How did I get this poll with almost 29k responses to balance perfectly? A thread. 👇
Assuming most people didn't secretly flip a coin, where's the randomness in the poll coming from? I think it comes from three sources:
1. Some folks were genuinely picking randomly
2. Based on the comments, even for folks who used a system, the method they used was very unique to them and therefore really random relative to other people
3. From the perspective of the Twitter algorithm, each new person that gets shown the poll is a toss up in terms of whether they favor heads or tails, much like flipping a coin. It doesn't matter if they picked non-randomly. From the perspective of the poll, they appear random.
So, now that we've established that the people answering the poll are probably going to act a little bit like a flipping coin, what does statistics have to say about flipping a coin 29k times?
Law of Large Numbers
The average of a large number of observations should get closer to a particular value as more observations are collected. This value is called the "expected value". If we code heads as 0 and tails as 1 then the expected value for a fair coin should be 0.5.
Central Limit Theorem
The average of a large number of observations tends to cluster around the "expected value" in an increasingly tightly-clustered pattern that resembles a bell curve. We can't see the pattern with just one experiment but we do see it with lots of experiments.
You might be wondering what's a bell curve? It looks like this. The previous tweet is saying that most of the experiments will have averages that cluster around the center with fewer and fewer as the averages get more extreme.
If the bell curve feels a little abstract, don't worry. It's a lot more familiar than you might think. Men's heights are roughly distributed like a bell curve and so are the heights of women. So we've actually all been experiencing bell curves our whole lives.
You also might be wondering why I'm talking about "experiments". We only did one poll. Often statistics means thinking about the multiverse. We don't just think about our universe but every other universe where randomness would have caused our experiment to turn out differently.
Looking at our experiment in the context of the "multiverse" is what allows us to see that the results become from predictable as we get more observations.
If we assume our poll is like a fair coin then using the math of the bell curve, we can figure out what kind of results we can expect to get after 29k answers. As you can see below, there was about a 95% chance that the percent of heads would be between 0.494 and 0.506.
The precise proportion of heads in this poll was 0.504 which was well within the realm of possibility!
The one thing I did get lucky on is that the preference for heads and tails seems to be symmetric in the polling population. So for every person that prefers heads, there's an equal and opposite person that likes tails, and vice versa.
This didn't have to be true but will tend to give a close to 50-50 split when you select people randomly, even if their choices aren't random.
The first time I tried this poll experiment, it was pretty biased. I think because there was lack of symmetry in beliefs. My statistics savvy audience thought that people would be more likely to select the first option so they tried to "unbias" the poll by selecting the 2nd one.
My solution was just to tell them they were biased which caused them to be confused about what would happen on this poll, which unbiased them.
So there you go. That's the magic trick. I'm not a wizard. I'm just a statistician. 😏
Hope you enjoyed the thread. If you like this content and want to support it, please like and retweet the thread so others can enjoy it as well, and follow me to get more threads like this one in the future.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
If you think about how statistics works it’s extremely obvious why a model built on purely statistical patterns would “hallucinate”. Explanation in next tweet.
Very simply, statistics is about taking two points you know exist and drawing a line between them, basically completing patterns.
Sometimes that middle point is something that exists in the physical world, sometimes it’s something that could potentially exist, but doesn’t.
Imagine an algorithm that could predict what a couple’s kids might look like. How’s the algorithm supposed to know if one of those kids it predicted actually exists or not?
The child’s existence has no logical relationship to the genomics data the algorithm has available.
These grants aren't charity. They're highly competitive contracts where the US government determines Harvard is the best institution for conducting specific research, and then pays Harvard for services rendered to US taxpayers.
Each grant represents a fair contract that a group at Harvard won after being in competition with hundreds or even thousands of other groups. These are not handouts.
The US government pays Harvard and other universities to provide answers to questions that aren't directly profitable in themselves, but which provide a foundation for private sector innovation, and help maintain American dominance over geopolitical rivals like China.
As a someone who translates ideas into math for a living, I noticed something weird about the tariff formula that I haven't seen anybody else talk about. 🧵
The formula defines the tariff rate as exactly the percent you need to charge on imports to make up for the trade deficit. Basically,
trade deficit = tariff rate x imports
It's constructed as if tariffs are a kind of compensation for trade deficits but this raises a question.
If tariffs are something foreign countries owe to the American people for having a trade deficit, then forcing US businesses to make up for the difference, by paying extra money to the US government, is kind of a weird solution.
Whenever I see students with good grades but lots of college rejections, my first thought is a bad personal essay. As predicted, this guy's essay was kind of a disaster.
Since I did get into Harvard, I'll give my two cents on the essay:
In honor of international women's day, let's take a moment to remember the most famous statistician in history.
You've definitely heard of her, but you probably have no idea she was a statistician.
It's Florence Nightingale.
Nightingale was first female member of the Royal Statistical Society and a pioneer in using statistical analysis to guide medical decisions and public health policy.
Florence Nightingale's most famous statistical analysis was her investigation into the mortality rates of soldiers during the Crimean War. She demonstrated that the majority of deaths among soldiers were due to preventable diseases rather than battlefield injuries!
Took one for the team and made a histogram of the Elon social security data. Not sure why his data scientists are just giving him raw tables like that.
It’s also weird that they keep tweeting out these extremely strong claims without taking a few days to do some basic follow up work.
It doesn’t come off like they even:
- plotted the data
- talked to any of the data collectors
- considered any alternative explanations