I hear a lot of theories around how to work optimally. “You shouldn’t work more than eight hours a day.” “You can work 12 hours a day and be fine.” “It’s important to take weekends or evenings off work entirely.” “It’s best to immerse yourself in your work 24/7 if you want to be an expert.”
Perhaps most well known is Cal Newport’s claim in Deep Work that “For a novice, somewhere around an hour a day of intense concentration seems to be a limit, while for experts this number can expand to as many as four hours—but rarely more.”
Many of these theories are asserted with surprising confidence...especially since they contradict each other. At least some have to be wrong or more nuanced, and it matters which are right.
I coach Effective Altruists who want to maximize the good they can do. So they want to know how much they can work before additional work is wasted (or just less valuable compared to extra time doing other things). They also want to know how much they can work before risking reducing their long-term productivity -- burning out from working too hard is a lose-lose for them and the world.
So, I dug into these questions to see if I could find an answer.
Short answer, there isn’t a lot of good research on the topic. Long answer, our best data comes from World War One factory workers (turns out you can do interesting research when your research subjects aren’t legally allowed to leave), but maybe we have enough information anyway to make educated guesses. At the least, we can run personal experiments.
Here’s a summary of my findings and opinion, followed by the actual research.
1. Limits on total hours. First, as you work more hours, each hour becomes less productive. If I had to guess based on the research, I’d say there are steeply diminishing marginal value around 40-50 hours per week, and negative returns (meaning less total output for the day per additional hour) somewhere between 50 and 70 hours. But, this is based on a grand total of two data sets. World War One studies of factory workers are the only source that experimentally tests the question. They didn’t even know to report sample sizes. (Also, a bunch of guys who wanted to prove workers should have Sunday off, found that having one day per week off was good. Give that as much credence as you think it’s worth.) The other data set shows a similar, correlational trend between CEO hours and company sales.
2. Limits on learning. Second, the spaced learning literature indicates that studying more than a few hours at a time may be useless. At least if you’re trying to memorize information. On one topic. If you’re producing output or switching up topics, this literature might be totally irrelevant. Still, it’s important to consider if additional hours of hard work might be wasted. I can only imagine what the control group guys in one study thought; “The guys studying four hours a day learned just a much as I did studying for seven? I worked my ass off an extra three hours a day for NOTHING?!?” (Also worth noting, a meta-analysis found that less rigorous studies had bigger effect sizes.) IF this literature applies to regular work, then it might be good to work on one priority for 1-4 hours a day, then switch to other topics.
3. Anecdotal reports. I’m fairly skeptical any of this research tells us how much to work (you can see more details below). I place more confidence on the anecdotal reports of productive people. It’s common for them to report three to five hours of deep work on a top priority each day, plus several hours more of lower energy or more “following curiosity”-type work (three more yet-to-be-released interviews also report in this range; one interview reports more). To be clear, I think they’re describing consistent, intense, “write a book chapter” levels of focus for those three to five hours.
When people talk about working longer days (e.g. 12 hours), they usually report that it takes enough of a toll on subsequent days that it’s not worth it. I know a number of people who consistently put in 8-12 hours of focused work a day, so it’s possible. But most of these people are chronically stressed about their work. I’m not sure if working a lot makes people stressed, or if stressed people work a lot to cope with the stress. Either way, I’m hesitant to recommend people try to work this much.
4. Long term sustainability. Additionally, limits to working hours might be necessary for long-term health and wellbeing, e.g. avoiding burnout. Setting aside vague “balance” like only working 40 hours a week (since you’re probably not a World War One factory worker) - what do you need to stay sustainably healthy, happy, and productive? If you’re unhappy, stressed, or tired, pay attention to this and check in about your base needs such as sleep, exercise, social, and leisure time.
If you haven’t before, try experimenting to see what makes you happy and productive at the same time. I didn’t know I needed eight hours of sleep every night until I tracked my sleep, and it took me a while to notice that exercise made me happier and more productive the following day. If you’re otherwise doing okay and want to make more time for productive work, you can try experimenting freeing up time in one area and see how that impacts you. If the change makes you feel worse, try something else instead. (It’s good for you to be happy. Please believe me here.)
5. Caveats. Additionally, I want to note that I still expect prioritization and deep work to be much more important than total hours spent working. E.g. if actions differ by 10x or more in importance regularly, then output is mostly determined by choosing the right things to work on, plus a smaller multiplier for time worked. In this case, having better processes for choosing the right work is way more important than hours spent working. Additionally, deep work in focused blocks is likely better for making progress on big priorities, so increasing deep work time is more important than total work time. Trying to work more hours may even be counter productive if you prioritize less (see The Five-Hour Workday and Owen’s interview).
6. Individual Experiments. Finally, the best process may vary dramatically from individual to individual. This is obvious in extreme cases, such as people with extreme fatigue who can only work a few hours a day (if this is you, focus on addressing your fatigue first). So pay attention to yourself. Test what works for you.
I wrote up the results of a simple experiment I did on myself here. TLDR; I did one hour of deep work each day for two days, then two days of four hours each, and finally two days of eight hours each. I spend the entire 26 hours writing, and tracked how many words I wrote each hour to calculate the diminishing returns to big blocks.
You can fairly easily do a similar experiment to play with your personal limits.
If you want to do more, here are some other experiments you might find valuable:
- Remember that feeling productive and actually accomplishing valuable work might be separate. If you don't already, you may need to track how much you work and your goals/output so you have a baseline for what you get done before you can meaningfully run experiments. Rough metrics are fine - reducing uncertainty is better than throwing your hands up. If you’re not sure what output to measure, here are some ideas: time spent on priorities, deep work time, writing pace and quality, novel ideas, difficulty of problem you can solve, amount of coding or bugs, emails answered. These are all rough, so try a couple and see which feel like they’re correlated with what you care about.
- Does having a full day off once a week help? Do you see a noticeable difference when you take a day off with zero work vs still checking email, etc.?
- Energy and focus also often change throughout the day - is there a particular schedule that works well for you? Are you more likely to get high quality, focused work done in the morning, afternoon, or evening? I estimate I’m about four times more productive in my peak work sessions than when I’m tired, which I know because I track my working time on Toggl.com and evaluated my writing output from a few blocks of time spent writing when I felt tired and when I felt energized.
- Are breaks, walks, or naps helpful? In particular, taking breaks to check you’re working efficiently is a good way to avoid wasting big blocks of time. E.g. pause every hour to check that you used the time to push forward your priority with the least wasted motion possible. If not, then make a better plan for the next hour to nail your goal.
- Do you need bigger breaks periodically? If you push yourself for a while, do you hit a wall and need to take a few days to recharge? If so, what does it feel like when you hit a wall and what ways are best for you to recharge?
And remember, you’re finding what works for you here.
7. Conclusion. So, to the effective altruists who want to push their ability to do more so they can have a bigger impact and to the effective altruists who want EAs to cut themselves some slack before they burn out -- you both have a valid point. We don’t know how much a motivated person at peak performance can work. It might be a lot more than four hours a day. It might not.
If you run the experiments and find ways that you can work more efficiently or more hours, great! As long as you’re still good. If you’re regularly stressed and unhappy about your work, that should be a big red flag. You burning out will not help your cause. (P.s. your happiness matters too.) In that case, it might be good to work less - at least temporarily - to force yourself to be more efficient and thoughtful about your work.
If you just wanted the high-level overview, feel free to stop here! The rest is the nitty gritty of the studies I looked at.
1. Diminishing returns on total time spent working
To start with, I recommend Elizabeth Van Nostrand’s epistemic spot check on The Role of Deliberate Practice in the Acquisition of Expert Performance, the basis for Newport’s claims about deep work time. His claim is among the more commonly cited reasons for believing people can only work so much. In turn, he cites Ericsson, Krampe, and Tesch-Römer’s 1993 paper, The Role of Deliberate Practice in the Acquisition of Expert Performance.
Elizabeth concludes, “Many of the studies were criminally small, and typically focused on singular, monotonous tasks like responding to patterns of light or memorizing digits. The precision of these studies is greatly exaggerated. There’s no reason to believe Ericsson, Krampe, and Tesch-Römer’s conclusion that the correct number of hours for deliberate practice is 3.5, much less the commonly repeated factoid that humans can do good work for 4 hours/day.”
I wanted to go beyond this pitiful citation trail. So, the following are the studies I dug up to see whether there was actually good evidence out there. You can jump back to the first page if you want to hear the summary of how, no, there’s really not much good evidence available.
1. Time-use diary study of CEOs
I found one dataset that seemed relevant to diminishing returns on output per hour worked doing thought work. It was a study of CEOs' time use, with data collected by phone calls with the CEOs’ personal assistants to reconstruct the CEOs’ days, which used company sales as the “output” measure.
According to the study, among 1,114 CEOs, a 10% increase in weekly hours worked was associated with a 3.3% increase in company productivity. The CEOs worked an average of 52 hours per week. It doesn’t say if or when additional hours worked become negatively correlated with productivity.
I briefly looked at the data set (obtained here). Take my analysis with a huge grain of salt – this is from eyeballing one graph. The scatter plot on hours worked and company sales (without controlling for anything) trended toward more hours worked correlating with more sales but there wasn’t a clear pattern, and a LOESS curve flattened shortly after 40 hours per week. This weakly inclined me to think the data set supports a positive relationship between more hours worked and more total output, with diminishing returns per additional hour after about 40 or 45 hours per week. I decided it wasn’t worth my time right now to buy SPSS and play around with the data set a bunch more, but I would be very interested in the results if someone else felt like doing so.
2. Studies of factory workers in World War One
The main dataset I saw cited for supporting diminishing marginal returns is a collection of studies done during World War One, mostly on factory munitions workers. The studies support that general conclusion to a limited degree, but take them with a big grain of salt: the sample sizes were generally not given but probably “criminally small”, the study quality was probably bad given it was done a hundred years ago, and it’s unclear that factory work generalizes to thought work. In so far as these studies generalize, they would support working <50 hours per week and taking one entire day off per week to rest.
I hunted down Fundamentals of Skill, Welford (1968) (cited in the deliberate practice paper) and found what seems to be the most relevant passage: “Presumptive evidence of some kind of fatigue effect during an industrial shift is contained in the classical reports of the Industrial Fatigue Research Board (later renamed Industrial Health Research Board). These reports showed not only that shorter working shifts led to higher hourly output (Osborne, 1919, Vernon, 1919), but also that a net reduction in working hours could sometimes lead to a net rise in total output (Vernon, 1920a, b).” (p.282)
I tracked down Vernon 1920a (since this was the only paper cited for the claim that a net reduction in working hours could sometimes lead to a net rise in total output). The paper supports both claims, but only at a factory-level (not individual-level) for the net reduction in productivity.
Reducing shifts from 8 to 4 hours increased output per hour by 11.5%, and reducing shifts from 8 to 6 hours increased output per hour by 4.7-10.6%. The factory could use more men for more, shorter shifts, so the total working time was nearly constant (2 hours less total for the 6-hour shifts). So in cases where more workers are unavailable (such as most thought work), working the extra hours would result in more total output even if lower hourly output.
Next, I read a review of the rest of research on munitions workers (Pencavel, 2014). Pencavel found that total output increased with each additional hour until 63 hours per week, then decreased with additional hours worked. Average output per hour decreased above 49 hours per week. In addition, holding hours per week constant, having one day off is about 10% better than working seven shorter days. (I’m additionally skeptical of this finding because it felt like the researchers were trying to prove that having Sunday off was worthwhile.) Between the diminishing marginal output and the benefit of a day off, working 48 hours per week across six days resulted in slightly more output than working 70 hours per week across seven days. So, capping work time at 50 hours per week would have resulted in little lost output. Finally, they claim the rate at which additional hours of work becomes less productive varies across workers and across types of work.
2. Diminishing returns on time spent learning
The numbers on hours of deep work time might have originated in the literature on spaced learning. It’s well studied that learning via spaced repetition is more efficient and effective than massed studying (e.g. cramming for a test). Below, I summarize the spaced learning studies that the deliberate practice paper cited, directly or indirectly via citing a reference in another work, plus another study from a meta-analysis that had particularly long-term tests for retention.
These studies do seem to support diminishing returns on time spent learning in one day. According to these studies, studying one hour per day, spaced out over more days, might shave a quarter or more off the total time required to learn a given thing, compared to studying for 2 or more hours per day. So more calendar time but fewer hours of studying. This meta-analysis suggested the benefit of spaced learning compared to massed as 15% on average, which might be a more realistic expectation.
Still, there are a couple reasons this literature might not generalize to time spent working.
First, spaced learning might only be relevant when memorizing new skills or material, e.g. if the benefit is due to how memories are formed. I also didn’t see anything on how the benefit of spaced learning would change as one becomes an expert – contrary to Newport’s claims that the amount of deep work done per day increases from one to four hours as one becomes an expert. (I suspect his claim is more likely to relate to one’s ability to stay motivated and focused instead.)
Second, even when relevant, it seems like the spaced learning benefit is for one topic, not all work done in a day. For example, the participants in the army study below did other things after their four hours of studying. In that case, a good rule of thumb would be to only work/learn on one subject for a limited number of hours, but you could study something else later in the day.
1. Telegraphy army study
From Elizabeth’s review: “An interesting army study showing that students given telegraphy training for 4 hours/day (and spending [the rest] on other topics) learned as much as students studying 7 hours/day. This one seems genuinely relevant, although not enough to tell us where peak performance lies, just that four hours are better than seven. Additionally, the students weren’t loafing around for the excess three hours: they were learning other things. So this is about how long you can study a particular subject, not total learning capacity in a day.”
The students studying for 4 hours per day spent an additional three weeks doing so, and ended up “markedly superior”. Breaking the four-hour period into four one-hour periods didn’t give further improvement.
2. Postmen learning to type
Study on learning to type: "Four groups of postmen were trained to type alpha-numeric code material using a conventional typewriter keyboard. Training was based on sessions lasting for one or two hours occurring once or twice per day.”
It took the people studying 2 x 2 (two hours per session, two sessions per day) 49.7 hours on average to learn the keyboard range, while the people doing 1 x 1 only took 34.9 hours. It took fewer days to study in mass (12 instead of 35), but the per hour learning was less efficient.
3. Assembly line training task
A second passage from Fundamentals of Skill referred to training session length, though again apparently in a factory setting: “A further, as yet unanswered, question is raised by the problem of optimum length of training session. It is clear that there are severe limits to the rate at which material can be learnt when considered on a time scale of seconds and minutes, but are there any additional limitations operating over periods of hours, days or even longer times? Common experience suggests that there may be, but the question does not seem to have been posed in a scientific context. An indication that it might be worth asking is contained in the finding by Henshaw and Holman (1933) in an industrial study, that 8o min training per day at a chain assembly task yielded as rapid improvement as 160 min.” I was unable to find the original study.
4. Learning a foreign language
Another small study that I saw found: “Four adults (aged 25–57 yrs at the beginning of the study) learned and relearned 300 English–foreign language word pairs. Either 13 or 26 relearning sessions were administered at intervals of 14, 28, or 56 days. Retention was tested for 1, 2, 3, or 5 yrs after training terminated. The longer intersession intervals slowed down acquisition slightly, but this disadvantage during training was offset by higher retention. 13 retraining sessions spaced at 56 days yielded retention comparable to 26 sessions spaced at 14 days.”
And that’s all folks. I hope you had fun reading the blog equivalent of a null finding.