Investigating the Pink Tax:
Evidence against a Systematic Price Premium for Women in CPG
∗
Sarah Moshary
University of Chicago – Booth
Anna Tuchman
Northwestern University – Kellogg
Natasha Bhatia
Cornerstone Research
October 29, 2021
Abstract
The pink tax refers to an alleged empirical regularity: that products targeted toward
women are more expensive than similar products targeted toward men. This paper
provides systematic evidence on price disparities for personal care products targeted at
different genders using a national dataset of grocery, convenience, drugstore, and mass
merchandiser sales, in combination with novel sources on product gender targeting. We
do not find evidence of a systematic pink tax: women’s products are more expensive in
some categories (e.g., deodorant) but less expensive in others (e.g., razors). Further, in
an apples-to-apples comparison of women’s and men’s products with the same active and
inactive ingredients, the women’s variant is less expensive in five out of six categories.
Our results call into question the need for and efficacy of recently proposed and enacted
federal and state legislation mandating price parity across gendered products in posted
price markets.
1 Introduction
The “pink tax” refers to an alleged empirical regularity that goods marketed toward
women are more expensive than their counterparts marketed toward men. Gender-based
∗
We would like to thank Sara Drango, Malika Korganbekova, and Jason Zhao for excellent research assistance.
Researchers’ own analyses calculated (or derived) based in part on (i) retail measurement/consumer data from Nielsen
Consumer LLC ("NielsenIQ"); (ii) media data from The Nielsen Company (US), LLC ("Nielsen"); and (iii) marketing
databases provided through the respective NielsenIQ and the Nielsen Datasets at the Kilts Center for Marketing Data
Center at The University of Chicago Booth School of Business. The conclusions drawn from the Nielsen data are those
of the researchers and do not reflect the views of NielsenIQ or Nielsen. Neither NielsenIQ nor Nielsen is responsible
for, had any role in, or was involved in analyzing and preparing the results reported herein.
Researchers’ own analyses calculated (or derived) based in part on data from The Numerator Company (US), LLC
and marketing databases provided through the Numerator Datasets at the Kilts Center for Marketing Data Center
at The University of Chicago Booth School of Business. The conclusions drawn from the Numerator data are those
of the researchers and do not reflect the views of Numerator. Numerator is not responsible for and had no role in
analyzing or preparing the results reported herein.
1
pricing of consumer packaged goods (CPG) is concerning because it would exacerbate well-
documented gender inequality in the labor market.
1
Investigative journalists and government
agencies report that price differences in CPG occur most frequently for personal care prod-
ucts, such as deodorant and razors, and peg price differences in this category at 13% (e.g.,
Bessendorf 2015; Consumer Reports 2010; Duffin 2019). Policymakers are keen to address
these perceived inequalities through legislation. For example, in 2019-2020, the NY State
Assembly and State Senate passed bill S2679 that bans pricing on the basis of gender. Since
2015, Congresswoman Jackie Speier has introduced the Pink Tax Repeal Act four times,
with the goal of implementing a similar ban nationwide.
Unfortunately, there is a dearth of systematic evidence on the the pink tax to guide
legislative action. Gender-based pricing likely operates differently in consumer packaged
goods compared to other settings studied in the literature, which typically involve price
negotiation: gender pay gaps in the labor market (e.g., Blau and Kahn 2017), price premiums
in automobile sales and repairs (e.g., Ayres (1991), Ayres and Siegelman (1995), Goldberg
(1996), and Busse, Israeli, and Zettelmeyer (2017)), and more recently, disparities in real
estate transactions (e.g., Goldsmith-Pinkham and Shue 2020). In these settings, a female
customer may be unaware that she is quoted a higher price than male customers for the
same product and/or may be unable to secure a lower price because her gender is observable
to the opposite party in the transaction. In contrast, personal care products are sold in
posted price markets, where a woman can typically observe the shelf prices of products
aimed at men, and there is no rule or regulation that bars her from buying a cheaper men’s
product. To prevent arbitrage, firms must therefore differentiate the products targeted at
different genders in such a way that consumers self-select into the product designed for their
group. As an example, a soap manufacturer might sell two versions of an otherwise identical
soap, a low-priced blue bar and a high-priced pink bar. Second degree price discrimination
of this sort can be profitable for firms if men and women have different demand for soap.
Gender-based price discrimination in CPG would therefore manifest as differences in shelf
price across products that are targeted at different genders, rather than differences in prices
charged to men and women for the same exact product.
It is difficult to measure the pink tax precisely because gender segmentation in CPG
necessitates product differentiation. If men’s and women’s products differ in appearance and
content, then any price differences could reflect differences in markups or costs. To this end,
proposed and adopted legislation bans price differences for “substantially similar” products,
but does not provide clear criteria for evaluating similarity. The legislation does rule out
color as a meaningful differentiator, with an exception in cases where color generates cost
disparities. While the legislation leaves room for interpretation, these provisions indicate
1
Blau and Kahn (2017) provide a recent review of the literature.
2
that regulators seek to crack down on differential markups, while permitting differential
pricing that is driven by costs.
The goal of this paper is to assess the prevalence of gender-based pricing for CPG goods.
We begin by evaluating existing evidence on the pink tax from Bessendorf (2015), a New
York City Department of Consumer Affairs (NYC DCA) report frequently referenced in state
and national pink tax legislation. Bessendorf (2015) hand-collected prices for 122 products
at three NYC drugstores and found that women’s products were more expensive in six of
seven personal care categories. When we compare prices of the same products but extend the
analysis to include other retail formats and retail outlets across the country, we too find that
average prices are higher for women’s products in those same six categories. However, we
hesitate to generalize these findings to the rest of the personal care market for two reasons.
First, the products considered in the report account for less than 6% of category sales and
were not selected at random. Second, while the sample was constructed by manually pairing
men’s and women’s products, we find that most pairs in the sample differ in their ingredients.
We then construct our own estimates of gender price differences for a wide array of per-
sonal care products at thousands of retail outlets across the United States from 2015-2018
using Nielsen RMS data. In our first set of estimates, we do not condition on product at-
tributes. Thus, we term this difference the pink gap because it could reflect either differences
in markups and/or costs between the products targeted at men and women. Of the nine
categories in this comparison, unit prices for women’s products are higher than those for
men’s products in only four: bar soap, body wash, deodorant, and razor blades. In three of
the other categories, the unit prices for men’s products are higher than the unit prices for
women’s products, and the remaining two categories do not have significant differences in
unit prices. In other words, while average prices for men’s and women’s products differ, the
pink gap is often negative.
Price differences shrink even more when we refine the comparison to substantially similar
products, which is how current and proposed legislation conceptualize the pink tax. Following
the Pink Tax Repeal Act, we operationalize substantial similarity as products made by the
same manufacturer that contain the same leading ingredients. Constructing such an apples-
to-apples comparison is important to rule out the possibility that price differences stem from
differences in quality and therefore costs across gendered products. For example, women
might face the same prices as men but their products could be lower quality and cost less to
produce. In this case, equal shelf-prices could mask a larger markup on women’s goods. To
the contrary, when we consider within-manufacturer comparisons of products with similar
ingredients, price differences decrease further, and women’s products are more expensive than
men’s in only one of the six categories where ingredient information is observed. Furthermore,
pooling the apples-to-apples estimates across categories, unit prices for women’s products
are 5% cheaper than for men’s products. Our findings imply that the Pink Tax Repeal Act
3
is unlikely to meaningfully change average prices in personal care; we show that men and
women already face similar prices for similar products.
The paper proceeds as follows: Section 2 describes current pink tax legislation and Section
3 details the data. Section 4 provides a replication and evaluation of existing evidence on
the pink tax from the NYC DCA. Section 5 describes our preferred estimates of the pink
gap and pink tax. Section 6 discusses policy implications.
2 Current Legislation
The Pink Tax Repeal Act is the principle federal legislation aimed at combating price dis-
crimination against women in CPG. The act was first introduced in 2015 by Congresswoman
Jackie Speier, who succinctly describes the act as:
“prohibit[ing] the sale of substantially similar goods or services that are priced
differently based on gender, allow[ing] the Federal Trade Commission to enforce
violations, and ensur[ing] that State Attorneys General have the authority to take
civil action on behalf of consumers wronged by discriminatory practices.”
In practice, the Pink Tax Repeal Act defines similar products as those produced by the
same manufacturer and that have no substantial differences in (a) the materials used in
the products, (b) the intended use of the products, or (c) the “functioning and features” of
the products. The bill specifies that differences in color do not qualify as substantial. The
legislation has 48 current signatories and is endorsed by Consumer Reports, the Consumer
Federation of America, and the National Women’s Law Center.
2
To motivate the bill, Con-
gresswoman Speier cites a report by the NYC Department of Consumer Affairs (Bessendorf,
2015) that found substantial price differences between the prices of men’s and women’s prod-
ucts. The NYC DCA study also features in the Joint Economic Committee 2016 report on
the pink tax.
3
We replicate and extend estimates from this study in Section 4. Other studies
of the pink tax include contemporaneous work by Gonzalez Guittar et al. (2021), which
finds mixed results in their study of scraped price data from four online retailers, and a 2018
study by the Government Accountability Office, which finds that average prices are higher
for women’s (men’s) products in five (two) of ten personal care categories.
4
Both California and New York have passed separate legislation aimed at eradicating
the pink tax. The 1996 California Gender Tax Repeal Act bans gender-based pricing of
consumer services, such as haircuts and dry cleaning. As Part of the FY 2021 Budget Bill,
2
https://speier.house.gov/press-releases?id=C2F060D1-0D84-4824-B9E5-40F879F22CFA
3
Report available at: https://www.jec.senate.gov/public/_cache/files/
8a42df04-8b6d-4949-b20b-6f40a326db9e/the-pink-tax---how-gender-based-pricing-hurts-women-s-buying-power.
pdf.
4
The report can be found at: https://www.gao.gov/products/gao-18-500.
4
Governor Andrew Cuomo of New York included a provision expressly banning the pink tax.
Similar to the Pink Tax Repeal Act, the bill defines the pink tax as gender-based pricing for
substantially similar products. The New York ban went into effect September 30, 2020. The
bill describes exemptions when differences in prices reflect differences in cost.
5
3 Data
Retail Prices
We use Nielsen Retail Scanner data from 2015 to 2018 to document price differences
between personal care products targeted at men and women. We examine nine categories:
bar soap, body wash, deodorant, hair coloring,
6
razor blades, disposable and non-disposable
razors, shampoo, and shaving cream. The data records the price and quantity sold for
products (UPCs) sold in 39,697 stores affiliated with 93 chains across the US. The data is
recorded at the store-UPC-week level, so that prices reflect the weekly average price paid
by consumers in a particular store during a particular week. A feature of the data is that it
reflects price promotions, whereas other studies of the pink tax, such as Bessendorf (2015),
focus on the full sticker price. The data also includes product characteristics, such as brand
name and product size.
The data does not indicate the price of a product in weeks when it earns no sales at that
store. These missing prices are not problematic for our analyses that focus on average price
paid. However, we must impute prices for our analysis of shelf prices (price charged). First,
we assume that a product with zero sales in a particular week was offered at its regular (non-
discounted) shelf price. Then we impute prices based on adjacent weeks when the product
was sold, in an approach similar to Hitsch, Hortaçsu, and Lin (2021). Web Appendix A
details the algorithm that we use to construct regular (non-discounted) shelf prices.
Gender
We extract information on product-level gender-targeting from the following sources:
1. Nielsen Brand and Product Module Descriptions: We search for gendered words
such as “his” and “hers” in Nielsen’s brand description for each UPC and product module
name.
7
5
https://www.governor.ny.gov/news/governor-cuomo-reminds-new-yorkers-pink-tax-ban-goes-effect-today
6
We exclude temporary, costume hair coloring products. These account for 1.8% of category market share.
Additionally, 92% of category market share is for hair coloring products measured in counts, and the remainder is
measured in ounces. To simplify our per unit analysis, we exclude hair coloring products measured in ounces.
7
To classify men’s UPCs, we searched for the following words: his, men, clubman, hombre, man, man cave, homme,
men’s choice, men’s select, monsieur, and Mr. To classify women’s UPCs, we searched for the following words or
abbreviations: her, lady, girl, ldy, women, femme, ladies, lady’s, and wmn.
5
2. Label Insight: We collect data on gender targeting from Label Insight, a market
research firm that records marketing claims for CPG brands. The database also includes
product pictures.
8
3. Walgreens Website: We scrape gender categorizations from Walgreens.com, the
website of the large American drugstore chain. Web Appendix B.1 displays screenshots
of gender categorization and filters on the Walgreens webpage. Scraping was performed
in Summer 2020.
4. Differential purchasing by all-male and all-female households in the Nielsen
consumer panel dataset from 2006 to 2018. We identify products whose consumer
base is significantly skewed towards one gender using data from the Nielsen Consumer
Panel on the purchases of single-gender households. These households account for more
than 25% of households in the panel.
9
For each UPC, we define the female (male) share
as the percent of single-gender household purchases that are made by female (male)
households. Finally, we identify women’s (men’s) UPCs as those whose female (male)
share is significantly larger than that gender’s representation in the panel via a binomial
test where the null hypothesis is that the female (male) share is equal to 71% (29%).
10
If we do not reject the null, the product is left uncategorized. Skew in purchasing could
indicate an explicit gender cue (e.g., a label or picture), or simply an attribute with
a gender-specific appeal. Empirically, these two sets turn out to be similar but not
identical.
11
5. Hand-coding Label Insight product images. We hired undergraduate research as-
sistants at the University of Chicago and Northwestern University to categorize product
images from Label Insight. Products are categorized as male, female, unisex, or un-
known. Web Appendix B.2 describes the recruiting and labeling process in detail.
We combine these sources to construct a single gender variable. In the event of conflicts, we
prioritize the classification from the RMS brand description and break remaining ties using
majority rule or in the case of even ties, the authors’ judgement.
12
In a final step, we fill in
the gender for unclassified UPCs for which the corresponding brand or brand-size pair has
i) at least 10 UPCs in the data and ii) at least 20% of those UPCs are labeled unanimously
as a single gender.
8
The gender field is populated for 12% of the deodorants observed in the Label Insight data.
9
Female-only households are more common – they represent 71% of single-gendered households.
10
One benefit of this approach is that it does not categorize UPCs with few household purchases.
11
For example, the brand Old Spice is solely marketed towards men, but it produces deodorants in scents like Fiji
and Citron that are also purchased by women in the Nielsen Consumer Panel dataset. However, we do find that
most panelists buy products that are marketed toward their own gender. Using one person households in the Nielsen
Consumer Panel, we find that 78% of female and 81% of male panelists purchase deodorants for their own gender,
where product gender is defined using all sources in this section. We draw a similar conclusion if we exclude the
categorization using the Panelist data.
12
Conflicts between sources are rare. For example, less than 0.1% of deodorant products have a conflict.
6
Table 1: Gender Targeting across Personal Care Categories
Nielsen Product Module % Qty Gendered Total % UPCs Gendered Count of
All for Women Qty (MM) All for Women UPCs
Soap - Bar 71.7% 57.3% 523 19.1% 54.7% 3,710
Soap - Liquid 46.3% 94.2% 508 13.7% 90.6% 3,100
Soap - Specialty 78.5% 63.0% 820 26.7% 58.8% 6,889
Deodorants - Personal 99.0% 49.7% 1,059 76.6% 46.1% 2,958
Hair Coloring 100.0% 88.9% 312 100.0% 95.8% 2,534
Hand & Body Lotions 73.4% 95.4% 472 18.7% 89.2% 7,593
Razor Blades 86.5% 32.9% 96 51.4% 39.3% 519
Razors Disposable 68.2% 51.2% 289 36.5% 49.9% 978
Razors Non-Disposable 88.5% 45.8% 68 48.8% 38.2% 451
Creme Rinses & Conditioners 81.1% 99.5% 551 26.4% 93.6% 5,916
Shampoo
–Aerosol/ Liquid/ Lotion/ Powder 71.1% 74.3% 895 26.8% 68.3% 7,746
–Bars/ Concentrates/ And Creams 65.9% 89.1% 9 21.4% 89.3% 131
–Combinations 49.8% 99.9% 20 18.1% 83.0% 519
Shaving Cream 100.0% 25.2% 271 100.0% 21.9% 942
Notes: This table describes the share of products available at Nielsen RMS stores between 2015-
2018 that we record as gendered. Gender targeting is determined based on data from five sources:
gendered words in the category, brand, or product description; gender claims in Label Insight;
gender labels on Walgreens.com; differential purchasing by all-male and all-female households in
the HMS panel; and hand-coding by undergraduate research assistants at the University of Chicago
and Northwestern University.
Table 1 shows the pervasiveness of gender targeting across product modules.
13
Consistent
with the focus on personal care in the media surrounding the pink tax, we find that personal
care categories are highly gendered. Our methods assign a gender to 37% of the personal
care products in the Nielsen data, although there is considerable variation across categories.
There is also substantial variation in the share of products targeted at men vs. women; for
example, the overwhelming share of hair coloring products are targeted at women, but most
shaving creams are targeted at men. One reason for the substantial share of products with
no gender label is that it is challenging to label niche products. Market shares reveal that
gendered products account for 80% of volume sales across all categories.
14
13
Nielsen has separate product modules for men’s and women’s hair coloring and shaving creams. We extract
gender information from these categorizations and then combine the gendered product modules together.
14
We note that a relatively high share (32%) of disposable razors sold are not assigned a gender. This lower share
may be driven by the dominance of private label products (27% market share, as shown in Web Appendix B.4). To
protect the identity of retailers in the data, Nielsen masks the UPC of private label products, so we cannot map these
products to our data sources for gender.
7
Ingredients
We use Syndigo Product Label Data on product ingredients. For each UPC, the data
includes the names and amounts of any active ingredients as well as the names of inactive
ingredients and their relative prevalence in the product. The data covers 7,020 of the 25,982
personal care products in our sample of gendered products from the Nielsen data.
4 NYC Replication
In this section, we revisit evidence from Bessendorf (2015), a NYC Department of Con-
sumer Affairs study that reports a 13% pink tax in personal care. We focus on this report
because it is cited as motivation both for proposed federal legislation and existing state reg-
ulation on the pink tax. We begin by replicating the results of the report using the original
data collected by the NYC DCA for the study. This data was collected for 122 UPCs sold
in NYC drugstores from July-October 2015. Next, in order to understand whether the 13%
price difference is peculiar to New York City or represents a broader phenomenon, we extend
the scope of the analysis by examining the prices charged for these same products by a large
sample of supermarkets, mass merchandisers, convenience stores, and drugstores across the
US. We then provide evidence on the comparability of the men’s and women’s products
studied in the report.
Table 2 reports estimates of price disparities calculated following Bessendorf (2015). The
report measures the so-called “pink tax” by pairing men’s and women’s products, calculating
the within-pair price difference, averaging price differences across pairs within a category,
and then scaling by the average price for men’s products in the category. In cases where the
men’s and women’s products are different sizes, the report re-scales prices using the ratio of
sizes (e.g., by multiplying the price per oz of the men’s product by the size of the women’s
product).
15
It arrives at a 13% pink tax via a simple average across categories. Column
(6) replicates Bessendorf (2015)’s estimates using the original data collected by the NYC
DCA.
16
Based on the NYC DCA data, women’s products are more expensive in five out of
six personal care categories. Our aim is to understand whether and to what extent these price
differences extend to other stores, retail formats, and geographies. Using the Nielsen RMS
data, we estimate price differences for the set of products (UPCs) considered in the report
for three samples: drugstores in New York City, all drugstores, and all retailers. Our analysis
15
The report does not rescale prices for body wash. Because our aim is to replicate their methodology, estimates
in Table 2 do not rescale in this category either.
16
We are able to replicate all values in Table 5 in Bessendorf (2015) except the average price of razors targeted to
women. Table 5 in Bessendorf (2015) reports an average price of $8.90 for women’s razors, while we find an average
price of $8.73. This difference in price leads them to report a price gap of 11% while we compute a price gap of
9%. We believe the discrepancy is most likely due to a typo in the product-level price data reported in Bessendorf
(2015)’s appendix or a mistake in computing the averages reported in their summary table.
8
Table 2: Replication and Extension of NYC DCA Report Pink Tax Estimates
(1) (2) (3) (4) (5) (6) (7)
Estimate Men’s Price Pink NYC Report Estimates
Category Channels Geographies ($) ($) Gap Reported Nielsen UPCs
Body Wash Drugstores NYC Only 0.45*** 5.73 7.9% 5.5% 5.5%
Drugstores National 0.54*** 4.85 11.1%
All National 0.74*** 4.53 16.4%
Deodorant Drugstores NYC Only 0.06*** 5.15 1.1% 3.3% 4.0%
Drugstores National 0.31*** 4.27 7.2%
All National 0.40*** 3.90 10.1%
Hair Care Drugstores NYC Only 2.35*** 7.88 29.9% 47.7% 29.7%
Drugstores National 0.80*** 6.38 12.6%
All National 0.22*** 5.09 4.3%
Razor Drugstores NYC Only 0.78*** 10.60 7.4% 9.3% 15.2%
Drugstores National 1.53*** 8.51 18.0%
All National 1.18*** 8.54 13.9%
Razor Cartridges Drugstores NYC Only 2.59*** 15.34 16.9% 10.9% 15.4%
Drugstores National 2.11*** 14.06 15.0%
All National 2.21*** 14.09 15.7%
Shaving Cream Drugstores NYC Only -0.48*** 4.09 -11.7% -4.1% -13.0%
Drugstores National -0.38*** 3.67 -10.5%
All National -0.35*** 3.46 -10.0%
Notes: The pink tax is measured as the ratio of the estimated price difference (column (3)) to the average
price of a men’s product in the same category (column (4)) multiplied by 100. Columns (6) and (7) present
estimates of the pink tax using the NYC DCA data, where column (7) subsets to the products that can
be matched to the Nielsen data. These prices of these matched products in the Nielsen data comprise the
sample in columns (3)-(5).
9
excludes 33 of the products in the NYC DCA sample (27 of them are private label products)
because we cannot match them to a product observed in the Nielsen data.
17
However, we
do not believe this substantively affects our estimates of price differences; column (7) shows
that this subset of UPCs produces similar estimates of the pink gap in the NYC DCA data.
Column (3) reports average price difference in dollars for different samples, and column (5)
reports the implied pink tax. The estimates echo Bessendorf (2015) in that five of the six
categories feature a price premium for women’s products.
We next consider the generalizability of these estimated price gaps beyond the prod-
ucts studied in Bessendorf (2015). The question of extrapolation is important because the
products in the sample comprise less than 6% of category sales and were not selected at ran-
dom.
18
Rather, the sample was constructed by manual identification of men’s and women’s
products that were perceived to be comparable. Correctly constructing an apples-to-apples
comparison is important to ensure that estimated price differences do not reflect differences
in marginal cost and also to evaluate proposed legislation, which mandates price parity only
in instances where men’s and women’s products are substantially similar. The report does
not provide its criteria for comparability, and perusal of product pairs included in the report
reveals salient differences: as an example, in two of eight shampoo comparisons, the price
of a single 2-in-1 men’s product is compared to the combined price of a women’s shampoo
and a women’s conditioner, producing price gaps over 100%. To provide systematic evidence
on the similarity of product pairs, we leverage data from Syndigo on product ingredients.
Table 3 reports the number of pairs in each category with matching ingredients. The crite-
ria for matching ingredients becomes more stringent from left to right in the table; column
(3) reports the number of pairs with the same active ingredient (which is only relevant in
certain categories), column (4) reports the number with the same active and first inactive
ingredients, etc.
19
Only one-third of product pairs comprise the same leading ingredients.
We note that the challenge of identifying similar products is compounded by the challenge of
identifying gender targeting. The NYC DCA report includes comparisons between explicitly
labeled men’s products and unisex products in cases where no women’s product could be
identified. These issues of comparability in Bessendorf (2015) hamper interpretation of the
price difference estimates in Table 2 as a pink tax. It is unclear whether the estimates reflect
differences in the attributes of men’s and women’s products or differences in the mapping
from attributes to prices for men’s and women’s products (i.e., markups) and, more broadly,
whether the 122 products considered are representative of personal care. In the next sec-
tion, we describe our preferred methodology for calculating price disparities in personal care
17
Because the comparison studies only within-pair prices, we lose a further 9 products that do not have a match
in our sample.
18
Appendix table A2 shows the market share of products in the NYC DCA sample by category.
19
The FDA mandates that active ingredients are reported first, then inactive ingredients in descending order of
predominance, and then any order is permitted for ingredients that comprise less than 1% of the product. [https:
//www.fda.gov/cosmetics/cosmetics-labeling-regulations/cosmetics-labeling-guide#clgl]
10
Table 3: Similarity of Product Ingredients for NYC DCA Report Product Pairs
Product N N Pairs w/ N Pairs Matching Up To
Category Pairs Active Active Inactive 1 Inactive 2 Inactive 3 Inactive 4 Inactive 5
Body Wash 9 0 - 9 7 7 5 2
Deodorant 9 9 9 9 9 6 6 6
Hair Care 6 2 1 5 3 2 2 2
Shaving Cream 6 0 - 6 4 4 1 0
Total 30 11 91% 97% 77% 63% 47% 33%
Notes: Column (1) reports the number of product pairs that we could identify in the Nielsen data
and column (2) reports the number of pairs that have an active ingredient. The remaining columns
report the number of pairs that match up to and including that ingredient. For example, the
last column reports the number of pairs that match on active ingredient and the first five inactive
ingredients.
that leverages a national dataset of prices and product ingredients, as well as a data-driven
approach to identifying gender targeting.
5 Measuring Price Disparities
We estimate the unconditional gender price gap as the difference in the average price of
products targeted at men and women across retail outlets from 2015-2018. This specification
includes all gendered products and so reflects both differences in the attributes of products
targeted at men and women and differences in the mapping between attributes and prices
for men’s and women’s products. To the extent that firms select the attributes of gendered
products to segment consumers, these price differences can therefore incorporate both second
and third degree price discrimination.
Our main specification models the price of product j sold by retail outlet s in year t,
p
jst
, as a function of its intended gender target, year fixed effects (Γ
t
), and retail outlet fixed
effects (Ω
s
). These fixed effects capture determinants of price that vary across location or
over time and are important to the extent that different types of stores offer a larger or
smaller assortment of men’s and women’s products. Our specification is
p
jst
= β · women
j
+ Γ
t
+ Ω
s
+ ε
jst,
(1)
where the object of interest, β, is the coefficient on an indicator for whether product j is
targeted at women, women
j
. This coefficient captures the national pink gap. We estimate
11
equation (1) separately for each personal care product module.
20
In contrast with Bessendorf (2015), we find that women face lower average shelf prices
in a majority of personal care product modules. Specifically, we estimate equation (1) with
shelf price per product as the dependent variable, where observations are weighted by the
number of weeks a product is available in a given store and year in order to mirror the
assortment available to consumers. Results are displayed in column (1) of Table 4. Bar
soap, body wash,
21
and deodorant products targeted at women are more expensive than their
counterparts targeted toward men, but hair coloring, razors and razor blades, shampoos, and
shaving creams targeted at women are less expensive.
A key difference between these estimates of the pink gap and those presented in Bessendorf
(2015) is the treatment of product size. If women’s products tend to be smaller than men’s
products, then lower product shelf prices might obfuscate higher per-unit prices. To address
this concern, we repeat this exercise using unit shelf price as the dependent variable (i.e. price
per ounce or count, depending on the category). Results are reported in column (2): on a
per unit basis, women’s (men’s) products are more expensive in four (three) of nine product
categories. The remaining two categories do not have statistically significant differences in
unit prices. While the split of categories seems fairly even between those where women fare
better/worse, we note that the magnitudes of price differences are larger in the categories
where women’s products are more expensive. For example, the average unit price of women’s
bar soap is 110% more expensive than the average unit price of men’s bar soap, while the
average unit price of women’s shaving cream is 17% less expensive than the average unit
price of men’s shaving cream.
We next consider the difference in the price paid for men’s and women’s products. This
measure gives a sense for how differences in offered prices translate to differences in expen-
ditures and whether price differences create an economically significant burden for women.
This analysis is novel in the pink tax literature. Of course, we must keep in mind that
higher prices paid do not per se indicate a higher burden if they are offset by higher qual-
ity products. We also recognize that the impact of high shelf prices for seldom-purchased
products could still be large if women prefer those products but cannot afford them. These
two possibilities are explored further below. Column (3) of Table 4 shows the difference in
the average unit price paid for women’s and men’s products, which we obtain by estimating
equation (1) using quantity sold as regression weights. We interpret these quantity-weighted
regressions as evidence on the prices that men and women pay for personal care products
because we find that women seldom purchase products targeted at men and vice versa; as an
example, 78% of women and 81% of men that participate in the Nielsen panel only purchase
20
We restrict our analysis to product modules that are not dominated by a single gender and modules with sufficient
sales volume. Specifically, we focus on modules in which neither gender accounts for more than 90% of gendered
volume sales and that have at least 10 million units sold.
21
Called “(Soap - Specialty)” in the Nielsen RMS data.
12
Table 4: Price Gap by Category, 2015-2018
(1) (2) (3) (4) (5)
Product Unit Unit Unit Unit
Module Shelf Price Shelf Price Price Paid Shelf Price Shelf Price
Bar Soap 0.37*** 0.25*** 0.22*** 0.28*** 0.03***
(0.04) (0.00) (0.00) (0.01) (0.00)
4.31 0.23 0.21 0.23 0.23
8.7% 110.3% 109.0% 122.6% 14.0%
Body Wash 0.73*** 0.18** 0.14** 0.11*** -0.02***
(0.06) (0.03) (0.02) (0.01) (0.00)
4.57 0.28 0.26 0.28 0.28
16.0% 64.3% 53.1% 40.5% -6.5%
Deodorant 0.29*** 0.50*** 0.44*** 0.50*** -0.07***
(0.02) (0.02) (0.02) (0.01) (0.00)
4.62 1.51 1.38 1.46 1.46
6.3% 33.1% 31.6% 34.2% -4.5%
Hair Coloring -0.81*** -0.52** -0.71** -0.69** 0.21
(0.03) (0.13) (0.17) (0.13) (0.28)
8.78 8.48 8.07 8.78 8.78
-9.2% -6.1% -8.8% -7.9% 2.4%
Razor Blades -4.14*** 0.70*** 0.59***
(0.31) (0.04) (0.05)
22.02 3.67 3.64
-18.8% 18.9% 16.2%
Razors Disposable -0.62*** -0.18* -0.17**
(0.04) (0.06) (0.03)
7.72 2.26 2.05
-8.1% -7.9% -8.1%
Razors Non-Disposable -1.05** -1.05** -0.95**
(0.21) (0.21) (0.25)
11.69 11.69 11.18
-9.0% -9.0% -8.5%
Shampoo -1.09** 0.00 0.02 -0.01 -0.06**
(0.20) (0.01) (0.01) (0.01) (0.01)
6.33 0.49 0.43 0.48 0.48
-17.2% 0.6% 4.5% -1.6% -12.1%
Shaving Cream -0.42*** -0.10*** -0.07*** -0.08*** -0.04***
(0.04) (0.01) (0.01) (0.01) (0.01)
3.60 0.59 0.51 0.54 0.54
-11.6% -17.4% -13.6% -14.2% -8.0%
Data All All All Syndigo Syndigo
Ingredients FE N N N N Y
Manufacturer FE N N N N Y
Notes: The sample in columns (1)-(3) comprises the full set of products. Column (4)-(6) subset
to products with observed ingredient and manufacturer information. For each category, the first
row reports the average price gap and the second row reports the standard errors in parentheses
(clustered at the store and year level). The third row reports the average price of men’s products.
The fourth row reports the percentage price gap, calculated as the ratio of row one to row three.
Regressions are estimated separately by product module and include store and year fixed effects.
Columns (4)-(5) exclude razors because their ingredients are not reported. *** p < 0.01, ** p <
0.05, * p < .01.
13
deodorants targeted at their own gender.
22
The estimates imply that women pay higher
prices for bar soap, body wash, deodorant, and razor blades on a per unit basis, while men
pay higher prices for hair coloring, disposable and non-disposable razors, and shaving cream.
In seven of the nine product categories, the magnitude of the differences in price paid is
smaller than the magnitude of the differences in price charged.
We now return to the question of comparability between men’s and women’s products.
The estimates in columns (1)-(3) show that women’s products are less expensive than men’s
products in several categories, but they do not rule out differential markups for women’s
products in those same categories. In other words, the results so far do not preclude the pos-
sibility that women’s products are lower quality, but are sold at higher markups. Recognizing
that the features of men’s and women’s products may differ, proposed legislation mandates
price parity specifically in instances where men’s and women’s products are substantially
similar and made by the same manufacturer; it is price differences for these comparable
pairs that legislation defines as the pink tax. The legislation provides only loose guidance
on how to determine substantial similarity, so we assess similarity on the basis of product
ingredients. The law does explicitly restrict to within-store price comparisons, limiting its
scope. Table 5 describes the extent of overlap in manufacturer-formulations across genders
within store.
23
As an example, Table 5 shows that the average retail outlet carries 32 unique
deodorant formulations targeted to women and 34 targeted to men. Approximately 28% of
formulations targeted to women have a comparable formulation targeted to men within the
same store and vice versa. There is substantial variation across categories: hair coloring
formulations have almost no overlap across genders, and deodorants have the most overlap.
Turning to unique UPCs, in the average store less than half of gendered UPCs have a com-
parable product targeted at the other gender. These patterns indicate that legislation like
the Pink Tax Repeal Act would not apply to most personal care products.
We estimate the pink tax for substantially similar products by incorporating data on
manufacturer and product ingredients into equation (1). This exercise subsets to the sample
of Nielsen products that merge to the Syndigo data on product ingredients. This merged
sample is similar to the full set of Nielsen products in terms of the unconditional pink gap,
as shown in column (4) in Table 4, which replicates the regression in column (3) for products
with ingredient information. Column (5) displays price gaps when we add fixed effects for
22
These statistics are based on the purchase of products where gender is defined using all data sources, but the same
patterns hold if we rely only on explicit gender targeting that excludes the panelist data.
23
This exercise is conducted for the subset of products with Syndigo ingredient data. A product formulation is
defined as the combination of manufacturer, active ingredient and the first five inactive ingredients, where the order of
ingredients matters. In using this subset of the ingredients, our goal is to identify products that are similar in overall
formulation without requiring otherwise similar products to be identical on more cosmetic features. We conduct
the analysis on 2018 data to ensure similar formulations were available at the same point in time, and we exclude
convenience stores from this analysis because these stores carry very limited assortments.
14
Table 5: Overlap in Manufacturer-Ingredients Across Genders, 2018
(1) (2) (3) (4) (5) (6)
Module Gender N Formula % Formula N UPCs % UPCs Unit Sales % Sales
Bar Soap men 9 14.5% 19 14.9% 962 11.6%
women 12 12.0% 24 13.9% 1,239 15.0%
Body Wash men 19 16.3% 32 16.5% 1,278 23.3%
women 38 8.5% 56 15.9% 1,973 25.6%
Deodorants men 34 28.1% 85 30.8% 3,213 25.5%
women 32 28.5% 80 47.8% 2,817 44.1%
Hair Coloring men 7 0.1% 15 0.1% 243 0.0%
women 32 0.0% 142 0.0% 1,514 0.0%
Shampoo men 16 25.6% 29 30.3% 791 31.7%
women 49 7.5% 73 18.0% 1,884 23.0%
Shaving Cream men 12 12.5% 21 13.5% 971 13.1%
women 6 28.3% 9 37.6% 290 43.5%
Notes: Columns (1), (3), and (5) report the number of unique formulations, number of UPCs,
and the unit sales for the average store in 2018. We define a formulation as the combination of
manufacturer, active ingredient, and top five inactive ingredients. Column (2) reports the fraction
of formulations targeted to one gender for which there is a comparable formulation targeted to the
other gender. Column (4) reports the fraction of UPCs targeted to one gender for which there is a
comparable formulation targeted to the other gender. Column (6) reports the fraction of unit sales
for one gender’s products for which there is a comparable formulation targeted to the other gender.
The analysis is conducted on the subset of products with ingredients information in the Syndigo
data and convenience stores are dropped because they have very small assortments.
15
the manufacturer, active ingredient, and the first five inactive ingredients. The gender price
gap shrinks towards zero in all but one product category. Controlling for manufacturer and
ingredients, women face lower prices in four of the six product categories; men face higher
prices for body wash, deodorant, shampoo, and shaving cream, while women face higher
prices for bar soap.
To compute the overall price difference for personal care products, we re-estimate equation
(1) stacking the data across categories, weighing categories by assortment size, and using
log price per unit as the dependent variable. Unconditional on attributes, unit prices for
women’s products are 18% more expensive than unit prices for men’s products (SE = 0.001,
p < 0.01), but conditional on manufacturer and ingredients, unit prices for women’s products
are on average 5% cheaper (SE = 0.01, p < 0.05).
Taken together, the estimates in Table 4 do not support the hypothesis that women
systematically face and pay higher prices in personal care. Rather, they tell a more complex
story: there are economically and statistically significant price differences across genders,
but the direction of these differences varies across product categories. Furthermore, price
differences shrink when we look within substantially similar products, which is the focus
of current legislation. These findings contrast markedly with popular press reporting on
the pink tax and highlight the importance of leveraging scanner data to provide systematic
evidence on pricing across a wide array of products, retailers, and geographies.
6 Discussion
Gender inequalities in the labor market have spawned a substantial literature in economics
and a spate of federal and state regulations.
24
In this paper, we show that recent concern that
gender price discrimination extends to personal care products is unfounded. We leverage a
national dataset of prices and sales of grocery, convenience, drug, and mass merchandise retail
outlets coupled with detailed data on product gender-targeting and ingredients to shed light
on the pink tax. We begin by evaluating existing evidence on the pink tax from Bessendorf
(2015), which compares 122 seemingly similar men’s and women’s products sold in NYC
drugstores. Focusing on the same set of products considered in the report, we find similar
price differences using a national panel of retail chains. However, we find that the men’s and
women’s products in the study differ in their ingredients, hampering interpretation of the
reported price differences as a pink tax. Further, the products in the report represent only
a small share of category sales and exclude many big national brands.
We construct our own measure of price differences to provide systematic evidence on
gender-based pricing in CPG. First, we estimate the difference in the average price of men’s
and women’s products within-store and time period. We term this comparison the “pink gap”
rather than the “pink tax” because it may confound differences in markups with differences
24
E.g., the Equal Pay Act of 1963.
16
in marginal costs, stemming, for instance, from differences in product attributes. We find
that the pink gap is often negative; men’s products command higher per-product prices in
six of nine categories that we study and higher unit prices in three of nine categories. We
then estimate the pink tax via an apples-to-apples comparison of products manufactured
by the same firm and comprising the same leading ingredients. Controlling for attributes
shrinks price differences in all but one category. Further, men’s products are more expensive
in four of six categories when we control for ingredients. Taken together, our findings do not
support the existence of a systematic price premium for women’s products, but our results do
reveal that gender segmentation in personal care is pervasive and operates through product
differentiation. A back-of-the-envelope calculation implies that the average household would
save less than 1% by switching to substantially similar products targeted to a different gender.
The potential savings are much larger – on the order of 20% – if a household were willing
instead to substitute to products with different gender-targeting and different formulations.
25
However, a revealed preference argument suggests that such switching would lower consumer
welfare.
The implications of our findings for current and proposed legislation are several. First,
our finding that women’s personal care products are not systematically more expensive calls
into question the role of government intervention to reduce the pink tax. We acknowledge
that our findings speak to average price differences, which may mask instances of a particular
retail outlet pricing in a way precluded by the Pink Tax Repeal Act. As an example, if one
store sets a 5% higher price for the men’s version of a product and a neighboring store sets
a 5% higher price for the women’s version, we would detect no gender price gap, but the
Pink Tax Repeal Act would require that both retailers change their pricing policy or alter
their product assortments. However, our analysis reveals that most women’s products do not
have a men’s analog sold in the same retail store, limiting the scope for such adjustments.
Even in cases where a retailer does sell a women’s variant at a higher price than its men’s
analogue, the Pink Tax Repeal Act might induce the retailer to drop the men’s variant, de
25
To approximate household savings, we first compute the dollar spending, average price, and total volume (mea-
sured in ounces or counts) of purchases made by each Homescan household for each product category/gender com-
bination analyzed in Table 4. Next, for each household/category/gender, we construct the counterfactual price a
household would pay if they were willing to switch to the cheaper gender within each product category. We do this
by adjusting the household’s price paid for the more expensive gender by the estimated price gaps reported in Table 4.
When estimating savings from switching to a comparable formulation, we use the estimates in column (5), and when
estimating savings when switching across formulations, we use the estimates in column (2). We then compute the
household’s counterfactual personal care spending by multiplying the counterfactual prices by the observed purchase
volumes and summing across categories. When estimating savings from switching to a comparable formulation, we
also need to account for whether a household’s purchases actually have a formulaic analog that is targeted to the
other gender. We do so by multiplying each household’s category-level purchase volumes by the fraction of each
gender’s unit sales that have a comparable formulation on the shelf in the average store (column (6) of Table 5). The
estimated savings from switching within formulation across gender (0.9%) are much lower than the potential savings
from switching across formulations (20%) both because most purchases don’t have a comparable formulation offered
to the other gender in the same store, and because the price gap within a formulation is substantially smaller than
the price gap unconditional on formulation.
17
facto increasing price dispersion by setting an infinite price.
Finally, while our findings do not support the existence of a pink tax as conceived by reg-
ulators, an economist might define the pink tax differently, perhaps as systematic differences
in markups across men’s and women’s products. This alternative definition departs from
the current and proposed legislation in that it classifies the following two cases as forms of
the pink tax: (1) if men’s products are cheaper to produce than their women’s counterparts,
e.g., due to economies of scale and (2) if markups for unique-to-men products are system-
atically lower than markups for unique-to-women products. We think the first scenario is
unlikely because, as shown in Table 5 column (3), we do not observe systematic differences in
assortment sizes that would engender differences in scale across genders in most categories.
Regarding the second point, given our finding that similar men’s and women’s products have
similar prices, we question whether differential markups for gender-unique products are rele-
vant to the Pink Tax debate.
26
One might see differential markups for differentiated goods as
unfair if they are sustained through frivolous or spurious attributes, as discussed in Shapiro
(1982) and Bronnenberg et al. (2015). On the other hand, retailers and manufacturers may
set markups in personal care to reflect differences in product performance that bring real
value to customers, employing a commonplace pricing strategy that is perhaps no different
than matinee pricing at movie theaters or early-bird specials at restaurants.
26
We note that the debate over the tampon tax is related but distinct in that the outcry surrounds differential tax
rates for tampons compared to other necessities rather than the markups for tampons.
18
References
Ayres, Ian (1991). “Fair Driving: Gender and Race Discrimination in Retail Car Negotiations”. In:
Harvard Law Review 104.4, pp. 817–871.
Ayres, Ian and Peter Siegelman (1995). “Race and Gender Discrimination in Bargaining for a New
Car”. In: American Economic Review 85.3, pp. 304–321.
Bessendorf, Anna (2015). From Cradle to Cane: The Cost of Being a Female Consumer. Tech. rep.
NYC Consumer Affairs.
Blau, Francine D. and Lawrence M. Kahn (2017). “The Gender Wage Gap: Extent, Trends, and Ex-
planations”. In: Journal of Economic Literature 55.3, pp. 789–865. doi: 10.1257/jel.20160995.
url: https://www.aeaweb.org/articles?id=10.1257/jel.20160995.
Bronnenberg, Bart J. et al. (2015). “Do Pharmacists Buy Bayer? Informed Shoppers and the Brand
Premium”. In: The Quarterly Journal of Economics 130.4, pp. 1669–1726. issn: 0033-5533. doi:
10.1093/qje/qjv024. eprint: https://academic.oup.com/qje/article-pdf/130/4/1669/
30637328/qjv024.pdf. url: https://doi.org/10.1093/qje/qjv024.
Busse, Meghan R., Ayelet Israeli, and Florian Zettelmeyer (2017). “Repairing the Damage: The
Effect of Gender and Price Knowledge on Auto-Repair Price Quotes”. In: Journal of Marketing
Research 54.1, pp. 75–95.
Consumer Reports (2010). Men Win the Battle of the Sexes. (Visited on 2010).
Duffin, Karen (2019). The Problem with the Pink Tax. (Visited on 12/06/2019).
Goldberg, Pinelopi Koujianou (1996). “Dealer Price Discrimination in New Car Purchases: Evidence
from the Consumer Expenditure Survey”. In: Journal of Political Economy 104.3, pp. 622–654.
Goldsmith-Pinkham, Paul and Kelly Shue (2020). “The Gender Gap in Housing Returns”. In: WP.
Gonzalez Guittar, Stephanie et al. (2021). “Beyond the Pink Tax: Gender-Based Pricing and Dif-
ferentiation of Personal Care Products”. In: Gender Issues.
Hitsch, Gunter, Ali Hortaçsu, and Xiliang Lin (2021). “Prices and Promotions in U.S. Retail Mar-
kets”. In: Working Paper.
Shapiro, Carl (1982). “Product Differentiation and Imperfect Information: Policy Perspectives”. In:
FTC Working Paper 70.
19
A Imputing Regular Price in the RMS Data
The Nielsen RMS data does not record the shelf price of a product (UPC) at a store in
weeks when that store does not sell any units of that product. We impute missing prices
using an algorithm that is motivated by the insititutional practice that retailers rarely change
their regular shelf price for a product, and instead create short-term variation in prices by
running temporary price promotions that discount off the regular price. Motivated by these
institutional pricing practices, we use prices of the same product at the same store location
in recent weeks to construct a “regular” price series, i.e., the price that would have been
charged if no discounts were available that week. We operationalize this approach by setting
the regular price to be equal to the maximum price observed in the current, preceding, and
subsequent 4 weeks. In any weeks with an unobserved price, we then set price equal to the
regular price. This is based on the intuition that zero-sales weeks are most likely to occur
when the product is not on discount.
B Gender-Targeting Data Sources
B.1 Walgreens
We extract gender information from the Walgreens website. The website explicitly cate-
gorizes certain product categories by gender. Figure A1 (a) presents one such example for
the Deodorant & Antiperspirant category. We also collect gender information from search
result page gender filters, as in Figure A1 (b).
20
Figure A1: Walgreens.com Gender Categorizations
(a) Primary Gender Classification
(b) Gender Filter on Search Results Page
Notes: Screenshots taken on Walgreens.com on September 1, 2020.
B.2 Hand-Coding Product Images
We recruited three undergraduates at the University of Chicago to assign gender labels
to 1,302 personal care product images from Label Insight. The research assistants were
selected based on their performance on a 25 image training dataset, where their answers
were compared to our own hand coding. RAs were directed to a webapp (https://task.
shinyapps.io/classify-products/) on September 1, 2020. Figure A2 provides snapshots
of the webapp. We take the modal gender label across the three RAs; we do not record a
gender label in the instances where all three RAs disagreed on their classification.
21
Figure A2: Webapp for Gender Classification
(a) Instructions
(b) Task
22
B.3 Panelist Purchases
The 2006-2018 Nielsen panelist data provides additional information on gender targeting.
Intuitively, we aim to infer a product’s intended gender target based on a significant skew
in purchasing toward men or women. Because the data does not include the identity of
the household member who purchases or consumes a product, we focus on single-gendered
households for this analysis. These households comprise approximately 30% of households
in the data: 14,421 all-women and 37,569 all-men households. For each product, we label
products as targeted at women (men) if the share of purchases from all-women (all-men)
households is significantly higher than would be expected from their preponderance in the
data. Formally, we treat the number of single-gendered households that purchase an item
as the number of trials in a binomial distribution, where the number of all-women (all-men)
households that purchase is the count of successes. The null hypothesis in our binomial test
is a one-tailed test that all-men and all-women households are equally likely to purchase the
product. A product is determined to be targeted at women (men) if the null is rejected at
the 5% level. This approach categorizes approximately 247,358 products (including, but not
limited to, personal care). It is particularly helpful for products in early years in the sample
and for products that use non-verbal cues to signal gender, such as brands like Old Spice,
Secret, and Axe.
B.4 Prevalence of Private Label Products by Personal Care Category
Because Nielsen masks the UPC of private label products, we cannot identify gender
targeting for these products, except through the Homescan panelist approach described in
Appendix B.3. To give a sense for the importance of private label products in the personal
care market, Table A1 summarizes the market share of the store brand across categories.
The market shares are modest overall, with the exception of disposable razors where private
label products hold a 27% market share. We acknowledge this limitation for this category.
Table A1: Market Share of Store Brand by Product Module
Nielsen Product Module Store Brand Share
Soap - Bar 4.15%
Soap - Liquid 21.96%
Soap - Specialty 7.75%
Deodorants - Personal 0.03%
Hair Coloring 1.05%
Razor Blades 9.49%
Razors Disposable 26.75%
Razors Non-Disposable 6.98%
Creme Rinses & Conditioners 0.62%
Shampoo-Aerosol/ Liquid/ Lotion/ Powder 2.40%
Shampoo-Bars/ Concentrates/ And Creams 11.66%
Shampoo-Combinations 0.44%
Total 8.09%
23
Table A2: Market Share of UPCs Studied in the NYCDCA Report
(1) (2)
Category UPCs Market Share Brands Market Share
Bodywash 4.1% 32.2%
Deodorant 5.3% 35.6%
Razors 12.5% 23.2%
Shampoo 2.5% 20.7%
Shaving Cream 19.7% 48.8%
Total 5.8% 30.3%
C Additional Details on New York City Report Replication
In Section 4, we analyze the prices of products included in Bessendorf (2015) using the
Nielsen RMS data. Table A2 reports the market share of these products in the Nielsen RMS
data. As shown in column (1), across all categories, the share is modest, ranging from 4.1%
of body wash sales to 19.7% of shaving cream sales. These figures indicate that the sample
of products omits much of the personal care product landscape. This concern is amplified
because the sample was not selected at random. For example, the sample omits products
from some of the most popular brands because they are produced by a manufacturer uses
different brand names for their men’s and women’s products(e.g. P&G’s Secret and Old
Spice brands). Column (2) reports the combined market share of brands represented in the
sample, which is less than 50% for all categories.
Replicating and extending the NYC DCA analysis using the Nielsen data requires identi-
fying the UPCs of the products in the survey, which are described on page 65 of the report.
We proceed in three steps:
1. Google search for product names and descriptions. We discern the UPC from images
of the back of products or from Amazon and Walmart third-party sellers. We used our
best judgement in cases where product descriptions are vague.
2. For UPCs recovered in step 1, we merge to the Nielsen data using the full UPC or
alternatively the UPC without the check digit. We remove any candidate matches
where the Nielsen and NYC DCA report product descriptions conflict on size or brand.
This left 76 matches between the NYC report and the Nielsen data.
3. For the thirteen remaining UPCs in the NYC DCA report without a match, we search
for the product directly in the list of products sold in NYC drugstores in the Nielsen
data.
We also emulate Bessendorf (2015) in the construction of prices using the following steps:
• In comparisons where a men’s 2-in-1 shampoo and conditioner is compared to two
women’s products, a shampoo and a conditioner, we collapse the latter into a single
observation. This requires filtering to stores and years that have both the shampoo
and conditioner for a given year.
24
• For product pairs where the women’s and men’s products are different sizes, we create
an “equivalent price” that is the max size within a pair multiplied by each product’s
unit price. Because the report does not re-scale for body wash products, we do not
rescale in the body wash category.
We estimate price disparities via regressions of equivalent price on an indicator for whether
the product is targeted at women. The estimates include store, year, and product-pair fixed
effects.
25
