Ethical Issues and Considerations Arising in Big Data Research

Image from Pixabay

Image from Pixabay

In 48 of 50 states in the US, new attorneys are required to pass a 60 multiple-choice question exam on legal ethics in addition to passing their state’s bar exam. This is known as the Multistate Professional Responsibility Examination (MPRE). I well recall taking this test myself.

The subject matter of this test is the professional ethical roles and responsibilities a lawyer must abide by as an advocate and counselor to clients, courts and the legal profession. It is founded upon a series of ethical considerations and disciplinary rules that are strictly enforced by the bars of each state. Violations can potentially lead to a series of professional sanctions and, in severe cases depending upon the facts, disbarment from practice for a term of years or even permanently.

In other professions including, among others, medicine and accounting, similar codes of ethics exist and are expected to be scrupulously followed. They are defined efforts to ensure honesty, quality, transparency and integrity in their industries’ dealings with the public, and to address certain defined breaches. Many professional trade organizations also have formal codes of ethics but often do not have much, if any, sanction authority.

Should some comparable forms of guidelines and boards likewise be put into place to oversee the work of big data researchers? This was the subject of a very compelling article posted on Wired.com on May 20, 2016, entitled Scientists Are Just as Confused About the Ethics of Big-Data Research as You by Sharon Zhang. I highly recommend reading it in its entirety. I will summarize, annotate and add some further context to this, as well as pose a few questions of my own.

Two Recent Data Research Incidents

Last month. an independent researcher released, without permission, the profiles with very personal information of 70,000 users of the online dating site OKCupid. These users were quite angered by this. OKCupid is pursuing a legal claim to remove this data.

Earlier in 2014, researchers at Facebook manipulated items in users’ News Feeds for a study on “mood contagion“.¹ Many users were likewise upset when they found out. The journal that published this study released an “expression of concern”.

Users’ reactions over such incidents can have an effect upon subsequent “ethical boundaries”.

Nonetheless, the researchers involved in both of these cases had “never anticipated” the significant negative responses to their work. The OKCupid study was not scrutinized by any “ethical review process”, while a review board at Cornell had concluded that the Facebook study did not require a full review because the Cornell researchers only had a limited role in it.

Both of these incidents illustrate how “untested the ethics” are of these big data research. Only now are the review boards that oversee the work of these researchers starting to pay attention to emerging ethical concerns. This is in high contrast to the controls and guidelines upon medical research in clinical trials.

The Applicability of The Common Rule and Institutional Research Boards

In the US, under the The Common Rule, which governs ethics for federally funded biomedical and behavioral research where humans are involved, studies are required to undergo an ethical review.  However, such review does not apply a “unified system”, but rather, each university maintains its own institutional review board (IRB). These are composed of other (mostly medical) researchers at each university. Only a few of them “are professional ethicists“.

To a lesser extent, do they have experience in computer technology. This deficit may be affecting the protection of subjects who participate in data science research projects. In the US, there are hundreds of IRBs but they are each dealing with “research efforts in the digital age” in their own ways.

Both the Common Rule and the IRB system came into being following the revelation in the 1970s that the U.S. Public Health Service had, between 1932 and 1972, engaged in a terrible and shameful secret program that came to be known as the Tuskegee Syphilis Experiment. This involved leaving African Americans living in rural Alabama with untreated syphilis in order to study the disease. As a result of this outrage, the US Department of Health and Human Services created new regulations concerning any research on human subjects they conducted. All other federal agencies likewise adopted such regulations. Currently, “any institution that gets federal funding has to set up an IRB to oversee research involving humans”.

However, many social scientists today believe these regulations are not accurate or appropriate for their types of research involving areas where the risks involved “are usually more subtle than life or death”. For example, if you are seeking volunteers to take a survey on test-taking behaviors, the IRB language requirements on physical risks does not fit the needs of the participants in such a study.

Social scientist organizations have expressed their concern about this situation. As a result, the American Association of University Professors (AAUP) has recommended:

  • Adding more social scientists to IRBs, or
  • Creating new and separate review boards to assess social science research

In 2013, AAUP issued a report entitled Regulation of Research on Human Subjects: Academic Freedom and the Institutional Review Board, recommending that the researchers themselves should decide if “their minimal risk work needs IRB approval or not”. In turn, this would make more time available to IRBs for “biomedical research with life-or-death stakes”.

This does not, however, imply that all social science research, including big data studies, are entirely risk-free.

Ethical Issues and Risk Analyses When Data Sources Are Comingled

Dr. Elizabeth A. Buchanan who works as an ethicist at the University of Wisconsin-Stout, believes that the Internet is now entering its “third phase” where researchers can, for example, purchase several years’ worth of Twitter data and then integrate it “with other publicly available data”.² This mixture results in issues involving “ethics and privacy”.

Recently, while serving on an IRB, she took part in evaluated a project proposal involving merging mentions of a drug by its street name appearing on social media with public crime data. As a result, people involved in crimes could potentially become identified. The IRB still gave its approval. According to Dr. Buchanan, the social value of this undertaking must be weighed against its risk. As well, the risk should be minimized by removing any possible “idenifiers” in any public release of this information.

As technology continues to advance, such risk evaluation can become more challenging. For instance, in 2013, MIT researchers found out that they were able to match up “publicly available DNA sequences” by using data about the participants that the “original researchers” had uploaded online.³ Consequently, in such cases, Dr. Buchanan believes it is crucial for IRBs “to have either a data scientist, computer scientist or IT security individual” involved.

Likewise, other types of research organizations such as, among others, open science repositories, could perhaps “pick up the slack” and handle more of these ethical questions. According to Michelle Meyer, a bioethicist at Mount Sinai, oversight must be assumed by someone but the best means is not likely to be an IRB because they do not have the necessary “expertise in de-identification and re-identification techniques”.

Different Perspectives on Big Data Research

A technology researcher at the University of Maryland 4 named Dr. Katie Shilton recently conducted interviews of “20 online data researchers”. She discovered “significant disagreement” among them on matters such as the “ethics of ignoring Terms of Service and obtaining informed consent“. The group also reported that the ethical review boards they dealt with never questioned the ethics of the researchers, while peer reviewers and their professional colleagues had done so.

Professional groups such as the Association of Internet Researchers (AOIR) and the Center for Applied Internet Data Analysis (CAIDA) have created and posted their own guidelines:

However, IRBs who “actually have power” are only now “catching up”.

Beyond universities, tech companies such as Microsoft have begun to establish in-house “ethical review processes”. As well, in December 2015, the Future of Privacy Forum held a gathering called Beyond IRBs to evaluate “processes for ethical review outside of federally funded research”.

In conclusion., companies continually “experiment on us” with data studies. Just to name to name two, among numerous others, they focus on A/B testing 5 of news headings and supermarket checkout lines. As they hire increasing numbers of data scientists from universities’ Ph.D. programs, these schools are sensing an opportunity to close the gap in terms of using “data to contribute to public knowledge”.

My Questions

  • Would the companies, universities and professional organizations who issue and administer ethical guidelines for big data studies be taken more seriously if they had the power to assess and issue public notices for violations? How could this be made binding and what sort of appeals processes might be necessary?
  • At what point should the legal system become involved? When do these matters begin to involve civil and/or criminal investigations and allegations? How would big data research experts be certified for hearings and trials?
  • Should teaching ethics become a mandatory part of curriculum in data science programs at universities? If so, should the instructors only be selected from the technology industry or would it be helpful to invite them from other industries?
  • How should researchers and their employers ideally handle unintended security and privacy breaches as a result of their work? Should they make timely disclosures and treat all inquiries with a high level of transparency?
  • Should researchers experiment with open source methods online to conduct certain IRB functions for more immediate feedback?

 


1.  For a detailed report on this story, see Facebook Tinkers With Users’ Emotions in News Feed Experiment, Stirring Outcry, by Vindu Goel, in the June 29, 2014 edition of The New York Times.

2These ten Subway Fold posts cover a variety of applications in analyzing Twitter usage data.

3.  For coverage on this story see an article published in The New York Times on January 17, 2013, entitled Web Hunt for DNA Sequences Leaves Privacy Compromised, by Gina Kolata.

4.  For another highly interesting but unrelated research initiative at the University of Maryland, see the December 27, 2015 Subway Fold post entitled Virtual Reality Universe-ity: The Immersive “Augmentarium” Lab at the U. of Maryland.

5.  For a detailed report on this methodology, see the September 30, 2015 Subway Fold post entitled Google’s A/B Testing Method is Being Applied to Improve Government Operations.

Google’s A/B Testing Method is Being Applied to Improve Government Operations

"A/B", Image by Alan Rothman

“A/B”, Image by Alan Rothman

June 30, 2017 Update: This post was originally uploaded on September 30, 2015. It has been updated with new information below. 


During my annual visit with my ophthalmologist, he always checks the accuracy of the prescription for my glasses by trying out different pairs of lenses and then asking me to read the letter chart on the wall. For each eye, he switches the lenses back and forth and asks me a series of times “Which is better, 1 or 2?”. This is called a refraction test. My answers either confirm that my current lenses are correct or that I need an updated prescription for new lenses.

I never realized until recently that this method of testing is very similar to one tech companies use to measure and adjust the usability of their products and services. (I view this as my own bit of, well, in-sight.) This process is called “A/B testing“, where test subjects are shown two nearly identical versions of something with one of them containing some slight variation.  Then they are asked to choose which one they prefer between the two.

What if this method was transposed and applied in a seemingly non-intuitive leap to the public sector? A new initiative founded upon this by the US federal government was reported on in a fascinating and instructive article in the September 26, 2015 edition of The New York Times entitled A Better Government, One Tweak at a Time, by Justin Wolfers*. I highly recommend reading it in its entirety. I will summarize and annotate it, and then ask some of my own non-A/B questions. (There is another very informative article on this topic, covering the US and elsewhere, in today’s September 30, 2015 edition of The New York Times entitled Behaviorists Show the U.S. How to Improve Government Operations, by Binyamin Appelbaum.)

Google makes extensive use of this method in their testing and development projects. Their A/B testing has confirmed an effect that social scientists have known for years in that “small changes in how choices are presented can lead to big changes in behavior”. Moreover, effective design is not so much about an aesthetically pleasing appearance as it is about testing competing ideas and gathering data to evaluate which of them works best.

Last year, this project team introduced the effectiveness and success of A/B to the public sector was launched when  the federal government organized a group of officials (enlisted from a wide variety of backgrounds and professions), called the Social and Behavioral Sciences Team (SBST). It is also referred to as the “Nudge Unit“. Their mandate was to “design a better government”. They set out to A/B test different government functions to see what works and what does not.

After a year in operation, they have recently released their first annual report, detailing the many “small tweaks” they have implemented. Each of these changes was subjected to A/B testing. Their results have been “impressive” and imply that their efforts will save $Millions, if not $Billions. Moreover, because these changes are so relatively inexpensive, “even moderate impacts” could produce remarkably “high cost-benefit ratios”.

Among the SBST’s accomplishments are the following:

  • Improving Printing Efficiency: Some, but not all, printers at the US Department of Agriculture presented users with a pop-up message to encourage two-sided printing. As a result, two-sided printing rose by 6%. While this sounds small, its magnitude quickly scales up because US government printers produce 18 billion pages each year. The SBST report suggests that implementing this for the entire federal government could potentially save more than half a billion pages a year.
  • Reminding High School Graduates to Finish Their College Enrollment: Text messages were sent by the researchers to high school students   during the summer after their graduation, urging them to follow-up on the next steps needed to enroll in college. The differential of those who received the texts and those who did not, in terms of completing their enrollment, was 68% to 65%, respectively. The positive effect was more pronounced for low-income students who got these texts. While this 3% improvement also might not sound so large, at a mere cost of doing this at $7 per student, it proved to be tremendously cost-effective as compared to the $Thousands it otherwise costs to offer “grant and scholarship aid”.
  • Increasing Vendors’ Honesty on Tax Forms: Prompts were randomly placed on some versions of a federal-vendor tax collection form asking vendors to be truthful in completing it. Those who used the form containing the prompt reported more taxable sales than those using the untweaked form. In turn, this resulted in vendors voluntarily paying “an additional $1.6 million in taxes”. Again, scaling up this experiment could potentially raise additional $Billions in tax revenue.
  • Raising Applications by Those Eligible for Student Loan Relief: The government knows, through their own methods, who is struggling to repay their federally funded student loans. Another experiment sent a selected group of them emails about applying for loan relief resulted in “many more” applying for it than those who did not receive this message.
  • Lifting Savings Rates for People in the Military: When members of the military service were transferred to Joint Base Myer-Henderson Hall in Virginia, they received a prompt to enroll in the military’s savings plan. The result was a significant rise in participants. This contrasts with no increase by other who were transferred to Fort Bragg in North Carolina and not prompted.
  • Other Successful Experimental “Nudges”:
    • Well written letters resulting in more health care sign-ups
    • Emails urging employees to join workplace savings plans
    • Shortened URLs encouraging more people to pay bills online
    • Telling veterans that they earned rather than were entitled to a program increased their participation in it.

Justin Wolfers, the author of this article, concludes that it is the testing itself that makes for these successes. He very succinctly summarizes this by stating:

“Experiment relentlessly, keep what works, and discard what doesn’t.”

He further asserts that if this is done as Google has done it, the US government might likewise become “clear, user-friendly and unflinchingly effective”.

My own questions about A/B testing by the government include:

  • Would it also produce cost-effective results for state and local governments? Are there any applications that could be done on a multi-national or even global level?
  • Could it be applied to improve electronic and perhaps even online public voting systems?
  • Could it bring incremental improvements in government administered health programs?
  • What would be the result if the government asked the public to submit suggestions online for new A/B testing applications? Could A/B testing itself be done by governments online?
  • Does it lend itself to being open sourced for test projects in the design, collection and interpretation of data?

An earlier and well-regarded book about using a variety of forms of nudges to improve public policies and functions is Nudge: Improving Decisions About Health, Wealth, and Happiness, by Richard H. Thaler and Cass R. Sunstein (Penguin Press, 2009).


June 30, 2017 Update: For a timely and valuable primer and update on A/B testing I highly recommend a click-through and full reading of A Refresher on A/B Testing, by Amy Gallo (@amygallo), posted 6/28/17 on the Harvard Business Review blog. The author expertly covers the definition, process, interpretation, applications and errors of this methodology.


*  The author’s bio in this article states he is “a senior fellow at the Peterson Institute for International Economics and professor of economics and public policy at the University of Michigan“. (The links were added by me and not included in the original text.)