Survey Results Pt. 1: Rationalists, furries, white guys, and more

A couple months ago, I started writing a survey of Manifold users, inspired by a similar one run by Pat Scott. After being open for a full month, from Sept. 7 to Oct. 7, the survey has finally concluded, with a total of 168 respondents. I already wrote another blog post about what’s on the survey when it was released. For this post, I’ll be going over the results of the first two sections (the first section only had a single question).

Section 1

Section 1 is mainly just the introduction section. It begins with the text, “Welcome to Plasma's Manifold Survey! This survey will include a variety of questions, from demographic information to your favorite movies to opinions on controversial Manifold subjects. There will also be some questions that are purely for fun. All questions on the survey are optional, and all of your answers are completely anonymous. It should save your answers, and you can edit your response later, so if you want to finish it in multiple sittings, you can.”

It then contains a single question:

Are Manifolders Human?

The first question on the survey is, “Before continuing with this survey, we need to check whether you are human. Are you human?” The only options were Yes and No. As expected, most respondents (162) said Yes. Maybe the 5 No’s were people who programmed bots to take the survey for them.

Section 2

Section 2 is called, “General Info” and asks respondents questions about themselves that don’t have anything to do directly with Manifold, and also don’t fit into the other categories (philosophy, favorite media, and fun questions). This might sound like the “boring” section, but in fact, the most popular market about the survey by a nearly threefold margin (and my most popular market overall) came from this section, as did many of the other popular markets.

How many people took Pat’s Manifold Survey?

As mentioned, Pat’s Manifold Survey was the precursor to mine, so I wanted to see what proportion of my respondents had also taken his. I knew it couldn’t be 100%, since I ended up getting more responses, but I didn’t know how many of Pat’s respondents had taken my survey until I looked at the results.

In total, that’s 50 Yes’s and 117 No’s.

Manifold’s Gender Ratio

Pat’s Manifold Survey drew a lot of attention by asking questions like, “Are there more than five women on Manifold?” Ultimately, he did get about a third of female respondents. I asked the exact same question on my survey (“What’s your gender?”), and here are the responses:

I was pretty surprised at how big the difference is. I don’t even have enough non-males to rule out “Is Manifold just a bunch of white guys?” resolving YES from this question alone, like I thought I would. But I guess the theory that Pat’s survey had a higher percentage of female responses because women were specifically drawn to taking it by the “Are there at least N women on Manifold?” markets could be correct.

To quantify how my results compare to Pat’s, I calculated the true diversity of order 2 for both. This is a measure of the effective number of categories that takes into account how evenly distributed members are between the categories (in this case, male, female, non-binary, and other). The idea behind the true diversity is that, if we had N categories that were all equally abundant, then the average proportional abundance of each category (by any notion of “average”) would be 1/N. So to find the “effective number of categories” for some unevenly distributed population, we take 1 divided by the mean proportional abundance, where the mean can be any type of power mean and is weighted by the proportional abundances of each category (this weighting means that it can also be thought of as a mean over all individuals of the proportional abundance of the category those individuals belong to). The exact formula is that the true diversity of order q is equal to the inverse of (q-1)th weighted power mean of the abundances, i.e.

\(^qD = \frac1{M_{q-1}} = \begin{cases} \frac1{\sqrt[q-1]{\sum_{i=1}^Np_i^q}} = \left(\sum_{i=1}^N p_i^q\right)^{1/(1-q)} & \text{if }q\ne1 \\ \frac1{\prod_{i=1}^N p_i^{p_i}} = \exp\left( - \sum_{i=1}^N p_i\ln(p_i) \right) & \text{if }q=1 \end{cases}\)

where pᵢ is the proportional abundance of each type. Note that the sum in the (q-1) power mean uses pᵢ^q, rather than pᵢ^(q-1), because it is also weighted by pᵢ.

There are four commonly used special cases of the true diversity that can each be understood in a natural way. With q=0, it is just equal to the number of categories, and the distribution of individuals has no effect. For q=1, the true diversity is the exponential of the Shannon entropy, a measure of how much uncertainty there is if you have to guess what category a random individual belongs to. For q=2, the true diversity is the reciprocal of the probability that two randomly chosen individuals (with replacement) come from the same category. This probability is just the sum of squares of the proportional abundances. There’s also q=∞, which is just the reciprocal of the proportional abundance of the most common category.

For my market comparing the results of the two surveys, I decided to use the q=2 true diversity, since it seems like it had the most natural interpretation. After all, a natural way to define the diversity of a group would be how likely you are to encounter someone in that group who is different from you in some specified way, and that number, the Gini-Simpson index, is just a monotonic function of ²D (specificially, 1-1/²D).

On Pat’s Manifold survey, there were 90 males, 31 females, 11 non-binary respondents, and 2 respondents who picked “Other”, so ²D is about 1.95 (¹D is about 2.40). On my survey, there were 138 males, 20 females, 7 non-binary respondents, and 1 respondent who picked “Other”, so my ²D is only 1.41 (¹D is about 1.77). Here is how the diversity index compares for different values of q:

Racial and Ethnic Demographics

The next question asked, “How would you describe your race or ethnicity?” with a note that, “Some of these categories overlap, and multiple may apply to you regardless. Choose all options that you believe apply to you.”

The responses were as follows:

Okay, so this one does guarantee that “Is Manifold just a bunch of white guys?” will resolve NO, based on the results from this question alone.

I can’t calculate a diversity index for it because people were allowed to select multiple options, so the concept doesn’t really apply.

Is Manifold just a bunch of white guys?

My most popular market asked about the intersection between the previous two questions, specifically, what proportion of respondents would say that they are white males. Although there was wide agreement that the proportion would likely be less than 80%, there was not so much agreement on what the exact proportion would be. Now I can finally resolve the uncertainty. From examining the responses, I found that 161 people answered both questions, and 107 said that they were white males, so the proportion is 66.46%. Out of those 107, 100 only selected “White”, while 7 others selected both “White” and a different option.

Favorite colors

I asked respondents what their favorite colors are, and these are all the answers, in almost alphabetical order (Python’s sort function still puts uppercase letters before lowercase ones):

BLACK
Black
Black
Black
Black
Black
Black
Black 
Blue
Blue
Blue
Blue
Blue
Blue
Blue
Blue
Blue
Blue
Blue
Blue
Blue
Blue
Blue
Blue
Blue
Blue
Blue (especially hyperbolic blue)
Blue, no yellow, ahhhhhhhh! :))
Blue, no yellow, ahhhhhhhhh!
Brown
Cerulean
Context-dependent, because I'm not a kid
Cyan
Cyan
Dark cyan
Do not have one. 
Forest Green
Forest green
Green
Green
Green
Green
Green
Green
Green
Green
Green
Green
Green
Green
Green
Green
Green
Green
Green
Green
Green
Green
Greenish orange
Grey
Grey
Grey
Grey
Hooloovoo
Lavender
Light Light Green (Something like #DAF7A6)
Mint
Navy blue
No favorite
None
None
Orange
Orange
Orange
Orange
Orange
Orange
Orange
Orange
Orange
Orange
Orange 
Orange 
Pink
Pink
Purple
Purple
Purple
Purple
Purple
Purple
Purple
Purple
Purple
Purple
Purple
Purple
Purple
Purple 
Purple 
Rationalussy
Red
Red
Red
Red
Red
Red
Sea green 
Sparkles
Teal
Violet
Whatever allows me to manipulate a market
Yellow
Yellow
Yellow
Yellow
all colours equally 
black
blue
blue
blue
blue
blue
blue
blue
blue
blue
blue 
blue-green
bluish purple
cyan
cyan
dark orange
dark teal
gray
green
green
green
green
gren
grey
indigo
metallic grayish tealish purplish
orange
pink
pink
pink
purple
purple
purple
red
red
scarlet
sea green
titanium
transparent
violet
violet
water blue

159 people answered the question, with the most popular color being blue, and green being a close second. If you’re wondering about the Monty Python references, that’s because I made a market asking how many people would answer, “Blue, no yellow, ahhhhh!”

What continent do you live on?

This is another one where a market I made on the topic could have influenced the number of people who chose, “Antarctica”. But it looks like only two people decided to choose the joke option. And hey, maybe they’re just physicists working at IceCube or something.

The exact number of responses for each is 95 for North America, 45 for Europe, 10 for Asia, 7 for Australia, 3 for Africa, 3 for South America, 2 for Antarctica, and 2 who don’t live on a continent.

Tech bros and crypto bros

It’s no secret that Manifold has a lot of users who are into tech. But just how many of these users would describe themselves as tech-bros? Given that the term is often pejorative, and that it would probably only be used by males to describe themselves, it’s not too surprising that the majority (101) did not identify as tech-bros.

I also asked whether users were crypto-bros, since I’ve seen a lot of markets related to cryptocurrency prices. However, very few Manifolders (only 10 here) actually describe themselves as crypto bros.

What countries are Manifolders from?

Here are the responses to the “What country do you live in?” question:

America
Argentina
Auroville
Australia
Australia
Australia
Australia 
Australia 
Belgium
Benelux
Brazil
Canada
Canada
Canada
Canada
Canada
Canada
Canada 
Chile
China
China
Czech Republic 
Denmark
Finland
France
France
France
France
Germany
Germany
Germany
Germany
Germany
Germany
Germoney
India
India
Italy
Kazakhstan
Kenya
Mozambique
N/A (Marie Byrd Land)
Netherlands
New Zealand
New Zealand
Poland
Poland
Russia
Russia
Russia
Scotland
Slovenia
South Africa
South Korea
Spain
Spain
Spain
Sweden
THE US OF A
The U.S., unfortunately
The United States
The United States
The United States of America
U.S.
UK
UK
UK
UK
US
US
US
US
US
US
US
US
US
US
US
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
USA
Uk
Uk
United Kingdom
United Kingdom
United Kingdom
United Kingdom/England. 
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States
United States 
United States of America
United States of America
Usa
Usa
aus
france
israel
nope
not US
united stated
united states of america
us
usa
usa
why do you need to know this for a survey on thanos's crimes

There were 156 responses in total, and 83 of them said they were from the U.S. (possibly a few more, depending on whether “aus” was a typo for “usa” or an abbreviation for Australia, and whether the person from “not US” is actually from the U.S.). Unsurprisingly, other English-Speaking countries also had a lot of people, with 5 (or 6, depending on the “aus” response) Australians, 7 Canadians, 7 people from the UK (including the one from Scotland), and 2 from New Zealand. There were also a lot of people from France (5) and Germany (7). There was also one person who claimed to live in Marie Byrd Land, presumably to commit to the bit of claiming to be from Antarctica. The other alleged Antarctican was the one who answered, “not US”.

Manifold’s political leanings

I already had a sense before making this survey that Manifold was liberal-leaning. This is to be expected from a group of mostly educated people, a lot of whom are rationalists or work in Silicon Valley, and all of whom are probably nerds of some sort. But I wanted to find out exactly how much. I only included U.S. politics here, since I didn’t want to compare apples to oranges (someone who is considered conservative because they vote for the Tories in the UK is not the same as someone who’s conservative because they vote for Republicans in the U.S.), and because I assumed that the majority, or at least a large plurality, of Manifolders are from the U.S. I also specifically asked about what party people usually vote for rather than what party they identify with or their political ideology. I think this is a better measure of people’s political leanings, since ideological labels mean different things to different people, and because a lot of people identify as moderates and independents but still in practice vote for the same party each election.

The ratio was even more overwhelming than I thought. There were 98 respondents, which is more than the number of people who said they were from the U.S. on the previous section, but it could include people who either didn’t answer that question or gave a joke answer. There were probably also some non-U.S. respondents who either answered which party they would vote for, or misinterpreted the question and answered, “I don’t vote”. There ended up being 58 Democrats, 2 Republicans, 10 Independents or third-party voters, and 28 non-voters. So the Democrat:Republican ratio was 29. I’m not sure what implications this has for Manifold. Is Manifold likely to overestimate the probability of Democrats winning an election due to wishcasting? Or maybe Manifolders are in general cyncial about politics and overestimate the probability of their preferred candidate losing?

Sexuality

I asked, “Which of these would you use to describe your sexuality?”, with checkboxes, like the race and ethnicity question. These were the results:

Most were heterosexual, but “Exclusively heterosexual” was just shy of an outright majority. The first seven options correspond roughly with the Kinsey scale, so we end up with an average rating of 1.33.

Monogamy vs. Polyamory

I asked a similar question about people’s feelings toward different numbers of relationship partners (monogamy, polyamory, and aromanticism), based on my impression that Manifold would be more accepting of polyamory than most of society. Here were the results:

The option that’s cut off there is, “Equally fine with polyamorous or monogamous relationships”.

Manifold and the rationalist community

Manifold is very closely adjacent with the “rationalist” movement, which has long favored prediction markets as a way to aggregate information. On the survey, I described it as follows: “This refers to the subculture focused on thinking and acting rationally through Bayesian reasoning, decision theory, game theory, being aware of cognitive biases and common fallacies, discussing epistemology, etc. It's commonly associated with ideas like futurism, effective altruism, and AI risk, and the websites LessWrong and Astral Codex.” The responses showed that most people considered themselves associated with the movement or adjacent to it to some extent.

The exact numbers of responses were:

“Yes, fully”: 28

“Pretty much, but I'm not fully associated with it.”: 36

“Maybe, I'm interested in their ideas and somewhat adjacent to it.”: 45

“No, but I do have a lot of overlapping ideas.”: 36

“No, not at all.”: 12

“No, and I find the rationalist community to be either annoying or bad in some way.”: 7

“No, and I strongly oppose the rationalist movement.”: 1

I used these numbers to determine how aligned Manifold is with the rationalist movement, with the answers above corresponding to “alignment scores” of 1, .75, .5, .25, 0, -.5, and -1, respectively. The average alignment score was .497.

Ages of Manifolders

I asked what year respondents were born in. Google forms gives me this graph to summarize the data, but it’s not as helpful as it looks. It’s not actually a histogram with a numerical scale - it’s just a chart of all responses and the number of people who gave each answer (including ones that weren’t years).

I used Python to create a better chart that actually used a proper scale. I also had to remove two non-numeric answers from the data set and turn “89” into “1989” for one of the responses.

The two non-numeric responses were “A Space Odyssey” and “Spain”. Assuming that “A Space Odyssey” means that the respondent was born in 2001, I corrected the histogram by adding that data point back in.

After doing this, I also found the average year of birth of all respondents: 1992.9. So the average age of Manifolder is something like 31.

True or false: More people have been to Russia than I have.

This was a completely nonsensical question that I put on the survey to see how many people would notice. I made a market on whether people would see through the illusion, with a Sha256 hash that describes how I would determine if people had seen through it.

The text used for the hash was

The linguistic illusion used in my survey is the Escher sentence, "More people have been to Russia than I have." It's an illusion because it sounds grammatically correct at first to most people who hear it, but upon trying to parse what it means, you realize that it's actually not a well-formed sentence (unless "have" is interpreted to mean "own" rather than as part of a past participle word form, in which case the sentence means, "There are more people who have been to Russia than there are people who I own."). On my survey, I will ask people whether this sentence is true or false as a free response question. If someone answers "true" or "false" (or some equivalent) without clarification, then it will be assumed that they fell for the illusion. Likewise, if their response indicates that they think the sentence is meaningful (other than the technically correct meaning mentioned previously), they have fallen for the illusion. Anyone who answers "Neither" or some equivalent without clarification, specifically states that the sentence doesn't make sense or isn't well-formed, or indicates that they are interpreting the sentence to mean, "There are more people who have been to Russia than there are people who I own," has seen through the illusion. Anyone who gives a response such that I can't tell if they've seen through the illusion or not (e.g., they just respond with a random string) will not be counted towards either side.

Out of the responses I got, 68 answered “True”, 9 answered “False”, 2 answered “Yes”, and 1 answered “No" without clarification. So that’s 80 who didn’t see through it right off the bat, which is already more than half.

Five of them gave responses that indicated confusion, though they didn’t specifically say what they thought it means or that it lacks meaning:

True? (x3)
True??? (x2)

Four them specifically pointed out that it was a comparative illusion/escher sentence:

comparative illusion 
Escher sentence
https://en.wikipedia.org/wiki/Comparative_illusion
Love me an Escher Sentence I do

Four more interpreted it as meaning, "There are more people who have been to Russia than there are people who I own."

I don't know how many people you have, it's illegal though
perhaps, i don't know how many people you have
slavery is bad
True (I imagine you don't have many people)

One person seemed to see that something was up, but still didn’t interpret the question in a way that makes sense:

I'm not telling you how many people have been in me

I’m going to count this as a case of not seeing through the illusion, since there’s no way to grammatically interpret the sentence as, “More people have been to Russia than have been in me.”

A lot (50) of people gave responses indicating that it was not a coherent sentence:

0.5
блядь
🤔
:(
Bro. 
haha
*head explodes*
huh?
huh? (I've been to Russia several times)
If it needs to be one of the two, then "false", because it doesn't resolve to a true statement.
I didn't understand the statement
Invalid
lol
Malformed claim
Maybe
meaningless
μ
Mu (x3)
Mu.
mu
N/A (x2)
Neither
Neither, fails to form a proposition.
Neither lol
Nice try. Fake question. 
nonsensical
not a comprehensible sentence, no truth value 
Not even wrong
Not well-formed
Syntax Error.
that doesn't make sense
This doesn't make grammatical sense.
this doesn't make sense, but if I must pick: false
This isn't a valid question
this is nonsense.
This makes no sense
Unclear
Wait a second, that doesn't make any sense
wat
What (x3)
What? (x2)
...what
wtf

Here I assumed that the responses that were expressions of amusement, incredulity, or “:(“ were people who saw through the illusion, but it doesn’t affect the final result of the market if I exclude some of them from the count (since the majority didn’t see through it). For people who are wondering, “mu” basically means “neither true nor false”, and блядь seems to be a swear word in Russian and Ukranian, but it’s inherited from an Old Slavic word meaning “to talk nonsense, err”, at least according to Wiktionary, so I assumed it is being used in that latter sense.

One person saw through the illusion, but disagreed that it was actually incoherent:

More people than I have been to Russia, at least, and more linguists have been ensnared by prescriptivism in the name of "that's ungrammatical" than you [have]

I disagree with the argument being made (I don’t think you have to be a prescriptivist to say it’s ungrammatical), but this clearly counts as seeing through the illusion regardless.

Another response was just “Da”, and I couldn’t find any interpretation of this response that makes sense, so I’m putting it in the “impossible to judge” category.

And finally, there were the joke answers.

I am Russia
Homestuck is a webcomic about four kids who play a videogame with absurd risks and consequences

I assume that both of these people probably saw throught the illusion and decided to submit a joke answer because of it, but I can’t know for sure since I’m not either of these people. By the standard set in my text for the hash code, they don’t count towards either side.

Anyway, that’s at least 59 people who saw through the illusion and 81 who didn’t, plus some ambiguous cases that wouldn’t be enough to change the result either way.

Furries

OwO, what’s this? A question about furries? (I’m sorry, I had to do that for the market). Anyway, I noticed that there were a lot of furries on Manifold (I mean, someone even walked around in a fursuit at Manifest. Okay, well not really…), so I decided to make a question to figure out how many.

That’s 11 definite furries, 7 furries in denial, 12 furry-adjacents, 21 people whose interests overlap with furries, and 115 non-furries. Treating this as a linear scale from 0 to 100%, that makes Manifold 16.6% furry, on average.

Crimes

Manifolders often make markets about whether other people will be charged with crimes, but what about the Manifolders themselves? Will they fess up to their own crimes? The scale for this question went from 0, “I have never comitted a crime,” to 10, “I commit serious crimes on a daily basis.”

As we can see from graph, 25 respondents are dirty liars. Well, either that, or they’re too young to drive. Even just slightly exceeding the speed limit is technically a crime, after all. Among the more honest people, the distribution tapers off as the number gets higher, except for the bump at 10 from the five people who picked it. I’m not too worried about the five people who picked 10, though - that was obviously a joke. The people I’m much more concerned about are the 6’s and 7’s.

From these numbers, we find that Manifold’s average criminality is 2.1375, though the average was skewed a bit by those 10’s. I can’t wait for the coming AI police state to use this number against us to profile our likelihood to commit crime.

Manifold’s Big Five personality traits

Pat’s Manifold Survey asked users about their Meyer-Briggs type, so I thought it would be fun to do something similar, but using the Big Five model. I allowed people to state their place on each axis on a scale, since that is a more faithful representation of personality than just stating a binary result on each axis. I didn’t actually ask for numerical values, though, since I assumed most people wouldn’t have those on hand, and that not everyone would want to take a survey within a survey just to determine where they are on each scale. Instead, I just asked whether they were “Very” or “Somewhat” low or high in each trait, or in the middle. These were the results:

The results of the openness to experience and extraversion axes were no surprise to me. In fact, I was already so confident that Manifold would be introverted and open to experience that I made markets asking just how introverted and open to experience Manifold would be, whereas I made binary markets for the other traits. To quantify exactly how much the average Manifolder embodies each of the Big Five traits, I made a spreadsheet to calculate the average score. To do this, I converted the responses on each trait to a linear scale from -2 to 2. In other words, “Very low/high” was converted into a score of ±2 (- for low, + for high), “Somewhat low/high” was converted into ±1, and “In the middle” was converted to 0. So the maximum score for each trait is 2, the minimum is -2, and a positive value means that the average Manifolder has the trait, while a negative value means the average Manifolder has the opposite trait. Here are the results:

As was already obvious from the graph, openness to experience is by far Manifold’s most extreme trait, the only one to have an average score larger than one, and introversion is the next most extreme. So my prediction that these two would be the most extreme traits was correct. The least extreme trait was conscientiousness, with Manifolders being just barely more conscientious than spontaneous (the name typically given to the opposite trait). This is kind of ironic, since prediction markets are in theory meant to help people prepare for the future, and high conscientiousness is associated with planning more. But it also doesn’t surprise me at all, since many Manifolders seem to be spontaneous, and betting on prediction markets requires some degree of flexibility.

Manifold’s Big Five “personality type” ended up being OCIAN: open, conscientious, introverted, agreeable, and neurotic.