Measuring Long-Term Causal Impact of Ads on LinkedIn Users
arxiv.org
arxiv.org
This preprint paper contains no indication of participation by an Ethical Review Board and lists 5 @linked.com email addresses as the sole authors.
How is this study ethically responsible?
I don’t trust LinkedIn to act ethically in such a regard. Do you?
Isn't "how does [decision architecture] permanently affect human behavior" the main starting point for all scientific evaluations of user interfaces?
A supermarkets behavioral experiments revealing that people permanently shift their dietary habits towards goods placed in the middle of the aisle appears to me to be structurally equal.
I never worried that I can't trust the supermarket in such a regard.
This is something psychology has had issues with in the past, and so it's a sensitive area. Several nations also have laws governing these kinds of experiments as a result.
The supermarket should be assessing whether their experiment is ethical. Increasing sales of fruit and veg is unlikely to raise any red flags, but if you drive people towards unhealthy options, then you may need to assess whether or not your study has the potential for detrimental long term effects.
Conducting a study on unknowing testsubjects which aims to influence the subject is an ethical question. Picture for example the nocebo effect, there are only limited studies on humans, as inducing these negative effects on unknowing subjects is extremely unethical. An ethics board has to, among other things, evaluate if your study will induce any negative effects in the participants. While negative effects can be accepted, its extremely unlikely for unknowing testsubjects which didnt give meaningful consent.
> [...] Conducting a study on unknowing testsubjects which aims to influence the subject is an ethical question. [...]
As far as I understood it you are just restating what our parent already said with:
> [...] How is this study ethically responsible? [...] They’re studying the long-term perceptual impact of advertising on test subjects who did not authorize being participants in this experiment. [...]
It's not sufficient to state that there's an ethical argument to make the argument an ethical argument. This is important, because it's the only way to productively talk about ethical concerns at all - irrespective of how one feels regarding this particular issue.
An argument based on ethics is usually founded in some set of value propositions. Often a case gets structured/applied along those propositions in the argument.
> [...] Picture for example the nocebo effect, there are only limited studies on humans, as inducing these negative effects on unknowing subjects is extremely unethical. [...]
That does appear to be the ethical standard case from clinical research. You've test subjects with a variation of intentions, knowledge and consent as well as (for the most part) clearly isolatable harm.
One would first find the obvious ethical fault in intentionally inducing the risk of harming someone and subsequently start healing/weighting through something like consent, individual and humanitarian utility, etc. (That's roughly what you described as part of an ethics board workflow.)
Now back to this case: How do we quantify harm induced by a change regarding LinkedIn ad-density? It's not clear to me that there is any risk for induced harm even if all (100%) posts visible on LinkedIn were ads.
Next question would be consent, but since I wanted to start with the strongest argument first, I'll stick with the harm.
This is why I would like to know how the ethical argument against this study works.
Just having an ethical consideration doesnt preempt the conclusion of that consideration. I am talking ethical consideration not ethical statement. The consideration part might be summarized like you did here
> How do we quantify harm induced by a change regarding LinkedIn ad-density?
For the consideration what negative effects there are, there are ethic boards, as its generally considered unethical to make that consideration alone as a stakeholder. Forgoing the conclusion of that consideration doesnt mean we didnt need an ethics board in the first place.
There are two distinct questions here, what are the ethical concerns for the study and should there be an ethics board involved in finding an answer to the first question. Since we are talking about long term impact on peoples behavior, i think you can make a case that not considering the ethical aspects of that impact on unknowing test subjects is generally unethical?
> For the consideration what negative effects there are, there are ethic boards, as its generally considered unethical to make that consideration alone as a stakeholder.
A well trained ethics board might be a nice institutional tool to provide a documented ethical evaluation one can start with, but it of course does not in any way relieve anyone involved from their moral agency or the need to assess their actions "alone".
I'm also in strong opposition to the idea that it's unethical to make ethical considerations alone, as a stakeholder or not. Generally: delegating ethical responsibility of ones own action to a hierarchy is IMO very unethical, no matter how the industry handles it.
> Forgoing the conclusion of that consideration doesnt mean we didnt need an ethics board in the first place.
In my earlier comments I wasn't writing (or thinking) about whether we(?) need ethics boards or not and I do think that this is a different topic, but: We don't strictly need ethics boards, right? It's not something like human agency. They're a nice attempt to increase checks and balances in an institutionalized manner, not some kind of moral requirement.
> There are two distinct questions here, what are the ethical concerns for the study and should there be an ethics board involved in finding an answer to the first question.
There I agree, except that the ethics board can not answer the first question for the researchers. But that aside: Yes, the demand of seeking counsel can be morally justified.
> Since we are talking about long term impact on peoples behavior, i think you can make a case that not considering the ethical aspects of that impact on unknowing test subjects is generally unethical?
I think this is the essence of the question. I don't know of any ethical difference between long and short term impact, so I'd cut that out. I don't think that the test subjects knowledge would change anything in that regard, but tell me, if I'm wrong here. All shortened it would leave us with:
You should ethically consider your impact on test subjects.
Which I would agree on. Let's hope they've considered it in an appropriate manner.
I just thought that there is some material concern - like an actual moral wrongdoing - they've done, instead of just a concern that they might not have been concerned.
I am unable to identify any statement in the preprint where they indicated that they had ethically considered their impact on test subjects. Did I miss a paragraph somewhere, or did they truly have nothing to say about such considerations?
They were going to influence human behavior anyway, the only difference to business as usual is that they were studying the effects rather than making changes blindly. Do you think that it's the studying that requires review by an ethics board?
If you require review for studies but not business decisions, you'll end up with people not evaluating the consequences to avoid having to get approval. If you require it for anything that influences behavior, you don't get that perverse incentive, but you'll have a hard time finding a reasonable threshold beyond which "influencing behavior" needs to be reviewed.
If what you were saying were true then just testing a joke at one stand-up event and testing a slight variation at a different one would require an ethics board. That's patently absurd.
It's only sailed if you give up and give in. Policies can change.
The problem is defining the boundaries. For example, does it count as an "experiment measuring human behaviour" to slowly roll out a new feature to see if engagement metrics change? What about A/B testing the colour of all the links on your page to see if they're clicked more?
https://www.linkedin.com/legal/user-agreement
>When you use our Services you agree to all of these terms. Your use of our Services is also subject to our Cookie Policy and our Privacy Policy, which covers how we collect, use, share, and store your personal information.
https://www.linkedin.com/legal/privacy-policy
>We use the personal data available to us to research social, economic and workplace trends such as jobs availability and skills needed for these jobs and policies that help bridge the gap in various industries and geographic areas. In some cases, we work with trusted third parties to perform this research, under controls that are designed to protect your privacy. We publish or allow others to publish economic insights, presented as aggregated data rather than personal data.
The criticism against this argument would be that those terms aren't sufficient to provide consent as people habitually ignore them.
Not sure what would count as the morally appropriate implementation of consent. Probably some middle-ground between buried in terms and continuously re-emerging full screen blocking warning labels.
The easy way out is probably found in some of the legal implementations of sufficiently informed consent.
I think these sorts of behind-the-scenes modifications happen quite frequently with social media feeds. Perhaps its not ethical, but at least we see some evidence of it happening.
I prefer to see the study than not.
Suppose you are ethically allowed to do A (show ads), and are ethically allowed to do B (not show ads). Then you are ethically allowed to run an experiment in which the only thing you do is A or B.
You can definitely argue that showing ads (or certain ads) is ethically bad. You can also argue that there should be legal requirements for ethics reviews on all research (which is not currently the law).
https://static.googleusercontent.com/media/research.google.c...