Ethics decisions, like statistical inferences, are informative only if they’re not too easy or too hard.
For the full story, read the whole thing.
Ethics decisions, like statistical inferences, are informative only if they’re not too easy or too hard.
For the full story, read the whole thing.
“The statistical tools of decision analysis should enable us to make useful recommendations
here, but as statisticians it is hard for us to get to that point, given the challenges we have
in addressing our own ethical issues.” (from the article)
Statisticians could adopt the stance of the bishops of the Catholic Church, which while still embroiled in pedophilia scandals nevertheless continued to lecture about the moral failures of others.
To a certain extent that’s inevitable, because bishops and statisticians never will have their house fully in order.
—–
I actually expected a different point about the similarity of logistic regression and ethics, which perhaps is more banal.
In logistic regression, the end decision is “1” or “0”, but the actual result is a probability of being “1” or “0”. Similarly, in ethics, we have to do “A” or “B”, but we can only get a feeling that A is probably right, or that A is more right than B. And certainly the weight we give to the predictors can move that probability around.
Zbicyclist:
I won’t touch that Catholic church thing! But, regarding your other point, yes, I agree that this is another good connection, that any sort of decision making can turn an underlying continuous problem into something discrete.
We could never teach anyone anything if we waited to first be perfect in that attribute / skill.
Here’s a case of how one sort of ethical decision –decision to punish– is actually modeled by Bayesian logistic regression:
Liddell, T. M. and Kruschke, J. K. (2014). Ostracism and fines in a public goods game with accidental contributions: The importance of punishment type. Judgment and Decision Making, 9(6), 523-547.
This is descriptive modeling of human choice, not prescriptive modeling of normative ethics.
What are examples of pharma / healthcare ethics decisions that are “too hard” versus “just right”.
i.e. What ought we focus on & what leave alone for now?
I will limit myself to statistical ethics dilemmas in Andrew Gelman’s framework. In pharma research aimed for regulatory review, there is an incredible amount of planning, with ad-hoc analysis only performed after the planned analysis.
Easy- Five weeks after study data is made available, the clinical coordinator finds out that a lab instrument was broken for two weeks at one of the centers. Since this data is collected for safety purposes and there were no adverse events, it is easy to decide that those two weeks of data should be switched to missing, despite any idea of the level of inaccuracy.
Medium- You arrive to analyze the data from a study that has been on-going for three years. The last statistician was at a contract research firm that is no being sued by your employer. So the only thing you have is the plan that was submitted with the new
drug application. But the method is a now faded non-parametric approach, and you believe it would be better to a random effects model with Bayesian estimators. And an interim analysis came up successful with the original plan. But your method will be better for secondary endpoints. Do you go with the old method, or ask your employer to submit an expensive addendum and create a lot of chaos with the regulatory department?
Hard- A drug is about to be released. A prestigious medical group reads the report and says look there is a high-rate of a severe adverse event. There was a recent scandal regarding this event at another company. However this adverse is correlated with the disease being treated. And the proportional hazard method shows it actually decreases over 2 years as people are treated. But that was not part of your original analysis and the market will say you are ‘playing games’? How do you handle this situation so you are not forced to do more extensive clinical work?
What a great piece, Andrew! Well done. I’ve long been concerned about the presentation of ethical “dilemmas” that either make them simple (albeit in a sort “Well, here’s where everyone would obviously not do what is obviously the right thing to do”) or too hard (your “coin toss” cases.) We need to construct cases for the ethics classroom that reward the hard work of searching for interesting solutions.
What does the classification of “easy/intermediate/hard” ethical problems refer to? Is it just how large a majority would agree about a certain “solution”? If not that what else?
The whole idea of having “true solutions” for ethical problems, some of which may be easy, some intermediate and some difficult looks very dodgy to me, although I’m not totally sure whether this is claimed here.
The problem with all these “dilemmas”/problems to be presented in ethics classes is that it seems to be somehow implied that it makes sense to try to solve them from the outside, just based on “storytelling”. But I think that ethics is something between people; what matters most in my opinion about most ethical problems is how to make sure that the real people involved get their say and are enabled to solve it for themselves. If it’s just storytelling, there is nobody involved, so this is fundamentally different.
Then still such stories at least serve the aim of making people think about their values and the message may just be that for this aim the stories are not the best about which either everybody agrees or where most people feel that there are very strong arguments either way, to the point of making the solution rather “arbitrary”. Not sure whether I agree with this message, though. Finding out what exactly makes the “easy” stories easy, and what to do with those who don’t agree about even those can be quite tough already.