The document analyzes demographic diversity in online advertising by major brands on social media platforms like Instagram and Facebook. It finds that while most brand posts depict white middle-aged individuals, some brands' posts are more diverse than their actual audiences. It also finds evidence that users tend to engage more with posts depicting individuals from their own demographic groups (in-group preference), such as females engaging more with female-featured posts. However, the analysis is limited by face detection errors and only considers single-faced images from major US brands.

1. Diversity in Online Advertising:
A Case Study of 69 Brands on
Social Media
Jisun An and Ingmar Weber
QCRI, HBKU

2. In 1960,
businessmen were afraid of
using non-white model in their ads
as that might result in
loosing their customers.

3. “If you don’t target,
you cannot sell.”
By Tom Burrell
Source: https://twitter.com/tomburrell_chi

5. Minority groups are still under-represented in
the ad world
As recent as 2016, Lloyds
reported that in the UK,
just 19% of people in ads
are from minority groups.
E.g. < 1% of people featured
in ads are disabled, < 1% are
single parents.
Lloyds Bank: 'next step' campaign celebrates diversity
Source: https://www.campaignlive.co.uk/article/less-20-uk-ads-feature-minorities-says-lloyds-bank-survey/1417861

6. Researchers have studied how different ethnic
groups perceive culturally embedded ads
Whites responded similarly or more favorably to ads with non-White
models [Bush et al., 1974][Martin et al., 2004]
Integrated or all-Black casts in ads may or may not elicit a backlash
among some White college students and adults [Cagley and Cardozo, 1970]
Relative to Whites, non-Whites seem to be more aware of and
responded more favorably to ethnically resonant ads [Forehand et al.,
2002][Martin et al., 2004]
High-income Asians and Hispanics are known to prefer to see Whites
in ads [Lee and Kim, 2017]
When choosing a doctor, Kenyans prefer to meet European-looking
doctors when a condition is a serious matter [Miller et al., 2011]

7. However, these previous studies
- have mostly been conducted in a lab-experiment settings
- have focused on ads on TV or in magazines, which are both forms
of one-way interaction
Since a brand can see the users’ responses on social media directly,
social media has an advantage when it comes to understanding what
role diversity can play in ads.
Researchers have studied how different ethnic
groups perceive culturally embedded ads

8. We aim to characterize the
demographic diversity among
ads of top US brands on
social media.

9. Research Questions
1. Diversity in Online Ads
How much demographic diversity is there in online advertising in
the content put out by major brands on social media?
2. Gap between demographics of ads and users
How does the demographic diversity in a brand’s posts compare to
the diversity of their engaging audience?
3. Impact of demographics on user engagement with ads
Is there resonance between the demographics depicted in a
particular post and the audience engaging with the post?

10. Data collection - Brands on social media
We get a list of 132 brands which are US based brands and officially
appear in at least one of the two social media: Instagram and Facebook.
Source: https://www.interbrand.com/best-brands/best-global-brands/2016/ranking/#?filter=The%20Americas, http://brandirectory.com/league_tables/table/global-500-2016

11. Data collection - Social media data
We collect social media posts of those brands with metadata (including
image URL) and comments

12. Data collection - Social media data
Instagram Facebook
Initial data collection
Number of brands 82 98
Number of posts 107,678 255,935
Number of comments 15.84M 1.68M

13. Face Detection by Face++
1 face has been detected

14. Demographic Inference by Face++
Age
Gender
Ethnicity

15. Accuracy of Face detection
Using Figure Eight, a crowdsourcing platform, we label a random
sample of single-faced images with their gender (Female or Male),
ethnicity (White, AF-AM, or Asian), and age group (Minor [<=17],
Middle, Elders [>=60]).
Accuracy of inferring demographics
Gender 86.7%
Age 87.3%
Ethnicity - White 84.22%
Ethnicity - AF-AM 77.78%
Ethnicity - Asian 21.77%
We exclude “Asian” category in our analysis.
We apply the same approach for both posts and users.

16. Data collection - Social media data
Instagram Facebook
1. Posts with single face by Face++
Number of posts 14,852 (13.7%) 17,196 (6.7%)
2. Remove False-positive (e.g., Cartoon)
3. Remove “Asian” category
4. Remove brands do not have at least 50 posts, a total of at least 100 users
5. Remove brands which receives largely negative comments or complaints (e.g., airlines
and telecommunication providers)
Number of brands 40 46
Number of posts 4,877 7,426
Number of users with demographic info. 171,179 8,841

17. Research Questions
1. Diversity in Online Ads
How much demographic diversity is there in online advertising in
the content put out by major brands on social media?
2. Gap between Demographics of Ads and Users
How does the demographic diversity in a brand’s posts compare to
the diversity of their engaging audience?
3. Impact of Demographics on User Engagement with Ads
Is there resonance between the demographics depicted in a
particular post and the audience engaging with the post?

18. Diversity in Online Ads
More females (55%) 86% are White 82% are Middle-aged
Fairly balanced 90% are White 77% are Middle-aged
and 21% are minors

19. Diversity in Online Ads
Those brands with
skewed fraction of
a demographic
group are ones
with a particular
target group.

20. Research Questions
1. Diversity in Online Ads
How much demographic diversity is there in online advertising in
the content put out by major brands on social media?
2. Gap between Demographics of Ads and Users
How does the demographic diversity in a brand’s posts compare to
the diversity of their engaging audience?
3. Impact of Demographics on User Engagement with Ads
Is there resonance between the demographics depicted in a
particular post and the audience engaging with the post?

21. Gap between demographics of ads and users
For each brand, we compute the difference between the ratio of a
certain demographic group in the ads and that of the engaging users.
For example, on Facebook, for Victoria’s Secret,
- the ratio of a female group in the ads is 95%
- the ratio of a female group in the engaging users is 74%
- the gap in a female group is 95-74 = 21%
- faces in Victoria’s Secret Facebook posts are 21% more female than
the users engaging with the brand

22. Is any lack
of
diversity
simply a
reflection
of the
audience?

23. Top 5 brands
for the
difference of
fraction of
each
demographic
between posts
and users.

24. Several tech
companies
over-represent
female faces.
Or, rather, the content
they put out is gender
balanced but their
audience skews toward
male.

25. Sprite and
Gatorade (IG)
over-represent
AF-AM in their
posts and
many Whites
engage.

26. There are
brand-specific
outliers like
Pampers (IG)
Pampers mostly uses
infants wearing its
products as models for
its posts while the
engaging users are the
Middle group.

27. KFC (IG) over-
represents the
older and
White
demographics
compared to its
user base.

28. Research Questions
1. Diversity in Online Ads
How much demographic diversity is there in online advertising in
the content put out by major brands on social media?
2. Gap between Demographics of Ads and Users
How does the demographic diversity in a brand’s posts compare to
the diversity of their engaging audience?
3. Impact of Demographics on User Engagement with Ads
Is there resonance between the demographics depicted in a
particular post and the audience engaging with the post?

29. Two theories on how users are reacting to the
demographics represented in ads
In-group preferences
When a female appears in the
ad, female users tend to
engage more.
Out-group preferences
When a female appears in the
ad, male users tend to
engage more.

30. Is there any in-group preference in user
engagement?
We observe in-group preference for gender.
For Instagram posts showing a female face, the fraction of female users
commenting on this post is 13.3% higher compared to posts not
showing a female face (i.e., those showing a male face).
Mean of (absolute) percentage differences between the two fractions

31. Is there any in-group preference in user
engagement?
We observe in-group preference for ethnicity.
For Instagram posts showing a white face, the fraction of white users
commenting on this post is 4.2% higher compared to posts not showing
a white face (i.e., those showing a AF-AM face).
Mean of (absolute) percentage differences between the two fractions

32. Limitations & Discussions
Limitations:
● Face++ misclassification errors
● Solely images with a single face
● Solely US-based global brands
● Number of comments as sole signal for engagement
Potential extensions:
● Other minority groups: people with disabilities, single
mom, ..
● Personalized advertising? Personalized media?

33. Thanks!
@jisunan
@ingmarweber

34. References
[Bush et al., 1974] Bush, R.F., et al.: White consumer sales response to black models. The
Journal of Marketing 38(2), 25–29 (1974)
[Martin et al., 2004] Martin, B., et al.: The influence of ad model ethnicity and self-
referencing on attitudes: Evidence from new zealand. Journal of Advertising 33(4), 27–37
(2004)
[Cagley and Cardozo, 1970] Cagley, J.W., Cardozo, R.N.: White response to integrated
advertising. Journal of Advertising Research 10(2), 35–39 (1970)
[Forehand et al., 2002] Forehand, M.R., Deshpand´e, R., Reed II, A.: Identity salience and
the influence of differential activation of the social self-schema on advertising response.
Journal of Applied psychology 87(6), 1086–1099 (2002)
[Lee and Kim, 2017] Lee, Y.J., Kim, S.: How do racial minority consumers process a model
race cue in csr advertising? a comparison of asian and white americans. Journal of Marketing
Communications 0(0), 1–21 (2017)
[Miller et al., 2011] Miller, A.N., Kinya, J., Booker, N., Kizito, M., wa Ngula, K.: Kenyan
patients’ attitudes regarding doctor ethnicity and doctor–patient ethnic discordance. Patient
Education and Counseling 82(2), 201–206 (2011)