Summary: I argue that, contra a recent post, even CEOs of tobacco companies do not do enough direct harm to offset the good they could do by donating their income.
Recently 80,000 Hours wrote a blog post about how CEOs of tobacco companies do so much harm that, even if they donated everything they earned to a highly effective charity, they would still be a net negative impact on the world.
Their argument is pretty simple
- In aggregate tobacco CEOs are paid $23 for every premature death caused by smoking
- Top CEOs can affect share prices by 5-40% vs the next best CEO
- In a commodity business share prices will track sales quite closely
- Hence top CEOs can affect sales by 5-40% vs the next best CEO
- Assume 50% of the extra sales will come from gains in market share, and 50% from expanding the market for cigarettes
- Hence the top CEO is paid $115 - $920 per counterfactual premature death.
- Premature smoking deaths cost about 10 QALYs as they tend to happen quite late in life.
- So a top tobacco CEO paid $1,000,000 would cost between 22,000 and 175,000 QALYs
- It costs AMF $3,000 to counterfactually save a life.
- The lives saved by AMD occur early in life, so AMF saves about 60 QALYs per life.
- So if the CEO donated the $1,000,000 they could save 20,000 QALYs
So in the best case scenario, even if the CEO donated 100% of their compensation, they would be approximately neutral. In the worst case, their actions would cost approximately 150,000 QALYs. As such, this lends considerable to support to Rob's conclusion, that:
"There are very likely jobs such as this, where even considering ‘replaceability’, and even if you donate all your earnings effectively, it would not be enough to offset the direct harm done."
However, I think that while 80,000 Hours substantially improved on the analysis in the original paper, they omit a number of factors. Unfortunately, almost all of these factors seem to pull in the same direction, causing them to over-estimate the amount of harm done by a tobacco CEO. These include both over-estimating the direct harm done and under-estimating the benefits donating your income would cause.
Over-estimating the harm.
There are many ways in which CEOs can enhance shareholder value that do not involve selling more units. To the extent they do these things instead, we will tend to over-estimate the harm they do as a byproduct of promoting shareholder value.
Sales vs Volume
The article argues that,
"in a commodity industry like cigarettes, share price and sales are likely to track one another fairly closely."
"Let’s say the top choice for CEO will raise sales 5-40% relative to their next best alternative.2 In that case, these CEOs are actually being paid $60-$460 for each death resulting from the cigarettes they cause to be sold."
However, Cigarettes are definitely not a commodity industry. For the sake of specificity, lets look at Altria, a large US tobacco company. In 2014 it made an operating profit of $6,900m on revenues of $21,900m - over a 30% operating margin. This is a very high margin, well above that of most companies, and far in excess of that earned by commodity manufacturers.
There are a number of other features of the tobacco market that also suggest it is not a commodity market:
- Tobacco companies earn a very high return on invested capital, where in a commodity industry this would be competed away by new entrants
- The retail price is very stable (compare to price of canonical commodities, like oil or corn)
- There are large price differences between brands
Indeed, cigarettes seem to be about as far as you can get from being a commodity, on a similar level to other consumer staples with a high degree of branding power.
As a result of this, the article omits a key way that CEOs can improve their company: raising revenues by increasing the average selling price (ASP), rather than selling more units.
A key part of the strategy of tobacco companies has been aggressively raising prices to take advantage of the low elasticity of demand. The high level of taxes of tobacco products also helps them, as a large price increase only increases the cost to the end-user by a small percentage - which many of them will blame on the government anyway.
To show how much value can be destroyed when tobacco companies fail to stick to this script, we only have to look at Marlboro Friday, where Philip Morris cut the price on their signature product by 20%. Their share price fell by 26% that day, as the market expected this would dramatically lower profitability. If cigarettes really were a commodity, producers wouldn't be at liberty to make pricing decisions like that; oil companies can choose how much to produce, but they cannot choose the price they sell at.
Improving production efficiency
Even if tobacco was a commodity industry, there is another simple way CEOs can enhance shareholder value without selling more units: cutting production costs. Indeed, the investors are frequently far more excited about plans to improve margins by making manufacturing more efficient than by expansion plans. This is especially true in a mature industry like tobacco: cost cutting is seen as predictable, low risk and within management's control; expansion plans require expensive upfront investment and high risk. You might have to lower prices, undermining margins on the existing business, especially if competitors lower prices in response.
Stock prices vary largely as a function of expectations of future profitability, but not entirely. They're also a function of the perceived riskiness of future profits. If you can make a company seem less risky to investors you can increase its worth, even if the expected profitability is unchanged.
This is especially evident in the tobacco Industry. Altria currently trades on a 22x trailing Price/Earnings multiple, well above the average for the market, in spite of the lack of growth in the underlying tobacco market. This is largely because investors perceive the company as very stable and low-risk. This perception is not a coincidence - it is (in part) the result of the decisions of management teams in the industry to reduce volatility.
Are you really the best choice for the job?
There are broadly two different ways of examining the effect of CEO quality on firm performance. One is the approach used in the HBS paper Robert mentioned, where you look at variation in various metrics cross-sectionally and temporally and attempt to correlate that with CEO tenure. The other is to look at market reactions to unexpected CEO terminations - for example this paper. A key distinction between these is that the former looks at how much variation can be explained by CEO choice, while the latter looks at how predictable the impact of different CEOs is. It should be of no surprise that the former type of studies tend to produce larger estimates of CEO impact than the latter: CEO choice might matter a lot, but it's hard to know ex ante how good the CEO you chose will be.
This introduces another factor we need to control for. Yes, if you really are better than the alternative CEO you might sell more cigarettes, and yes the board clearly thought you were the best choice for CEO - but what if they're wrong? We need to adjust by the probability that you are indeed the best choice for CEO, conditional on the board thinking you were.
This seems like a pretty hard probability to estimate. My guess is it is quite low - I would expect many potential applicants, and a relatively poor ability to discriminate between them - but in lieu of actual analysis lets just say 50%.
Under-estimating the Benefit
There are probably charities better than AMF, even ex ante
GiveWell recommends AMF so highly not because they think that AMF is the best charity in the world. They don't even think it's the best charity in expectation - it's just the charity they're most confident has a very high level of effectiveness. If you're happy taking on a little risk (uncorrelated to the stock market) in return for a better expected outcome, you can probably get at least another factor of two from your donations.
Right to choose
Sometimes people criticise EA for being paternalistic. Who are we, busybody altruists, to decide what is best to others? What makes us think we are so smart? Isn't this just paternalism?
I think in this case the response in the EA FAQ is excellent:
Some people support GiveDirectly because it gives cash to people in poverty, leaving it entirely up to them how they spend the money. This is seen as empowering people in poverty to a greater extent than choosing services that may ultimately not be desired by the local community.
Others provide basic health services, such as vaccinations or micronutrients, that they regard as so clearly good it is very unlikely the recipients wouldn’t value them (and in any case, could reject if they wanted). They then hope that better health will empower people to improve aspects of their own circumstances in ways we as outsiders cannot.
In cases where the above don’t apply, you have to conduct detailed impact evaluations to see how the recipients actually feel about the service that purports to help them. Of course such surveys won’t always be reliable,
In other cases where people are trying to help non-human animals or future generations, these issues can be even more difficult, and people do their best to predict what they would want if they could speak to us. Obvious cases would include pigs not wanting to be permanently confined to ‘gestation crates’ in which they cannot turn around, or future generations not wanting to inherit a planet on which humans cannot easily live.
This response is amazingly reasonable, and it covers basically everything EAs do. What's notable is that nowhere do we tell people that their preferences are wrong. We aim to help people promote their goals by giving them money, or trying to work out what they would want, or even helping those who cannot help themselves.
Yet in the tobacco cost-benefit analysis, this is exactly what we do. We count as a cost the negative side-effects of smoking, but give no credence at all to any benefit to smoking. Some people just like smoking! They enjoy it. Perhaps smoking is a mistake - maybe smokers have miscalculated the personal cost-benefit analysis, or maybe they're too addicted to be able to quit - but there is a consumption good that they get from smoking, and it has to be accounted for in any cost-benefit analysis. Failing to do so stacks the deck against tobacco, and against the tobacco CEO.
Here are the adjustments we have to make to the estimate of the harm caused by a Tobacco CEO's direct work
- Price vs Volume - maybe a factor of 2?
- Sales vs Margin - maybe a factor of 2?
- Equity Risk - uncertain impact
- CEO choice uncertainty - maybe a factor of 2? (very uncertain)
- Better charities than AMF - maybe a factor of 2?
- Right to Choose - uncertain impact
Just eye-balling the length of the list, it is clear that this is going to make a big impact!
Several of these factors seems to require an adjustment by at least a factor of two. Using these highly unscientific numbers, we adjust down the harm caused by at least a factor of 2*2*3 = 8, to between 1,360 and 5,500 QALYs, and increase the good done from donations by a factor of at least 2, for 40,000 QALYs. Net this leaves us with a 34,000-39,000 QALY positive impact from a career as EA tobacco CEO, and without giving any credit for the two of the factors. Of course in practice few would actually donate 100% of their income - but your net impact could still be clearly positive even at much lower donation rates, especially if we consider that you might also be able to differentially promote less harmful types of cigarettes.
Now, this doesn't mean that 80,000 Hour's conclusion - that there are some jobs so bad that even replaceability concerns won't make Earning to Give in them net beneficial - is false. But Tobacco CEO does seem like it had a strong prima facie case. Who could be more evil? If even selling cigarettes isn't harmful enough to offset the good done by your earnings, what is?
I'm not sure why people use this estimate, given that the effect of anti-malarial nets is primarily on avoiding the disease itself, the grief of family members, economic costs, and other downsides of having malaria, rather than on creating more years of happy life. This is because population tends to adjust for the death rate, i.e. "I think the best interpretation of the available evidence is that the impact of life-saving interventions on fertility and population growth varies by context, above all with total fertility, and is rarely greater than 1:1."
Not all EAs, or even all utilitarians, believe in total utilitarianism. The person-affecting view is much more intuitive for most people, and by that metric, saving QALYs of people already alive is qualitatively distinct from adding more QALYs by popping out more kids (even after adjusting for the second-order effects).
I am one of the authors of the original paper. Thanks for all the good discussion on offsetting harm. But really that was a small part of the paper-the main idea was setting up a pay structure for the CEO to incentivize saving lives. Anyone care to comment on that? Maybe a separate post?
Hey Larks, thanks for the response to the article. As always when doing a Fermi estimate like this, at some point you have to stop and hope that the remaining unexamined factors roughly balance out. This brings us closer to an accurate estimate. The original estimates were within a few factors of one another, which left open the possibility of them flipping if there were significant changes made.
Here are the headings I basically agree with:
Right to choose There are probably charities better than AMF, even ex ante
I think Michael is right about the question of whether you're the best CEO.
I think that these ones are less clear:
Sales vs Volume Improving production efficiency
While your evidence convinces me that cigarettes aren't a commodity at the sales end because of branding (in developed countries anyway), presumably they are on the production side (you allude to this by calling the industry 'mature'). If you can introduce an innovation that lowers cigarette production costs for your firm, that will presumably in time be implemented across the entire industry. The weakness here is that most of the price of cigarettes in rich countries is tax, so a 50% reduction in production costs leads to a much more modest decline in sales prices.
The case study where prices were reduced and the share price also went down suggests inelastic demand with respect to price on this product (in the USA anyway), so little temptation to compete on price.
I'll give this some more thought.
"But Tobacco CEO does seem like it had a strong prima facie case. Who could be more evil? If even selling cigarettes isn't harmful enough to offset the good done by your earnings, what is?"
I agree - I was surprised that the answer I got was as close as it was. One qualification is that while selling cigarettes may be bad, being CEO is an unusually lucrative position. There could be poorly paid roles that do a lot of harm (for example, being a soldier in a bad military). Though of course such jobs are less relevant to anyone considering 'earning to give'.
Sorry to rain on the party, but who even cares? No one is going to become a tobacco CEO to earn to give. 80k is supposed to be an effective altruist nonprofit organization that helps people with their careers; they should have better things to spend their time and donor money on than such absurd and obscure questions. Not to detract from your post - analyze it all you want, although again I don't see why you would determine this to be a topic worthy of time and attention.
The people who think finance is harmful won't be persuaded by adding up tobacco deaths and doing this kind of math. They believe that donating to charity is not nearly as good as the numbers suggest, or they believe that finance is worse than being a tobacco CEO, or they believe that charity doesn't morally offset direct harm, or they believe that the entire paradigm of calculating how many people die or live and adding those numbers up is totally misguided from a methodological standpoint. Moreover, this wouldn't even counter the steelmanned argument which is that even if earning to give is not harmful, it's still not nearly as good as direct work, and that is what the most authoritative critics (e.g. Deaton, Acemoglu) have been saying. Seriously, how many people are there who would seriously think "Wall Street financiers are exploiting the poor more than their charity will help - but, wait, if you add up tobacco deaths and subtract it from AMF lives saved, you get a positive number, so working in finance must be okay after all" and then make a decision accordingly?
If it gets people interested in relevant content, then I suppose that's fine, as long as everyone keeps in mind how fine the line is between pretending to be clickbait and being clickbait.
If there was a good article concluding that financial work did so much counterfactual harm you couldn't more than cancel it out through donations I'd be happy to write a response to their methodology. But at least I have yet to come across any articles that even attempt to do the calculation. The ones I've seen basically just assume that the reader has a very strong anti-finance prior.
I've genuinely found it pretty useful in the past when people say 'you could justify any career no matter how evil that way' to say 'actually there are industries that 80k doesn't think are that great, e.g. tobacco'*. I haven't had an easy way of linking to that opinion before; now I do. So as an 80k donor I'm glad they also see the value of distancing themselves from the 'it's all good as long as you donate' position.
*For a example from less than a month ago: https://www.facebook.com/groups/effective.altruists/permalink/990753924314298/
OP: "Several points in this article intrigued me, but what I would like reaction to is that there is a massive flaw in the seeming innocence of the work to give model, if that employment is based on an inherently exploitative system. As I see it, very few of our high paying positions in the US are not based on a system which exploits the labor and/or resources of people on other parts of the globe. Is it possible to calculate for the most good, while adjusting for exploitation?"
Me: "...An alternative would be for me to decide that my job does so much bad that it can't be outweighed by even the most-targeted donations. I think that claim fails a basic sanity check as Tilde has already highlighted, but again if the author and those who think like her would like to actually engage quantitatively with that analysis, I'd love to hear from them. It's worth noting that there are a small number of industries that 80k largely avoids (e.g. Tobacco) due to exactly this concern."
I don't know if that convinced the person in the thread, but rationally that doesn't really look like a sound argument. If finance exploits people too much then it's still a problem regardless of what 80k thinks tobacco CEOs happen to do. Moreover, the article about tobacco will make 80k look even worse to these types of critics, because it's implying that something even more morally repugnant than finance could be justified if the math worked out a little differently.
What exactly is rationally unsound about responding to 'You/80k could justify any career that way' with 'Actually, I/they still don't think it justifies this career'? You might disagree with the response or prefer to just focus on the direct career in question*, but it certainly seems like a valid response.
I feel like I'm missing something.
*I only didn't do this because someone else (Tilde) had already done it; if you look you'll see I did reference their response.
Edit: I realise that my OP makes it look like I'm trying to make an original argument. I'm not; Rob makes pretty much the exact same point in bold on the 80k post:
"The key point I want to highlight is that there are very likely jobs such as this, where even considering ‘replaceability’, and even if you donate all your earnings effectively, it would not be enough to offset the direct harm done.
If you are trying to do the most good by earning to give, you will probably do better by looking for a job that both pays well, and has either neutral or positive direct impacts."
The point still holds because you still could justify being a tobacco salesperson in this way - if the math went a little differently, if you decided charity had better flow through effects, if you decided that smoking was not as bad as the article's assumptions, etc - then you would justify being a tobacco salesman. If you wanted to alleviate animal suffering or x-risk then your calculations could come out in favor of that. That's the problem: people have a moral issue with the idea that it could be acceptable to harm some to donate to others.
I'm sure he means well, but it does nothing to answer the criticisms: again, just because it works out this way for this one job with this one calculation doesn't necessarily imply anything about other jobs with a negative impact. It doesn't answer the moral objection to the assumed framework.
"That's the problem: people have a moral issue with the idea that it could be acceptable to harm some to donate to others."
Oh, I see. You think I'm addressing this argument to someone coming from a non-consequentialist framework who has inherent issues with dealing small damage to A through your work in order to deliver large benefits to B through your donations. If that were the case I agree this would be an unsound or at least barely-relevant response.
But that's not the case. I'm addressing this to people who just have a feeling that the damage you do is greater than the benefits, i.e. they don't have a problem with the framework, they are just suspicious of the seemingly fully general nature of the earning-to-give argument. E.g. the OP in the example I gave explicitly refers to doing a 'calculation' for the most good. If the OP had been about the moral un-virtuousness of EtG instead, then for sure I should respond differently.
Apart from that misunderstanding I don't have any substantive disagreement with your points.
To be clear, you think it's worth spending a few hundred quid apiece for linkable resources like this? (I.e. you'd pay others a similar amount in an impact purchase if they did so.)
Given that this particular straw man attack tends to be addressed to 80k and their public platform is relatively stronger than this forum or random blogs, there's obviously higher value in it being hosted on the 80k blog. I wouldn't be averse to someone else writing it and them hosting it though.
Btw, where are you getting a few hundred quid from? This strikes me as very much the type of thing Rob might spend some of his own time working on and thinking about anyway; the estimated cost of refining that into a blog post seems very high to me. Rob should obviously feel free to contradict me if I'm off-base.
Nice analysis, Larks!
You don't actually care about the probability that you're the best CEO. If you chose not to work there, the company wouldn't necessarily hire the best CEO in your place. Instead you care about how much better you are in expectation versus who the board would hire instead of you. Even if you're unlikely to be the best CEO, you might still be better in expectation than any other candidate.
You're right, you'll presumably be better in expectation than the next candidate. But the research that yields the high CEO impact numbers tends to look at actual performance differentials, rather than expected - clearly their expectation will be much lower, so we need to scale down the statistics these studies provide.
You seem to shift here between p(Best among applicants) and p(Better than the guy who would have been hired in lieu of you). Guesstimating 50% for the former sounds reasonable-ish to me, but I would guess it's substantially higher for the latter.
Maybe this comes out in the wash, since the difference between you and your actual replacement is smaller in expectation than the difference between you and the best among all the applicants.