Update

Update

More votes are in! Numbers have changed! And I'm gonna do a followup on some of the previous stats I did and also some new tests! More graphs and numbers under the cut! This one's a little more text heavy and deals more with stats lingo, but I tried to keep it digestible. tl;dr at the end.

First and foremost, last test I did assuming everyone who voted in this poll was AFAB, there was an approximately 9% chance of seeing a vote for "exactly the same height" as high as we did. That vote has gone up, at the time of writing this it's 10.4%. Furthermore, the total number of votes has gone up a lot, it's around 83k at the time of writing but I rounded it up. I used the same methods to redo the simulations for an all AFAB voter group using the new conditions and voila!

Now 0% of the simulations had a proportion of the votes be exactly the same height as prev as high as we are currently seeing. There's no way to know for sure, but my preferred theory is seeing that your prev has the same height as you catches your eye and makes you more likely to vote. But for all I know, the standard deviation of heights of tumblr users is 1 inch. Can't go much further than this without rounding all of you up and measuring you >:).

Now for new stuff!

Since my last post, the gap between the taller and shorter votes has gotten bigger. At the time of writing, 44.6% of the votes are for shorter and 45% are for taller. That's only a difference of 0.4%, but on a scale of 80k+, it might mean something.

My next hypothesis is that you are less likely to vote on this poll if you are shorter than prev than if you are taller than prev.

There's a couple ways we can go about looking at this. I'll start with some of the more general ones (including same height votes) and go narrower.

Method 1: Bar Charts

Bar charts! Generally a pretty good way to visualize the distributions, but it may not be too helpful here.

On the left is what we should be seeing if all of my hypotheses (including the one about same height) are false, and you vote regardless of your height relationship to prev. The proportions here are 46.4% shorter, 7.2% the same height, and 46.4% taller. On the right is what we actually see. You can see that the same height bar is taller and can maybe see that the taller bar is slightly taller than the shorter bar, but it's pretty hard to eyeball.

At the very least we have a baseline of what the proportions should be if (given the tumblr population reflects the general population (big given here)) the likelihood of voting on this poll is independent of your height relationship with prev. This'll be my expected distribution, the good ol' aitch nought if you will.

Method 2: Chi-Square Test for Goodness of Fit

This is a good test to use to see if the distributions of categories in a sample that you are observing (y'all poll voters) matches the expected distribution of categories. I don't have a nice looking graph for this one, you basically just input your observations and expectations into a math formula, you get a number and stick that number in a graph, and the number it spits out (called the p-value) tells you how likely, in a very large number of simulations, that you get results as different from what is expected as the ones you observed, given that your hypothesis is false.

In this case, I'm just looking for if the distribution we see in our sample is likely given that the true distribution is 46.4 - 7.2 46.4. And lo and behold, my p-value was 3.4e-292. This is zero. For all intents and purposes, there is a 0% chance of seeing the results we did if all of my hypotheses are false, but we already knew this.

Method 3: Exactly the Same Thing I Did Before

Beyond this point, the actual underlying distribution of tumblr voters' heights don't matter, which is more conceptually fun for me.

This time, I redid my simulations and graphs for the percentages but using taller and shorter rather than the same height. Since the proportion of same height answers are so much higher than expected, I just factored them out completely. This left me with an expected 50/50 for what remains, and an observed 50.2/49.8 difference in taller versus shorter:

And these look pretty promising! In both cases (this makes sense, they're symmetric), only 1.3% of the simulations fell into the areas as extreme as our observations got. In the showbiz we call this statistically significant, but it doesn't win any asterisks. So it looks like this supports my hypothesis. But wait! There's one more thing we can do!

Method 4: Z-test for Proportions

Since we're just looking at two groups now, taller and shorter, we can pretty much assign a boolean flag to each observation. Let's just say that we're looking at the proportion of voters who are taller than prev. Again, we would expect this to be 50%. No fun visualizations for this one either, pretty much the same procedure as method 2 but using a different formula and different graph.

I got a p-value of 0.12. Aw shucks. In the showbiz we call this not significant and I get fired and shunned forever. Basically if we ran this simulation a whole bunch of times where voters actually were equally likely to answer "taller" as they were "shorter", in 12% of these simulations, we would see results that deviate at least as far as 50.2% voting "taller". Not that many but not enough to make any claims. I've failed you all, sorry.

SURPRISE REPLICATION CRISIS MENTION

Now, see, I could have chosen not to share that last test with all of you, and in science, a lot of the time, you wouldn't have seen this. This is because if that little number, if that p-value doesn't reach 0.05, you're not getting published. And if you're not getting published, you're not getting paid. And so there's all sorts of incentives for scientists to do a little this and a little that to their data and their experiments that end up showing you something that might not reflect what's really happening. And the only way to really tell is when other scientists try to replicate those studies (not very often because this isn't profitable), they can't get the same results.

This is a real problem in science, called the replication crisis, and in a lot of big name journals across many fields (even medical ones oh no), the number of studies that don't replicate vastly outnumber the ones that do.

Anyways I attended a talk once about ways in which we could counter the replication crisis, and for every suggestion someone made, someone else came up with a method you could use to circumvent that suggestion. People are always going to be able to find ways to wriggle their ways into filling all those requirements to get paid. This isn't to scare you from science or demonize scientists (most probably? aren't doing anything wrong), just point out that the machine is not the best.

Disclaimer: I'm not a professional, I'm just a student right now, so any number of things I said may be wrong. I could also just be lying to you and so could all the poll voters.

tl;dr

It is even more unlikely at this point that given the tumblr population is representative of the global population, we'd see a percentage of 10.4% the same height just by random chance, people are only maybe more likely to vote if they are taller than prev than if they are shorter than prev (legally I can't say), and replication crisis jumpscare!

tdp season 6 shitpost doodles

I've taken the other posts down, it seems like Tumblr has finally let my post live in public. Again, apologies for the kerfuffle if anyone was able to see it

Apologies if you guys have seen the same post a bunch of times. Tumblr kept hiding and unhiding and hiding the post so I was running a bunch of tests to see if I could figure out what was going on. The latest version of the post on THIS BLOG, not any of the ones on @eggsistential-basket is the one I'm intending to keep. If the post is still visible in a couple hours ALL of the other versions will be deleted so I can have it consolidated into one post. I'm going to keep the others up for a bit in case I can't get a stable version on this blog. Sorry for the confusion

It has literally just occurred to me after three years that there is a benign interpretation (you can use an onion for culinary purposes) and you can in fact use an onion non violently

Is... is this a threat?

pspspsps

do any of my followers like statistics