# Subjunctive Analyses and the Conditional Fallacy

Subjunctive analyses of what we ought to do often appeal to what we would do (or what we would want ourselves to do) if we were both fully informed and fully rational. Many such analyses commit what is called the Conditional Fallacy and are, therefore, subject to counterexample. The counterexamples all involve cases where full information (and/or full rationality) affects what one has reason to do and, thus, what one ought to do. To illustrate, consider the following very simple analysis:

(SA) S ought to do x in C iff a fully rational and fully informed version of S would want to do x in C.

To see that this analysis commits the Conditional Fallacy, suppose that S is facing a choice of picking only one of either box A or box B. One contains a very large sum of money and the other contains none at all. In C, S has no information about which box contains the money, but, in C, S can perform x, which is the act of S’s paying a small sum of money to peek inside each box. Now, a fully informed and fully rational version of S (viz., S+) would not want to perform x, because doing so costs money and, being fully informed, S+ would learn nothing by peeking inside them. But clearly S ought to do x in his current uninformed state. So SA fails due to this counterexample.

The problem arises because coming to have full information and/or coming to be fully rational can itself affect what one has reason to do. So, we should give our subjunctive analyses, not in terms of full information (that is, knowledge of all the facts), but in terms of knowledge of only the relevant facts. What are the relevant facts? Well, let C be S’s actual circumstances and let C+ be exactly like C except that, in C+, S knows that p. For any proposition p, that p is a relevant fact for S in C iff S’s objective reasons for action in C+ do not differ from S’s objective reasons for action in C. Clearly, then, neither the fact that the money is in box A nor the fact that there is no money in box B are relevant facts, for knowledge of these fact would affect what one has reason to do. So an S that knew only the relevant facts (viz., S*) would still want to do x.

Instead of SA, then, perhaps we should adopt something like the following analysis (Doug’s analysis):

(DA) S ought to do x in C iff S would want S to do x in C if S were informed of all relevant facts and if S were to respond appropriately to the reasons that those facts provide, coming to desire only what S has, given the relevant facts, sufficient reason to want as well as coming to desire everything that S has, given the relevant facts, decisive reason to want.

I know that there are other ways of potentially avoiding the Conditional Fallacy, but I wonder if this one is satisfactory. This is all just preliminary and very half-baked. I have a lot to read still on this sort stuff, but I just want to know whether this is a non-starter or not. Thoughts?

## 28 Replies to “Subjunctive Analyses and the Conditional Fallacy”

1. Ummmmm.
All of us sophisticates recognize an agent-info-relative kind of ought and one that is not agent-info-relative. Right?
And it looks like the one you are trying to analyze must be agent-info-relative, because you think the verdict that the agent ought to decline x and just take box A (which, I happen to know, has the big sum in it) is a bad verdict.
But then, why do you want a “full information” clause in your analysis?

2. Right. Revise the example such that if he chooses box A without knowing that the money is in there, he will be killed by the nefarious Black of philosophical lore. Thus, S objectively ought (the not agent-info-relative ought) to do x and then choose A, but S+ objectively ought to decline to do x and just choose A.

3. Jamie,
Usually you write “Hmmm” in response to my posts and comments, which I take to indicate that you find what I said implausible. The more m’s the more implausible, I take it.
Does “Ummmm” indicate something different?

4. If you can appeal to reasons on the RHS, why the need for a subjunctive analysis at all? (I thought they were generally used by those who wanted a reductive analysis of the normative, i.e. without presupposing a prior account of normative reasons. Perhaps I’m misunderstanding the dialectic here though?)

5. Then I don’t see what’s wrong with (SA).
I assume C includes facts about what information S has. So, a fully informed version of S would want to perform x in C. She would not want to perform x in her fully informed circumstance, but that’s irrelevant, because her fully informed circumstance is not C.
Look, here’s the thing. We’ve kind of been through all of this for a similar problem, namely, the one that Michael Smith tackles with the ‘advice model’. (That one has to do with connecting reasons to desires under ‘fully rational’ conditions.) But once you have added “in C”, you have already moved to the advice model, which is supposed to get around the Conditional Fallacy problem.
Of course, many people think this doesn’t work. (Did you want references for this issue?) But it doesn’t look like the problem you’re raising is a problem for the advice model, and you’ve already got the advice model in (SA).

6. Doug, it indicated puzzlement, not that what you said was implausible.

7. Hi Richard,
Yes, normally, the people giving these subjunctive analyses (internalists) are interested in giving a reductive analysis. But I’m just interested in whether one can avoid the Conditional Fallacy in the way that I’m suggesting. I don’t care if the analysis will be reductive or not.
The reason that I’m interested in this is that I give the following analysis in my forthcoming book:
“Scrupulously securable: A set of actions, αj, is, as of ti, scrupulously securable by S if and only if there is a time, tj, that either immediately follows ti or is identical to ti, a set of actions, αi (where αi may, or may not, be identical to αj), and a set of background attitudes, B, such that the following are all true: (1) S would perform αj if S were to have at tj both B and the intention to perform αi; (2) S has at ti the capacity to continue, or to come, to have at tj both B and the intention to perform αi; and (3) S would continue, or come, to have at tj B (and, where αi is not identical to αj, the intention to perform αi as well) if S both were at ti aware of all the relevant reason-constituting facts and were at tj to respond to these facts/reasons in all and only the ways that they prescribe, thereby coming to have at tj all those attitudes that she has, given these facts, decisive reason to have and only those attitudes that she has, given these facts, sufficient reason to have.”
Now, in an earlier draft of the manuscript, I used “all reason-constituting facts” instead of “relevant reason-constituting facts.” But, as Peter Graham rightly pointed out to me, that earlier analysis commits the Conditional Fallacy. I’m hoping that the one in the book (the one above) doesn’t. And that the way of specifying the relevant facts as I have in the above post is plausible.

8. Hi Jamie,
Okay, fair enough. Yes, references would be helpful.
In any case, I’m less interested in whether SA in fact commits the Conditional Fallacy and more interested in whether DA avoids the conditional fallacy in a way that wouldn’t be worrisome to either those who think that merely appealing to C isn’t enough to get you to the advice model or those who think that the advice model doesn’t succeed in avoiding the Conditional Fallacy. Aren’t there people who think that the advice model doesn’t work?
Again, I’m only starting to get up to speed on this stuff, so pointing out that some people don’t think that even SA commits the Conditional Fallacy is helpful. I guess that I was thinking that C didn’t include the information state of the agent, but that’s probably bad and isn’t consistent with what I go on to say about C and C+.

9. Well, I was thinking of two PEA Soupers: Robert Johnson and Eric Wiland.
Eric’s worry is that the advice model is not really internalist, so I think it’s probably not relevant to you. Robert’s worry (at least one of them — I’m probably forgetting something) is that it’s unclear what a highly idealized agent would want to happen to her in very unidealized circumstances.
But SA does get around the basic, Shope-style conditional fallacy problems. The basic problems come from cases in which an agent does have the feature being analyzed, but would no longer have it under the idealized circumstances. SA gets around this problem. It’s a good trick.
(As you probably suspected, I find the ‘scrupulously securable’ thing impenetrable, so as things stand I’ll just take your word for it that there’s a conditional fallacy problem there.)
Robert Johnson’s paper is Reasons and Advice for the Practically Rational and Eric Wiland’s is Good Advice and Rational Action.

10. Doug,
One of the things I like about Robert’s paper which Jamie cites is that it says a good bit about why you might want something like an example version of the subjunctive model if you are moved by the considerations that pushed Williams to internalism. I think you can get what these motivations push one to say by restricting the ways in which we idealize (both in terms of the information a person might have and in the features of the agent’s) but stick with the example model. On your version we both move to the advice model and move to less than full information. From my perspective that seems like two moves where only one is needed (and if Robert’s even earlier paper (“Reasons and Advice for the Practically Rational,” Philosophy and Phenomenological
Research, LVII, No. 3, September 1997: 619-625) and Eric’s paper are right you’d do better if you could avoid abandoning the example model).
That’s pretty abstract, but the basic idea is that you can handle conditional fallacy worries within the example model by not idealizing too much.
FWIW . . .

11. Jussi Suikkanen says:

There’s a lot to be said about this proposal but here’s at least one cost. The definition of relevant facts makes DA unable to capture certain intuitive oughts.
So, consider the intuitively true statement: ‘Doug [S] ought to go home [to do x] when there is a surprise party organised for him there [in C]. Can we find a right-hand side of the analysis that gives this intuitively true claim?
To do so, we would need to specify what the relevant facts are that the subject knows in the counterfactual situation. The question is, do these facts include the fact that there is a surprise party organised at Doug’s home? Either they do or they don’t. If they don’t, then there need not be any fact known by Doug that makes Doug want to go home and so the left-hand side comes out as false.
Yet, it is not clear how the relevant facts could include that fact either. In the situation in which Doug knows that there is a surprise party organised him Doug has different objective reasons. It might be that Doug only enjoys surprise parties but hates all other parties. So, given that this piece of information changes the objective reasons, it cannot count as relevant information either by the lights of the definition.
This means that we don’t yet have a right-hand side that gives us the left-hand side that Doug ought to go home for a surprise party. This means that there seems to be oughts that cannot be captured with the definition of relevant facts and DA. So, getting rid of the counter-examples gets rid of some intuitive oughts too.

12. Some thoughts:
I for one like DA better than SA, or the advice model version of SA, though perhaps from different motivations than Doug’s own.
What SA does not explain, it seems to me, is why the idealized version of S (on the RHS) constitutes a relevant version of S (on the LHS), relevant, that is, for the task of correctly determining what S ought to do in C.
To elaborate with the sort of consideration that Schroeder has made famous, two different agents S and S* can have reasons to do different things in the similar sets of circumstances, e.g., S ought to go to the party because she very much likes dancing, whereas S* ought not to because he very much hates loud music, and so on. And let’s say there are two distinct idealized agents, S1+ and S2+, one of whom correctly determines what S ought to do in C, and the other one correctly determining what S* ought to do in C. SA has nothing to say about which of S1+ and S2+ is the relevant idealized version of S or S*. And if you pick the incorrect idealized version, you get incorrect normative results.
Now it might be proposed that there is one idealized version of both S and S*, more abstract and bloodless as it were than S1+ and S2+. Call this hyper-idealized agent S++. Take all the features of S1+ or S2+ that enabled either of them to deliver correct normative results for either S or S*, and incorporate them into the circumstances of S or S*. Then S++ is enabled to deliver correct normative results, and different ones, for S and S* because S++ takes the relevant circumstances, which are different for S and S*, into account.
Against this move of hyper-idealization, I think it needs to be pointed out that all the important legwork is done by identifying the relevant features/circumstances that generate the correct norms for S and S*. The introduction of S++ seems superfluous. And it is reasonable to hope, by the same token, that we can equally dispense with the first idealizing move to S1+ or S2+, and speak simply of the relevant features of S and S* and of their circumstances that correctly generate the norms, or determine which norms are applicable to S or S*.
p.s. Doug, as for references I remember very much liking the discussion between Van Roojen and Robertson (van Roojen, “Motivational Internalism: A Somewhat Less Idealized Account” and Robertson, “Internal Reasons: Reply to Brady, van Roojen and Gert”). This takes place mostly in the context of internalism, but still it seems to me that van Roojen’s piecemeal approach is valuable step away from idealization and towards identifying the relevant features/circumstances determining the existence of norms and reasons.

13. @Mark: Thanks, that’s helpful. My thought is that if the general problem (and not any specific problem for any specific subjunctive analysis—e.g., some internalist analysis) arises because knowledge of certain facts can, in certain situations (like those where Black is going to kill us if we have certain knowledge), affects what reasons we have, then the general solution ought to be to restrict the relevant facts to those whose knowledge of which doesn’t affect our reasons for action.
@Jussi: I take it that the fact that S would receive some added enjoyment if S were to go home constitutes an objective reason for S to want herself to go home whether S knows that she would receive this enjoyment or not. So the fact that S would receive some enjoyment if S were to go home is a relevant fact, and if S were informed of this fact (and the other relevant facts) and were to respond appropriately to it (and the others), then S would want herself to go home. So DA seems to get the intuitively correct result.
@Boram: Thanks for the references and thanks for the interesting comments. I’m still pondering them.

14. Jussi Suikkanen says:

Doug,
I don’t think that that can be right – especially the first sentence. This is because the enjoyment in question must be spontaneous and unexpected. To see this,here’s your definition again:
“For any proposition p, that p is a relevant fact for S in C iff S’s objective reasons for action in C+ do not differ from S’s objective reasons for action in C.”
[there’s one formal problem with this as it claims that proposition p is a relevant fact but let that be…]
Now, assume that p is ‘that there is added enjoyment for Doug at home – only if the party remains a surprise for him’. According to the definiion, this can be a relevant fact for Doug in his actual ignorant circumstances iff there’s no difference between Doug’s objective reasons in the actual circumstances in which he doesn’t know about the party (C)and Doug’s objective reasons in the counterfactual circumstances in which he knows that there is added enjoyment at home because of the party that is there (C+).
The problem, however, is that, given that Doug knows about the party and potential enjoyment in the counterfactual situation, that fact that p ceases to provide a reason for Doug – there no longer is enjoyment at home because the surprise is spoiled. So, the objective reasons of the ignorant Doug (in C) and the cognisant Doug (in C+) must differ. In the latter situation, there is one less reason. And, by the lights of your definition, this makes the fact irrelevant.
The point is that some reasons are ‘ignorance dependent’ for their existence. Because of this, in C and C+, there will be different reasons for the agents who have these reasons. This makes the given facts not relevant by your definition. And, given DA, these considerations could not make a difference to what one ought to do. This will lead to unintuitive consequences. There’s plenty of such reasons as became clear in the debates about Williams and also about virtue ethics.

15. Hi Jussi,
I’m probably just missing something but I don’t understand this: “I don’t think that that can be right – especially the first sentence. This is because the enjoyment in question must be spontaneous and unexpected.”
Are you claiming that in order for the fact that S’s doing x would increase S’s enjoyment to constitute an objective reason for S to do x the enjoyment itself must be unexpected? Increased enjoyment from doing x seems to be a reason to do x whether the enjoyment is expected or not. Why would the enjoyment have to be unexpected?
I would admit that the subjective reasons that S has for doing x in his actual circumstances can be ignorance dependent, but I would deny that the objective reasons that S has for doing x in his actual circumstances can be ignorant dependent.
And what’s the formal problem? DA doesn’t claim “that proposition p is a relevant fact.” Rather it claims that p is a relevant fact if and only if certain conditions are met.

16. Jussi Suikkanen says:

Hi,
the formal problem: the quantifier ranges over propositions [‘for any proposition p’] and then some of those propositions are said to be facts [‘that p is a relevant fact…’] if certain condition is satisfied [‘iff…’]. That seems like a category mistake. Maybe the thing should start with ‘for any fact p, …’. This is just nitpicking.
I can slightly modify the example to bring up the problem more. Imagine that an agent for whom the surprise party has been arranged suffers from an extreme case of the paradox of hedonism. He cannot get any enjoyment whatsoever in circumstances in which he has before that event thought about future enjoyment. He can only enjoy things that come to him as a complete surprise. Any prior thoughts of enjoyment make him unable to experience any enjoyment at all. This is true for this agent even with such abstract and general thoughts as ‘future enjoyment provides a reason for me’. After thinking this, the agent will not enjoy anything for awhile. This is like the classic shy chameleon case in the colour discussion. Call this agent Jim.
Now, assume also that our theory of oughts (and reasons) is hedonistic. Only enjoyment grounds oughts (and reasons) for agents. Intuitively, even this does not rule out it being the case that Jim ought to do many things and that there are many reasons for him (like going to surprise parties, being amazed by the sunset, buy chocolate, and so on – all the things that lead him to experience enjoyment).
DA would then require for this being the case that Jim would want to do go to surprise parties, be amazed by sunsets, buy chocolate, and so on if he were aware of all ‘the relevant facts’ and reacted to them as reasons.
But, now, the relevant facts that serve as reasons here would need to be propositions that p which presumably included thougths about enjoyment in them. However, there is no more knowledgeable circumstance C+ for Jim such that he would be aware of these enjoyment-facts in them and they would continue to obtain. Starting to think about such enjoyment involving considerations makes Jim unable to experience any enjoyment. Whenever Jim becomes more informed, the future enjoyment facts change and his reasons change. So, there’s no more informed C+ for us to go to where the Jim’s reasons remain the same, and so there’s no relevant facts that could do the work in DA.
In any case, it seems enough for me that you deny that objective reasons that S has for doing x in his actual circumstances can be dependent on ignorance about them. This is hugely controversial. For instance, in the classic virtue ethics cases, the objective reasons to improve one’s character always depend on the fact that one doesn’t know that one’s character is not good enough. Knowing those facts requires a good character in which case one would not need any improvements. So, you must deny that such cases exist.

17. Hi Jussi,
I minor quibble about this: “the objective reasons to improve one’s character always depend on the fact that one doesn’t know that one’s character is not good enough.” I would have thought that it, at least, often (if not always) depends on one’s character needing improvement and that one has a reason to improve a character that needs improving whether one knows that it needs improving or not. I know that my character needs improving, but that doesn’t mean that my character doesn’t need improving. Although it may be that I can’t know that my character needs improving without having some good character traits, knowing this doesn’t require a *perfect* character “in which case one would not need any improvements.”

18. Hi Jussi,
If I understand things correctly, you can make your point more simply with the following case:
S has two boxes to choose from: A and B. S knows that one of the boxes contains a lot of money and the other contains none, but S has no idea which box contains the money. Now, if S chooses box A while not knowing that box A (not B) contains the money, he will receive the money in box A. But if he chooses box A while knowing that box A contains the money, Black will intervene and vaporize all of the money in A, thereby seeing to it that S receives nothing.

19. Boram Lee means Robert Johnson, in the comment above where it says ‘Robertson’ – which may be a compression of Robert’s name. Boram may also be thinking of Theresa Robertson’s paper (“Internalism, (Super)Fragile Reasons, and the Conditional Fallacy.” Philosophical Papers 32 (2003):171-184. which makes some points similar to mine (and more besides) in the paper Boram cites but Robertson is responding to a paper of David Sobel’s.

20. David Sobel says:

Mark: I wondered if that was the conflation too.

21. Jussi Suikkanen says:

Hi Doug,
well, I’m sure there are character cases like the ones you describe. I shouldn’t have written ‘always’ but rather ‘sometimes’. All we need are some cases in which to have the reasons for character building one needs insensitivity to those reasons in the first place.
I think your case does have the right structure too. You might think that S ougght to choose A and that he has objective reasons to do so. DA guides us to a counterfactual in which S considers all the relevant facts. Does the fact that there is money in A belong to these facts?
If it doesn’t then S would not want to choose A necessarily in the counter-factual case in which he considers the relevant facts. But, it doesn’t seem like it could be a relevant fact either by the definition of relevant facts. There’s no situation C+ in which S knows that there is money in A. If S finds out that there’s money in A, the money disappears. And, so there are no C and C+ where the reasons would be the same. And this makes the fact not a relevant one. And so we are back to the DA giving the wrong result.

22. Hi Jussi,
Having thought about it, it seems to me that your case (and my variant on it) are decisive counterexamples to DA combined with the above account of the relevant facts. Now, what I need to think about is whether I can combine the above account of relevant facts with my analysis of scrupulous securability without running into counterexamples like the one that you suggest. So, thanks, this has been very helpful.

23. I think that if we combine my account of scrupulous securability (below) with the my account of the relevant facts (also below), we don’t get any of the counterexamples of the sort that Jussi is proposing.
Scrupulously securable: A set of actions, αj, is, as of ti, scrupulously securable by S if and only if there is a time, tj, that either immediately follows ti or is identical to ti, a set of actions, αi (where αi may, or may not, be identical to αj), and a set of background attitudes, B, such that the following are all true: (1) S would perform αj if S were to have at tj both B and the intention to perform αi; (2) S has at ti the capacity to continue, or to come, to have at tj both B and the intention to perform αi; and (3) S would continue, or come, to have at tj B (and, where αi is not identical to αj, the intention to perform αi as well) if S both were at ti aware of all the relevant reason-constituting facts and were at tj to respond to these facts/reasons in all and only the ways that they prescribe, thereby coming to have at tj all those attitudes that she has, given these facts, decisive reason to have and only those attitudes that she has, given these facts, sufficient reason to have.
Relevant facts: Let S+ and her circumstances be just like S and her circumstance except that S+ knows in addition to whatever else S knows all other truths. Now, for any true proposition p, the fact that p is a relevant fact for S iff S+’s objective reasons for action do not differ from S’s objective reasons for action.
To illustrate how there is no problem for my analyses, take my variant on Jussi case:
Two boxes: Jones has two boxes to choose from: A and B. Jones knows that one of the boxes contains a lot of money and the other contains none, but Jones has no idea which box contains the money. Now, if Jones chooses box A while not knowing that box A contains the money, he will receive the money in box A. But if he chooses box A while knowing that box A contains the money, Black will intervene and vaporize both him and all of the money in A, thereby seeing to it that Jones dies and receives nothing.
The fact that the money is in A would not count as a relevant fact, for whereas Jones+ has an objective reason to refrain from choosing A (viz., that choosing A will result in Jones+’s death), Jones has no objective reason to refrain from choosing A and a good objective reason for choosing A (viz., that choosing A will make him rich). This means that it is not the case that Jones would come to have at tj the belief that the money is in A if Jones both were at ti aware of all the relevant reason-constituting facts and were at tj to respond to these facts/reasons in all and only the ways that they prescribe, thereby coming to have at tj all those attitudes that she has, given these facts, decisive reason to have and only those attitudes that she has, given these facts, sufficient reason to have. Therefore, Jones’s choosing A is scrupulously securable option. Indeed, it’s Jones’s best scrupulously securable option, which account for why he should choose A.

24. Jussi Suikkanen says:

Doug,
sorry – I just cannot follow any of this. There’s a point in which things get just too complicated for me. One problem is that I don’t have any pretheoretical intuitions about ‘scrupulously securable’. Without them, it’s hard to construct counter-examples.
Here’s something that would help. Take a version of Scrupulously securable that still includes the full information condition and explain to us a counter-example to that view that has the form of the conditional fallacy. If such cases exist, I predict that there are similar counter-examples to the ‘relevant facts’ view. This is just because of the structural features of the view. But, given that I don’t understand the initial view and what kind counter-examples there were against that view, unfortunately I cannot help you here.

25. Hi Jussi,
Yes, I know, the complexity is boarding on the absurd. (I spell out the need for all this complexity in the book, but don’t have space here to do so.) Hence, I tried to avoid talking about scrupulous securability and focused on DA instead. But since I have no interest in defending DA, that didn’t work out so well either. If you don’t have time for what’s below, I understand. But since you asked, here’s the counterexample to the full information version of scrupulous securability.
If you change clause (iii) of my definition of scrupulous securability such that it presumes that S is aware of all facts and not just the relevant facts, then the view that S should perform the best scrupulously securable option (viz., securitism) falls victim to the following counterexample, which Peter Graham suggested to me:
Two Boxes v2: “Before Jones are two boxes, A and B. [He may choose one or none of the two and, if he chooses a box, he will receive whatever money it contains.] Jones knows that there is a lot of money in A—he knows that there is more than \$1,000,000 in it—and none in B, but Jones does not know exactly how much money is in A. There is in fact \$1,786,444.23 in A. Black, however, is on the scene and will detonate a bomb killing Jones if he chooses A while at the same time knowing that there is \$1,786,444.23.”
The fact that there is \$1,786,444.23 in A is an objective reason for Jones to believe that there is \$1,786,444.23 in A. Thus, if Jones were fully informed and responded appropriately to this information, Jones would come to know that there is \$1,786,444.23 in A. And if he knew that, he would be unable to choose A without being killed by Black. So if I were to define ‘scrupulously securable’ in terms of full information, it would turn out that Jones’s choosing A is not his best scrupulously securable option. And so securitism would imply that Jones ought not to choose A. And this seems counterintuitive.
But if we keep clause (iii) as it is (so that it appeals to only the relevant facts and define relevant facts as I have above), then the fact that there is exactly \$1,786,444.23 in A would not count as a relevant fact, for whereas Jones+ has an objective reason to refrain from choosing A (viz., that choosing A will result in Jones+’s death), Jones has no objective reason to refrain from choosing A. This means that it is not the case that Jones would come to have at tj the belief that there is \$1,786,444.23 in A if Jones both were at ti aware of all the relevant reason-constituting facts and were at tj to respond to these facts/reasons in all and only the ways that they prescribe, thereby coming to have at tj all those attitudes that she has, given these facts, decisive reason to have and only those attitudes that she has, given these facts, sufficient reason to have. Therefore, Jones’s choosing A is scrupulously securable and, indeed, his best scrupulously securable option. So securitism rightly implies that he should choose A.

26. Dowell says:

Sobel also has several papers defending subjunctive analyses connecting desires with well-being and reasons for action. In (I think) “Subjectivism and Idealization”( Ethics, 2009) he defends them against various objections. I know Arthur Ripstein and David Enoch are among those who have offered objections to such accounts and am pretty sure that’s Dave’s reply paper.
Two other papers worth a look are his “Full Information Accounts of Well-being” (Ethics, 1994) and “Subjective Accounts of Reasons for Action” (Ethics, 2001).