This is the fourth of four posts on DPR. In this part I provide some simplistic but illustrative cost-effectiveness estimates comparing an imaginary campaign for DPR against current interventions for poverty, physical health and mental health; I also consider what EAs should do next.
Links to the articles in this series:
Part 1 (1,800 words): Introduction and Summary.
Part 2 (8,000 words): Six Ways DPR Could Do Good And Anticipating The Objections
Part 3 (3,000 words): Policy Suggestions, Tractability and Neglectedess.
Part 4 (3,500 words): Estimating Cost-Effectiveness vs Other Causes; What EA Should Do Next.
6. Speculative cost-effectiveness calculations
For the sake of argument, assume there is an effective campaigning organisation we could fund, or set up, if we wanted to bring about drug policy reform. How cost-effective would it need to be to be more cost-effective than other things effective altruists currently fund, such as Give Directly (unconditional cash transfers to those in poverty) or the Against Malaria Foundation (‘AMF’; bednets to stop children dying from malaria)?
In addition to Give Directly and AMF, I’m also going to add Basic Needs, a mental health charity that operates in the developing world, to the list, even though it isn't a GiveWell top charity. This is because I(’m an outrageous heretic and) think that, if you want to make people happier, it's probably easier to target misery directly by treating mental illnesses such as depression. By contrast, targeting disease and poverty seem a nuch less direct way to achieving the same goal. I’ve made this point elsewhere and won’t restate the case.
Technical paragraph non-philosophers may want to skip:
I’m also assuming we’re interested in happiness, where happiness is understood in a roughly Benthamite way as positive conscious states – those that feel good to you, that you enjoy – and unhappiness the opposite, as negative conscious states. All plausible moral theories think this matters, even if they don’t think it’s the only thing that matters, so let’s start with a concept of valuable mental states everyone at least shares (even if you think well-being consists in having your desires realised, one desire everyone presumably shares is feeling good). I’ll assume we can do interpersonal cardinal comparisons of happiness, i.e. we can compare units of happiness, where one unit to you feels as good as it does to me. This is just the QALY-approach, but using happiness instead of health. Let’s also say 1 is maximum sustainable happiness, 0 is neutral and -1 is minimum sustainable happiness.
If we’re going to do the comparison, what we need to know is:
- The cost of the intervention
- The number of people it could affect
- How long it affects each of them for
- The amount it increases their happiness by (on the -1 to 1 scale)
I’ll explain my thinking in words, but I’ve also put the figures in this google sheet I invite people to copy and use to create their own estimates.
Let’s consider AMF’s cost-effectiveness. Give Well estimate it costs $7,500 to save each child under 5 from dying from malaria. Assume that child lives another 45 years (developing world life expectancy), each year at full happiness (very improbable), then that’s one happiness-adjusted life year (‘HALY’) for each $166.6 spent. Or, that’s 60 HALYs/$10,000. (You need to have some quite implausible beliefs about population ethics and the badness of death to get these numbers, which I’ve discussed here, but let’s leave those aside for now.)
I’m less sure how to do the numbers for Give Directly. Suppose recipients are given $1/day extra and this increase their happiness by 1 (i.e. very improbably taking them from neutral to full happiness) for a year. This is $365 per HALY or 27.4 HALYs/$10,000. (I think this is extremely generous to Give Directly and I’m sceptical it increases happiness at all, which I’ve discussed here, but let’s leave this aside too.)
Basic Needs estimate it costs $14/participate/month to run their programme, which is $168/participant/year. Assume the effect lasts one year and increases happiness by 0.3 over that year. It costs Basic Needs $560 to generate one HALY. Or, that’s 17.8 HALYS/$10,000.
Currently, AMF is well in the lead with $166.6 per HALY, so that’s the one to beat. Now, let’s assume we could fund a rescheduling campaign in just the UK to change the schedule on just the psychedelic drugs, LSD and psilocybin (magic mushrooms). This would make it easier to use them to research and treat depression and anxiety, which together affect around 1/6th of the UK’s 66m population each week. Let’s round down the 11m figure down to 10m, to be conservative. Assume the research caused by the rescheduling reveals ways to increase the happiness of each of these 10m people by 0.1 for a single year (I think this is a conservative figure). [Update 14/08/2017: following Tom Sittler's comment below, I should have said that this rescheduling would increase the happiness of those 10m by 0.1 on average; it's unreasonable to assume all will get this treatment. However, we might assume that if some people get the new treatment, which I presume is cheaper, that frees up resources for more people, who wouldn't have received the old treatment, to get that in the first instance]
We don’t know how much it would cost to run a successful campaign, but we could ask the question the other way: how much could we spend on a successful rescheduling campaign – I.e. one that worked, got the law changed and allowed research to happen – and for that campaign to still be as cost-effective as our currently most cost-effective pick, AMF?
On the assumptions made above, the rescheduling campaign would generate 1m HALYs (0.1 HALY for each of 10m people). As AMF generates 1 HALY for $166, a successful rescheduling campaign costing less than £166m would be more cost-effective than AMF. If we spend $83m on the campaign it would be twice as cost-effective, if we spend $333m on the campaign before it succeeded that would be half as cost-effective as AMF, and so on.
To be clear, and before the figure is anchored in the mind, I am not suggesting this is how much a successful rescheduling campaign would, in fact, cost. The point is that if you had £166m to spend and thought you could pull off the rescheduling campaign for less than that, you should do that instead of giving you money to AMF as the rescheduling campaign is, in expectation, going to do more good.
What I suggested is a simple, relatively conservative cost-effectiveness estimate. I’ll now add a few more considerations.
You might worry the chance of success really varies with the size of campaign: you can’t believe a campaign for less than $10m would have any impact; thus, adding your $100 once there's a $10m pot is much more effective, in expectation, than adding your $100 to a £1m pot. This could be true, but if it is, this is a reason for a big funder to kick-start the campaign before smaller donors add their money, not a reason to give up on the project.
You might also worry that an extra $100 here or there can’t make a difference between a policy change happening and not happening. This is a ‘sorites’-type problem (e.g. "how many grains do you need to add before it become a heap of sand?") If you think $100 wouldn’t change people’s minds but $1bn would, you have to accept these extra $100s will matter somewhere because if you keep adding $100s, as eventually you’ll get to $1bn. Also, this objection applies just as easily to other interventions, such as AMF: how do you know any of your additional nets made the difference between life and death for someone? Presumably you accept the idea that $7,500 saves a life in expectation.
Having noted those two worries, I’ll now suggest my earlier estimate how much we should be prepared to pay for the rescheduling campaign was conservative.
I assumed the happiness impact was 0.1. We might expect someone who is depressed or anxious to be below 0 (as 0 is the neutral point, this means there are mostly unhappy) on a -1 to +1 scale. Potentially, someone could go up 1 point if they’re going from -0.5 to 0.5, so my 0.1 should be upgraded ten times to 1.
However, counted against this is consideration the size of the impact is really counterfactual, not absolute. We should be asking: how much better will the treatment they received due to the rescheduling campaign be than what they would have received without it? This can get a bit complicated: maybe the psychedelic-based treatments would be cheaper and more effective, so the government could treat more people; maybe the drugs offer only a tiny improvement; maybe many of those people would never sought or received treatment in either case; maybe the fact some people get cheaper, more-effective psychedelic-assisted therapy frees up resources to treat those who wouldn't have been treated with current therapy; etc. Having considered the counterfactuals, let's now divide the cost-effective estimate by three. As we increase the estimate by 10 in the last paragraph and reduce it be 3 in this one, the net increase is 3.3 (i.e. 10/3).
Next, we should recognise that the rescheduling policy, if it occurred, wouldn’t last for just one year, but would last in perpetuity and continue to provide better treatment than the status quo. This also needs to be counterfactual: if some group of EAs didn’t campaign for rescheduling, how many years would it be before it happened anyway? 5? 10? 50? Let’s say the campaign counterfactually makes the policy happen 10 years earlier, so it has an effective duration of 10 years. An additional confusing thought is that, if we 'solved' DPR, then current drug policy campaigners would probably move on to do something else good instead. Let's ignore this additional fact as it's unclear how this 'replacement' feature is going to play out.
If Britain changed its drug policies, this might have a knock-on effect around the world as other countries took note and copied. Alternatively, this might be isolated just to Britain. Total isolation seems unlikely, so let’s assume a spillover effect and double the number of people affected.
In considering the rescheduling of drugs for mental health, I haven’t factored in any of the other 5 arguments I mentioned in section 2. It seems, if you’re going to campaign for drug policy reform, once you’re campaigning for rescheduling, the additional cost of campaign for decriminalisation and/or legalisation is presumably quite small and, if the rescheduling did happen, it might do so alongside other drug policy reforms. I haven’t assessed the impact of the other 5 arguments but think the impact of mental health is likely to be by far the biggest anyway. Let’s cautiously add 5% on top of the effect.
For this more optimistic estimate we multiply our original £166m figure by 3 (counterfactual impact) x 10 (counterfactual duration) x 2 (spillover) x 1.05 (other benefits of drug policy reform) which is a multiplier of 63, giving is £10.5bn, a rather high number. While the number is large, I don’t find it implausible. It’s the result of bringing about a systemic change could potentially improve the lives of very many people by a sizeable amount.
What the estimated figure of £10.5bn means is that, if you believed an imaginary rescheduling campaign would succeed with less than £10.5bn of funding, then you should think that campaign would be more cost-effective than giving your money to AMF and, therefore, you should support it over AMF. I accept this is currently hypothetical – I haven’t identified a place readers could send their money to (yet) – but I would ask the reader to make their own guess about how much money it would require to run a successful rescheduling campaign. This will give them a sense of how cost-effective they think drug policy reform is compared to the other charities I mentioned (AMF, Give Directly, Basic Needs). If you believe a cool £100m is all that’s required to get the laws changed, the rescheduling campaign would be 105 times more cost-effective than AMF.
If we want an even bigger number to chew on, consider that around 500m people have either depression or anxiety worldwide. If we could improve their happiness level by 0.3 and do so 10 years earlier, we could spend up to $250bn on that campaign and it would be as cost-effective as I’ve assumed AMF is.
I’ll now anticipate two objections to my cost-effectiveness estimates and the conclusion I reach.
First, one could accept everything I’ve said is true, but argue another cause is still more cost-effective. I’ve claimed DPR looks particularly good at increasing happiness for presently existing people. However, you might, for instance, be a total utilitarian whose wants to maximise the happiness of the history of the universe. If you were such a person, you’d care not just about the happiness of those alive today, but about the happiness of all future, possible people too, and might think existential threats to humanity, such as AI safety, as more pressing.
For those who think something else is more important, I would be very grateful if you could produce some (very rough) estimates of how many times more cost effective money to their preferred cause is than DPR. As far as I’m aware, there is only one cost-effectiveness estimates comparing near-term causes like Give Directly and AMF to far-future ones, (Michael Dicken's, which I'm not smart enough to use) so I don’t know how much better X-risk charities are supposed to be. As all plausible moral theories hold improving the happiness of existing people is good, even total utilitarian X-risk advocates should be prepared to support near-term altruism if it can be done cheaply enough (e.g. if you think MIRI, an AI safety research charity, is 5x more effective than AMF, but then conclude DPR is 10x more effective than AMF, you should switch to DPR). Certainly, even if you’re largely uninterested in the happiness of present people, the long-run effects benefits of DPR are considerable: stopping the War on Drugs with its associated crime, corruption and instability, as well as helping potentially half a billion mentally ill or drug addicted people get back to work would be quite an economic and societal boon.
Second, one might object DPR only looks attractive because I’ve used a suspicious mechanism to generate the expected value calculations: systemic change campaigns look (delusionally) effective because they have a small chance of affecting so many people. Roughly, the complaint is that I’ve found a new Pascal’s Mugging. I’m not sure how suspicious this kind of mugging is: I’m at least talking about real, concretely existing people, rather than conjuring up a near infinity of possible people. There doesn’t seem to be anything strange about systemic changes per se; everyone should accept the abolition of slavery, which affected millions, was a substantial systemic change that had a large positive impact. For those who think my estimates are too generous I would welcome them pointing out exactly which part they disagree with; that would be helpful and allow me to improve them.
An alternate way of pressing the second objection would be to accept the type of expected value calculations I’ve used but claim they don’t favour DPR more than any another cause. The idea here is to claim “fine, but all systemic change campaigns look ludicrously effective”. The critic could then generate some additional numbers to show, for instance, it has a higher expected value for private individuals to support a campaign that lobbies governments to increase international development spending than it does for those donors to send their straight cash to Give Directly (or AMF or Basic Needs). On this sort of analysis, one could object what my argument has really done is push EA towards systemic change interventions and away from ‘sticking plaster’ interventions (those which help one person at a time), such as Give Directly, in general, rather than push EAs towards drug policy reform in particular.
I think it is possible that systemic change campaigns could have higher expected value than their ‘sticking plaster’ alternatives. This would be a very interesting result and I would like to see people producing worked out systemic changes estimates for say, poverty and physical health. However, my sense is that DPR is uniquely well placed to be a good systemic change intervention. This is because not only would it not cost governments any more money, it could generate lots of revenue. Drug policy reform is the in the enlightened self-interested of taxpayers and governments. In contrast, raising taxes to fund greater aid spending runs counter to the self-interest of taxpayers, and taking money from one part of government spending to increase international aid will similarly meet resistance from whoever loses by this redistribution. There may be some, but I can’t think of any other policy changes that would simultaneously reduce costs and increase happiness and so do without even an initial investment from the government.
7. What should EAs do now?
I admit I don’t have a top charity EAs should give to, nor an ironed-out plan for DPR campaigning; I haven’t got that far. I thought the prudent thing to do, in the first instance, would be to write this up and see if others agree it is an important, unrecognised cause. If there are some crucial considerations that render this area unpromising, it makes sense to establish that now before spending time trying to plan the next steps are in detail.
Somewhat glibly, my answer to the question “what should EAs do now?” is “Answer that question.” Supposing people do agree this is important, I think what’s needed is more research to figure out what to do. I don’t think I fully understand what the best way to tackle this problem is. More concretely, some obvious next steps would be to talk to the charities in the area, get their thoughts and try to assess where money, time and research could be best used.
There are also a whole host of questions I’ve littered through this document I think need answers. I’ve collected them here in case anyone wants to help. These are in the order I raised them, not necessarily of importance:
- How effective and how expensive are treatments of anxiety? Do people relapse? I only have information for depression currently.
- What is the worldwide scheduling system on drugs? How does it differ from place to place?
- How much happiness might be gained from arguments 2-6 that I didn’t really included in my cost-effectiveness speculations? Questions that need to be answered to find this out include:
- How valuable is the illicit drug trade? Does the drug trade fuel other crimes including terrorism? How much crime, corruption, etc. would be removed by legalisation? How much happiness would this create?
- How much do governments spend on locking up drug users? How much does getting a criminal record impact one’s life prospects and happiness?
- How much could governments raise from legalising and taxes drugs? How much good could this extra money do?
- What would the recreational benefits, if any, be from legalising (some) drugs? How big is this compared to the other benefits?
- What are good ways of thinking about tractability? How effective are public opinions or lobbying campaigns? What are good comparisons to make?
- What are the best campaigning organisations working on this? How do we assess how effective such organisations are? Should we just fund them or try and start something of our own? If so, what?
- What should effective altruists do next?
 J Bentham, An Introduction to the Principles of Morals and Legislation, 1789.
 ‘Maximum ‘sustainable’ happiness refers to the highest average happiness level one can keep up over a lifetime, which contrasts with maximum ‘peak’ happiness, the most intensity happiness could can feel at a given point, which is presumably higher than maximum sustainable happiness.
 GiveWell, “Against Malaria Foundation | GiveWell,” 2016, http://www.givewell.org/charities/against-malaria-foundation.
 Basic Needs, “Basic Need Annual Report.”
 NHS, “Adult Psychiatric Morbidity Survey: Survey of Mental Health and Wellbeing, England,” 2014, http://content.digital.nhs.uk/catalogue/PUB21748.nnhs
 The paradox of the heap: if you take away a grain of sand from a heap of sand, when does the heap stop being heap? No single grain seems to make a difference but if you keep taking grains away eventually they’ll be nothing left, at which point there must be no heap.
 Michael Dickens’ cause prioritisation model can be found here (http://mdickens.me/causepri-app/). Sadly I am not mathematically competent enough to tweak so it matches my assumptions. I also understand CEA are working on a model comparing near-term to far-future causes.