Conversation

Aella
@Aella_Girl
Women in science need to learn more about how statistics work. If I published results this low in significance as tho they were real I could have 'proved' eighty different insane things by now. To explain why this is bad:
Quote Tweet
Emily Khazan
@EmilyKhazan
It can be hard being a #womaninscience, and teaching evaluations often don't help. We show gender bias in teaching reviews of graduate students; another subtle force making the leaky pipeline leakier: insidehighered.com/news/2020/11/0
Show this thread
13
145
Aella
@Aella_Girl
Her sample size is low. There's calculations you can do to figure out exactly how meaningful a result (basically, what's the probability it was an "accidental" positive) is based on the sample size and the correlation strength. It's irresponsible to publish results as significant
1
59
Aella
@Aella_Girl
if there's too high a probability that your correlation was just random chance. If you test a lot of things for long enough, you're absolutely gonna find lots of correlations that are just random chance! This means you have to be extra careful if you're checking a ton of these.
2
Retweets
68
Likes
Aella
@Aella_Girl
Replying to
@Aella_Girl
But also it's extra terrible that this author reported a meaningful result when she admits it's not statistically significant! I'm physically recoiling from this idea. How are they training people in school that this is something they think is ok?
3
106
anton 🏴‍☠️
@atroyn
Replying to
@Aella_Girl
this is really what should be taught in science ethics classes;
Quote Tweet
anton 🏴‍☠️
@atroyn
a large part of the problem is people mentally substitute “science” with “randomized double blind study on 12 people a psych grad student slapped a p value on and interpreted to mean whatever gets them published”
Show this thread
1
3
Kowalski Nicolas
@kowalski_nico
Replying to
@Aella_Girl
This is so, so important in statistical analysis! If you consider your results significant with a 95% chance — which is the default — by DEFINITION one in 20 experiments will be a false positive. If you keep testing random stuff, you will eventually find “positives”.
1