of the coupon value to the product’s price. These savings are
then averaged across all purchases in that product category. Bulk discounts are computed using coefficient
estimates obtained from Equation (1) relating log unit prices to log package sizes. Bulk savings are the
estimated savings obtained from moving from the second to the fourth quintile of the size distribution for
each product category.
Source: Nielsen Consumer Panel (2004–2017) and Nielsen Retail Scanner (2016)

over $100,000 are more than twice as likely to buy in bulk compared to households
making under $25,000. In popular food categories like milk and eggs, there is little
relationship or even a negative relationship between income and bulk buying (See
Figure A2).

A4. Alternative Calculation of Missed Quantity Discounts

An alternative way of calculating savings from quantity discounts is to calculate

first-best savings obtained from purchasing the lowest unit-priced item available,
since even high-income households may not be buying at the lowest unit price.
I compute this by taking the difference between the unit price paid by each
household and the lowest unit price available in the store, given the household’s
brand preference. I get this information through linking the Nielsen Consumer
Panel with the Nielsen Retail Scanner data.
I compute the first-best savings a household could obtain for toilet paper,
diapers, milk, and eggs using the following approach. First, for each shopping trip,
I compute the lowest unit price the household could have paid given its brand
 INCOME DIFFERENCES IN BULK BUYING 49

Figure A2. : Bulk Purchasing by Household Income (Selected Products)

Difference in Bulk Purchasing (Percentage Points) Food Non−Food

20

10

●
0 ●●●
●●
●●
●● ●
●
●
−10
●

0 25 50 75 100 0 25 50 75 100
Household Income

Product Diapers Eggs ● Milk Paper Towels Toilet Paper

Note: Figure plots the income bin coefficients from Equation (2), which regresses the share of annual
purchases that were bulk packages on household characteristics as well as market and year fixed effects.
This regression is estimated for milk, eggs, diapers, toilet paper, and paper towels. Nielsen projection
weights are used to ensure national representativeness. Households making $5–$8k are the reference group.
Standard errors are clustered at the DMA level. Coefficient values are reported in Appendix Table A10
Source: Nielsen Consumer Panel (2004–2017)

and store choice in that week. The difference in unit prices relative to the unit
price chosen is a household’s first-best savings for that purchase. Then, to get
the average savings for a household, I compute the expenditure-weighted average
savings across all purchases for each household. Based on this measure, Table A3
reports average excess spending by income group, computed for a family of four.
Overall, households could save over 30% by buying in bulk and low-income
households could save even more. I estimate the differences in savings between
households from the following regression:
X
(A1) Yimt = α + β q Incomeimt + γXimt + λmt + imt ,
q

where Yimt is the excess spending of household i in market m in year t. Incomeimt

is the household’s income bin and Ximt consists of household characteristics. λmt
is a market-year fixed effect. Table A4 shows that low-income households miss out
on 1.7–1.8 percentage points more savings than high-income households and the
excess spending is primarily in non-food categories like toilet paper (36% savings)
and diapers (33% savings) as opposed to food categories like milk (31% savings)
50 INCOME DIFFERENCES IN BULK BUYING

Table A3—: First-Best Savings by Household Income and Product

Non-Perishable Perishable
Income Toilet Paper Diapers Milk Eggs
<25k 0.36 0.33 0.31 0.17
25-50k 0.35 0.33 0.30 0.17
50-100k 0.34 0.33 0.31 0.17
>100k 0.33 0.31 0.33 0.18

Note: Table reports average savings a household could achieve given its brand and store choice. Average
savings for a family of four is reported above. For example, a household making <$25k could save 36%
by purchasing at the lowest unit price available.
Source: Nielsen Consumer Panel (2006–2016) and Nielsen Retail Scanner (2006–2016)

and eggs (17% savings). Given the perishability of food items, these savings may
not be realized if the product perishes before it can be consumed.

Table A4—: Regression Results of First-Best Savings Across Household Income

and Products

Diapers Toilet Paper Eggs Milk

(1) (2) (3) (4)
25-50k −0.010∗∗ −0.005∗∗∗ 0.001 −0.002
(0.005) (0.001) (0.001) (0.001)
50-100k −0.015∗∗∗ −0.013∗∗∗ 0.004∗∗∗ 0.002∗∗
(0.005) (0.001) (0.001) (0.001)
>100k −0.018∗∗∗ −0.017∗∗∗ 0.018∗∗∗ 0.010∗∗∗
(0.005) (0.002) (0.002) (0.001)
Demographics Y Y Y Y
Market-Year FE Y Y Y Y
Observations 36,903 182,415 194,413 247,451
Adjusted R2 0.012 0.071 0.117 0.231

Note: Table reports the income coefficients of Equation (A1), which regresses savings on household
characteristics as well as a market and year fixed effect. Nielsen’s projection weights are used for national
representativeness. ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Source: Nielsen Consumer Panel (2006–2016) and Nielsen Retail Scanner (2006–2016)

Overall, low-income households could benefit substantially from buying in bulk

and obtaining lower unit prices. Furthermore, these savings are likely to be more
important for low-income households since the marginal utility of an additional
 INCOME DIFFERENCES IN BULK BUYING 51

dollar of savings is likely to be higher than for high-income households. This

analysis also provides evidence that all households could benefit from purchasing
at the lowest unit price.

A5. Bulk Buying by Store Type or Chain Size

In this section, I analyze whether the effect of unit pricing differs by store type
or chain size. Unit price regulations are only at the state level, but retailers are
free to post (or not post) unit prices as long as they are within the boundaries
of the law. Large chains may post prices uniformly across all stores in a way
that meets the strictest requirements they are subject to. On the other hand,
regional chains or independent stores may more closely mirror the laws of the
state they are located in. I estimate Equation 4 using annual household bulk
buying at specific stores types or within different chain sizes. Each observation is
at the household-year-channel (or chain) level. For example, bulk items accounted
for 50% of Household A’s grocery store purchases while bulk items accounted for
100% of Household A’s warehouse club purchases.

Table A5—: Unit Price Regulations and Bulk Buying by Store Type

Grocery Drug Discount Dollar Warehouse

(1) (2) (3) (4) (5)
Vol. Disp 0.011 −0.011∗∗∗ −0.006 −0.006 −0.004∗
(0.009) (0.003) (0.005) (0.005) (0.002)
Mand. Disp 0.029∗∗∗ 0.009 0.014∗ −0.022∗∗ 0.010∗∗∗
(0.005) (0.007) (0.008) (0.010) (0.003)
Mand. Disp, Strict 0.054∗∗∗ 0.018∗∗∗ 0.002 −0.006 0.006∗∗∗
(0.009) (0.003) (0.004) (0.006) (0.002)
Avg Bulk 0.36 0.29 0.49 0.35 0.95
Demographics Y Y Y Y Y
Omit California Y Y Y Y Y
Observations 618,029 298,166 562,749 328,607 267,759
Adjusted R2 0.011 0.003 0.005 0.002 0.001
Note: Table shows the results of estimating Equation 4. The dependent variable is the annual share of
bulk purchases made by households in a particular channel (i.e. store type) and the independent variables
are an ordered measure of regulatory stringency. California is omitted because it is the only state that
has voluntary unit price, but strict requirements on how unit prices are displayed. Standard errors are
clustered by state. ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Source: Nielsen Consumer Panel (2004–2017)

Table A5 shows that in the cross-section, stricter unit price regulations are
associated with more bulk buying primarily for grocery stores, drug stores, and
52 INCOME DIFFERENCES IN BULK BUYING

warehouse clubs. Households in states with strict unit price regulations buy in
bulk five percentage points more at grocery stores compared to households in
states without any pricing regulations. Since grocery stores tend to be regional
or independent, the large positive relationship provides strong evidence that unit
price regulations can increase bulk purchasing. Grocery stores also have the richest
variety in Nielsen’s data with over 900 unique retailers being captured compared
to 65 drug stores, 25 discount stores, 17 dollar stores, and 10 warehouse clubs.45
Other store types exhibit smaller or insignificant effects, which could be because
these are generally large chains that have more uniform pricing practices across
all locations.

Table A6—: Unit Price Regulations and Bulk Buying by Retailer Size

Local Regional National

(1) (2) (3)
Vol. Disp 0.063∗∗∗ 0.030 0.010
(0.022) (0.025) (0.009)
Mand. Disp −0.067 −0.018 0.039∗∗∗
(0.053) (0.014) (0.009)
Mand. Disp, Strict 0.064∗∗∗ 0.070∗∗∗ 0.026∗∗∗
(0.016) (0.014) (0.006)
Avg Bulk 0.33 0.32 0.49
Demographics Y Y Y
Omit California Y Y Y
Observations 1,578 43,756 668,566
Adjusted R2 0.008 0.005 0.037
Note: Table shows the results of estimating Equation 4. The dependent variable is the annual share
of bulk purchases made by households at different retail chain sizes and the independent variables are
an ordered measure of regulatory stringency. California is omitted because it is the only state that
has voluntary unit price, but strict requirements on how unit prices are displayed. Standard errors are
clustered by state. ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Source: Nielsen Consumer Panel (2004–2017)

Table A6 shows the results by chain size. Following Jarmin, Klimek and Miranda
(2009), I define a “local” chain as only having locations in one state, a “regional”
chain has locations in two to ten states, and a “national” chain has locations
in more than ten states. In the cross-section, stricter unit price regulations are
associated with more bulk buying across all chain types. The effect is strongest
45 Nielsen’s categorization includes a “catch-all” category that is not unique to a particular retailer, so
it actually uniquely captures 64 drug stores and purchases at other drug stores are assigned to the last
“catch-all” drug store. Generally, larger retailers are uniquely tracked and smaller ones may fall into the
“catch-all” category.
 INCOME DIFFERENCES IN BULK BUYING 53

for local and regional chains, exhibiting a six to seven percentage point increase in
bulk buying relative to states without unit price regulations. National chains still
have significant differences, but they are a more moderate three to four percentage
point difference relative to states without regulations. Overall, the relative effect
is strongest for the smaller chains that are likely to only be subject to a limited
set of regulations and the effect is weaker for national chains which may be more
likely to adopt pricing practices that satisfy the strictest requirements nationwide.

A6. Annual Consumption Analysis

I show that income is not predictive of a household’s toilet paper consumption

rate first using basic OLS regressions. I then formalize the result using a 100-fold
cross-validated elastic net regression to select the most predictive variables. If
income and toilet paper consumption are related, then an OLS regression will
extract the correlation.
First, I compute a household’s daily consumption by aggregating the total
number of sheets purchased by a household in a given year, excluding the final
purchase of the year since it may not be consumed within the year. I divide this
total by the number of days between the first and last purchase of the year to get
a household’s average daily consumption rate. This method avoids complications
where end of the year inventory may be carried over to the following year or a
household may start the year with some inventory.
Given a household’s average daily consumption rate, I estimate an OLS regression
of consumption on household characteristics:

(A2) Yi = α + βXi + i ,

where Yi is household i’s average daily consumption and Xi is a vector of household

characteristics. Figure A3 plots the income coefficients of an OLS regression
including only income covariates and the coefficients when household characteristics
are included. The graph illustrates that after controlling for covariates that
plausibly cause increased consumption, income is not significantly correlated with
consumption.
The above specification omits many other possible covariates that could be
correlated with average daily consumption. When there are many possible variables
that can be included, there is a risk of over-fitting. Elastic net regularization is
a machine learning method that penalizes over-fitting and selects only the most
predictive variables.
The elastic net solves the following minimization problem:

min ky − Xβk2 + λ αkβk1 + (1 − α)kβk22 ,

(A3)
β

where k · k1 is the L1 norm and k · k2 is the L2 norm. The OLS estimate is the
β that solves the minimization problem with only the first term. The second
54 INCOME DIFFERENCES IN BULK BUYING

Figure A3. : Average Daily Consumption by Household Income

Average Consumption Rate (Sheets / Day)

40

30

20

10

● ●
0 ●
● ● ● ● ●
● ● ● ●
●
−10
25 50 75 100
Annual Household Income ($000s)

Covariates ● Income and Demographics Income Only

Note: Figure plots the income bin coefficients from Equation (A2), which regresses average daily household
toilet paper consumption on household characteristics. Average daily consumption is computed by dividing
total quantity purchased in a year by the number of days a household was active in the panel.
Source: Nielsen Consumer Panel (2004–2017)

term and third term provide penalties to shrink and select for the most predictive
variables.
I set the mixing parameter α to be 0.5. When covariates are correlated in
groups, lasso regression (α = 1) tends to only select one and discard all other
members of the group while ridge regression (α = 0) tends to shrink correlated
coefficients towards each other (Zou and Hastie, 2005). Because some of the
possible covariates form natural groups (e.g., all income bins or all markets), I
chose α = 0.5 since this tends to include or exclude groups together.
I estimate a 100-fold cross-validated elastic net regression to select the most
predictive covariates. The resulting estimates selects many household characteris-
tics including household composition, age, marital status, and race, but excludes
almost all income and geographic coefficients.46

A7. Appendix Tables

46 Elastic net results are available upon request.

 INCOME DIFFERENCES IN BULK BUYING 55

Table A7—: Correlation of Bulk Buying and Demographics (Food Products)

(1) (2) (3) (4) (5) (6) (7) (8) (9)

8-10k −0.005∗∗∗ −0.004∗∗ −0.004∗∗ −0.003∗ −0.002 −0.002 −0.002 −0.004∗∗ −0.003
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
10-12k −0.004∗∗ −0.005∗∗∗ −0.004∗∗ −0.006∗∗∗ −0.006∗∗∗ −0.006∗∗∗ −0.006∗∗∗ −0.008∗∗∗ −0.008∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
12-15k −0.0001 −0.002 −0.001 −0.002 −0.004∗∗ −0.005∗∗∗ −0.005∗∗∗ −0.005∗∗∗ −0.004∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
15-20k 0.001 −0.003∗ −0.002 −0.004∗∗∗ −0.006∗∗∗ −0.006∗∗∗ −0.006∗∗∗ −0.006∗∗∗ −0.005∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.001) (0.001)
20-25k 0.007∗∗∗ 0.0003 0.001 −0.002 −0.005∗∗∗ −0.006∗∗∗ −0.005∗∗∗ −0.006∗∗∗ −0.005∗∗∗
(0.002) (0.002) (0.002) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
25-30k 0.013∗∗∗ 0.004∗∗∗ 0.004∗∗∗ −0.0002 −0.004∗∗∗ −0.005∗∗∗ −0.004∗∗∗ −0.004∗∗∗ −0.004∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.001)
30-35k 0.016∗∗∗ 0.005∗∗∗ 0.005∗∗∗ 0.0004 −0.004∗∗ −0.004∗∗∗ −0.003∗∗ −0.004∗∗ −0.003∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.001) (0.001)
35-40k 0.021∗∗∗ 0.008∗∗∗ 0.008∗∗∗ 0.003∗∗ −0.002 −0.002 −0.001 −0.001 −0.0005
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
40-45k 0.024∗∗∗ 0.010∗∗∗ 0.009∗∗∗ 0.004∗∗∗ −0.001 −0.001 0.0002 0.0002 0.001
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
45-50k 0.028∗∗∗ 0.011∗∗∗ 0.011∗∗∗ 0.005∗∗∗ −0.0001 −0.0003 0.001 0.001 0.002
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
50-60k 0.029∗∗∗ 0.011∗∗∗ 0.010∗∗∗ 0.005∗∗∗ −0.001 −0.001 0.001 0.001 0.002
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
60-70k 0.035∗∗∗ 0.015∗∗∗ 0.014∗∗∗ 0.008∗∗∗ 0.002 0.001 0.003∗∗ 0.004∗∗ 0.004∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.001) (0.002) (0.001) (0.001)
70-100k 0.038∗∗∗ 0.015∗∗∗ 0.015∗∗∗ 0.009∗∗∗ 0.002 0.002 0.005∗∗∗ 0.005∗∗∗ 0.006∗∗∗
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
>100k 0.041∗∗∗ 0.015∗∗∗ 0.015∗∗∗ 0.010∗∗∗ 0.002 0.003∗ 0.006∗∗∗ 0.007∗∗∗ 0.008∗∗∗
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Married 0.043∗∗∗ 0.042∗∗∗ 0.022∗∗∗ 0.018∗∗∗ 0.016∗∗∗ 0.016∗∗∗ 0.017∗∗∗ 0.018∗∗∗
(0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004)
Age −0.0002∗∗∗ 0.0002∗∗∗ 0.00005∗∗∗ 0.0001∗∗∗ 0.00003∗∗ 0.0001∗∗∗ 0.0001∗∗∗
(0.00001) (0.00001) (0.00001) (0.00001) (0.00001) (0.00001) (0.00001)
Adults 0.017∗∗∗ 0.016∗∗∗ 0.017∗∗∗ 0.016∗∗∗ 0.015∗∗∗ 0.015∗∗∗
(0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002)
Children 0.016∗∗∗ 0.015∗∗∗ 0.015∗∗∗ 0.015∗∗∗ 0.014∗∗∗ 0.014∗∗∗
(0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002)
Single-Family Home 0.033∗∗∗ 0.026∗∗∗ 0.026∗∗∗ 0.025∗∗∗ 0.025∗∗∗
(0.0004) (0.0004) (0.0004) (0.0004) (0.0004)
(0.002) (0.002) (0.002) (0.002)
College −0.006∗∗∗ −0.006∗∗∗ −0.006∗∗∗
(0.0004) (0.0004) (0.0004)
Market FE’s N N N N N N N Y Y
Year FE’s N N N N N N N N Y
Observations 733,894 733,894 733,894 733,894 733,894 733,894 733,894 733,894 733,894
Adjusted R2 0.011 0.030 0.030 0.050 0.058 0.062 0.062 0.101 0.102
Note: ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01

Source: Author calulations from Nielsen Consumer Panel.

56 INCOME DIFFERENCES IN BULK BUYING

Table A8—: Correlation of Bulk Buying and Demographics (Non-Food Products)

(1) (2) (3) (4) (5) (6) (7) (8) (9)

8-10k 0.008∗∗ 0.008∗∗∗ 0.008∗∗∗ 0.008∗∗∗ 0.009∗∗∗ 0.009∗∗∗ 0.009∗∗∗ 0.006 0.002
(0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.007) (0.007)
10-12k 0.023∗∗∗ 0.022∗∗∗ 0.022∗∗∗ 0.021∗∗∗ 0.020∗∗∗ 0.020∗∗∗ 0.020∗∗∗ 0.016∗ 0.013
(0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.008) (0.008)
12-15k 0.023∗∗∗ 0.021∗∗∗ 0.021∗∗∗ 0.019∗∗∗ 0.018∗∗∗ 0.018∗∗∗ 0.017∗∗∗ 0.016∗∗ 0.013∗∗
(0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.007) (0.007)
15-20k 0.033∗∗∗ 0.030∗∗∗ 0.029∗∗∗ 0.028∗∗∗ 0.026∗∗∗ 0.026∗∗∗ 0.025∗∗∗ 0.022∗∗∗ 0.019∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.007) (0.007)
20-25k 0.052∗∗∗ 0.046∗∗∗ 0.046∗∗∗ 0.044∗∗∗ 0.041∗∗∗ 0.041∗∗∗ 0.039∗∗∗ 0.035∗∗∗ 0.032∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.006) (0.006)
25-30k 0.057∗∗∗ 0.049∗∗∗ 0.049∗∗∗ 0.046∗∗∗ 0.042∗∗∗ 0.042∗∗∗ 0.041∗∗∗ 0.036∗∗∗ 0.034∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.006) (0.006)
30-35k 0.069∗∗∗ 0.061∗∗∗ 0.061∗∗∗ 0.057∗∗∗ 0.053∗∗∗ 0.053∗∗∗ 0.051∗∗∗ 0.045∗∗∗ 0.042∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.005) (0.006)
35-40k 0.076∗∗∗ 0.065∗∗∗ 0.065∗∗∗ 0.061∗∗∗ 0.057∗∗∗ 0.057∗∗∗ 0.054∗∗∗ 0.048∗∗∗ 0.046∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.006) (0.006)
40-45k 0.084∗∗∗ 0.072∗∗∗ 0.073∗∗∗ 0.068∗∗∗ 0.064∗∗∗ 0.064∗∗∗ 0.060∗∗∗ 0.054∗∗∗ 0.052∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.006) (0.006)
45-50k 0.095∗∗∗ 0.081∗∗∗ 0.081∗∗∗ 0.077∗∗∗ 0.072∗∗∗ 0.072∗∗∗ 0.068∗∗∗ 0.060∗∗∗ 0.057∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.006) (0.006)
50-60k 0.099∗∗∗ 0.084∗∗∗ 0.085∗∗∗ 0.080∗∗∗ 0.075∗∗∗ 0.075∗∗∗ 0.070∗∗∗ 0.062∗∗∗ 0.059∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.005) (0.005)
60-70k 0.117∗∗∗ 0.101∗∗∗ 0.101∗∗∗ 0.096∗∗∗ 0.091∗∗∗ 0.091∗∗∗ 0.085∗∗∗ 0.075∗∗∗ 0.071∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.006) (0.006)
70-100k 0.136∗∗∗ 0.118∗∗∗ 0.118∗∗∗ 0.113∗∗∗ 0.107∗∗∗ 0.107∗∗∗ 0.099∗∗∗ 0.088∗∗∗ 0.085∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.006) (0.006)
>100k 0.172∗∗∗ 0.151∗∗∗ 0.151∗∗∗ 0.146∗∗∗ 0.139∗∗∗ 0.140∗∗∗ 0.128∗∗∗ 0.111∗∗∗ 0.102∗∗∗
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.006) (0.006)
Married 0.035∗∗∗ 0.035∗∗∗ 0.024∗∗∗ 0.020∗∗∗ 0.020∗∗∗ 0.021∗∗∗ 0.026∗∗∗ 0.025∗∗∗
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.002)
Age 0.0001∗∗∗ 0.0003∗∗∗ 0.0002∗∗∗ 0.0002∗∗∗ 0.0003∗∗∗ 0.0002∗∗∗ 0.0002∗∗
(0.00002) (0.00002) (0.00002) (0.00002) (0.00002) (0.0001) (0.0001)
Men 0.016∗∗∗ 0.015∗∗∗ 0.015∗∗∗ 0.016∗∗∗ 0.013∗∗∗ 0.014∗∗∗
(0.0004) (0.0004) (0.0004) (0.0004) (0.002) (0.002)
Women 0.003∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.004∗∗∗ 0.001 0.001
(0.0004) (0.0004) (0.0004) (0.0004) (0.001) (0.001)
Children 0.008∗∗∗ 0.007∗∗∗ 0.007∗∗∗ 0.007∗∗∗ 0.005∗∗∗ 0.005∗∗∗
(0.0003) (0.0003) (0.0003) (0.0003) (0.001) (0.001)
Single-Family Home 0.027∗∗∗ 0.026∗∗∗ 0.027∗∗∗ 0.038∗∗∗ 0.039∗∗∗
(0.001) (0.001) (0.001) (0.003) (0.003)
(0.003) (0.003) (0.009) (0.008)
College 0.020∗∗∗ 0.018∗∗∗ 0.017∗∗∗
(0.001) (0.002) (0.002)
Market FE’s N N N N N N N Y Y
Year FE’s N N N N N N N N Y
Observations 731,762 731,762 731,762 731,762 731,762 731,762 731,762 731,762 731,762
Adjusted R2 0.049 0.054 0.054 0.056 0.058 0.058 0.060 0.087 0.094
Note: ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01

Source: Author calulations from Nielsen Consumer Panel. Columns (7) and (8)
cluster standard errors at the market level
 INCOME DIFFERENCES IN BULK BUYING 57

Table A9—: Bulk Buying Within Zip Codes or Retail Chains by Income

Zip Code Store/Retailer Type

8-10k 0.003 0.006 0.006∗∗∗ 0.006∗∗∗ 0.006∗∗∗
(0.003) (0.004) (0.002) (0.002) (0.002)
10-12k 0.013∗∗∗ 0.006∗ 0.013∗∗∗ 0.011∗∗∗ 0.011∗∗∗
(0.003) (0.003) (0.002) (0.002) (0.002)
12-15k 0.013∗∗∗ 0.008∗∗ 0.014∗∗∗ 0.008∗∗∗ 0.008∗∗∗
(0.003) (0.003) (0.002) (0.002) (0.002)
15-20k 0.020∗∗∗ 0.015∗∗∗ 0.019∗∗∗ 0.013∗∗∗ 0.012∗∗∗
(0.002) (0.003) (0.002) (0.002) (0.002)
20-25k 0.032∗∗∗ 0.027∗∗∗ 0.028∗∗∗ 0.017∗∗∗ 0.017∗∗∗
(0.002) (0.003) (0.002) (0.002) (0.002)
25-30k 0.035∗∗∗ 0.030∗∗∗ 0.028∗∗∗ 0.015∗∗∗ 0.014∗∗∗
(0.002) (0.003) (0.002) (0.002) (0.002)
30-35k 0.044∗∗∗ 0.038∗∗∗ 0.036∗∗∗ 0.019∗∗∗ 0.019∗∗∗
(0.002) (0.003) (0.002) (0.002) (0.002)
35-40k 0.047∗∗∗ 0.044∗∗∗ 0.040∗∗∗ 0.019∗∗∗ 0.019∗∗∗
(0.002) (0.003) (0.002) (0.002) (0.002)
40-45k 0.053∗∗∗ 0.049∗∗∗ 0.041∗∗∗ 0.018∗∗∗ 0.017∗∗∗
(0.002) (0.003) (0.002) (0.002) (0.002)
45-50k 0.059∗∗∗ 0.052∗∗∗ 0.047∗∗∗ 0.022∗∗∗ 0.021∗∗∗
(0.002) (0.003) (0.002) (0.002) (0.002)
50-60k 0.061∗∗∗ 0.056∗∗∗ 0.047∗∗∗ 0.020∗∗∗ 0.019∗∗∗
(0.002) (0.003) (0.002) (0.002) (0.002)
60-70k 0.073∗∗∗ 0.066∗∗∗ 0.053∗∗∗ 0.020∗∗∗ 0.020∗∗∗
(0.002) (0.003) (0.002) (0.002) (0.002)
70-100k 0.087∗∗∗ 0.081∗∗∗ 0.063∗∗∗ 0.025∗∗∗ 0.025∗∗∗
(0.002) (0.003) (0.002) (0.002) (0.001)
>100k 0.105∗∗∗ 0.096∗∗∗ 0.074∗∗∗ 0.025∗∗∗ 0.025∗∗∗
(0.002) (0.003) (0.002) (0.002) (0.002)
Demographics Y Y Y Y Y
ZIP-Year FE N Y N N N
Channel-Year FE N N N Y Y
Chain-Year FE N N N N Y
Observations 731,762 731,762 3,799,404 3,799,404 3,799,404
Adjusted R2 0.093 0.270 0.013 0.223 0.241
Note: Table displays the regression coefficients from estimating Equation 7 which regresses a household’s
annual share of bulk purchases of non-food products on household characteristics (household composition,
age, marital status, and education) and includes a DMA and year fixed effect (Columns 1 and 3). Columns
2, 4, and 5 also include a zip code-year, store type-year, or retail chain-year fixed effect. ∗ p<0.1; ∗∗ p<0.05;
∗∗∗ p<0.01
Source: Nielsen Consumer Panel (2004–2017)
58 INCOME DIFFERENCES IN BULK BUYING

Table A10—: Bulk Purchasing by Household Income (Selected Products)

Toilet Paper Paper Towels Diapers Eggs Milk

8-10k 0.003 0.004 −0.020 −0.013∗∗∗ 0.013∗∗
(0.006) (0.007) (0.019) (0.005) (0.006)
10-12k 0.002 0.007 −0.002 −0.001 −0.005
(0.005) (0.006) (0.017) (0.005) (0.005)
12-15k −0.002 0.024∗∗∗ −0.028∗ −0.007 −0.011∗∗
(0.005) (0.006) (0.016) (0.004) (0.005)
15-20k 0.010∗∗ 0.038∗∗∗ 0.017 −0.012∗∗∗ −0.011∗∗
(0.005) (0.005) (0.015) (0.004) (0.004)
20-25k 0.026∗∗∗ 0.055∗∗∗ 0.016 −0.020∗∗∗ −0.021∗∗∗
(0.004) (0.005) (0.014) (0.004) (0.004)
25-30k 0.029∗∗∗ 0.056∗∗∗ 0.017 −0.021∗∗∗ −0.025∗∗∗
(0.005) (0.005) (0.015) (0.004) (0.004)
30-35k 0.042∗∗∗ 0.063∗∗∗ 0.005 −0.019∗∗∗ −0.034∗∗∗
(0.005) (0.005) (0.014) (0.004) (0.004)
35-40k 0.041∗∗∗ 0.082∗∗∗ 0.021 −0.021∗∗∗ −0.043∗∗∗
(0.005) (0.005) (0.014) (0.004) (0.004)
40-45k 0.054∗∗∗ 0.094∗∗∗ 0.024∗ −0.021∗∗∗ −0.047∗∗∗
(0.005) (0.005) (0.014) (0.004) (0.004)
45-50k 0.065∗∗∗ 0.107∗∗∗ 0.015 −0.018∗∗∗ −0.052∗∗∗
(0.005) (0.005) (0.014) (0.004) (0.005)
50-60k 0.068∗∗∗ 0.111∗∗∗ 0.019 −0.011∗∗∗ −0.056∗∗∗
(0.004) (0.005) (0.014) (0.004) (0.004)
60-70k 0.092∗∗∗ 0.138∗∗∗ 0.026∗ −0.006 −0.063∗∗∗
(0.004) (0.005) (0.014) (0.004) (0.004)
70-100k 0.112∗∗∗ 0.162∗∗∗ 0.048∗∗∗ −0.008∗∗ −0.075∗∗∗
(0.004) (0.005) (0.014) (0.004) (0.004)
>100k 0.158∗∗∗ 0.210∗∗∗ 0.056∗∗∗ −0.019∗∗∗ −0.109∗∗∗
(0.004) (0.005) (0.014) (0.004) (0.004)
Married 0.045∗∗∗ 0.040∗∗∗ 0.027∗∗∗ 0.028∗∗∗ 0.062∗∗∗
(0.001) (0.001) (0.003) (0.001) (0.001)
Age −0.0003∗∗∗ 0.002∗∗∗ −0.003∗∗∗ −0.001∗∗∗ −0.002∗∗∗
(0.00004) (0.00004) (0.0001) (0.00003) (0.00004)
Adults 0.016∗∗∗ 0.010∗∗∗ 0.009∗∗∗ 0.055∗∗∗ 0.065∗∗∗
(0.001) (0.001) (0.002) (0.001) (0.001)
Children 0.010∗∗∗ 0.001 −0.014∗∗∗ 0.040∗∗∗ 0.088∗∗∗
(0.001) (0.001) (0.001) (0.0005) (0.001)
Single-Family Home 0.068∗∗∗ 0.085∗∗∗ −0.004 0.024∗∗∗ 0.056∗∗∗
(0.001) (0.002) (0.004) (0.001) (0.001)
(0.007) (0.008) (0.022) (0.006) (0.007)
College 0.029∗∗∗ 0.019∗∗∗ 0.045∗∗∗ −0.047∗∗∗ −0.029∗∗∗
(0.001) (0.001) (0.003) (0.001) (0.001)
Market FE’s Y Y Y Y Y
Year FE’s Y Y Y Y Y
Observations 671,131 597,542 111,945 683,671 683,567
Adjusted R2 0.094 0.084 0.041 0.137 0.171
Note: Table reports coefficients in Figure A2. Standard errors are clustered at the market level. ∗ p<0.1;
∗∗ p<0.05; ∗∗∗ p<0.01
Source: Nielsen Consumer Panel (2004–2017)
 INCOME DIFFERENCES IN BULK BUYING 59

Table A11—: Nielsen Consumer Panel Summary Statistics by Unit Price Regula-
tion

Without Regs With Regs

Variable Mean SD Mean SD
Household income ($000s) 55.65 30.75 59.60 31.70
Household size 2.53 1.43 2.61 1.49
Age 52.34 14.37 53.02 14.43
College Educated 0.37 0.48 0.41 0.49
Child present 0.33 0.47 0.32 0.47
Married 0.52 0.50 0.49 0.50
N (Household-Years) 488,461 246,263

Note: Unweighted means and standard deviations are reported.

Table A12—: Nielsen Consumer Panel Summary Statistics for Households that
Move

Reg to Reg Reg to No Reg

Variable Mean SD Mean SD
Household income ($000s) 60.03 29.14 62.32 28.60
Household size 2.17 1.20 2.24 1.20
Child present 0.18 0.39 0.17 0.38
Married 0.54 0.50 0.66 0.47
College Educated 0.56 0.50 0.60 0.49
Age 56.24 13.10 58.09 12.53
N (Household-Year) 26,393 6,027
No Reg to Reg No Reg to No Reg
Variable Mean SD Mean SD
Household income ($000s) 59.75 29.02 57.11 28.77
Household size 2.17 1.14 2.25 1.26
Child present 0.15 0.35 0.21 0.41
Married 0.64 0.48 0.58 0.49
College educated 0.52 0.50 0.55 0.50
Age 59.30 12.34 55.36 13.08
N (Household-Year) 4,260 56,059
Note: Unweighted means and standard deviations are reported.
60 INCOME DIFFERENCES IN BULK BUYING

Table A13—: Robustness Test: Warehouse Club Entry on Bulk Buying (Different
Radii)

5 Mi 10 Mi 15 Mi 20 Mi
(1) (2) (3) (4)
Post-Entry −0.005 −0.007 0.001 0.005
(0.006) (0.007) (0.008) (0.008)
Post-Entry : 25-50k 0.012∗∗∗ 0.012∗∗ 0.019∗∗∗ 0.014∗∗
(0.005) (0.006) (0.006) (0.005)
Post-Entry : 50-100k 0.014∗∗∗ 0.022∗∗∗ 0.025∗∗∗ 0.023∗∗∗
(0.004) (0.005) (0.005) (0.006)
Post-Entry : >100k 0.017∗∗∗ 0.030∗∗∗ 0.035∗∗∗ 0.043∗∗∗
(0.006) (0.008) (0.009) (0.010)
Household-Zip FE’s Y Y Y Y
Year-Qtr FE’s Y Y Y Y
Demographic Controls Y Y Y Y
Observations 2,400,344 2,401,665 2,401,038 2,400,924
Adjusted R2 0.428 0.428 0.428 0.428
Note: Table reports coefficient estimates for Equation (9) which regresses households’ quarterly bulk
purchase shares on an indicator for warehouse club entry, an indicator for whether the household shops at
a warehouse club, and an interaction term as well as household characteristics. Different distance cutoffs
defining an “entry” are used for each regression. Household-Zip code and year-quarter fixed effects are
included. Projection weights are not used. ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
Source: Nielsen Consumer Panel (2004–2015)