Tuesday, December 04, 2012

Second- vs. Third-Person Presentations of Moral Dilemmas

Is it better for you to kill an innocent person to save others than it is for someone else to do so? And does the answer you're apt to give depend on whether you are a professional philosopher? Kevin Tobia, Wesley Buckwalter, and Stephen Stich have a forthcoming paper in which they report results that seem to suggest that philosophers think very differently about such matters than do non-philosophers. However, I'm worried that Tobia and collaborators' results might not be very robust.

Tobia, Buckwalter, and Stich report results from two scenarios. One is a version of Bernard Williams' hostage scenario, in which the protagonist is captured and given the chance to personally kill one person among a crowd of innocent villagers so that the other villagers may go free. If the protagonist refuses, all the villagers will be killed. Forty undergrads and 62 professional philosophers were given the scenario. For half of the respondents the protagonist was "you"; for the other half it was "Jim". Undergrads were much more likely to say that shooting the villager was morally obligatory if "Jim" was the protagonist (53%) than if "you" was the protagonist (19%). Professional philosophers, however, went the opposite direction: 9% if "Jim", 36% if "you". Their second case is the famous trolley problem, in which the protagonist can save five people by flipping a switch to shunt a runaway trolley to a sidetrack where it will kill one person instead. Undergrads were more likely to say that shunting the trolley is permissible in the third-person case than in the second-person "you" case, and philosophers again showed the opposite pattern.

Weird! Are we to conclude that undergrads would rather let other people get their hands dirty for the greater good, while philosophers would rather get their hands dirty themselves? Or...?

When I first read about these studies in draft, though, one thing struck me as odd. Fiery Cushman and I had piloted similar-seeming studies previously, and we hadn't found much difference at all between second- and third-person presentations.

In one pilot study (the pilot for Schwitzgebel and Cushman 2012), we had given participants the following scenario:

Nancy is part of a group of ecologists who live in a remote stretch of jungle.  The entire group, which includes eight children, has been taken hostage by a group of paramilitary terrorists.  One of the terrorists takes a liking to Nancy.  He informs her that his leader intends to kill her and the rest of the hostages the following morning.  He is willing to help Nancy and the children escape, but as an act of good faith he wants Nancy to kill one of her fellow hostages whom he does not like.  If Nancy refuses his offer all the hostages including the children and Nancy herself will die.  If she accepts his offer then the others will die in the morning but she and the eight children will escape.
The second-person version substituted "you" for "Nancy".  Responses were on a seven-point scale from "extremely morally good" (1) to "extremely morally bad" (7). We had three groups of respondents, philosophers (reporting Master's or PhD in philosophy), non-philosopher academics (reporting graduate degree other than in philosophy), and non-academics (reporting no graduate degree). The mean responses for the non-academics were 4.3 for both cases (with thousands of respondents), for academic non-philosophers 4.6 for "you" vs. 4.5 for "Nancy" (not a statistically significant difference, even with several hundred respondents in each group). And the mean responses for philosophers were 3.9 for "you" vs. 4.2 for "Nancy" (not statistically significant, with about 200 in each group). Similarly, we found no differences in several runaway trolley cases, moral luck cases, and action-omission cases. Or, to speak more accurately, we found a few weak results that may or may not qualify as statistically significant, depending on one how one approaches the statistical issue of multiple comparisons, but nothing strong or consistent. Certainly nothing that pops out with a large effect size like the one Tobia, Buckwalter, and Stich found.

I'm not sure how to account for these different results. One difference is that Fiery and I used internet respondents rather than pencil-and-paper respondents. Also, we solicited responses on a 1-7 scale rather than asking yes or no. And the scenarios differed in wording and detail -- including the important difference that in our version of the hostage scenario the protagonist herself would be killed. But still, it's not obvious why should our results be so flat when Tobia, Buckwalter, and Stich find such large effects.

Because Fiery and I were disappointed by seeming ineffectuality of switching between "you" and a third-party protagonist, in our later published study we decided to try varying, in a few select scenarios, the victim rather than the protagonist.  In other words, what do you think about Nancy's choice when Nancy is to shoot "you" rather than "one the fellow hostages"?

Here we did see a difference, though since it wasn't relevant to the main hypothesis discussed in the final version of the study we didn't detail that aspect of our results in the published essay. Philosophers seemed to treat the scenarios about the same when the victim was "you" as when the victim was described in the third person; but non-philosophers expressed more favorable attitudes toward the protagonist when the protagonist sacrificed "you" for the greater good.  In the hostage scenario, non-academics rated it 3.6 in the "you" condition vs. 4.1 in the other condition (p < .001) (remember, lower numbers are morally better on our scale); non-philosopher academics split 4.1 "you" vs. 4.5 third-person (p = .001); and philosophers split 3.9 vs. 4.0 (p = .60, N = 320). (Multiple regression shows the expected interaction effects here, implying that non-philosophers were statistically more influenced by the manipulation than were philosophers.) There was a similar pattern of results in another scenario involving shooting one person to save others on a sinking submarine, with philosophers showing no detectable difference but non-philosophers rating shooting the one person more favorably when the victim is "you" than when the victim is described in third-person language. However, we did not see the same pattern in some action-omission cases we tried.

Fiery and I didn't end up publishing this part of our findings because it didn't seem very robust and we weren't sure what to make of it, and I should emphasize that the findings and analysis are only preliminary; but I thought I'd put it out there in the blogosphere at least, especially since it relates to the forthcoming piece by Tobia, Buckwalter, and Stich. The issue seems ripe for some follow-up work, though I might need to finish proving that the external world exists first!


Farshad said...

Dear Prof.
I am an Iranian MA graduate in philosophy. Now, I'm applying for the doctorate degree in philosophy to some U.S. universities. I just got my Gre scores: 165 in Verbal, 164 in Quantitative, and just 3 in Analytical writing! I expected 5 or at least 4. I'm thinking about requesting a review from ETS, but anyway, could you please let me know that how these scores can affect my application?


Eric Sotnak said...

One obvious possible confound is age. The brain is still undergoing a lot of changes until around age 25, making a comparison between pre-25 undergrads and post-25 professional philosophers possibly problematic.

Jorgen said...

I have a couple of (merely conjectural) thoughts on the matter.

First, there are several differences between the vignette you and Fiery gave vs. the vignette Tobia, et al. gave. E.g., theirs paints a picture with more concrete detail ("heavy man in a sweat stained khaki shirt...") and theirs builds in more emotional content (the "villagers, understand the situation, and are obviously begging you to accept [the deal]"). The Schwitzgebel/Cushman vignette, however, is left much more abstract. So if we take construal level theory seriously (which I tend to), then the third-person case will be more psychologically distant than the second person case, but in the Tobia vignette, there are psychologically relevant features built in: namely, emotional proximity (villagers begging) and concrete details (the captain's shirt/size). Now, I don't know what affect this will have on philosophers vs. non-philosophers, but there appear to be more variables (more complex construals) going on, possibly with features that philosophers may ignore or pick up on, which non-philosophers won't. For example, non-philosophers might construe the third-person more distantly, and apply general principles, whereas philosophers might feel the emotional weight of the volunteers pleas in the second-person case, but not the third-person case, and have their judgments altered in that way (mere speculation, who knows?). The important point is that there appears to be a lot more going on in their vignette than in yours, Eric, and these differences might complicate the construals and the explanation as to why philosophers differ from non-philosophers.

Secondly, Amit & Greene (2012), they provided some evidence, applying construal level theory, that in high-level construals (psychological distant cases), participants (all non-philosophers) give utilitarian responses, whereas in low-level construals (psychological proximate cases) participants give deontological responses. Amit/Greene argued that this is because high-level construals apply general goals/principles (and thus focus on ends/consequences), whereas low-level construals focus more upon the context of the situation (and thus the means to the ends). If there is weight to Amit/Greene's argument, then it would appear that, according to Tobia, et al's paper, philosophers would weigh the means in higher-order (third person/more distant) construals but give weight to the consequences in lower-level (second person/more proximate) construals. And IF THIS IS SO, then this would be an odd thing indeed! Since it would mean philosophers' representations of events simply don't match up in the way the body of psychological literature suggest. Why this would be, who knows? (Maybe philosophers make things too abstract, maybe they have trouble applying abstract principles to particular cases when construed more proximally, or some other thing, or maybe the results presented in this paper aren't really representative - personally, I'd like to see them replicated.)

In any case, this paper is interesting and presents some interesting dilemmas, I'm excited to see how this debate continues to develop.

Eric Schwitzgebel said...

@ Eric: As it happens, we have self-reported age of our respondents. The "non-academics" have a mean age of 28.2 years in our later dataset, the "philosophers" have a mean age of 33.7 years, and the "non-philosopher academics" have a mean age of 36.6 years. Our main intended comparison was the philosophers vs. the non-philosopher academics. Although the latter are a bit older, they're not a whole lot older.

Eric Schwitzgebel said...

Interesting thoughts, Jorgen! I'm inclined to hold interpretations in abeyance pending more data (not that you're suggesting otherwise). I do agree that the higher level of concrete detail in the Tobia et al. is a potentially relevant psychological difference.