Sustained Overtime is Often a Mistake

INTRO

A few weeks ago I came across a discussion of whether an individual contributor doing long hours should be promoted before someone of equal productivity (output/hour) doing fewer hours. I declined to vote in the accompanying poll because I thought the author was trolling.

However, because they might have been serious, this article highlights 100 years of data indicating that sustained overtime, the behavior being rewarded, is often a mistake, both for the employee and by their manager.

TL;DR - The data show that the amount of work achieved during a fixed time frame is negatively affected by working overtime beyond an 8-hour, 5-day (5x8) work week. Go too far past either of those magic numbers and productivity can decline so far that the team produces less total output than in a standard 5x8 work week. You get negative returns.

My goal here is to clearly communicate the consequences of sustained overtime for your team. If you know someone who needs to be made aware, share this article or some of the referenced sources with them.

Especially if those folks are in a position to encourage, coerce, or mandate overtime or worse, a toxic overtime culture.

CORE FACTS

Most people recognize from personal experience that sustained overtime eventually results in decreased productivity. Well, there’s lots of evidence confirming your experience. Literally, data collected since 1914 by businesses and academics, and across many disciplines of work. You can [Google, Bing, DuckDuckGo, AltaVista] for yourself, but I include links and data below because sometimes you have to dig deep.

I’ll summarize the findings from my readings linked in the references, but include details and links below.

This article assumes the same definition for overtime predominantly used in the literature: exceeding an 8-hour day, a 40-hour 5 day week, or both.

• Productivity — the rate of creating a measurable result — decreases immediately upon beginning an overtime schedule. There ain’t no such thing as a free lunch.
• Productivity will eventually continue to decrease further with each week of sustained overtime.
• The rate of decrease will eventually negate the gains provided by the extra hours of work.
• The magnitude of productivity loss is affected by both long days and long weeks:
• Like a sleep deficit, burnout and fatigue incurred by overtime do not go away after a good night’s sleep. The team has to rest to get out of the reduced-productivity hole.
• The data collected over the past 100 years include both unskilled manual labor (e.g., widgets on an assembly line) and skilled manual labor (e.g. electrical work, brick laying).
• I did not find specific studies of today’s knowledge work — where mostly we use our minds and not our hands, KBM aside — but I expect it is even more impacted, because writing code, creating art, and designing gameplay aren’t driven by muscle memory.

PROPOSED CAUSES

Mental and physical fatigue increase along with the hours spent working. As work days grow longer, the periods of unproductive time increase and the work rate slows. People start having to do IRL things, like shopping and errands, during the work day.

Excessive overtime adversely affects both physical and mental health, correlating with issues such as back or joint pain, hypertension, heightened stress levels, increased alcohol consumption, weight gain, significant work-family conflicts, elevated divorce rates, and depression.

Extended work hours are associated with heightened job-related risks. As hours accrue, attention and concentration diminish, potentially leading to more errors and accidents that jeopardize the safety of employees, customers, or patients.

Increased hours lead to sleep deprivation, which on it’s own creates cognitive deficits and can increase rates of interpersonal conflict. Today’s knowledge work is highly sensitive to this issue.

These things all combine to generate presenteeism (being physically present but mentally absent), increased absenteeism, and higher turnover.

MAKING IT WORSE

The disconnect over the effects of overtime originates from too simple an observation: when people work more hours, they make more things. If you stop there then it seems obvious that overtime is a great way to make more money or reach a goal sooner.

But the data provide a different answer.

• Some individuals or teams can be affected less than others, but on average there is always an inevitable decrease in productivity.
• Team culture can magnify this problem. Lauding the overtime martyr with attention, raises, or bonuses, will encourage others to do the same even though they are effectively reducing productivity.
• Hourly workers being less productive actually increase the marginal cost of the resulting work.
• Managers can be rewarded — unbeknownst to the team — for driving the team to hit a deadline. If that gets out, or even is inferred, morale will go into the toilet.
• Overtime can be driven by the business being in trouble, instilling fear in management and possibly the team (if they are informed).
• It is difficult to convince investors that driving the team beyond a certain point is counterproductive and is likely to reduce their odds of success.
• Individuals can gain positive feelings when heading home after a protracted work day, a rush of serotonin. In medicine, doctors on long (30+ hour) shifts refer to this phenomenon as “post-call euphoria”.

SCIENCE THINGS

Methodology

I began by searching online for papers and then recursively searching for references. One paper referenced a review that was skeptical of past results — a thing often required by peer reviewers — so I also followed that trail.

I did not spend money to get past paywalls. It was often possible, however, to find copies of the papers from other sources.

Multiple papers reference the same data sets. These core data sets are referenced for your own inspection, not just the papers that draw conclusions.

Meta reviews point out drawbacks and deficiencies in the data collection process of various studies. Understanding these things is important, but we also have to acknowledge and accept that this type of data is hard to collect. Several of the data sets are collected over years or even a decade. Some are unearthed via archaeology (data from 1914). Some data is better than no data.

The gold standard used repeatedly is a Business Roundtable (BRT) paper, Scheduled Overtime Effect On Construction Projects - A Construction Industry Cost Effectiveness Task Force Report, which covers data collected in 1974 and then updated in 1980.

One source, the Revay Report, indicates that across all teams, productivity losses on average tend to follow the BRT report, but that individual teams can sometimes do better or worse than that average.

I’ll note that skeptics usually attack methodology, but those same skeptics also never provide any data to counter the conclusions. A good scientist knows that absence of evidence is not evidence of absence, but thus far the preponderance of data paint the same picture.

1914

The following is reproduced from the 2014 article, Proof you should get a life. I include the original data graph from the paper under discussion, The Productivity of Working Hours, by John Pencavel. Those data were collected in 1914 by investigators of the British “Health of Munition Workers Committee” (HMWC) during the first world war. Description of conclusions are quoted below, but the takeaway is this:

• These workers reached a point where total productivity was capped. Greater hours failed to produce more product after about 50 hours/week.
• The data do not report the number of weeks of sustained overtime, so are some sort of average. Longitudinal data are presented below.

Mr Pencavel crunches the data and concludes that there was a “non-linear” relationship between working hours and output. Below 49 weekly hours, variations in output are proportional to variations in hours. But when people worked more than about 50 hours, output rose at a decreasing rate. In other words, output per hour started to fall (in the jargon, “the marginal product of hours is a constant until the knot at [about 50] hours after which it declines”).

1924

Here are conclusions from page 348 of the 1924 book, The Economics of Fatigue and Unrest under “The Findings of Statistical Investigation Summed Up”.

These sound familiar, right? This book was published 100 years ago.

A reduction of hours increases hourly output and decreases absence and accidents per hour.

Reduction of hours to eight per day increases daily output in occupations where speed depends mainly on the human factor or in a factory of mixed operations, such as the Zeiss Optical Works; but may fail to do so where the machine sets the pace or the completion of the operation depends on chemical processes, e.g. in the class of occupations numbered 5 e and f in Table 2.

Reductions of hours below eight per day does not increase hourly output sufficiently to increase the daily total, unless, possibly, speed depends purely on the human factor and work is of a heavy type.

The effects of a reduction of hours just described may follow only after a period of adjustment.

Increase of hours has the reverse effect. Hourly output falls and daily output also, at any rate if the working day increased was already of eight hours or more.

The fall is immediate, i.e. there is usually no period of adaptation.

Rates of absence, especially of unavoidable absence due to sickliness, tend to rise when the scheduled hours are increased, so that the number of hours actually worked may not as a net result be materially advanced.

Unavoidable absence, presumed to be due to sickness and accident, is generally lower in munition factories working shorter hours.

1974-1980

Here is a graph reproduced repeatedly from the Business Roundtable Data. It shows productivity measures by Proctor and Gamble of projects for two different overtime schedules.

The Business Roundtable (BRT) collected data over a ten-year period from a Proctor and Gamble processing plant in Green Bay, Wisconsin. The productivity of the construction was recorded as the ratio of the standard man-hours per unit to man-hours per unit achieved (which reduces to earned/actual hours) (Hanna, 2003). Thomas (1992) reported that the Proctor and Gamble data was from a single project, consisting of shorter jobs with overtime, and that the true type of construction performed is unknown.

The metric used is such that it can be converted directly from hours spent into “equivalent production hours” of a 40-hour week. Here is a graph with that conversion.

I added a reference to the standard 40-hour work week. When a curve falls below that reference, workers were producing less so the overtime schedule was making the project even later.

Doing the math, we can look at the number of 8-hour schedule days gained through overtime:

• The 50-hour schedule gained about 2.5 days of time after four weeks. The next four weeks gained only an additional 1/3 of a day of production and then ran into negative returns.
• The 60-hour schedule gained a bit less than 6 days of time after four weeks. The next four weeks gained only an additional 2 days of production and then ran into negative returns.

Applying this guideline to today: you can only gain between 3 and 8 days of schedule reduction over two months, assuming that the work is all being applied to the critical path. Most of that is in the first four weeks. Note that this only works if you can also allow the team to rest after 7-9 weeks of sustained overtime.

1989

Below we reproduce figure 11 from a comprehensive review document, Construction Productivity Impacts due to Extended Work Weeks, Shift Work, and Fatigue (2017).

Here is the description of this graph from the source.

In 1989, NECA published a second edition of “Overtime and Productivity in Electrical Construction”.

The study provides information on low, average and high productivity loss for 5-, 6- and 7-day work weeks and 9, 10 and 12 hours per day for sixteen successive work weeks, based on data gathered by NECA since 1969 for journeymen electricians. The origin of the data and the work environment are unknown. Figure 11 summarizes the data for average productivity for successive weeks of overtime.

These curves follow a similar pattern to the BRT data but without a temporary plateau at the beginning.

I digitized the above graph to plot the corresponding effective hours in the same manner as the BRT data. Here is that plot:

The “point of no return” — when productivity falls below that of the the work done in a typical 40-hour week — falls in line with the BRT data in the range of 5 to 8 weeks. Beyond that week the teams were in negative return territory, making the project even later than if they were on a normal schedule.

Doing the math once again, we can look at the number of 8-hour schedule days gained:

• These data show no “grace period” in the first few weeks, instead having a steady decline without an initial plateau.
• All schedules gain the bulk of their time in the first four weeks.
• The first four weeks provides the best “return on overtime” (ROO), defined as the ratio of hours gained to overtime hours spent varies. The worst ROO is 7-8s at 50%, and the best is 6-8’s at 88%.
• The 7-12s schedule accumulates the most gained hours but only at 82% ROO.

Graphs showing this type of relationship occur throughout the literature. I have yet to find original data for any of these, only the published graphs. It is likely I’d have to locate a physical journal or other document .

The review sharing these data generally provides reasonable criticisms, but do not question the results beyond this statement:

The origin of the data and the work environment are unknown.

SKEPTICS

The skeptic will point out that game developers are not laying brick or doing electrical work. And a good thing, because I don’t want my house built by workers doing crunch. I’d like the walls to remain standing and to avoid electrical fires in my kitchen.

The skeptic will also assert that these data don’t apply, because the work is so different, or that little is known about critical details of older data.

These are valid questions, and they recur in the literature on construction tasks. From a Korean journal on Civil Engineering:

However, it was found that most studies reviewed in this paper have pitfalls. Most of them are out of date, based on a small sample size, and largely developed from questionable or unknown sources. Where the data source is known, little is known about other pertinent information, such as the characteristics of the project, work environments, quality of management and supervision, type of work, and trades involved.

The good scientist agrees with, “sure, this is something to evaluate.”

On the other hand, we keep seeing the same overtime productivity patterns in multiple studies. From what I have been able to find, data were generally collected through projects of opportunity. What economists would refer to as a natural experiment, and we never get to structure conditions of a natural experiment.

Being Skeptical of the Skeptics

Some scientific endeavors can be driven by a desire to support a specific idea. In this case there are some who believe that overtime can be bad when it reduces productivity too far. There are others (the skeptics) who want to believe the opposite, or are at least not convinced that cost-free overtime is unachievable.

On the other hand, these skeptics don’t offer counter evidence. They instead seem to reach for arguments to discredit results. For example, implying that we just need a better work environment, better management or supervision, better logistics, or something else. Or casting doubt on the reliability of the data collection, or how long ago the study was conducted.

There are likely factors that make overtime more tolerable, less of a burden on the team. But based on my experiences, I find it hard to believe that there are effective ways to avoid the observed patterns of overtime productivity loss.

Also, how can human fatigue responses be “out of date”? I don’t think our physiology or neuroanatomy has evolved a lot in the past 100 years.

KNOWLEDGE WORK AND YOUR TEAM

All overtime situations are affected by issues outside of individual human experiences and situations. Undoubtedly there are some people who can sustain longer work hours than the average person. Perhaps this is similar to the very small number of people (1-5%) who can function on 6 hours of sleep or less.

But we are talking about your entire team, and not a singular individual who is standard deviations away from the mean. And unless you measure that person, you can’t know if they are an overtime martyr, an overtime theater actor, or an actual outlier.

In the end, however, the consequence of overtime we face is not for an individual but for your team as a whole. And the issue is accumulated fatigue.

Worse, common sense and a lot of research says that knowledge work productivity should be even more sensitive to fatigue than unskilled or skilled manual labor. Our brains burn calories (about 20% of the daily calories) and even more of them when we are thinking hard. Regardless of our ability to reach a state of flow, we need to take breaks to maintain productivity.

There is also this from a Forbes article:

In a recent Blind poll, nearly 45% of tech workers said they spend four hours or fewer on “focused work”—uninterrupted time spent in a flow state, concentrating on high-priority tasks.

IMO, sitting more hours in the chair is not gonna stretch that a lot, if at all, assuming “tech worker” maps to “knowledge worker”.

WHERE’S MY DATA, MAN?

Why don’t we have better metrics on knowledge work productivity? I can only speculate.

Certain kinds of work are easier to measure. If someone builds a widget, we count widgets. If someone wires houses, we measure the number of receptacles, switches, and light fixtures together with the rate of passed inspections. If someone lays bricks, we count bricks and the rate of passed inspections.

But knowledge work is hard to measure — a controversial topic in today’s RTO world — because the work product can be hard to quantify. Why? At least one reason is that we are always trying to create new things. Unlike, say, installing that 512th electrical receptacle, or the 10,000th brick.

The metrics people do try can backfire, like code commits, lines-of-code, or god forbid, mouse movements or the time of day an email is sent.

I believe we can measure productivity in a knowledge work world, but it would require focused attention by subject matter experts who can be objective and fair. Like grading an exam. But then we also know that grading is rife with biases.

OUTRO

We need to be both clear and fair. The takeaway from these data is not that overtime is never useful. There will always be situations where the team needs to sprint to make that deadline.

But scheduled, sustained overtime should be used strategically, not as a replacement or fix for all the potential causes of schedule overrun. Especially bad management.

Going further, if you have an employee chronically working long hours then they are almost assuredly working well below their peak productivity. They are also potentially headed to burnout. If you are their manager, you should be asking why, and looking to ensure their health.

If you see other manager’s team members working chronically long hours, again, find out why. I’ve known people who can’t say no to any request, and who reported to a manager inclined to take advantage of that agreeableness to make their own selves look good, employee health be damned.

Those agreeable people are usually high performers who eventually get so burned out they quit. And that is a loss for your team and your project.

RESOURCES

?? IZA Institute of Labor Economics (PDF): The Productivity of Working Hours (2014; evaluating data collected in 1914)

?? The Economics of Fatigue and Unrest. By P. Sargant Florence. New York: Henry Holt and Company (1924)

?? National Electrical Contractors Association (NECA). “Overtime and productivity in electrical construction”, NECA, Washington D.C. (1969)

?? The Business Roundtable (PDF): Scheduled Overtime Effect On Construction Projects - A Construction Industry Cost Effectiveness Task Force Report (1980; including data collected in the 1960’s)

?? Construction Industry Institute (CII), Source document 43: The Effects of Scheduled Overtime and Shift Schedule on Construction Craft Productivity. (1988)

?? ASCE Library: Effects of Scheduled Overtime on Labor Productivity (1992)

?? ASCE Library: Scheduled Overtime and Labor Productivity: Quantitative Analysis (1997)

?? Concrete Construction: Effect of Overtime on Worker Productivity (1997)

?? The Revay Report: Calculating Loss of Productivity Due to Overtime Using Published Charts – Fact or Fiction (2001)

?? Harvard Business Review: Sleep Deficit: The Performance Killer (2006)

?? Greater Toronto Electrical Contractors Association (PDF): Impact of Overtime on Electrical Labor Productivity: A Measured Mile Approach (2011)

?? The Economist: Proof that you should get a life (2015)

?? CNBC: Memo to work martyrs: Long hours make you less productive (2015)

?? Harvard Business Review: The Research Is Clear: Long Hours Backfire for People and for Companies (2015)

?? Harvard Business Review: Why Some Men Pretend to Work 80-Hour Weeks (2015)

?? Entrepreneur: Studies Show Working Overtime Is Basically Pointless (2017)

?? Springer Link, KSCE Journal of Civil Engineering: Critical review of previous studies on labor productivity loss due to overtime (2017)

?? High Bridge Associates: Compilation of Analyses of Public Domain White Papers, Construction Productivity Impacts due to Extended Work Weeks, Shift Work, and Fatigue (2017)

?? Great Lakes Skilled Trades: The Impact of Overtime on Construction Workers (2023)

?? NIH - National Library of Medicine: Sleep Deprivation Impairs the Accurate Recognition of Human Emotions (2010)

?? Social Psychology and Personality Science: The Role of Sleep in Interpersonal Conflict: Do Sleepless Nights Mean Worse Fights? (2013)

?? Scientific America: Why Do Some People Need Less Sleep? It’s in Their DNA (2019)

?? NIH - National Heart, Lung, and Blood Institute: How Much Sleep Is Enough

?? Forbes: Can you be productive working only 45% of the day? (2023)

?? Wikipedia article defining Natural experiment

?? InnerDrive: Teacher assessments: 11 cognitive biases to keep in mind

?? Australian Journal of Education: Bias in grading: A meta-analysis of experimental research findings