What percent of people work in moral mazes?

[Raemon]

Epistemic status: A quick Fact Post on moral mazes. This is me trying to get my hands on some data and think through stuff, not meant as a definitive reference.

I wanted a sense of the proportion of people who work in moral mazes.

The middle manager hell hypothesis says “the more layers of middle management you get, the more your company will be Goodharty, deceptive, and optimized for upper management political games” (among other things).

Whether you buy the middle-manager-hell-hypothesis, I wanted to figure out how many people work in a highly hierarchal org.

This blogpost was a originally going to be a short aside in Recursive Middle Manager Hell, where I claimed “People in modern society are more likely to work in moral mazes. Large companies tend to become moral mazes, company size is probably heavy-tail distributed, therefore probably most employee-hours are spent working in companies with lots of employees.”

But, is that true? Size of companies is gonna be heavy-tailed, but, number of small companies is also probably heavy-tailed, and I wasn’t sure how the numbers checked out. This seemed like a good opportunity for me to practice being more numerically literate and getting in contact with some facts-on-the-ground about company size.

So:

I found this webpage claiming to have data for number of US companies for each size in 2022. Unfortunately, it only buckets up to “1000+” employees. I have a feeling that heavy tails are pretty important here.

Employees Per Org	Number of orgs
1 − 4 employees	12,737,231
5 − 9 employees	1,913,721
10 − 19 employees	817,604
20 − 49 employees	414,381
50 − 99 employees	154,255
100 − 249 employees	89,365
250 − 499 employees	33,467
500 − 999 employees	19,058
1,000+ employees	24,036

(It also includes a “uncoded companies” of which there are 1,771,725. But, as a first approximation, the coded companies are hopefully representative as a proportion of the US population)

I have a feeling heavy tailed orgs are pretty important here. As a quick workaround I found this page listing the very top companies. I asked ChatGPT to reformat the list so I could paste it into a spreadsheet. It… seemed to only go down to “Walgreens” before it seemed just start making stuff up. (I’m guessing it ran out of working memory to think?). But there were only a few more US companies anyway.

[Fake edit: oh goddammit the fr is maybe world data instead of US data, which would be fine except I assumed I was looking at US data and so filtered the next table for US companies only. I was gonna timebox this post to “today” so I’m going to ship it with the current numbers, and then later either me or Some Other Enterprising Amateur Scholar can try to get a more apples-to-apples set of numbers and see if the trends are roughly the same. Walmart and Amazon are still the largest companies even when not filtering for US, and there’s only 5 more companies between UPS and Amazon in size]

Company	Employees
Walmart	2,300,000
Amazon	1,544,000
United Parcel Service	500,000
Kroger	500,000
Home Depot	500,000
Target	450,000
Starbucks	402,000
Berkshire Hathaway	372,000
UnitedHealth	350,000
FedEx	345,000
Cognizant Technology Solutions	341,300
TJX Companies	340,000
Pepsico	309,000
Costco	304,000
Lowe’s Companies	300,000
Concentrix	290,000
JPMorgan Chase	288,474
IBM	282,100
Jabil	250,000
Aramark	248,300
Wells Fargo	239,209
Citigroup	238,000
Microsoft	221,000
CVS Health	216,000
Bank of America	213,000
HCA Healthcare	204,000
Walgreens Boots Alliance	200,000

I’m not quite sure what a serious scholar would do with this, but for immediate future I through this into one table. I collapsed everything between UPS and Walgreens Boots into a “200-500k” bucket, and treated Walmart and Amazon as two additional outliers counted separately.

To wrangle the buckets into a point-estimate of employees, I took the geometric mean of the upper and lower bounds of each bucket. I’m not sure that was the right assumption, and especially for the 1000+ bucket I could imagine all kinds of things turning out to be a better approximation if I found more data. But, it seemed like an okay first pass guess.

I assume each manager can manager 10 people, so the threshold for each additional level of hierarchy is roughly:

Levels of Hierarchy	Employees
1	1
2	11
3	111
4	1,111
5	11,111
6	111,111
7	1,111,111
8	11,111,111

So, that gets us this:

Employees Per Org	Number of orgs	Employees per org (geomean)	Levels of Hierarchy	Total employees	Percent population
1 − 4 employees	12,737,231	2	1	25,474,462	5.5%
5 − 9 employees	1,913,721	7	1	12,837,631	2.8%
10 − 19 employees	817,604	14	2	11,269,893	2.4%
20 − 49 employees	414,381	31	3	12,972,177	2.8%
50 − 99 employees	154,255	70	3	10,852,801	2.3%
100 − 249 employees	89,365	158	3	14,101,559	3.0%
250 − 499 employees	33,467	353	3	11,820,533	2.5%
500 − 999 employees	19,058	707	3	13,469,301	2.9%
1,000-199k employees	24,036	14,142	5	339,919,522	73.2%
200k − 500k employees	25	316,228	6	7,905,694	1.7%
Amazon Specifically	1	1,544,000	8	1,544,000	0.3%
Walmart Specifically	1	2,300,000	8	2,300,000	0.5%

So… this says 73% of people are in the approximately “1000+” (but less than the top US companies) bucket. And this is sort of annoying because that was definitely the most poorly defined bucket. It contains a wide range of numbers of employees and I’m not at all confident my geometric mean estimate is reasonable.

The 1000+ cluster would range from 3.5ish layers of hierarchy, up to 6.5ish. The Immoral Mazes theory suggest than things Start Getting Mazey around 3 layers of hierarchy, and are Solidly Mazey by the top you have 5 layers (where there’s at least one middle-management layer that never is directly connected with object-level reality).

This rough pass seems to support the original “most people work in a moral maze” claim, but I think I may have basically made simplifying assumptions that “made that true.”

Shrug?

[johnswentworth]

Before reading the post: I’ve heard somewhere that roughly 25-30% of the US workforce is self-employed, so based on that I’d be surprised if much more than a third of working Americans work in moral mazes. On the flip side, any state or federal government job is presumably a maze job, and presumably almost-all of the Fortune 500 have >10k employees each, so I’d be surprised if much less than than 20% of working Americans work in moral mazes.

After reading the post: there are not 340M workers in the US. Even the whole population isn’t quite that large (though very close), and labor force participation rate is something like 60%. So the 1k-199k bucket estimate is off by at least a factor of 2. I applaud the effort, but don’t think I can update much from this data.

[Ericf]

1:10 management ratio is a very bad assumption as things get large. First level supervisors in Retail or Manufacturing often have 30-100 direct reports, regardless of how many levels of management exist above them (with 2-6 direct reports each).

Also, modern supply chains mean that medium to big companies often have entire additional companies “reporting into” the management structure.

[Unnamed]

Is Walmart that mazy? My impression is that stores that are better of a big chain are generally better run than single location mom & pop stores. Which doesn’t mean that a huge chain like Walmart is completely free of maze-style dynamics, but does imply that the big management structure is doing more to make the stores functional than to make the dysfunctional.

Being organized into a bunch of separate stores seems like it could help fight off maziness, since it means that there’s a local reality (at their particular store) which most employees are in some contact with. It also creates some buffer from whatever culture develops among the higher layers of management above the levels of individual stores. Also, I’d expect Walmart to have a lot of metrics that are pretty good & concrete, and don’t tend towards much of the Goodhart stuff you’d worry about.

[Raemon]

Yeah. I guess I didn’t state explicitly but meant it to be implied: this is just meant as a first-pass approximation to get even remotely oriented in the space. Obviously it’d require followup work of “okay but what specifically makes things mazey?” to sus out more concrete claims.

(this is after dealing with things like “it’s not even clear where these numbers are coming from, whether they’re double-counting, etc)

[shminux]

Hmm, I wonder how much of a correlation there is between the number of middle management layers and competition for the scarce internal resources. The latter is about not having much Slack in an organization, and middle managers are good at competitively overfishing this particular resource to extinction. On the other hand, growth rates, corporate culture and other factors are also important. The Musk shops seem unusually low on slack, and yet they don’t give me an impression of being unusually immoral mazes. For example, if you compare glassdoor reviews for Tesla and Rivian, two growing electric car manufacturers, Rivian comes across as much more like an immoral maze, with the upper management out of touch and the middle management playing political games.

[Mojo]

Here’s a source with number of employees for businesses in ten bins from <500 to 20000+, US 2018 : https://​​www2.census.gov/​​programs-surveys/​​susb/​​tables/​​2018/​​us_naicssector_large_emplsize_2018.xlsx

[Raemon]

Thank you!

[Dagon]

Shrug?

Yeah. This is an interesting exploration, and contains some surprises—I would have guessed that the 1k-199k group was largest, but didn’t realize just HOW much larger.

But I think the hard part is “how maze-y is the experience?” Levels of management don’t really tell the story. It’s more about levels of ineffectual management, which is going to vary a lot based on the business(es) and culture within the org.

[Stephen McAleese]

From my experience, the 1:10 management ratio is a good approximation of reality.