We study the "privacy-for-content" tradeoff using the 2019 YouTube COPPA settlement.
"COPPAcalypse? The YouTube settlement’s impact on kids content” w/ @TesaryLin, James Cooper, & Liang Zhong
➡️ ssrn.com/abstract=44303… 1/9
Data sharing increases ad revenue, which pays for free content, & helps personalize websites to better find the content we want. On the other hand, people want more privacy online: especially for kids.
The YouTube settlement shows the consequences of strict privacy regulation. 2/
In Sept. 2019, YouTube paid a record $170M to settle charges it violated children’s privacy law (COPPA). Beginning Jan. 2020, YouTube identified kids content and eliminated all related personalization including: personalized ads, search, content recommendations, & commenting. 3/
YouTube creators worried these changes amounted to the 💣"COPPAcalypse"💣. YouTube is a top content destination for children. Made-for-kids (MFK) content creators planned to create less content, exit, or pivot away from MFK content. 4/
We study 5,066 top U.S. YouTube channels by comparing child-directed content creators (& mixed content creators) to their non-child-directed counterparts using a difference-in-differences design.
Below, we see a clear drop-off in total child-directed content creation in 2020. 5/
Our analysis shows child-directed content creators produce 13% less content (below) and pivot towards producing non-child-directed content.
Ad price data is not public, but one creator shared with us that ad prices fell 73% without personalized ads. 6/
On the demand side, views of child-directed channels fell by 22% even though viewers could substitute to other channels or previously released content.
The drop comes despite a moderate & transient bump in March 2020 that coincides with COVID lockdowns. 7/
Consistent with the YouTube's degraded capacity to match viewers to content, we find that content creation (L) and especially content views (R) become more concentrated among the top child-directed (and mixed) YouTube channels. 8/
Pres. Biden’s SOTU address called for a ban on personalized ads for kids & EU’s Digital Services Act bans this for kids <18.
The social benefit of kids’ privacy may exceed the cost to content supply & demand, but regulators should keep this in mind. END
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Regulators & researchers seek to balance privacy & the data economy. The EU’s #GDPR is a landmark & influential regulation that defines personal data expansively. GDPR establishes:
-rules for data processing,
-rights for EU residents,
-responsibilities for firms, &
-BIG fines. 2/
#GDPR is hard to study:
A) Finding a suitable control group is hard because the GDPR had global spillovers. E.g., it affects EU firms & non-EU firms serving EU residents.
B) GDPR can screw with personal data: e.g., you may only see data from consenting users. 3/
🧵Explainer for the Topics API 🧵
Google announced the Topics API for Privacy Sandbox🏖️. Topics is basically FLoC v2.0. Google is deftly replacing FLoC v1 with a more anodyne technology and name...
Details: developer.chrome.com/docs/privacy-s… 1/8
Topics allows for interest-based ads without 3rd party cookies. Most research, including our own, shows ad prices are 2-3X higher with cookies.
Put concretely: Interest-based ads help fund content that is socially valuable, but uninteresting to advertisers. 2/
In Topics, the browser classifies begins by classifying the sites that users visit into topics from a list of ~350 readable & benign topics like cats🐈 or hockey🏒.
To do so, sites must opt in and users can opt out. 3/
Thread explaining FLEDGE (formerly TURTLEDOVE).
Online advertising generates value for publishers, advertisers, & users. Now, Google & others are proposing alternatives that preserve this value while better protecting user privacy under the "Privacy Sandbox" proposals. 1/12
The public discussion of #PrivacySandbox is dominated by #FLoC, but many tech solutions are required to satisfy advertising use cases while protecting privacy. In particular, #FLEDGE propose more fundamental & interesting changes to the status quo. 2/
The key to FLEDGE is to move user targeting information onto the *browser*, rather than broadcasting a cookie ID to the adtech ecosystem so advertisers can bid on ad opportunities based on what they know about that cookie ID. The prototypical FLEDGE use-case is retargeting. 3/
🧵We recently released a big update to our working paper examining how the #GDPR affected the site traffic💻 & ecommerce revenue💶 of EU users. (w/ @samgarvingold & Scott Shriver) 1/11 ssrn.com/abstract=34217… (image: Digiday)
We partnered with @Adobe to study the GDPR’s impact on EU users across 1,084 firm site analytics dashboards. Our data contain 77 of the top 1K sites & >700 long tail sites below the top 100K. We see >$0.75 billion in EU spending and 4.4 billion EU pageviews in total per week. 2/
We observe rich site outcomes in Adobe data. But, after the GDPR, we stop observing data from non-consenting users. If recorded site traffic falls post-GDPR, we want to know:
A) how much did GDPR hurt real outcomes of websites?
B) how much did consent limit data recording? 3/
In 2020, the Dutch public broadcaster NPO got rid of cookies and saw its revenues improve. This is held up as a hopeful example that privacy and publisher monetization can coexist. I want to share three comments. 🧵 1/8 wired.com/story/can-kill…
NPO claims that revenue rose 70% year-over-year in the first 2 months without cookies. This is an impressive achievement unlocked by building their own ad server, next generating contextual targeting, marketing to advertisers, etc. 2/ brave.com/publisher-3rd-…
Comment #1: This case study is hard to assess because year-over-year changes are far from apples-to-apples comparisons. Much can change between years that would confound our interpretation--beyond the many changes NPO implemented. 3/
🤔DO ADS ACTUALLY WORK?🤔
🧵We tackle this question using a large collection of display ad field experiments in a working paper with @EconInformatics & @enub
"The Online Display Ad Effectiveness Funnel & Carryover: Lessons from 432 Field Experiments” ssrn.com/abstract=27015… 1/8
Recently, the effectiveness of advertising has been called into question. Some of this skepticism is healthy. Measuring the effect of advertising properly is *really* difficult, as I detail here:
On average, our ad experiments run for 20 days & contain 4M users: 1.6B user-campaign observations in all! Our data come from the 1st months of Google’s Conversion Lift & Brand Lift products, which are powered by our (predicted) ghost ads methodology: doi.org/10.1509/jmr.15… 3/