The map of hate...pretty surprising results!

[Acer]

If it's a population/access map, then why do LA, Miami and New York City not show up in the OP's map?

[davester]

Chakravartin wrote:
[quote=davester]
WRONG! Before jumping to conclusions, perhaps you should have clicked the link and looked at the methodology.

Re-read it.

Not one word about normalizing for INTERNET ACCESS.

All that they are saying is that counties with small populations of "hate-tweeters" are left off the map. So, the dark spaces correspond to a paucity of tweets. And that makes perfect sense since there's also a paucity of Internet access in those same places.

'Can't tweet when you can't get online.

Monica Stephens has been doing this with at least one class every year for the last several years. Every time she makes one of these population maps and claims to have discovered some deeper meaning, she and her school get lots of free publicity. But regardless of what her class is studying, every year the map looks almost exactly the same.

...Which may be the subtle point of it.
It's obvious that the dark spaces mean there are insufficient tweets. That doesn't really matter unless the reader is ascribing some meaning to them. I can see how this might be an issue for folks who don't understand what they are looking at, and perhaps it would have been better to map things differently so that the percentage of rural area was somehow accounted for (e.g. warp the map area to account for population, or apply the findings of western population centers to surrounding regions based on a population function).

Also, I don't think it is true that counties with small populations of "hate tweeters" are left off the map. I think what they are showing is that counties with small populations of tweeters in general are left off the map...a very large difference

That said, the map is very useful as is for people who understand what it represents, and the major population centers of the west have plenty of data and show significant contrasts with the population centers of the east. That is the point. Now, if you can demonstrate some systematic error that could account for that, then you would be able to criticize the map. However, you have yet to do that.

That said, I do have a problem with the map. They don't distinguish between areas of "no hate" (which would no doubt be below a certain threshold) and "no data". This is a major flaw and makes it difficult to evaluate.

[Chakravartin]

Geocoded "hate speech" after the Obama election:
http://4.bp.blogspot.com/-jBeieVub6yw/UJ...apture.PNG

Tweets about republican candidates:
http://3.bp.blogspot.com/-wdIirNznTwU/Ty...120129.jpg

Tweets about bars:
http://4.bp.blogspot.com/_x-FKwdGnxic/S2...100122.jpg

Tweets about church and beer:
http://2.bp.blogspot.com/-C1Sc-8q_yWw/T_...hfixed.jpg

Same kinds of "corrections" applied.

Astonishing how they all seem to cluster in the same places.

It's almost as if the rest of the country doesn't tweet very much.

Almost like Internet access is limited in some places and skewing the results.

Nah. Everything you read is true and complete and you don't ever have to think about it.

[davester]

Chakravartin wrote:
Astonishing how they all seem to cluster in the same places.

It's almost as if the rest of the country doesn't tweet very much.

Almost like Internet access is limited in some places and skewing the results.

Nah. Everything you read is true and complete and you don't ever have to think about it.

Actually they don't. Perhaps you're not understanding the maps. The only things that look the same are the distribution of data points. the colors or symbol sizes that represent the data being analyzed don't cluster the same way at all. The only point I think that you have is that these maps take some thought to interpret and so are not useful to the lay reader who does not look at them with a critical eye regarding what they represent. I agree with you that the data presentations are not optimal for general public consumption, but the data do appear to provide insights.

[Acer]

If you go to the linked site to view all the maps for the different keywords, the patterns fluctuate significantly from keyword to keyword. Whole metro areas come and go based on the keyword. It is not all just correlation to population or internet access or use of twitter.

[Chakravartin]

Acer wrote:
If you go to the linked site to view all the maps for the different keywords, the patterns fluctuate significantly from keyword to keyword.

Yes.

It's amazing that people in New Jersey use different cuss words than people in Washington.

Utterly astonishing.

[Acer]

Going from 'all homophobic' to 'all racist', entire metro areas appear and disappear.

[Ted King]

I noticed if you zoom in as far as you can that the maps show odd really "hot" spots in rural areas. In an effort to track down why that might be the case I found this:

http://www.floatingsheep.org/2013/05/hatemap.html

Even when normalized, many of the slurs included in our analysis display little meaningful spatial distribution. For example, tweets referencing ‘nigger’ [N word] are not concentrated in any single place or region in the United States; instead, quite depressingly, there are a number of pockets of concentration that demonstrate heavy usage of the word. In addition to looking at the density of hateful words, we also examined how many unique users were tweeting these words. For example in the Quad Cities (East Iowa) 31 unique Twitter users tweeted the word “nigger” in a hateful way 41 times. There are two likely reasons for higher proportion of such slurs in rural areas: demographic differences and differing social practices with regard to the use of Twitter. We will be testing the clusters of hate speech against the demographic composition of an area in a later phase of this project.

The quad cities isn't a highly populated area, but there are probably at least a few thousand twitter users in that area. Yet, they found only 31 unique Twitter users who used the "N" word in a hateful way. It's quite possible that they belong to inter-related groups and once one person broke the norm of not using the word on Twitter, then like-minded people in the inter-related groups felt okay about doing the same. It seems quite possible to me that in such a situation, if one person hadn't started it, then that area may not have shown up at all. If you have the same dynamic taking place in a large metropolitan area where you get a cluster of 30 people or so belonging to inter-related groups firing off twitters with the "N" word, that cluster may not show up in the metropolitan area at all because of the normalization of the data - even though it would involve exactly the same number of people in similar inter-related groups. Perhaps that isn't the explanation, but it does seem plausible.

At any rate, the authors of the maps say that there is little meaningful spatial distribution conclusions that one can draw from the data. They should have put that disclaimer front and center on the map.

[Lemon Drop]

wonder how that compares with Southern Poverty Law Center's hate group maps? (they've tagged 4 groups in Iowa)

http://www.splcenter.org/get-informed/hate-map


Like Ted my first thought too upon looking more closely at the map was this is concentrated activity by specific groups who communicate via Twitter and doesn't really say anything meaningful about the behavior or attitudes of people in certain geographical regions.

[gabester]

As I suggested before and couldn't quite put my finger on... it seems like the larger cosmopolitan cities have relatively low hate speech measured while smaller regional population centers seem to be loci for individual modes of hate speech - some words pop up more than others; seeing now that this may be skewed by fewer than 100 unique users out of the millions of users, well, it makes sense completely.

In other words, individual hate words seem to blow up at random, most likely due to a local/regional influence accelerating the use of that word relative to the overall population and that is disproportionate to the population normalization.

Does that make sense?