More on the Facebook study.

[mme_hardy]

 The Laboratorium says that it unequivocally  violates The Federal Policy for the Protection of Human Subjects, that it passed an IRB (!!!), and that it was funded with Federal money.   Why did it pass?  “on the grounds that Facebook filters user news feeds all the time, per the agreement.”

Comments

[julian]

I'm confused on how it can violate the Federal policy and yet pass an IRB. This is clearly me failing to grasp intricacies of procedure, though.

[mme_hardy]

Because the IRB defined it as "a pre-existing dataset". Basically, analyzing data that is already out there doesn't require the amount of approval as doing new experiments. In this particular case, "already out there" seems pretty tortured to me (follow the link), but I'm not an expert.

IRBs are supposed to enforce ethical rules, but "supposed" is the key word. I don't know who is responsible for checking that the enforcers are enforcing the Federal rules.

[rydra_wong]

Because the IRB defined it as "a pre-existing dataset". Basically, analyzing data that is already out there doesn't require the amount of approval as doing new experiments.

*bangs head repeatedly against desk*

It was an experimental study. The abstract repeatedly calls it an experiment. They manipulated the news feed to see if it produced effects, some of which they knew had the potential to be harmful. There is no meaningful sense in which the data they're describing (as "experimental evidence") existed prior to the experiment. It would not have been the same without the experimental intervention.

WORDS HAVE MEANINGS.

[mme_hardy]

I know. FLAMES ON THE SIDE OF MY FACE.

edit: whoops, already posted that link.

[mme_hardy]

" some of which they knew had the potential to be harmful. " Seriously. Screwing with people's mood *in an attempt to lower it*? That's gunpowder, right there.

[rydra_wong]

And something where it would be reasonable to expect that many/most people would refuse consent if given the choice.

[rachelmanija]

That would only be the case if they sifted users' feeds for coincidental levels of depressing/cheerful material, then correlated it with users' subsequent stated moods. That study would have been ethical and involved a pre-existing data set.

I hope an independent review board audits Cornell's IRB. That study clearly violates ethical guidelines on human experimentation, and should not have been approved.

One key factor in informed consent is that you can't just say, "By signing this paper you agree to take part in a study." You have to be told what the study is, and what the potential risks are.

In this particular case, they were trying to see if they could lower people's mood. That is a clear risk factor for people with depression (or other mental illnesses), and possibly for anyone. But no one was told that they were participating in a study which could lower their mood and possibly exacerbate depression. Nor were depressed people screened out.

In the unusual instances in which the experiment would be ruined if people knew too many details in advance, the requirements for "no danger of harm" are much higher, and people must be individually debriefed and explained what the experiment actually is about immediately afterward.

I mean, they'll probably get away with this because they're a powerful corporate entity. But they shouldn't.

When I was involved in a study where all I did to the subjects was interview them (admittedly on the topic of trauma they'd already gone through), I had to walk them through an 11-page consent form explaining exactly what the study was about, what their rights were, what the risks were, etc, before they agreed to take part. And that was a purely observational study. We weren't doing anything to them intended to affect them at all.

[mme_hardy]

Along with the idiots going "Well, don't use Facebook, then" I saw a really good essay pointing out that the big data companies are doing A/B tests *all the time*, that realistically many of those are human experimentation, and that the only difference in this one is that they actually published. It also points out that most of the people doing A/B testing don't come out of a psych/socio background and aren't even aware that ethics issues arise.

Fantastic analysis here:
http://codingconduct.tumblr.com/post/90242838320/frame-clashes-or-why-the-facebook-emotion-experiment

Edit:But thats only an explanation its not an excuse. I totally agree with you that it was unethical and should be penalized.

[rachelmanija]

Interesting.

It's true that all advertising is intended to affect unsuspecting people. What I find disturbing is that the rules of a largely unregulated field (advertising) have been applied to an extremely tightly regulated field (human experimentation.)

While it's philosophically true that advertising does experiment-like things, the reason human experimentation by actual scientists is tightly regulated is that it has an absolutely horrific history of abuse - abuse that is not the same as whatever harm has been done by advertising. So it's disturbing to see IRBs apply advertising rules to human experimentation.

I hate to bring up Nazis and I am in no way saying that Facebook is Hitler, but I think one reason people find this viscerally disturbing is that a big part of the reason for IRBs is, in fact, Nazis. And the Tuskegee syphilis experiment. Etc. Unregulated human experimentation has a history of horrific violations that unregulated advertising doesn't. (Cigarette advertising is harmful and misleading and causes many deaths (and so is often regulated) but doesn't quite hit the subjective horror level of Nazis.)

[mme_hardy]

I absolutely and totally agree with you. Tuskeegee, Japan, Germany, Milgram -- there are enough bad examples to go around.

I found it viscerally disturbing, myself, because of the possible damage. The experimenters are saying now "Well, we didn't show a BIG effect", but the point is that they did an experiment on manipulating mood without, apparently, giving any thought to adverse effects on the population.

I think that (again from that article) there are two separate failures. One is that the IRB utterly and totally fell down on the job. I saw an interview elsewhere with the editor of the journal; she said that she didn't feel it was appropriate to overrule somebody else's IRB, but now, thinking about it, admits the experiment was a bit creepy. http://www.theatlantic.com/technology/archive/2014/06/even-the-editor-of-facebooks-mood-study-thought-it-was-creepy/373649/

Failure two: there's a big group of people performing human experiments who are not only unaware they're performing human experiments but completely unaware of the history of experimental ethics having to do with humans. I don't know how you solve that one.

[mme_hardy]

For instance, if they'd been testing to see if they could make people post "fish" more often by adding stuff to their timeline about aquariums, I'd still think it was sleazy, but I wouldn't be as furious as I am.

With mood, they couldn't know beforehand what kind of shape their experimental population was in, they couldn't monitor what kind of shape they were in afterwards, there was no way to halt the experiment early and ensure that affected people were safe. That's what makes it visceral for me. On the other hand, I've never gone through experimental ethics training.

[rachelmanija]

No, that's a very real factor. There's no obvious way people could be harmed by seeing aquariums or by being influenced to post on fish. There is an extremely real risk of potentially lethal harm by deliberately attempting to lower people's mood.

I wonder if anyone has access to the list of people who were on the "lower mood" group. If any of them committed suicide, attempted suicide, were hospitalized for depression, etc... well, it would be impossible to prove that it was caused by the experiment. But depending on how big the two groups were, it might be possible to get a result like 0.5% of the "low mood" group had [x negative occurrance in time frame] vs. 0.1% of "high mood" group and 0.2% of control group.

[rydra_wong]

I wonder if anyone has access to the list of people who were on the "lower mood" group. If any of them committed suicide, attempted suicide, were hospitalized for depression, etc... well, it would be impossible to prove that it was caused by the experiment.

Also, unless those people happened to post about it publicly on Facebook, it'd be impossible to find out. Short of tracking down circa 300,000 people IRL and trying to find out their private medical details.

I mean, it's a set-up where any severe negative outcomes for experimental participants are going to be nicely hidden. Which I suspect the experimenters would not regard as a problem.

The only chance of it coming out now would be if people who had bad mood events in January 2012 and read about the study come forewards. And even then, there's no way -- AFAIK -- for people to find out if they were in the study or not.

(This gets creepier and creepier the more I think about it.)

But I really hope some angry people get lawyered up.

[rachelmanija]

No, I meant that you could literally look at their Facebook posts. You'd miss anything that wasn't posted on Facebook, but that sort of thing often is. If anyone went through the data, they could get the sort of percentage breakdown I suggested above. Facebook would be able to access that data, because it's the exact same stuff they used to monitor whether people were putting up happy or sad posts.

Your point still completely stands, of course.

[mme_hardy]

Who *does* watch the IRBs?

[rachelmanija]

IIRC, the Department of Health and Human Services. And the FDA, if drugs or other medical issues are involved. Though I think there are often some intermediate steps between the IRB and HHS that depend on the individual study, its location, etc.

[julian]

I find myself wanting to chant "Informed consent!" all over the place on this one, yes.

[mme_hardy]

This is the grant proposal. And this
Is the Hadoop script, whose use you will see,
When you are given your dataset. And this is the user's consent,
Which in your case you have not got.

[kore]

HAH that is awesome.

[lnhammer]

Bwah!

[mme_hardy]

I have the best commenters, because they get jokes like this.

[meara]

Yeah, that is a seriously shitty IRB. I mean, damn.

[ithiliana]

In my IRB experience, pre-existing datasets are usually defined as data already collected and stored by someone else, not the "past" data you collected before getting IRB approval.

*grrrrrrrrrrrrrrrr*

[mme_hardy]

I was hoping you'd weigh in! I think the argument was that the user's Facebook feed was pre-existing data because they'd already chosen whom to follow?

[mme_hardy]

See this very fine Tumblr discussion of the academic researcher's POV on this as opposed to the person-in-the-street's POV. It lays out the (IMHO) substantial violations of established consent practice.

http://codingconduct.tumblr.com/post/90242838320/frame-clashes-or-why-the-facebook-emotion-experiment

[ithiliana]

Wow, fantastic piece--thanks! Am saving for a future digital humanities course.

[mme_hardy]

Michelle N. Meyer suggests that, because the lead author was at Facebook and the two other authors just helped design the experiment and mung the data, neither of the two university IRBs legally had to be involved.

I find her analysis interesting, but I don't know enough to see whether it's valid. She moves on from analyzing informed-consent rules to saying "Well, if we let Facebook do it, we should certainly let academics analyze it", which seems to me to be glomming the fruit of a poisoned tree.

[ithiliana]

Sorry, have been so swamped.

At my university, if a faculty member is doing a collaborative research project with people off campus, they have to go through our training and meet our standards for IRB approval, whether they're lead or not.

See, the thing is, that different IRB committees can interpret Federal law and guidelines differently.

Now that the data is collected (and presumably archived? and available to others?), one might argue it's an existing collection.

I wouldn't touch it, but then I have issues with FB on many levels.

The whole internet, human subjects, privacy, consent, etc. etc. etc. is still in flux at the moment.

[neotoma]

Since IRBs are usually based at the institutes doing human research, it's those institutes that are suspect. Basically, I think someone is playing fast and loose with the rules to further their institutes' research goals.

[mme_hardy]

Here's even more analysis, plus more quotations from the IRB. Looks like fancy footwork to me.

https://thebrokenspokeblog.wordpress.com/2014/06/29/that-facebook-study-update/

[neotoma]

...Facebook's news feeds are built on what the user selects, aren't they?

Since the researchers apparently manipulated the news feeds, instead of just observing them, that seems pretty suspect to me.

[thebrokenspokeblog.wordpress.com]

Hey, thanks for the link. Thought you'd like to know that now Cornell has scrubbed the "we're funded by the military" bit from its press release, and another source is saying it was "only internal review," not an IRB. Cornell hasn't confirmed or denied being the IRB; Fiske isn't commenting on whether or not she was mislead.

It's a giant clusterfuck of information; I'm sure Forbes'll have another theory posted by the end of the day. :-|

[mme_hardy]

Oh wow, oh wow. No IRB? You can just see Cornell delicately picking up its skirts and backing away.

See also my next post on the formal educational backgrounds of the authors. The lead author, the Facebook guy, would certainly have done experimental ethics training.

[mme_hardy]

You did a great job of analysis; thanks for pulling it all together.

[ankaret]

My dentist manipulates my teeth all the time, but I'd still be outraged if she turned out to be experimenting with giving me more dental pain than usual.

I don't do Facebook, but I don't think the 'what did you expect' commenters are helping their own cause much. Congratulating yourself on being clever enough to avoid a lion in the road because you work from home is no consolation to people who have been mauled and doesn't do anything about the problem that THE ROAD IS FULL OF A LION.

Er, I seem to be full of metaphors this morning. Sorry.

[mme_hardy]

I have a big grin on my face from that metaphor, so carry on!

[nagasvoice.livejournal.com]

I'm wondering if folks like Electronic Freedom Foundation, ACLU and other various civil rights groups might be concerned about this too.

[mme-hardy.livejournal.com]

I'd expect some of the big-name privacy researchers to weigh in. Maybe Gene Spafford?