Privacy Notice and Choice in Practice Thesis Defense Pedro Giovanni Leon Department of Engineering and Public Policy Carnegie Mellon University August 28, 2014 Committee: Lorrie Faith Cranor, EPP & SCS, CMU Alessandro Acquisti, Heinz College, CMU Jon M. Peha, EPP & ECE, CMU Joel Reidenberg, School of Law, Fordham University Engineering & Public Policy Today’s Privacy Challenges • Extensive data collection – – – – – Online tracking Smartphone / Smart home / Smart store / Smart … Social media Web transactions Credit card transactions • Data aggregation and linkage from different data streams • Extensive sharing CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 2 Source: BlueKai CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 3 Source: BlueKai CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 4 Privacy Enables Fundamental Values • Freedom of speech • Anonymity • Innovation • Reputation • Autonomy • Equality CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 5 Transparency-based Regulations Notice Company’s response Company’s assessment Users’ perceptions and assessment Users’ choices Source: Fung A., Graham M, and Weil D. “Full Disclosure: The Perils and Promise of Transparency.” 2007 CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 6 Vehicle’s 5-Star Safety Ratings Source: http://www.safercar.gov/ CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 7 Restaurant’s Hygiene Source: http://www.nyc.gov/html/doh/html/services/restaurant-inspection.shtml CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 8 CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 9 CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 10 Regulators Ask for Better Privacy Protections White House – Feb. 2012 FTC – March 2012 CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 11 Current Protections Strongly Rely on Notice and Choice Notice • Consumers should be given notice of an entity’s information practices • Before any personal information is collected from them • Notice required for informed decisions as to whether and to what extent to disclose personal information Choice • Giving consumers options as to how any personal information collected from them may be used Source: http://www.ftc.gov/reports/privacy3/fairinfo.shtm CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 12 Thesis Goals [Need to work on content] • To investigate the effectiveness of today’s most widely deployed privacy protection mechanisms based on notice and choice • To evaluate new notice and choice alternatives • To propose both interface design improvements and policy recommendations CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 13 Thesis Scope and Approach Scope • [Need to work on content] Approach • Laboratory and field studies, usability testing, semistructured interviews, and online surveys • Evaluation of companies’ privacy practices and uses of privacy notice and choice mechanisms CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 14 Thesis Outline Chapter 1: Introduction Chapter 2: The Misrepresentation of Website Privacy Policies Through the Misuse of P3P Compact Policy Tokens Chapter 3: A Large-Scale Evaluation of U.S. Financial Institutions Privacy Notices Chapter 4: An In-depth Analysis of Online Advertising Companies’ Privacy Chapter Policies 5: Perceptions of Online Behavioral Advertising Chapter 6: A Usability Evaluation of Tools to Limit Online Behavioral Advertising Chapter 7: What Do Online Behavioral Advertising Privacy Disclosures Communicate to Users? Chapter 8: Factors That Affect Users’ Willingness to Share Information with Online Advertisers Chapter 9: A Field Trial of Privacy Nudges for Facebook Chapter 10: Public Policy Implications Chapter 11: Conclusions CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 15 Thesis Outline Chapter 1: Introduction Chapter 2: The Misrepresentation of Website Privacy Policies Trough the Misuse of P3P Compact Policy Tokens Chapter 3: A Large-Scale Evaluation of U.S. Financial Institutions Privacy Notices Chapter 4: An In-depth Analysis of Online Advertising Companies’ Privacy Chapter Policies 5: Perceptions of Online Behavioral Advertising Chapter 6: A Usability Evaluation of Tools to Limit Online Behavioral Advertising Chapter 7: What Do Online Behavioral Advertising Privacy Disclosures Communicate to Users? Chapter 8: Factors That Affect Users’ Willingness to Share Information with Online Advertisers Chapter 9: A Field Trial of Privacy Nudges for Facebook Chapter 10: Public Policy Implications Chapter 11: Conclusions CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 16 Scope of this presentation A Large-Scale Evaluation of U.S. Financial Institutions Privacy Notices [To be submitted to the Journal of Legal Studies 2014, WEIS 2013] What Do Online Behavioral Advertising Privacy Disclosures Communicate to Users? [WPES 2012] Factors that Affect Users’ Willingness to Share Information with Online Advertisers [SOUPS 2013, TPRC 2014] Public Policy Implications CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 17 A Large-Scale Evaluation of U.S. Financial Institutions Privacy Notices [To be submitted to the Journal of Legal Studies 2014, WEIS 2013] What Do Online Behavioral Advertising Privacy Disclosures Communicate to Users? [WPES 2012] Factors that Affect Users’ Willingness to Share Information with Online Advertisers [SOUPS 2013, TPRC 2014] Public Policy Implications CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 18 Research Questions • What is the impact of government regulation on US financial institutions’ privacy practices? • What is the spectrum of choices that customers have with respect to sharing practices of their personal information? • What are the internal and external factors that impact US financial institutions information sharing practices? CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 19 CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 20 CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 21 CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 22 Methodology • Public list of 19K+ FDIC-Insured financial institutions • Crawled the Internet and retrieved 6K+ notices • Parsed notices to extract information • Merged extracted info with companies’ additional info (size, geographical location, regulator, etc.) • Built binary logistic regression model to predict sharing practices CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 23 Important Differences in Practices • Need to add a graph here CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 24 Company’s Size And Location Affect Sharing Practices • Need to add a content showing the regression coefficients CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 25 Results • Substantial differences between sharing practices • Minimum compliance • A few instances of non-compliance • Larger institutions are more likely to share consumers’ personal information • Geographical location impacts sharing practices • Weaknesses in the regulation lead to reduced transparency of collection practices CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 26 Conclusions • TBD CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 27 A Large-Scale Evaluation of U.S. Financial Institutions Privacy Notices [To be submitted to the Journal of Legal Studies 2014, WEIS 2013] What Do Online Behavioral Advertising Privacy Disclosures Communicate to Users? [WPES 2012] Factors that Affect Users’ Willingness to Share Information with Online Advertisers [SOUPS 2013, TPRC 2014] Public Policy Implications CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 28 Ads Based on Online Behavior www.webmd.com User www.webmd.com Search: “Scalp conditions” Ad Network www.expedia.com CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 29 Study Goal Do OBA disclosures empower users to make privacy choices? CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 30 Methodology • Between-subjects online study • Recruitment: Amazon Mechanical Turk • Random assignment to: • • • • 1 of 2 priming conditions 1 of 2 icons 1 of 7 taglines 1 of 5 landing pages • Browsing scenario with online survey CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 31 Two Priming Conditions Imagine that you are… 1) Planning your next vacation to Paris… 2) Planning to buy a new car… CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 32 Two Icons Advertising option icon Asterisk man icon CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 33 Seven Taglines “AdChoices” “Interest based ads” “Why did I get this ad?” “Learn about your ad choices” “Configure ad preferences” “Sponsor ads” Blank (no tagline) CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 34 Five Landing Pages CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 35 Demographics • 1,505 participants • Mean age = 32, SD = 11.5 • 59% female • > 200 participants per tagline treatment CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 36 First Exposure to OBA Disclosures CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 37 Recalling Ads and OBA Disclosures • More likely to remember specific ads (49.3%) than icons (27.6%). p<0.001 • Tagline recall rate was only 11.9% • “Why did I get this ad?” recalled more often (22.3%) than others. p<0.05 • “AdChoices” (7.9%) and “Sponsor ads” (7%) recalled at about the same rate CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 38 OBA Disclosures in Isolation CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 39 To what extent, if any, does this combination of the symbol and phrase [icon+tagline placed here], placed on the top right corner of the above ad suggest the following? “This ad has been tailored based on websites you have visited in the past” Definitely not Probably not Not sure Probably Definitely CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 40 This ad has been tailored based on websites you have visited in the past. TRUE • “Why did I get this ad?” correct at a higher rate (p < 0.05) • No statistically significant difference between icons CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 41 What do you think will happen if you click on that symbol or that phrase? • It will take you to a page where you can tell the advertising company that you do not want to receive tailored ads. TRUE (27% agreement) • It will take you to a page where you can buy advertisements on this website. FALSE (30%) • You will let the advertising company know that you are interested in those products. FALSE (51%) • More ads will pop up. FALSE (53%) CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 42 It will take you to a page where you can tell the advertising company that you do not want to receive tailored ads. TRUE “Configure ad preferences” correct at a higher rate (p < 0.01) CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 43 It will take you to a page where you can buy advertisements on this website. FALSE “Configure ad preferences” and “Why did I get this ad?” correct at a higher rate (p < 0.0005) CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 44 More ads will pop up. FALSE Advertising option icon more likely to be incorrect than asterisk man icon (p = 0.003 ) CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 45 Notice Provided by Landing Pages To what extent, if at all, does the information on the ‘landing page’ suggest to you that… • The ads you see in the news website are based on your visits to this news website and other websites. TRUE (77% agreement) • This news website protects your privacy by not sharing your information. FALSE (24% agreement) CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 46 Indicate your agreement with the following statements defining what ‘opt out’ means in the context of Internet advertising: • Stop advertising companies from collecting information about your browsing activities. FALSE (63% agreement) • Stop seeing ads based on your browsing activities. TRUE (80% agreement) CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 47 Conclusions • OBA disclosures are not noticed • “AdChoices” one of the worst taglines • Users are unlikely to click on disclosures • Landing pages provide proper notice • Users misunderstood the meaning of opting out • Current OBA disclosures are falling short • Plenty of opportunities for improvement, but • User education is needed CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 48 A Large-Scale Evaluation of U.S. Financial Institutions Privacy Notices [To be submitted to the Journal of Legal Studies 2014, WEIS 2013] What Do Online Behavioral Advertising Privacy Disclosures Communicate to Users? [WPES 2012] Factors that Affect Users’ Willingness to Share Information with Online Advertisers [SOUPS 2013, TPRC 2014] Public Policy Implications CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 49 Benefits Concerns • Relevant ads • What info is collected? • Find things of interest • How collected information is used? • Personalized Internet experience • Inaccurate assumptions • Discounts • Lack of transparency and control User CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 50 Users often make decisions without having relevant information or under the wrong assumptions What actually matters to users? CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 51 What Matters to Users? • Type of data collected? • Extent of data sharing? • Retention period? • Familiarity with the visited website? • Control mechanisms? CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 52 Methodology • Between-subjects Mturk study [N = 2,912] – 15 scenarios representing different advertising-related data practices • Simulated web-browsing scenario • Explained value proposition of OBA • Collected willingness to share 30 types of information • Collected sharing preferences under different hypothetical controls CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 53 Treatments • Website familiarity – or CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 54 CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 55 Treatments • Website familiarity – or • Scope of Sharing – Only WebMD / WebDR – WebMD / WebDR and other visited sites – WebMD / WebDR and Facebook • Retention Period – One day or Indefinitely • Access to view and edit collected information – Access or No access CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 56 Analysis • Dependent variables: 30 types of personal information • Independent variables: Different data usage scenarios • Performed factor analysis to group types of personal information into categories • Built MANOVA regression model to investigate factors affecting disclosure preferences CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 57 Willingness to disclose information 60% 50% 40% 30% 20% 10% 0% CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 58 Willingness to disclose information 60% 50% 40% 30% • • • • • Country Web browser Gender Operating system State 20% 10% 0% CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 59 Willingness to disclose information 60% 50% 40% 30% 20% 10% • • • • • Credit score Exact current location Phone number Address Credit card # 0% CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 60 Categories of Data Types Category Data Types Browsing Pages visited, search terms, time spent on pages, + 2 more CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 61 Categories of Data Types Category Data Types Browsing Pages visited, search terms, time spent on pages, + 2 more Computer Web browser and OS CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 62 Categories of Data Types Category Data Types Browsing Pages visited, search terms, time spent on pages, + 2 more Computer Web browser and OS Location Country, State, Town/City, and ZIP code CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 63 Categories of Data Types Category Data Types Browsing Pages visited, search terms, time spent on pages, + 2 more Computer Web browser and OS Location Country, State, Town/City, and ZIP code Demographic Gender, political views, income bracket, religion, sexual orientation, + 4 more CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 64 Categories of Data Types Category Data Types Browsing Pages visited, search terms, time spent on pages, + 2 more Computer Web browser and OS Location Country, State, Town/City, and ZIP code Demographic Gender, political views, income bracket, religion, sexual orientation, + 4 more PII Email and name CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 65 What Factors Matter? Data Category Scope of Sharing Retention Period Browsing ✔ ✔ ✔ ✔ ✔ Access Site familiarity Computer Location Demographic PII ✔ CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 66 Example: Browsing (Pages visited) 80% 70% 60% 50% 40% Indefinitely One day 30% 20% 10% 0% WebMD + FB WebMD + Other Visited Scope of Sharing Only WebMD * Scope and retention statistically significant treatments (p-value < 0.001) CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 67 Example: Browsing (Pages visited) 80% 70% 60% 50% +15% +11% 40% Indefinitely One day 30% 20% 10% 0% WebMD + FB WebMD + Other Visited Scope of Sharing Only WebMD * Scope and retention statistically significant treatments (p-value < 0.001) CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 68 Example: Browsing (Pages visited) 80% 70% 60% 50% 40% +25% Indefinitely One day 30% 20% 10% 0% WebMD + FB WebMD + Other Visited Scope of Sharing Only WebMD * Scope and retention statistically significant treatments (p-value < 0.001) CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 69 Example: Browsing (Pages visited) 80% 70% +15% 60% 50% +25% 40% Indefinitely One day 30% 20% 10% 0% WebMD + FB WebMD + Other Visited Scope of Sharing Only WebMD * Scope and retention statistically significant treatments (p-value < 0.001) CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 70 Example: PII (name) 30% 25% 20% 15% +12% Indefinitely One day 10% 5% 0% WebMD + FB WebMD + Other Only WebMD Visited Scope of Sharing * Scope statistically significant treatment (p-value < 0.001) CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 71 Limitations • Survey data can differ from behavioral data • Mturk recruitment biases • Limited scope (a health related site) • The purpose of collection was limited to targeted ads CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 72 Conclusions • Notice and choice tools should allow users to allow or block trackers on the basis of factors users find important (e.g., retention and scope) • Ad companies should be transparent about their data practices • Standardized (and machine-readable) notices would allow for automation • Research on user preferences can assist the design of usable interfaces and policy frameworks CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 73 A Large-Scale Evaluation of U.S. Financial Institutions Privacy Notices [To be submitted to the Journal of Legal Studies 2014, WEIS 2013] What Do Online Behavioral Advertising Privacy Disclosures Communicate to Users? [WPES 2012] Factors that Affect Users’ Willingness to Share Information with Online Advertisers [SOUPS 2013, TPRC 2014] Public Policy Implications CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 74 Better Notices • Understand users’ expectations • Understand what information is relevant for users • Standard models similar to “Financial Notices” can help to facilitate comprehension and comparison • Usability testing is needed • Are worthless without meaningful choices • Are not sufficient, additional protections are CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 75 Standardization • Facilitate automatic checking of compliance with notice requirements • Facilitate automatic implementation of policies • Facilitate use of privacy agents • Facilitate comparison of notices (creating incentives to do better) • Standardization is also needed to enforce and automate information flow policies (e.g., sticky policy) CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 76 Incentives • Government regulation is necessary to achieve accountability and enforcement • Reduce information asymmetries – Enhance transparency mechanisms • Guarantee real alternatives • Data minimization and limited retention • Limit and control secondary uses CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 77 Regulation requirements may include – – – – – – Establishment of privacy program Establishment of Internal Review Boards Ongoing risk assessment (PIA, Audits, etc.) Employee education and training Limits and constraints for data retention and sharing Usable transparency and control mechanisms (awareness, standard notices, redress procedures) – Real sanctions for violations – Establish first parties responsibility for the tracking that happens on their websites – Alternative options requirements (offer free of-tracking paid services) CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ 78 Acknowledgements Thank you to: My advisor: Lorrie Faith Cranor My committee members: Alessandro Acquisti, Jon M. Peha, Joel Reidenberg Co-authors and CUPS Lab Members: Idris Adjerid, Lujo Bauer, Rebecca Balebako, Laura Brandimarte, Cristian Bravo Lillo, Justin Cranshaw, Mihai Christodorescu Jim Graves, Alain Forget, Manoj Hastak, Kelly Idouchi, Patrick Kelley, Saranga Komanduri, Abigail Marsh, Robert McGuire, Aleecia McDonald, Eyal Peer, Norman Sadeh, Rich Shay, Sonam Smat, Florian Schaub, Many Sleeper, Blase Ur, Yang Wang, Guzi Xu My family: Alejandra Penilla, Josué León, Mario León. Maria Antonia Nájera, Bella León, and Yonathan León Institutions: Carnegie Mellon University, Mexican Council of Science and CyLab Usable Privacy and Security Laboratory http://cups.cs.cmu.edu/ Technology, the Central Bank of Mexico, (other funding agencies) 79