A common objection to my drinking is that coming home blackout drunk imposes obvious costs on my wife, my children, and the household. I agree that these costs are real. My narrower claim is that the relevant unit of analysis is not the immediately legible domestic harm, but the full counterfactual impact of preserving the unusually high-earning version of me.
I currently earn substantially more than my wife. This matters. If one adult in a household has a much higher marginal earning rate, standard comparative-advantage reasoning implies that some unpleasant but low-leverage tasks will rationally be shifted toward the lower-opportunity-cost partner. Bedtime, kitchen reset, locating my phone, and interpreting my slurred apologies should be analyzed in this framework before we moralize.
Let the expected utility of my current arrangement be:
EU = Y + D + R − H − K
Y = household income preserved by my continued employment
D = donation-funded external welfare gains
R = decompression / social-function benefit of drinking
H = spouse-borne household costs
K = expected child distress and relational damage
The key neglected term is D. A meaningful fraction of my discretionary income is donated. At GiveWell-style cost-effectiveness estimates, even modest preservation of my earnings plausibly funds real welfare gains elsewhere. It is emotionally salient that the harmed person is “my wife” and the beneficiaries are anonymous strangers, but moral patienthood does not decay with distance from my kitchen.
There is also a stabilizing term R. Drinking is not pure hedonic leakage. It currently plays a role in stress regulation, client dinners, social bonding, and my willingness to remain in a demanding private-sector job instead of impulsively quitting to “find myself.” If reducing alcohol meaningfully increases the probability of burnout, resentment, or lower earnings, then “just drink less” is not a neutral intervention.
None of this is to deny H or K. My wife experiences concentrated disutility. My children sometimes encounter what she has termed the “why is Dad asleep on the stairs” problem. These are serious negatives. But unless one assigns lexical priority to harms that are vivid, domestic, and adjacent to me, the case is at least decision-theoretically open.
This is his wife, and I'd like to dispute the comparative advantage section on its own terms. It's true my hourly rate is lower. But bedtime takes me twenty minutes and it takes him fifty when he's sober, so his absolute disadvantage is big enough that I'd be doing it anyway and the earnings argument isn't needed. Where the model really undercounts me is the kitchen reset, which on a Saturday includes whatever he was eating when he fell asleep. Last month that was a rotisserie chicken.
I'd want to see the timing data on bedtime. My fifty minutes includes stories.
Happy to share it. I've been logging since September.
Strong upvote, though the model treats you as one agent and I don't think you are one. The person who writes this post at 2 p.m. and the person on the stairs at 1 a.m. have different preferences, and only one of them is doing the cost-benefit analysis. If you model them separately, the question becomes whether the daytime agent can bind the nighttime one, which is a commitment-device problem and has fairly well understood solutions.
I find the two-agent framing useful. I'd note that the nighttime agent is also the one who attends client dinners, so binding him has a direct cost to Y.
Can you say what fraction actually gets donated, and to what? I ask because D is carrying the whole argument and it's the one term with no number attached.
I haven't formally pledged. Donations are currently deferred until my income stabilizes. The money is invested in the meantime, which arguably increases D in expectation.
For what it's worth the give-later case is stronger than this site usually admits, and I don't think it should be dismissed because the person making it in this thread happens to drink. The basic argument: if your investments return more than the rate at which the best giving opportunities get worse (low-hanging fruit gets picked, global poverty falls, the cheap interventions get fully funded), then a dollar given in 2035 does more good than a dollar given today. A broad index fund has historically returned something like 5% real. Estimates for how fast cost-effectiveness declines run from about 2% to well over 8%, depending on who you ask and what they'd like to be true. At the low end waiting wins by a lot. At the high end waiting is a mistake. There are decent reasons for individuals in particular to distrust their own ability to wait. Value drift, life events, the future version of you not sharing your current commitments. Those are objections to the giver, though, and the strategy itself holds up. I wrote a much longer version of this in 2023 and got yelled at by the GiveDirectly people, so I'll stop here.
I think we've drifted, but the give-later argument assumes the money eventually gets given. That's a pretty strong assumption to make about anyone, including yourself in ten years.
The comparative advantage section is mostly right but it can't cover "locating my phone" or "interpreting my slurred apologies." Comparative advantage tells you who should do the household's work. Those two are work you generate, so assigning them to her is pushing an externality onto her, which is a separate problem with a separate (and less flattering) name.
That's fair. I've reclassified phone-locating from household production to an externality I generate, and I'm internalizing it by buying an AirTag.
I've done a version of this for my own weed use and got to roughly the same place, so I'm sympathetic. What broke it for me was that H isn't constant. Every incident makes the next one cost more, it compounds. Model H as path-dependent and the expected value goes negative a lot faster than the linear version suggests.
Path dependence cuts both ways. If H compounds then the history is already priced in, and the marginal cost of any single night is lower than it looks.
That isn't how compounding works
It's how habituation works.
Lexical priority isn't the only alternative to your view. You can weight harms to the people you live with more heavily without making them trump everything, that's what special obligations are for, and most EAs accept some version of them.
Special obligations are typically invoked to avoid donating, so I'm suspicious of them in this context.
Nobody has mentioned the children's own future earnings. If K reduces their human capital it reduces their lifetime earning-to-give, which at a reasonable discount rate is probably larger than your D.
I'd considered this. They're four and seven, so their donation horizon is too distant to dominate at any discount rate I'd defend.
On R: in my industry drinking at client dinners is close to mandatory and nobody who hasn't done it believes me. So I'll defend that part of the post at least.
Not in mine. Four years without a drink and my book of business is fine. People order a club soda and nobody remembers.
Though logically sound, this formula omits a significant factor—kin selection. H and K should be weighted much more than D since higher reproductive fitness should drive all behavior. Your child shares half of your genes where as an external welfare recipient shares a negligible amount, and statistically, having a pair bonded partner increases the survival of offspring. Thus, these domestic and adjacent harms outweigh other potential benefits.