search

giantoak / bias-exploration

0 stars 0 forks source link

sane scaling #1

Open samzhang111 opened 9 years ago

samzhang111 commented 9 years ago

copied from Jeff's email:

It turns out that scaling is super important for looking at changes in the correlation. Because otherwise there's this scale piece that affects the slope, so you misread the change as being a negative change instead of a positive one.

The way we want to scale it is to take the results of the adjust function and multiply them by

mean(y) / mean(adjusted y)

We don't want to compute the mean N times, or start doing it recursively or anything.

jborowitz commented 9 years ago

Makes me feel bad about not actually just filing bugs, and instead just complaining by email...

On Fri, Oct 31, 2014 at 11:57 AM, Sam Zhang notifications@github.com wrote:

copied from Jeff's email:

It turns out that scaling is super important for looking at changes in the correlation. Because otherwise there's this scale piece that affects the slope, so you misread the change as being a negative change instead of a positive one.

The way we want to scale it is to take the results of the adjust function and multiply them by

mean(y) / mean(adjusted y)

We don't want to compute the mean N times, or start doing it recursively or anything.

— Reply to this email directly or view it on GitHub https://github.com/giantoak/bias-exploration/issues/1.

samzhang111 commented 9 years ago

A friend and I joked once that we should make a repo for our friendship and only communicate through issues. Now that you bring this up, I think I'll do it