7mo ago

Nobody will question you

25 comments

Relevant xkcd
https://xkcd.com/2048/
- Of course there's always a relevant XKCD
  Love it
- It took too long for me to realize it was the same data.
Unless you're in a college statistics course, then if your line is off by a pixel your grade drops a full letter.
- If you're in a college statistics course and you're doing graphs by hand and not generated entirely be statistics software, the skills you're learning are useless anyway.
  
  My bitterness lingers from the 90s.
And make sure you use linear regression, nobody thinks linear regression is bad.
- Yeah that would be bad practice, industry standard is to run all the tests simultaneously and if something comes out statistically significant make up a narrative then try to split it into 4 papers.
  
  Tell that to the reviewers who constantly ask my wife why she didn't do linear regression in her analysis. She rages against linear regression constantly. But some people swear by it, which i think is weird.
- Folks in observation and analytics are gonna be real mad when they realize you're giving away their secrets.
Just saw the scatter plot and line and my mind immediately screamed "bullshit" without knowing what this was about at all. Only then I read the text.
Look at that choice of axis scale tho
- Could be valid. Now if it had been logarithmic the pro tip might still be true, since many don't look at the axis either.
Label your x and y, you dirty heathen. Such offense, you're lucky you're not catching a b&.
What's the r² on this, like ... 0.3 ish?
Less?
- My guess is lower. I'd put the correlation at about -.35 to -.45, so that'd correspond to an R² of .1225 to .2025. But eyeballing correlations is hard.
- Assuming it's a correction line, I don't think you can tell from the slope of that line alone as the clustering will matter and correlations are finicky. Now, if it was a regression coefficient, that sexy line can be calculated just by looking at it (although we'd want to know if it was significant, lol).
  
  I was assuming its a simple linear regression fit, and attempting to eyeball the r², haha.
- Delete enough data points and it will be 1. You'll only have two data points, but you'll have bragging rights.
Actual graph used to inform government decisions
Nobody questioned Hubble so why would they question you?
Zoom out so it looks better.

25 comments