You're Weirder Than You Think

Subscribe

Made by this blog’s author

Person Do Thing

Get your friends to guess a word.... but you can only use<br>33 extremely simple words to do it.

&ldquo;Person want thing, other person make person have thing, person feel good!&rdquo;<br>That&rsquo;s gift.<br>Can you do better?

Card game &middot; 2+ players &middot; Ages 8+ &middot; 15 min+

Get it on Amazon &rarr;

One of my long-time hobby-horses is that people are way weirder than they realize, and that our politics would look different if more people understood how massively unrepresentative they are of the rest of the population. So here's a little quiz designed to demonstrate how unrepresentative you are of the US population.

How Unrepresentative Are You?

How unrepresentative are you?

You probably don't look much like the average American.

1,000 randomly chosen Americans<br>1,000 match you

all 1,000 still match — answer questions below to narrow the field

Answer seven questions to see how many Americans share your exact combination of traits. The data comes from the General Social Survey.

Begin →

Question 1 of 7

Press Y or N — or click

Findings

Where each answer puts you

↺ Take it again<br>Copy result

Methodology & sources<br>Data comes from the General Social Survey (GSS) cumulative file, 1972–2024, Release 3. We use respondents from the four most recent waves (2018, 2021, 2022, 2024), restricted to those who answered all seven questions — 4,326 people, weighted by wtssps (NORC's recommended post-stratified weight). Three of the seven questions (about God, gun ownership, and views on same-sex relations) are asked of only a sub-sample of GSS respondents in each wave, which is why the complete-case sample is smaller than the full recent wave count.

The widget stores a 128-cell joint distribution (one cell for every combination of yes/no answers). After each answer, it sums the cells consistent with your answers so far — giving the true conditional probability that a random American still matches you, with all correlations between answers preserved. After all 7 answers, the sum collapses to a single cell.

Some rare combinations have very few observations, so percentages below ~0.1% are noisy. One cell with no observations is given a 0.05% floor.

An important note: the "easy" way to make a tool like this would be to get a bunch of interesting stats about the American population (32% of Americans have a bachelor's degree! 50% of Americans believe that God exists, without any doubts!) and then just multiply those numbers together, implicitly assuming that each result is independent. But of course, these things are not independent: for example, people with bachelor's degrees are less likely to believe with certainty in the existence of God.<br>This tool avoids that problem because it uses data directly from the General Social Survey, where thousands of Americans were asked a long series of questions, and researchers can access all the results in detail (huge thanks to the GSS team!) So when the tool says that 13% of Americans have a bachelor's degree and are certain about God, that's based on counting all the actual humans who specifically replied with those two answers out of the total respondents to the survey.<br>If you want to go deeper on the data, check out this explorer tool that lets you break down the data by age & see how many weirdness-points you'd get by changing each of your answers.<br>[I did shockingly little of the work on building this tool, so compliments should largely go to Claude, but if you have complaints or think this is wrong please do let me know. If you're interested in the process you can see the conversation here.]

Open-source hardware

RETINA

A node on your roof uses echoes from existing TV broadcasts to detect<br>aircraft up to 35 miles away. No transmitter. No license.

Passive radar &middot; Open source &middot; Community-powered

Explore at retina.fm &rarr;

AgentCAD

Lets coding agents "see" whether or not a design works. Free and open source.

Try AgentCAD Free &rarr;