drwex: (Default)
[personal profile] drwex
https://www.youtube.com/watch?v=R-sIb-b3yWE&list=TLGGgz7PEDd2hZ4wOTA2MjAxNw&index=2

In this "Big Think" video, Seth Stephens-Davidowitz (formerly data scientist at Google) talks about some of the analyses he's done on the mountains of search data. The results are not encouraging.

First, people make what might be termed "racist" searches quite often. That's looking for searches using derogatory terms for racial and ethnic groups, searches for jokes that include racist stereotypes and slurs, and so on.

Second, you can see from the geolocation data on where searches originate you see (he reports) a striking east-west divide. East of the Mississippi you get a lot more of these searches than you do west of it. The areas he lists, though, are what we generally think of as "post-industrial" parts of America, such as western PA, eastern Ohio, which makes me think there are more correlations there to be explored.

Regardless of how you characterize those areas, Stephens-Davidowitz notes that if you go back to Obama's first election and map his vote totals against these racist searches - SURPRISE Obama does worst in districts that have a high number of racist searches reported. He notes that this correlation beats out other possible factors including church attendance, gun ownership, etc.
The main factor that predicts where Obama did worse than other Democrats is how frequently they made racist searches on Google

If that doesn't depress you (as it did me) then I'd love to know why.

Now scroll forward to 2016. I am shocked, shocked I tell you to discover that
...the single highest predictor of where Trump was doing well was the measure of racist searches on Google

This is, just to be clear, not me saying that everyone who voted for Trump is or was a racist. But if you voted that way, that is the majority of your fellow travelers. That is the pack you chose to run with. It's an the unexamined undercurrent in American society. Nothing in Google's data can help us understand why these attitudes exist in the places they do and not elsewhere. That task falls to us.

The data do support the hypothesis that Trump didn't create this racist pack - it was there at least as far back as data analysis has gone. It's also clear that, at least so far, these kinds of analyses are best done post facto. For example, searches for information on how to vote in traditionally black areas were much lower in 2016 than in previous election cycles. Post facto we can see that this has some explanatory power for Trump's victory - Clinton did not motivate black voter turn-out the way Obama did. But there aren't enough examples to know which searches are going to be relevant predictors and which are spurious correlations. At least, not yet.

Date: 2017-06-13 06:04 pm (UTC)
rmd: (Default)
From: [personal profile] rmd
it's not clear if he's adjusting for population density.

Date: 2017-06-14 03:43 pm (UTC)
flexagon: (Default)
From: [personal profile] flexagon
I'm not depressed by searches that use racist terms. Lots of Google searches are really "what does this mean", and plenty of people are writing papers ABOUT racist terms and jokes, doing research for novels, etc. Working for the big G has taught me to explicitly not assume a user feels positive about the thing they're searching for.

(Overall correlation, fine. But a given search? Whatever.)

Profile

drwex: (Default)
drwex

July 2017

S M T W T F S
      1
2 3456 78
9 1011 12131415
1617 1819 202122
23242526272829
3031     

Most Popular Tags

Style Credit

Expand Cut Tags

No cut tags
Page generated Jul. 21st, 2017 04:45 pm
Powered by Dreamwidth Studios