Showing posts with label Risk. Show all posts
Showing posts with label Risk. Show all posts
Wednesday, February 4
Risk/Benefit Vs. Principle
One point from Tuesday's class keeps coming to mind. I'm curious what you all think. It seems that our arguments fell into two general categories, one that focused on the fact that a study needed to minimize risks to animal subjects and mazimize benefits to human health and knowledge, and another argument which stated that in principle research on animals was in unacceptable. I felt that I was mixing these two up during the discussion and am curious what other people think. Did we come to a consensus that animal research was wrong in principle or just that the risk to the dogs in that particular experiment was too high for the potential benefit tp humans? What do you say?
Saturday, January 31
Risk
One of the interesting issues that I think we should come back to next time is the issue of risk. Rollin mentions a number of times the need for researchers to communicate the risks associated with a certainly line of research and discuss whether these risks are balanced off, to a certain extent, by its benefits. This plausible point raises a host of questions though. There are at least two components to risk: probability and outcome. I was recently an experimental subject in an experiment involving single malt scotch (seriously!). I was asked to report on my subjective taste experiences before and after being instructed on how different aspects of these experiences were conceived. I signed a waiver acknowledging that, as I was to be drinking several measures of 100–110 proof liquor, there was a significant risk that I'd get a headache later (yeah. . . .). For the good of science I went forward.
Let's say that there was a 50% chance of getting headache. I don't like headaches. . . . I'd rather not get them. But they're not THAT bad. If she had told me that there's a 20% chance that my thumbnails might fall off, I'd probably think twice (or ask what KIND of of scotch it was — hey, they'd regrow!). Suppose I valued getting a headache at -20 "happiness units" (HUs) and having my thumbnails fall off at -50 HUs. By multiplying chance and outcome, we get something like the "expected value" of taking that risk: -10 HUs in each case. Supposing that I value the contribution to human knowledge (or just like single malt scotch), this expectation might be balanced off by a greater positive expectation — e.g., the 80% chance that this researcher has chosen some quite lovely malts whose enjoyment would give me 50 HUs (x 80% = +40 HUs). Overall, then, I expect to be 30 "units" happier. . . . I agree.
You probably see where this is going. What the hell are Happiness Units? Do my happiness units and yours compare? Do they even compare in an individual? What happens if an experimenter encounters a volunteer with extremely odd HU assignments (risk tolerances)? For example, suppose that a voluntee agrees to undergo an experiment that carries a 50% chance of total paralysis on the grounds that he's not a very active person. Should we even accept him as a volunteer? Can we rationally evaluate whether different HU assignments are rational?
There's another interesting problem who students from the Philosophy of Biology may remember connected to the St. Petersburg Paradox. How much would you be willing to pay to enter the following contest/bet: I flip a coin until it comes up tails. For each head I get before then, I give you $2n dollars. So if I flip three heads before finally getting a tails, I give you $8. But since the probability that I'd do that is .5n (=.125), you should expect to get $1. But in computing the expected value for this bet, since it's possible that I keep flipping heads arbitrarily many times, the expected payoff is arbitrarily much. If I were to flip a mere 10 heads in a row, I'd have to pay you over a thousand dollars! In fact, the expected value of the gamble is infinite! Thus on a simpleminded approach to rational decisionmaking, you should be rationally permitted to spend any finite amount of money to buy in. It's worth your entire life savings to take a chance on this bet!
Of course, this just goes to show that the simpleminded approach is simpleminded. There's no way you should spend very much to buy into this bet. I'd be hard-pressed to spend more than $10, say. Perhaps this shows that we're less interested in expected value than in what probably will happen (however that should be interpreted).
There's another side of this sort of phenomenon. Some physicists worried that there was a small but non-zero chance that when we fired up the Large Hadron Collider (at CERN), we'd produce a black hole that would destroy the earth. That'd be bad. How many HUs should we assign this outcome? I recall reading in Richard Posner's book that a reasonable monetary assignment would be -$600 trillion. I find that hilarious. Why that number? Does it even make sense to think of a dollar amount? Whatever. But if it's bad enough, then no matter how small the chance is, the expected cost will outweigh the benefit and we should shut down the experiment. (This might remind you of Pascal's Wager.)
Anyway, these sorts of questions are clearly relevant to our thinking about using humans (or animals) as experimental subjects.
Let's say that there was a 50% chance of getting headache. I don't like headaches. . . . I'd rather not get them. But they're not THAT bad. If she had told me that there's a 20% chance that my thumbnails might fall off, I'd probably think twice (or ask what KIND of of scotch it was — hey, they'd regrow!). Suppose I valued getting a headache at -20 "happiness units" (HUs) and having my thumbnails fall off at -50 HUs. By multiplying chance and outcome, we get something like the "expected value" of taking that risk: -10 HUs in each case. Supposing that I value the contribution to human knowledge (or just like single malt scotch), this expectation might be balanced off by a greater positive expectation — e.g., the 80% chance that this researcher has chosen some quite lovely malts whose enjoyment would give me 50 HUs (x 80% = +40 HUs). Overall, then, I expect to be 30 "units" happier. . . . I agree.
You probably see where this is going. What the hell are Happiness Units? Do my happiness units and yours compare? Do they even compare in an individual? What happens if an experimenter encounters a volunteer with extremely odd HU assignments (risk tolerances)? For example, suppose that a voluntee agrees to undergo an experiment that carries a 50% chance of total paralysis on the grounds that he's not a very active person. Should we even accept him as a volunteer? Can we rationally evaluate whether different HU assignments are rational?
There's another interesting problem who students from the Philosophy of Biology may remember connected to the St. Petersburg Paradox. How much would you be willing to pay to enter the following contest/bet: I flip a coin until it comes up tails. For each head I get before then, I give you $2n dollars. So if I flip three heads before finally getting a tails, I give you $8. But since the probability that I'd do that is .5n (=.125), you should expect to get $1. But in computing the expected value for this bet, since it's possible that I keep flipping heads arbitrarily many times, the expected payoff is arbitrarily much. If I were to flip a mere 10 heads in a row, I'd have to pay you over a thousand dollars! In fact, the expected value of the gamble is infinite! Thus on a simpleminded approach to rational decisionmaking, you should be rationally permitted to spend any finite amount of money to buy in. It's worth your entire life savings to take a chance on this bet!
Of course, this just goes to show that the simpleminded approach is simpleminded. There's no way you should spend very much to buy into this bet. I'd be hard-pressed to spend more than $10, say. Perhaps this shows that we're less interested in expected value than in what probably will happen (however that should be interpreted).
There's another side of this sort of phenomenon. Some physicists worried that there was a small but non-zero chance that when we fired up the Large Hadron Collider (at CERN), we'd produce a black hole that would destroy the earth. That'd be bad. How many HUs should we assign this outcome? I recall reading in Richard Posner's book that a reasonable monetary assignment would be -$600 trillion. I find that hilarious. Why that number? Does it even make sense to think of a dollar amount? Whatever. But if it's bad enough, then no matter how small the chance is, the expected cost will outweigh the benefit and we should shut down the experiment. (This might remind you of Pascal's Wager.)
Anyway, these sorts of questions are clearly relevant to our thinking about using humans (or animals) as experimental subjects.
Subscribe to:
Posts (Atom)