The Facebook experiment controversy

Facebook data scientists are being blasted for a social psychology experiment they ran in 2012 in which they varied the amount of positive/negative content exposed to users in newsfeeds and measured whether this affected the positive/negative content posted by those users. (link to WSJ report; link to paper)

I'm perplexed by the reaction. Boing Boing's Cory Doctorow calls it "likely illegal", who links to James Grimmelmann, a law professor. Slate slams it as "unethical". NPR proclaims that "we" are "lab rats, one and all".

***

About Consent

The biggest gripe is that users who were randomly selected to be part of the test were not informed. Facebook argues that consent is given in the overarching terms and conditions which all users must agree to.

I am against all forms of forced consent that is practiced by Internet companies like Facebook and Google but that's a much larger issue of which scientific experiments like this is but a small part. The same critics don't seem to mind if the experiment are conducted for financial gain (like generating advertising revenues) but they grumble about an academic exercise designed to verify a theory of social psychology.

Critics are emotionally charging the conversation by claiming that "Facebook is manipulating users' emotions". The truth is every form of marketing and advertising is a form of manipulating users' emotions. If you are against this particular experiment, you should be against all the other experiments (i.e. so-called A/B tests) that are conducted every day by the same companies on the same users who are not informed and have not agreed to be "lab rats".

In fact, the same critics should have been incensed more by those advertising experiments, for which the businesses are looking for "actionable" insights, meaning that they are looking for ways to manipulate not just your emotions but your actions and behavior, such as what you click on, what you view, and what you spend money on.

Scientific Progress

Playing the devil's advocate for the moment, I'd suggest that this type of large-scale randomized experiments has the potential of bringing a revolution to psychology experiments.

I have never been a fan of the typical psychological experiment that we are forced to accept as legitimate "science": you know, those experiments in which a professor recruits 10 or 50 students via campus posters offering $20 for participation. The severe limitation of the sample, both size and composition, does not stop researchers from generalizing the results to all humans. The standard claim is that the observed effect is so large as to obviate the need for having a representative sample. Sorry - the bad news is that a huge effect for a tiny non-random segment of a large population can coexist with no effect for the entire population.

I believe scientists should be actively addressing the privacy and ethical concerns of such experiments but not dismissing them categorically.

The Opt Out Solution

There are at least two components of the possible solution. One is having an organized review board like institutional review boards. This should be created across industry, not within a company. Perhaps a list of ongoing experiments should be made available for those who care enough to review it.

More importantly, users of Facebook or other websites should be allowed to opt out of any experiments (again, not just science experiments but also advertising-motivated experiments).

I think many critics are being incredibly facetious by asking for prior consent. For many psychology experiments, we need to use double blinding. The experiment at the center of this controversy would have been neutered if the participants know what is being changed.

The Issue of Harm

One of the weakest arguments raised against Facebook is the allegation of harm. James Grimmelmann, who has been cited by various articles, wrote definitively: "The study harmed participants." (link) How so? He explains:

The unwitting participants in the Facebook study were told (seemingly by their friends) for a week either that the world was a dark and cheerless place or that it was a saccharine paradise. That’s psychological manipulation, even when it’s carried out automatically.

That's it? And he condones advertisers and politicians who manipulate our emotions because somehow we should accept lower standards in those arenas.

The Fallacy of Paying for Free Service

The WSJ commits the same fallacy as a lot of other journalists when it comes to explaining why users should submit to experimentation. It says:

Companies like Facebook, Google Inc. and Twitter Inc. rely almost solely on data-driven advertising dollars. As a result, the companies collect and store massive amounts of personal information.

The same argument has been used to support massive invasion of privacy and indiscriminate data collection.

There are several problems with this argument:

Firstly, most companies that are collecting massive amounts of data on their users today are not solely advertisers. Amazon is not an advertiser. Netflix is not an advertiser. Cable and phone companies are not advertisers. Banks are not advertisers. Your doctor is not an advertiser. (I mean, not primarily advertisers; some of these actually do earn advertising dollars.)

Secondly, advertising dollars are there whether or not there is data. I have yet to see a proper study that shows that digital marketing dollars are incremental spending, not just repurposed spend shifting from offline channels. Indeed, the frequent claim that digital advertising solves the half-the-traditional-advertising-spend-is-wasted problem is evidence that the digital marketing play is mostly spend shifting, not spend creation. The alternative world in which these digital advertisers and data collectors do not exist is not one in which the advertising market is half the size of what it is today.

Thirdly, advertisers do not need personal information. If advertising is the purpose of the data collection, anonymous data is just as good. This is because brands have an identity. Brands do not want to be different things to different people. Brand messaging, just like other forms of messaging, benefits from simplicity. Nike wants everyone to use "just do it"; Nike is never going to want a million slogans for a million people. Thus, the idea that collecting massive amounts of personal data is a "result" of data-driven advertising is bogus.

Fourthly, the business model of Facebook, Google, etc. is a choice. No one but themselves is forcing them to rely on advertising dollars. Google for instance makes Android, Chrome, etc., all of which are products that they can charge money for.

The journalists who keep printing the argument that we have to accept mass surveillance in order to support the business model of "free" are merely repeating a marketing message without thinking about it. They have fallen victim to emotional manipulation.

 

Comments

[zbicyclist]

Google, Facebook and others do thousands of A/B tests each year.

In addition, advertisers do them as well. Even universities with review boards are unlikely to review A/B tests of their university web advertising ("should we show smiling multiethnic students at a football game, or students diligently studying in the library?")

But it's easy to spend practically no time writing a column expressing outrage.

[Ken]

I've also been amazed by the response. It hasn't taken the internet to allow this sort of project to run. I know of academics who have been involved with credit card companies mail outs to convince people to use increase their credit card limits. These are a simple randomised trial. Just send different deals to different customers and watch the responses. I expect they do the same thing to increase usage.

I actually find Amazon's and others efforts to sell me a book very worthwhile, and I'm certain that they do A/B testing to attempt to get people to buy a book. If I search them I'm more likely to find something worthwhile than I would be at my university library which doesn't care at all whether I find a book.

Last word might be XKCD http://xkcd.com/1390/ as almost always.

[ZBicyclist]

Supporting my earlier point that media are unfairly targeting the Facebook study as a cheap way to fill airtime:

http://chronicle.com/blogs/ticker/33-ethicists-defend-facebooks-controversial-mood-study/82091?cid=wc&utm_source=wc&utm_medium=en

Title:
33 Ethicists Defend Facebook’s Controversial Mood Study
by Andy Thomason

"A group of bioethicists wrote in a column published on Wednesday that Facebook’s controversial study of mood manipulation was not unethical, and harsh criticism of it risks putting a chill on future research. The article was written by six ethicists, joined by 27 others."

[Andrew Gelman]

Well put: "a huge effect for a tiny non-random segment of a large population can coexist with no effect for the entire population."